Auxiliary Selection Revisited: Gradience and Gradualness 9783110348866, 9783110347371

Citation preview

Auxiliary Selection Revisited

linguae & litterae

Publications of the School of Language & Literature Freiburg Institute for Advanced Studies

Edited by Peter Auer, Gesa von Essen, Werner Frick Editorial Board Michel Espagne (Paris), Marino Freschi (Rom), Ekkehard König (Berlin), Michael Lackner (Erlangen-Nürnberg), Per Linell (Linköping), Angelika Linke (Zürich), Christine Maillard (Strasbourg), Lorenza Mondada (Basel), Pieter Muysken (Nijmegen), Wolfgang Raible (Freiburg), Monika Schmitz-Emans (Bochum)

Volume 44

Auxiliary Selection Revisited

Gradience and Gradualness

Edited by Rolf Kailuweit and Malte Rosemeyer

DE GRUYTER

ISBN 978-3-11-034737-1 e-ISBN (PDF) 978-3-11-034886-6 e-ISBN (EPUB) 978-3-11-038643-1 ISSN 1869-7054 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. © 2015 Walter de Gruyter GmbH, Berlin/Munich/Boston Typesetting: Meta Systems Publishing & Printservices GmbH, Wustermark Printing and binding: Hubert & Co. GmbH & Co. KG, Göttingen ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Malte Rosemeyer and Rolf Kailuweit 1 Introduction

Section 1: (Limits of) Semantic and syntactic gradience Antonella Sorace The cognitive complexity of auxiliary selection: from processing to grammaticality judgements

23

Michele Loporcaro Perfective auxiliation with reflexives in Medieval Romance: syntactic vs. semantic gradients 43 Pierre-Don Giancarli Auxiliary selection with intransitive and reflexive verbs: the limits of gradience and scalarity, followed by a proposal

79

Section 2: Between constructional variation and auxiliary selection Artemis Alexiadou On the irrealis effect on auxiliary selection

123

Ida Larsson The HAVE/BE alternation in Scandinavian – perfects, resultatives and unaccusative structure 145 Jaume Mateu and Mar Massanell i Messalles A constructional approach to auxiliary selection: evidence from existential constructions 183 Peter Öhl Periphrasis as a precursor of analytic inflection: auxiliation in the (pre-)history of German 213

VI

Contents

Section 3: Mechanisms of Gradual Change: BE > HAVE and HAVE > BE Rolf Kailuweit BE or HAVE in Contemporary Standard French – residua of semantic 249 motivation Steffen Heidinger The auxiliary selection of French monter ‘move upward’ from the 16th to the 20th century 277 Malte Rosemeyer Entrenchment and discourse traditions in Spanish auxiliary selection Melitta Gillmann Auxiliary selection in closely related languages: the case of German and 333 Dutch

General index Language index

359 363

301

Malte Rosemeyer and Rolf Kailuweit, Albert-Ludwigs-Universität Freiburg

Introduction The papers collected in the present volume deal with the so-called phenomenon of auxiliary selection, which has been studied extensively since the second half of the 20 th century (see also Levin and Rappaport Hovav 1995; Müller 1999; Alexiadou, Agnostopoulou, and Everaert 2004; Mackenzie 2006: chapter 1). The volume provides new insights into a series of characteristics of both the synchronic rules and the historical development of auxiliary selection that are common to most of the European languages in which auxiliary selection is or was attested, i.e. Catalan, Corsican, Danish, Dutch, English, French, German, Greek, Icelandic, Italian, Norwegian, Spanish and Swedish. In the first section of these introductory remarks, a brief overview of previous research on auxiliary selection is given. Subsequently, we will outline the structure of our volume, underlining the main idea of each chapter. We will conclude with a short summary and an outlook on perspectives for further research.

1 Auxiliary selection between gradience and gradualness Many Modern European languages exhibit an alternation between the auxiliaries HAVE and BE in the perfect tense in intransitive sentences, as exemplified by the French, German and Italian tokens in (1–2).1 (1) Nous avons dormi. we have.prs.1pl sleep.ptcp.m.sg Wir haben geschlafen. we have.prs.1pl sleep.ptcp (Noi) abbiamo dormito. we have.prs.1pl sleep.ptcp.m.sg ‘We have slept.’

Fr.

Ger.

It.

1 Like many papers in this volume, we use HAVE and BE in capital letters to refer to the corresponding (language non-specific) lexemes; italicised small letters are used to refer to the language-specific lexeme (e.g. It. avere ‘have’).

2

Malte Rosemeyer and Rolf Kailuweit

(2) Nous sommes allés à Fribourg. we be.prs.1pl go.ptcp.m.pl to Freiburg

Fr.

Wir sind nach Freiburg gefahren. we be.prs.1pl to Freiburg go.ptcp

Ger.

(Noi) siamo andati a Friburgo. we be.prs.1pl go.ptcp.m.pl to Freiburg ‘We have gone to Freiburg.’

It.

In the world’s languages, perfect constructions typically derive from the copula verb BE or dynamic verbs with meanings such as ‘come’, ‘come from’ or ‘finish’ (Bybee, Perkins, and Pagliuca 1994: 64–65). In contrast, the perfect formed with the auxiliary HAVE is an areal feature typical of Romance and Germanic languages as well as Albanian, Czech, Greek and Macedonian (Dahl 1995; Dahl and Velupillai 2013). This is why Haspelmath (2001: 1495–1496) characterises the presence of HAVE-perfects as a defining feature of Standard Average European. Drinka (2013) attributes the spread of HAVE as a perfect auxiliary to the Carolingian scribal tradition, which would explain the almost complete restriction of HAVE-selection to the Charlemagne Sprachbund (van der Auwera 1998: 824). Perlmutter’s (1978) influential approach to auxiliary selection, developed in the framework of Relational Grammar, attributes the variation in auxiliary selection to the Unaccusative Hypothesis, i.e. the fact that “certain intransitive clauses have an initial 2 but no initial 1” (1978: 160). In other words, Perlmutter assumes that the sole argument of a verb such as go is an underlying object that gets promoted to subject. Contrastingly, the sole argument of a verb such as sleep is an underlying subject. As a result, no promotion of the argument takes place in sentences including verbs such as sleep. In Perlmutter’s terms, verbs such as go are called unaccusatives, whereas verbs such as sleep are called unergatives. The Unaccusative Hypothesis was integrated into the Government and Binding Theory by Burzio (1981, 1986). He argues that while the argument of an unergative verb is base-generated in the Spec-position of IP, the argument of an unaccusative verb is generated in the complement position of VP and subsequently rises to the Spec-position of IP. Burzio explains this movement process with case assignment: since an unaccusative verb cannot assign structural accusative case to the sister of V°, the argument of an unaccusative verb can only receive its case after moving to the Spec-position of IP (structural nominative). BE-selection is thus the result of the binding relation between the derived subject in the Spec-position of IP and the trace of the NP in the original position.

Introduction

3

Burzio’s proposal had the advantage of accounting for (a) the similarity of sentences containing unaccusative verbs to passival sentences and (b) the auxiliary selection behaviour of reflexive constructions in Italian and French, as illustrated in (3).2 (3) Ci siamo sposati questa mattina. pro.refl be.prs.1pl marry.ptcp.m.pl this morning Nous nous sommes mariés ce matin. We pro.refl be.prs.1pl marry.ptcp.m.pl this morning ‘We got married this morning.’

It.

Fr.

Work on the Unaccusative Hypothesis in Generative Grammar focused on identifying other syntactic parameters sensitive to the unaccusative – unergative distinction.3 Thus, it has been claimed that unaccusative verbs can be used in absolute constructions (4), allow for partitive cliticisation in Italian (5), and appear in sentences with postverbal subjects with low definiteness (6), unlike unergative verbs (Perlmutter 1978, 1983; Burzio 1986: 23–26; Belletti 1988: 4; Cinque 1990: 24) (4) Arrivata in ritardo, Maria non si scusò arrive.ptcp.f.sg in delay Maria not pro.refl excuse.pst.pfv.3sg neppure. even ‘Having arrived late, Maria did not even excuse herself.’ (Cinque 1990: 24) *?Telefonato a casa, Maria seppe che phone.ptcp.m.sg to home Maria know.pst.pfv.3sg that era stata promossa. be.pst.ipfv.3sg be.ptcp.f.sg promote.ptcp.f.sg ‘Having phoned home, Maria learned that she had passed.’ (Cinque 1990: 24)

It.

It.

2 Nonetheless, further research has shown that the syntactical behaviour of unaccusatives and reflexives in French is not identical (see Abeillé and Godard 2002). 3 Burzio’s proposal has also been taken up by Kayne (1993) and Cocchi (1994), who argue that in the underlying syntactic representation, there is only a BE auxiliary which is realised as HAVE in certain syntactic contexts.

4

Malte Rosemeyer and Rolf Kailuweit

(5) Ne arrivano molti. part arrive.prs.3pl many ‘Many of them arrive.’ (Burzio 1986: 20)

It.

*Ne telefonano molti. part phone.prs.3pl many ‘Many of them phone.’ (Burzio 1986: 20)

It.

(6) Il est arrivé trois filles. expl be.prs.3sg arrive.ptcp.m.sg three girls ‘Three girls have arrived.’ (Belletti 1988: 4)

Fr.

*Il est arrivé la fille. expl be.prs.3sg arrive.ptcp.m.sg the girl ‘The girl has arrived.’ (Belletti 1988: 4)

Fr.

However, neither the unaccusativity diagnostics exemplified in (4–6), nor perfect auxiliary selection always neatly mirror the unaccusative–unergative distinction (Alexiadou, Agnostopoulou, and Everaert 2004). Van Valin (1990) is concerned with examples such as (7–8), which illustrate variation in auxiliary selection: (7) Luisa ha corso (di proposito) nel parco. Luisa have.prs.3g run.ptcp.m.sg (on purpose) in.the park ‘Luisa ran in the park (on purpose).’ (Van Valin 1990: 235)

It.

(8) Luisa è corsa (di proposito) a casa. Luisa be.prs.3g run.ptcp.f.sg (on purpose) to house ‘Luisa ran home (on purpose).’ (Van Valin 1990: 235)

It.

The variable auxiliary selection of verbs such as It. correre ‘run’ suggests that such verbs cannot be classified as either unaccusative or unergative. In order to account for this observation, Van Valin develops a semantic approach to auxiliary selection. Essentially, he proposes that auxiliary selection is not an invariable feature of verbs, but rather depends on the sentence aspect. In (8),

Introduction

5

therefore, the goal a casa leads to a telic interpretation of the situation expressed in the sentence. This means that the subject referent Luisa is affected by the resultant state of a situation, namely being at home. While ha corso in (7) is interpreted as an activity predicate, è corsa in (8) is interpreted as an accomplishment. In Italian, HAVE is selected with activities, whereas BE is selected with accomplishments, achievements and states. This is due to the presence of a state predicate in the logical representation of these three Aktionsart classes (Van Valin 1990: 233). Other authors have claimed that the semantic role of the subject argument triggers the choice of the auxiliary. Dowty (1991: 606) showed that the subject of Dutch intransitive verbs that select hebben (HAVE) is a prototypical agent while the subject of verbs that select zijn (BE) is not a prototypical agent. In the LFG framework, Schwarze (1996) argues in the same direction. According to Schwarze, there are cases in which the variation between essere (BE) and avere (HAVE) for the same Italian intransitive verb is related to agentivity (9– 10). The proto-role approach was also applied to the historical development of auxiliary selection in Spanish by Aranovich (2003), who claimed that verbs that have a prototypical patient as the only argument display a greater longevity in the BE + PtcP construction than verbs with a prototypical agent. (9)

Il temporale era cominciato a notte fonda the storm be.pst.ipfv.3g begin.ptcp.m.sg at night deep ‘The storm had begun in the middle of the night.’

It.

*Il temporale aveva cominciato a notte fonda The storm have.pst.ipfv.3g begin.ptcp.m.sg at night deep ‘The storm had begun in the middle of the night.’

It.

(10) Il panettiere aveva cominciato a lavorare a the baker have.pst.ipfv.3g begin.ptcp.m.sg to work at notte fonda night deep ‘The baker had begun to work in the middle of the night.’ *Il panettiere era cominciato a lavorare a notte the baker era.pst.ipfv.3g begin.ptcp.m.sg to work at night fonda deep ‘The baker had begun to work in the middle of the night.’

It.

It.

Although the Aktionsart and the proto-role approaches to auxiliary selection explain some cases of variation in auxiliary selection, variable auxiliary selec-

6

Malte Rosemeyer and Rolf Kailuweit

tion behaviour is more widespread than suggested by these approaches. Consider the examples in (11–12) given by Sorace (2000). (11) La pianta ha fiorito due volte the plant have.prs.3sg blossom.ptcp.m.sg two times quest’=anno. this=year La pianta è fiorita due volte quest’=anno. the plant be.prs.3sg blossom.ptcp.f.sg two times this=year ‘The plant has blossomed twice this year.’ (Sorace 2000: 865) (12) Il presidente ha durato in carica due anni. the president have.prs.3sg last.ptcp.m.sg in post two years Il presidente è durato in carica due anni. the president be.prs.3sg last.ptcp.m.sg in post two years ‘The president has lasted in post for two years.’ (Sorace 2000: 868)

It.

It.

It.

It.

The variation in auxiliary selection in (11–12) does not seem to result from a difference in the expressed situation. Given that the adverbial due volte ‘twice’ in (11) provides a goal and thus telicises the situation, both sentences express accomplishments. As indicated by the durative adverbial due anni ‘for two years’ in (12), these sentences express states. According to Van Valin’s (1990) approach, essere ‘be’ should be selected in all of the tokens in (11–12). As a result, this variation escapes the aspectual–semantic approach to auxiliary selection. Dowty’s (1991) proto-role approach fares slightly better. The subjects in (11–12) show mixed semantic entailments regarding their role as proto-agents or proto-patients, which is why variation would be expected. However, given that the semantic entailments in the proto-role approach are not ordered (see the criticism in Mateu 2009), the approach can only predict the existence of this variation without making predictions regarding the degree of this variation. In addition to the variation regarding auxiliary selection in languages such as Italian, there are systematic differences between genetically related languages regarding auxiliary selection with the same type of predicates, as illustrated in (13).

Introduction

7

(13) I dinosauri sono esistiti 65 milioni di anni the dinosaurs be.prs.3pl exist.ptcp.m.pl 65 millions of years fa. before

It.

Les dinosaures ont existé il y the dinosaurs have.prs.3pl exist.ptcp.m.sg expl there a 65 millions d’=ans. have.prs.3sg 65 millions of=years. ‘The dinosaurs existed 65 million years ago.’ (Sorace 2000: 869)

Fr.

Sorace (2000) observes that the degree of variation in auxiliary selection, both within a language and between languages, differs according to the involved predicate class. This observation is integrated in the Auxiliary Selection Hierarchy (ASH). As illustrated by the examples in (13), variable auxiliary selection behaviour is common with state predicates. By contrast, activity and mannerof-motion predicates select HAVE more consistently, while change of state and change of location predicates select BE more consistently (see Sorace, this volume, for a more precise account of how the ASH is modelled). The crucial insight from Sorace’s approach (taken up, for instance, in Keller and Sorace 2003; Bentley and Eythórsson 2004; Cennamo and Sorace 2007; Larsson 2009; Mateu 2009; Kailuweit 2011; Rosemeyer 2014) is that the variation in auxiliary selection is modelled in gradients. Although Sorace (2000) proposes that to some degree, this variation is due to the greater susceptibility of state predicates to template augmentation processes. The exact reasons for the gradience in auxiliary selection across the languages in Europe are still under investigation. While Kayne and Cocchi claimed that HAVE and BE derive from the same auxiliary and thus posit an allomorphy between the two auxiliaries, a series of recent studies take the opposite view, proposing that HAVE and BE cannot always be characterised as allomorphs (Mackenzie 2005, 2006; McFadden and Alexiadou 2006, 2010; Loporcaro 2007; Rosemeyer 2012, 2014). Thus, auxiliary selection is interpreted as an opposition between two construction types or syntactic configurations. For instance, Earlier English have + PtcP is a perfect construction in which the auxiliary HAVE has attained a temporal function, while be + PtcP is a copula + participle construction whose temporal meaning arises compositionally and often has a resultative function (14–15). (14) I am come as ʒe bade Me I be.prs.1sg come.ptcp as you ask.pst.3.sg Me ‘I have come as you asked me.’ (Mirk, 75.2015, apud McFadden and Alexiadou 2010: 391)

E.En.

8

Malte Rosemeyer and Rolf Kailuweit

(15) he haþe foghten wyth þe Fend he have.prs.3sg fight.ptcp with the Enemy ‘He has fought with the enemy.’ (Mirk, 116.3171, apud McFadden and Alexiadou 2010: 391)

E.En.

One reason for this assumption is the so-called irrealis or counterfactual effect, attested in older stages of Dutch (Kern 1912), English (McFadden and Alexiadou 2010), German (Magnusson 1939; Shannon 1990, 1995), Neapolitan and Sicilian (Formentin 2001; Ledgeway 2003), Spanish (Stolova 2006; Rosemeyer 2014) and Swedish (Larsson 2009: 157–168). In earlier forms of these languages, split auxiliary selection appears to have been conditioned to some degree by modality. HAVE + PtcP is used relatively more frequently in contexts marked by counterfactuality than BE + PtcP. McFadden and Alexiadou (2010: 412–415) argue that these sentence-level constraints – previously neglected both by syntactic and semantic approaches to auxiliary selection – suggest that BE + PtcP is a copula construction with resultative function. In their analysis, HAVE has been grammaticalised to a perfect tense marker, whereas BE has not. Assuming that the functional category T(ense) can only appear once per clause and BE does not carry tense morphology, BE + PtcP clauses cannot have a past counterfactual meaning. Rather, they are interpreted as having a present counterfactual meaning, as in (16). (16) The Fellow looks as if he were broke out of E.En. the fellow look.prs.3sg as if he be.pst.3sg break.ptcp out of Bedlam Bedlam ‘The fellow looks like he broke out of Bedlam (and is still loose).’ (Farq, 60.477, apud McFadden and Alexiadou 2010: 406) This short summary of the research on auxiliary selection from the second half of the 20 th century to today illustrates that from a typological perspective, both unifying and diverging tendencies are discernible. Although the Unaccusative Hypothesis does not apply universally and without exception, there is a strong tendency for unaccusative verbs across languages to select BE and for unergative verbs to select HAVE. Variable auxiliary selection behaviour is sometimes, but not always, the result of template augmentation processes such as telicisation. While languages vary regarding the question as to which predicate class occurs with which predicate, this variation itself displays regular trends, as predicted by the ASH; state predicates typically display highly variable auxiliary selection behaviour, whereas activity and manner-of-motion predicates, as

Introduction

9

well as change of state and change of location predicates, are typically consistent regarding their use with HAVE or BE. Lastly, an irrealis effect has been documented in the earlier stages of many European languages that exhibit or exhibited auxiliary selection. The similarities between the European languages concern not only synchronic rules of auxiliary selection, but also extend to the historical development of auxiliary selection. Many European languages in which auxiliary selection is or was common exhibit a trend towards HAVE-selection. Thus, in Standard Catalan, English, Portuguese and Spanish, HAVE has replaced BE as an auxiliary. Likewise, there are indications that BE-selection was more common in earlier stages of French (Förster 1908: 69–100; Mackenzie 2006: 129–144; also cf. Heidinger this volume) and Swedish (Larsson 2009: 233–299). The stability or even extension of the use of BE-selection in other European languages such as Dutch, German and Italian appears to be correlated to a general high frequency of use of the HAVE + PtcP construction (Sapp 2011). For some of the languages where the frequency of use of HAVE expanded to the detriment of BE, there is evidence that the expansion process followed a similar course. At least in Catalan (Mateu 2009), Neapolitan (Cennamo 2008), Spanish (Aranovich 2003; Mateu 2009; Rosemeyer 2014) and Swedish (Larsson 2009: 233–299), the process of HAVE replacing BE was gradual, affecting certain predicate classes before others.4 With the exception of Aranovich’s (2003) work, all of these studies propose that the spread of HAVE followed the Auxiliary Selection Hierarchy, going from state predicates to change of state and change of location predicates. The gradualness of the expansion of the use of HAVE appears to concern not only the semantic but also the syntactic perspective. In particular, it has been claimed that the spread of HAVE affected certain syntactic contexts earlier than others, especially sentences containing reflexive constructions and dative arguments (Loporcaro 2011; Rosemeyer 2014). The working hypothesis of this volume is that these data can be analysed in terms of the interplay between gradience and gradualness recently described by Traugott and Trousdale (2010). The fact that synchronic variation is often ordered in gradients results from diachronic gradualness, i.e. the wellattested fact that historical changes typically proceed in small steps, affecting one usage context before another (Timberlake 1977; Andersen 2001b, 2001a; Hopper and Traugott 2003: 45–50; Brinton and Traugott 2005: 150; de Smet

4 Regarding the situation in Neapolitan, however, Loporcaro (this volume: section 8.1) argues that Cennamo’s analysis is flawed because she does not control for modal constraints – in particular, irrealis modality – on (Old) Neapolitan auxiliary selection. This refers to the “irrealis effect” on auxiliary selection mentioned above.

10

Malte Rosemeyer and Rolf Kailuweit

2012). Gradience in turn leads to gradualness. In Traugott and Trousdale’s (2010: 39) words, “variation over time involves the emergence of grammatical constructions: a gradual, global process, but one which involves a series of local micro-reanalyses”. This means that it might be possible to explain at least a part of the variation in auxiliary selection described in this chapter as a result of gradual language change processes.

2 Structure of the volume The papers presented in this volume are divided into three separate sections that explore three aspects of this working hypothesis. First, they discuss semantic and syntactic gradience in auxiliary selection, as well as the limits of the concept of gradience. Second, they deal with the question of whether the opposition between HAVE + PtcP and BE + PtcP can always be interpreted as an allomorphic relationship. Third, they investigate mechanisms in the gradual change from BE to HAVE.

2.1 (Limits of) semantic and syntactic gradience The first three papers discuss gradience in auxiliary selection. Antonella Sorace’s paper “The cognitive complexity of auxiliary selection: from processing to grammaticality judgements” describes the cognitive underpinnings of gradience, arguing that semantic gradience as modelled in the Auxiliary Selection Hierarchy does not exclude syntactic explanations to auxiliary selection. Sorace presents extensive evidence from experimental research supporting the hypothesis that certain predicates are more susceptible to variable auxiliary selection behaviour, and that this gradience results from the underspecification of the event-semantic template of these predicates. Predicates that are atelic and have non-agentive subject arguments, such as state predicates, allow properties common to both unergative and unaccusative verbs and therefore appear in variable auxiliary selection contexts. However, this gradience does not contradict the Unaccusative Hypothesis, as the existence of variable behaviour verbs does not imply that these verbs are mixed forms: rather, their meaning is flexible enough for contexts that trigger an unaccusative or unergative interpretation. Sorace proposes to reconceptualise the Unaccusative Hypothesis in terms of a model of the syntax-semantics interface in which the binary syntactic outcome is determined by the compatibility of certain verb meanings with usage context, thus leading to gradience.

Introduction

11

In “Perfective auxiliation with reflexives in Medieval Romance: syntactic vs. semantic gradients”, Michele Loporcaro argues that it is impossible to reduce unaccusativity to semantic features such as telicity and agentivity. Instead, he suggests that between the two poles of unaccusative and unergative verbs, reflexive verbs of different types occupy intermediate positions. This scale leads to syntactically motivated gradience between the selection of BE and HAVE, a fact which is not predicted by semantic approaches to auxiliary selection. Loporcaro demonstrates the relevance of this observation in his analysis of the development of auxiliary selection in Medieval Romance languages. Thus, Loporcaro’s syntactic scale is implicational in that in order for BE-selection to spread to unaccusative verbs, it first needs to spread to (different types of) reflexives. Loporcaro analyses the replacement of BE by HAVE in the history of Old Spanish, Old Neapolitan and Old Sicilian, suggesting that, in contrast to the findings from Aranovich’s (2003) study, the spread of HAVE to reflexive verbs is not secondary to the spread of HAVE to verbs whose subject is a proto-patient argument. Although the semantic factors modelled in e.g. the Auxiliary Selection Hierarchy indeed determine the directionality of the spread of HAVE, these semantic factors are subordinate to the syntactic factor of reflexivity. Like Loporcaro’s contribution, Pierre-Don Giancarli’s paper “Auxiliary selection with intransitive and reflexive verbs: the limits of gradience and scalarity, followed by a proposal” provides a non-scalar global representation of auxiliaries including both reflexive verbs and intransitive (non-reflexive) verbs. After a general discussion of the nature of gradience in syntactic phenomena, Giancarli adduces data from Corsican, Acadian and Standard French on modal and aspectual auxiliaries used in the perfect tense, e.g. in auxiliary + can + ptcp constructions. He argues that because these data demonstrate syntactic constraints, they escape approaches to auxiliary selection that make use of scalarity. The ellipsis or displacement of the infinitive in such constructions leads to HAVE-selection with unaccusatives, as in s’elli avianu pussutu (vultà) ‘if they had been able to return’. Giancarli proposes an account of auxiliary selection in terms of semantic macro-roles in which subjects with the macrorole “Source” cause HAVE-selection, while subjects that simultaneously satisfy the two macro-roles “Source” and “Goal” cause BE-selection.

2.2 Between constructional variation and auxiliary selection The papers presented in the second section of the volume provide evidence for the assumption that HAVE and BE in perfect constructions cannot always be

12

Malte Rosemeyer and Rolf Kailuweit

considered allomorphs and discuss the implications of these findings for the analysis of the historical development of auxiliaries. In “On the irrealis effect on auxiliary selection”, Artemis Alexiadou addresses the observation that in the older stages of the European languages, counterfactual or irrealis modality typically leads to the selection of HAVE over BE. She explores two questions: (a) does the irrealis effect apply to auxiliary selection in the strict sense, i.e. between two allomorphic expression types, and (b) is the switch from BE to HAVE in irrealis contexts responsible for the historical process of HAVE replacing BE? Alexiadou adduces data from Earlier English and Old Greek that prove the existence of an irrealis effect in auxiliary selection in these languages. In line with her earlier research (see section 1 of this introduction), Alexiadou argues that the irrealis effect evinces a difference in the function of HAVE + PtcP and BE + PtcP. In addition, she claims that the assumption of a constructional difference between HAVE and BE can explain why English and German differ with regard to the historical development of auxiliary selection. While the loss of BE in Early English was caused by the grammaticalisation of HAVE to an experiential perfect, BE-selection was not lost in German since BE + PtcP was also grammaticalised to an experiential perfect. Consequently, her approach leads to the prediction that across languages, BE-selection is only lost if HAVE + PtcP is grammaticalised to an experiential perfect. Ida Larsson takes a similar approach in her contribution “The HAVE/BE alternation in Scandinavian – perfects, resultatives and unaccusative structure”, arguing that in Swedish, Norwegian and Icelandic, auxiliary selection must be characterised as a contrast between two constructions with different functions: perfects and resultatives. She bases this hypothesis on the distribution of HAVE and BE in these languages regarding counterfactual/irrealis modality, pseudoclefts and manner adverbials. Larsson also addresses the directionality of the spread of HAVE-selection in the Scandinavian languages, illustrating that the actualisation of HAVE followed the pattern modelled in the Auxiliary Selection Hierarchy. She suggests accounting for these two findings – constructional difference and directionality of spread of HAVE – in terms of Ramchand’s (2008) decomposition of a verb phrase in three subeventualities (initP, procP and resP). Only prototypical unaccusative verbs like arrive denote complex events in which every subeventuality is specified. These verbs have a highly specific lexical entry and, as a result, display very stable auxiliary selection. This model allows Larsson to explain the interaction between gradience and gradualness in the development of auxiliary selection in the Scandinavian languages. The third paper in this section, Jaume Mateu and Mar Massanell’s “A constructional approach to auxiliary selection: evidence from existential construc-

Introduction

13

tion”, approaches the functional difference between HAVE + PtcP and BE + PtcP from a slightly different perspective. Mateu and Massanell observe that in Old Catalan, HAVE-selection is typical of sentences with indefinite subjects, whereas BE-selection is typical of sentences with definitive subjects. They account for this effect in terms of a construction difference between HAVE + PtcP and BE + PtcP: existential constructions usually involve indefinite subjects, while resultative constructions usually involve definite subjects. Consequently, Mateu and Massanell argue that contexts involving existence were catalysts of the replacement of BE with HAVE: Old Catalan tokens of core unaccusatives such as venir ‘come’ in HAVE + PtcP constructions are existential constructions. This explains the high position of verbs of existence and appearance on Sorace’s Auxiliary Selection Hierarchy. As a result, the authors suggest that (a) unaccusativity is not a property of verbs but of constructions and (b) auxiliary selection does not depend on the involved verb but rather on the involved construction type. Peter Öhl’s paper “Periphrasis as a precursor of analytic inflection” analyses the development of auxiliary selection in German. Öhl argues that the German HAVE-perfect originated as a periphrastic construction that over time acquired the function of analytic inflection. He distinguishes between periphrasis and inflection in terms of paradigmaticity: due to grammaticalisation processes, periphrases evolve into a paradigmatic means of functional marking. Öhl’s proposal of modelling this process aims at combining insights from generative and functional approaches to grammaticalisation. Thus, he claims that the Old High German HAVE + PtcP construction originally had a predicative function. Since the usage frequency of this construction increased until the 9th century, it became the input for the learners’ reanalysis. The verb haben ‘have’ in the periphrastic construction was reanalysed as an auxiliary representing anteriority, while the semantic value of perfectivity results from the participle. This led to the creation of the new paradigm of analytic tense. According to Öhl, this mechanism explains the interplay between gradual historical change and abrupt syntactic reanalysis.

2.3 Mechanisms of gradual change: BE > HAVE Given that Öhl’s paper discusses the interplay between grammatical function and processes of historical change, it already has a bearing on the topic discussed in the last section of the volume, i.e. the mechanisms of gradual change that have caused HAVE to spread to contexts previously occupied by BE in some European languages. The four papers in this section focus on the inci-

14

Malte Rosemeyer and Rolf Kailuweit

dence of the semantic features of locomotion, telicity and agentivity on the gradual development of auxiliary selection. In “BE or HAVE in Contemporary Standard French – residua of semantic motivation”, Rolf Kailuweit analyses auxiliary selection in French. He argues that in earlier French, auxiliary selection mirrored a functional split between an anterior and resultative construction. However, this functional split is obsolete in Contemporary Standard French: the variation between HAVE and BE is no longer driven by the anterior-resultative opposition. Kailuweit discusses unaccusativity as modelled by the Auxiliary Selection Hierarchy and reaches the conclusion that the semantic factor “change of location” is only a weak unaccusativity criterion. However, in Contemporary Standard French, this factor is highly relevant, as all BE-selecting verbs can be characterised as change of location verbs in a literal or metaphorical sense. Kailuweit explains this finding as a frequency effect: due to their high usage frequency, verbs expressing change of location conventionalised BE-selection over time. BE-selection in Contemporary Standard French is thus a residue of previous semantically motivated constraints. With his paper “The auxiliary selection in French monter ‘move upward’ from the 16th to the 20th century, Steffen Heidinger sheds light on the question of gradualness in the development of French auxiliary selection. As already mentioned above in the context of Kailuweit’s contribution, BE-selection was more widespread in earlier stages of French than Modern French. Surprisingly, Heidinger’s data suggest that monter ‘move upward’ follows a different path. In particular, the relative frequency of use of monter in the BE + PtcP construction expands over time. Heidinger analyses this process as a function of the three features “locomotion”, “telicity” and “agentivity”, which can be used for a semantic decomposition of the Auxiliary Selection Hierarchy. He demonstrates that the use of BE + PtcP with monter is conserved in contexts where monter has a change-of-location acceptation. In Heidinger’s words, the development of monter shows “the consolidation of a decreasing form in a restricted domain”. The paper thus suggests that the development of auxiliary selection in the European languages does not necessarily proceed unidirectionally from HAVE to BE. Similarly to the previous two papers, Malte Rosemeyer’s contribution “Entrenchment and discourse traditions in Spanish auxiliary selection” investigates the gradual development of Spanish auxiliary selection as a function of semantic verb features. Rosemeyer claims that due to the high usage frequency of verbs that typically express a change-of-location semantics, these verbs resisted the replacement process of BE with HAVE longer than other verbs. He compares data from Early Modern Spanish letters, historiographical texts and

Introduction

15

administrative documents. The results from the analysis suggest that this conserving effect led to the establishment of a new rule of auxiliary selection, i.e. “select BE when expressing a change-of-location event”. Rosemeyer claims that the conserving effect of frequency is context-dependent: given that in the corpus of Early Modern Spanish letters, change-of-location predicates are particularly frequent, verbs with this semantics exhibit a stronger conserving effect in this discourse tradition than in the others investigated. This assumption offers an explanation for the apparently disproportionately high relative usage frequency of BE + PtcP in these texts. In “Auxiliary selection in closely related languages: the case of German and Dutch”, Melitta Gillmann compares the development of auxiliary selection in German and Dutch. In a similar fashion to the papers in section 3, she claims that in earlier stages of these languages, auxiliary selection has to be characterised as an opposition between a resultative and a perfect/anterior construction. Gillmann argues that the distributional differences in the auxiliary selection in these languages result from differences in its historical development: although in both German and Dutch HAVE + PtcP has grammaticalised to a perfect construction, this process is more advanced in German than in Dutch. While in German the semantic feature [+ locomotion] has become a predictor of auxiliary selection, the aspectual feature [+ completion] remains the most important predictor of auxiliary selection in Dutch. Gillmann claims that the expansion of the originally intransitive zijn ‘be’ + PtcP construction to transitive verbs such as vergeten ‘forget’ is closely associated with the resultative function of zijn ‘be’ + PtcP and consequently the feature [+ completion]. According to Gillmann, this process of host class expansion is due to the higher degree of case syncretism in Dutch than in German.

3 Summary and future directions In this section, we give a concise summary of the conclusions that can be drawn from the content of this volume and indicate future directions for research on auxiliary selection. Most of the papers in this volume analyse split-intransitivity in terms of the interplay between synchronic gradience and diachronic gradualness which is central to research in historical linguistics (see, e.g., Hopper and Traugott 2003: 45–50). Thus, they explain variation in synchrony as the result of gradual changes in diachrony. This procedure enables the identification of both converging and diverging trends in the development and use of auxiliary selection in European languages.

16

Malte Rosemeyer and Rolf Kailuweit

Although most papers agree that gradience is common to auxiliary selection in all languages studied, this gradience is caused not only by semantic but also by syntactic features (Giancarli, Loporcaro, Mateu and Massanell, Sorace). To give an example, according to Mateu and Massanell, the high position of predicates that express “existence or appearance” is the result of the affinity of HAVE + PtcP constructions for existential constructions, in other words a syntactic feature. It is not necessary (and probably impossible) to find a monocausal explanation for gradience. The challenge for future approaches to auxiliary selection is to provide models that can unify semantic and syntactic explanations for gradients in auxiliary selection. In addition, many papers share an interest in whether or not HAVE + PtcP and BE + PtcP stand in a paradigmatic relationship. These papers emphasise the functional differences between HAVE + PtcP and BE + PtcP, suggesting that particularly in earlier stages of the languages studied auxiliary selection is not auxiliary selection in the strict sense, i.e. allomorphy between two semantically synonymous auxiliaries (Alexiadou, Gillmann, Kailuweit, Larsson, Mateu and Massanell, Öhl). This claim is substantiated by the existence of an irrealis effect in many of these languages (Alexiadou, Loporcaro). Several of these authors take a diachronic perspective in the investigation of the paradigmatic relationship. Over time, the degree of functional similarity between the two constructions appears to have increased (Gillmann, Kailuweit, Larsson, Öhl). Consequently, the results of this volume suggest that the analysis of auxiliary selection must account for a wide range of contextual factors as indicators of the distribution of HAVE and BE. Finally, a series of papers address the interplay between semantic gradience and gradualness in auxiliary selection (Gillmann, Heidinger, Kailuweit, Larsson, Rosemeyer). These papers demonstrate the predictive power of synchronic gradience for the diachronic actualisation of HAVE + PtcP in several European languages. The results in this chapter suggest that (a) the Auxiliary Selection Hierarchy can serve as a model for the actualisation of HAVE + PtcP in the European languages, (b) frequency effects further modified the course of actualisation of HAVE + PtcP and (c) discourse traditions had an influence on the actualisation of HAVE + PtcP. In modelling the historical trajectory of the replacement of BE, all of the papers in the section attribute a high relevance to the semantic parameter of “change of location” (Kailuweit, Heidinger, Rosemeyer, Gillmann). Thus, it appears that in German, French and Spanish, BEselection came to be associated with change of location contexts, while in Dutch telicity remained the most important predictor of auxiliary selection. Such developments could lead to local increases in the usage frequency of BE + PtcP and thus to rising syntactic productivity.

Introduction

17

Earlier versions of the papers collected in this volume were presented at a workshop at the Freiburg Institute for Advanced Studies (FRIAS) in June 2011. We would like to thank Peter Auer, the former director of the school of Language and Literature, for hosting the workshop and making this publication possible. Further thanks go to the FRIAS staff for organising the workshop and especially to Elin Arbin, Hannah Davidson, Eva Mieth and Sophie Routen for correcting and formatting the papers. We also want to thank the editors of the series linguae & litterae, Peter Auer, Gesa von Essen and Werner Frick, as well as the collaborators of the publishing house De Gruyter for their assistance in the publishing process.

References Abeillé, Anne and Danielle Godard. 2002 The syntactic structure of French Auxiliaries. Language 78(3): 404–452. Alexiadou, Artemis, Elena Agnostopoulou and Martin Everaert. 2004 Introduction. In: Artemis Alexiadou, Elena Agnostopoulou and Martin Everaert (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 1–21. Oxford: Oxford University Press. Andersen, Henning. 2001a Actualization and the (uni)directionality of change. In: Henning Andersen (ed.), Actualization: Linguistic Change in Progress, 225–248. Amsterdam, Philadelphia: John Benjamins. Andersen, Henning. 2001b Markedness and the theory of linguistic change. In: Henning Andersen (ed.), Actualization: Linguistic Change in Progress, 21–57. Amsterdam, Philadelphia: John Benjamins. Aranovich, Raúl. 2003 The semantics of auxiliary selection in Old Spanish. Studies in Language 27(1): 1–37. Belletti, Adriana. 1988 The case of unaccusatives. Linguistic Inquiry 19(1): 1–34. Bentley, Delia and Thórhallur Eythórsson. 2004 Auxiliary selection and the semantics of unaccusativity. Lingua 114(4): 447–471. Brinton, Laurel J. and Elizabeth Closs Traugott. 2005 Lexicalization and Language Change. Cambridge: Cambridge University Press. Burzio, Luigi. 1981 Intransitive verbs and Italian auxiliaries. PhD thesis, MIT. Burzio, Luigi. 1986 Italian Syntax. A Government-Binding Approach. Dordrecht: Reidel. Bybee, Joan L., Revere D. Perkins and William Pagliuca. 1994 The Evolution of Grammar. Tense, Aspect and Modality in the Languages of the World. Chicago: University of Chicago Press. Cennamo, Michela. 2008 The rise and development of analytic perfects in Italo-Romance. In: Thórhallur Eythórsson (ed.), Grammatical Change and Linguistic Theory: the Rosendal Papers, 115–142. Amsterdam, Philadelphia: John Benjamins. Cennamo, Michela and Antonella Sorace. 2007 Auxiliary selection and split intransitivity in Paduan: variation and lexical-aspectual constraints. In: Raul Aranovich (ed.), Split Auxiliary Systems: A Cross-Linguistic Perspective, 65–99. Amsterdam, Philadelphia: John Benjamins.

18

Malte Rosemeyer and Rolf Kailuweit

Cinque, Guglielmo. 1990 Ergative adjectives and the lexicalist hypothesis. Natural Language & Linguistic Theory 8(1): 1–39. Cocchi, Giulia. 1994 An explanation of the split in the choice of perfect auxiliaries. Probus 6: 87–102. Dahl, Östen. 1995 Areal tendencies in tense-aspect systems. In: Pier Marco Bertinetto, Valentina Bianchi, Östen Dahl and Mario Squartini (eds.), Temporal Reference, Aspect and Actionality, Vol. 2: Typological Perspectives, 11–28. Turin: Rosenberg & Sellier. Dahl, Östen and Viveka Velupillai. 2013 The perfect. In: Matthew S. Dryer and Martin Haspelmath (eds.), The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. Available online at http://wals.info/chapter/68. Last access 6 June 2013. de Smet, Hendrik. 2012 The course of actualization. Language 88(3): 601–633. Drinka, Bridget. 2013 Sources of auxiliation in the perfects of Europe. In: Hendrik de Smet, Lobke Ghesquière and Freek van der Velde (eds.), On Multiple Source Constructions in Language Change (Special issue of Studies in Language 37:3), 599–644. Amsterdam, Philadelphia: John Benjamins. Formentin, Vittorio. 2001 L’ausiliazione perfettiva in antico napoletano. Archivio Glottologico Italiano 86(79): 79–117. Förster, August. 1908 Avoir und être als Hilfsverba beim intransitiven Zeitwort in ihrer Entwicklung vom Alt- zum Neufranzösischen. Darmstadt: Otto. Haspelmath, Martin. 2001 The European linguistic area: Standard Average European. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language Universals − Sprachtypologie und sprachliche Universalien − La Typologie des langues et les universaux linguistiques. An International Handbook − Ein internationales Handbuch − Manuel international 1, 1492–1510. Berlin, New York: De Gruyter. Hopper, Paul J. and Elizabeth Closs Traugott. 2003 Grammaticalization. 2. Ed. Cambridge: Cambridge University Press. Kailuweit, Rolf. 2011 Le choix de l’auxiliaire: Être ou avoir en français standard contemporain. In: Renata Enghels, Machteld Meulleman and Clara Vanderschueren (eds.), Peregrinatio in Romania: Artículos en homenaje a Eugeen Roegiest con motivo de su 65 cumpleaños, 397–420. Gent: Academia Press. Kayne, Richard S. 1993 Towards a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. Keller, Frank and Antonella Sorace. 2003 Gradient auxiliary selection and impersonal passivization in German: an experimental investigation. Journal of Linguistics 39(1): 57– 108. Kern, Johan Hendrik. 1912 De met het participium praeteriti omschreven werkwoordsvormen in ’t Nederlands [Periphrastic verb forms with past participles in Dutch]. Amsterdam: Müller. Larsson, Ida. 2009 Participles in time. The development of the perfect tense in Swedish. PhD thesis, University of Gothenburg. Ledgeway, Adam. 2003 L’estensione dell’ausiliare perfettivo avere nell’antico napoletano: intransitività scissa condizionata da fattori modali. Archivo Glottologico Italiano 88: 29– 71. Levin, Beth and Malka Rappaport Hovav. 1995 Unaccusativity. At the Syntax-Lexical Semantics Interface. Cambridge, Mass.: MIT Press. Loporcaro, Michele. 2007 On triple auxiliation in Romance. Linguistics XLV: 173–222.

Introduction

19

Loporcaro, Michele. 2011 A euroversal in a global perspective: auxiliation and alignment. In: Peter Siemund (ed.), Linguistic Universals and Language Variation, 55–91. Berlin, New York: De Gruyter. Mackenzie, Ian. 2005 Achievement verbs in Medieval and Modern Spanish. In: Roger Wright Wright and Peter Ricketts (eds.), Studies on Ibero-Romance Linguistics: Dedicated to Ralph Penny, 375–390. Newark: Juan de la Cuesta. Mackenzie, Ian. 2006 Unaccusative Verbs in Romance Languages. Basingstoke, Hampshire, New York: Palgrave Macmillan. Magnusson, Erik Rudolf. 1939 Syntax des Prädikatsverbums im Mittelniederdeutschen. Von der ältesten Zeit bis zum Anfang des fünfzehnten Jahrhunderts. Lund: Gleerup. Mateu, Jaume. 2009 Gradience and auxiliary selection in Old Catalan and Old Spanish. In: Paola Crisma and Giuseppe Longobardi (eds.), Historical Syntax and Linguistic Theory, 176–193. Oxford: Oxford University Press. McFadden, Thomas and Artemis Alexiadou. 2006 Auxiliary selection and counterfactuality in the history of English and Germanic. In: Jutta M. Hartmann and Lászloacute Molnárfi (eds.), Comparative Studies in Germanic Syntax, 237–262. Amsterdam, Philadelphia: John Benjamins. McFadden, Thomas and Artemis Alexiadou. 2010 Perfects, resultatives and auxiliaries in Early English. Linguistic Inquiry 41(3): 389–425. Müller, Natascha. 1999 Ergative und unergative Verben aus romanistischer Sicht. Available online at http://www.fb10.uni-bremen.de/iaas/workshop/ergativ/mueller.pdf, retrieved 28. 8. 2012. Perlmutter, David M. 1978 Impersonal passives and the Unaccusative Hypothesis. In: Jeri J. Jaeger, Christine Chiarello, Henry Thompson and Farrell Ackerman (eds.), Proceedings of the 4 th Annual Meeting of the Berkeley Linguistics Society, 157–189. Berkeley: UC Berkeley. Perlmutter, David M. 1983 Personal vs. impersonal constructions. Natural Language and Linguistic Theory 1: 141–200. Ramchand, Gillian. 2008 Verb Meaning and the Lexicon. A First Phase Syntax. Cambridge: Cambridge University Press. Rosemeyer, Malte. 2012 How to measure replacement: auxiliary selection in Old Spanish bibles. Folia Linguistica Historica 33(1): 135–174. Rosemeyer, Malte. 2014 Auxiliary Selection in Spanish. Gradience, Gradualness, and Conservation. Amsterdam, Philadelphia: John Benjamins. Sapp, Christopher. 2011 Auxiliary selection in the Early New High German tenses. Groninger Arbeiten zur germanistischen Linguistik 53(2): 29–43. Shannon, Thomas F. 1990 The Unaccusative Hypothesis and the history of the perfect auxiliary in Germanic and Romance. In: Henning Andersen and Konrad Koerner (eds.), Historical Linguistics 1987, 461–488. Amsterdam, Philadelphia: John Benjamins. Shannon, Thomas F. 1995 Toward a cognitive explanation of perfect auxiliary variation: some modal and aspectual effects in the history of Germanic. American Journal of Germanic Linguistics & Literatures 7(2): 129–163. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76(4): 859–890. Stolova, Natalya I. 2006 Split intransitivity in Old Spanish: irreals and negation factors. Revue roumaine de linguistique 51(2): 301–320. Timberlake, Alan. 1977 Reanalysis and actualization in syntactic change. In: Charles N. Li (ed.), Mechanisms of Syntactic Change, 141–177. Austin: University of Texas Press.

20

Malte Rosemeyer and Rolf Kailuweit

Traugott, Elizabeth Closs and Graeme Trousdale. 2010 Gradience, gradualness and grammaticalization. How do they intersect? In: Elizabeth Closs Traugott and Graeme Trousdale (eds.), Gradience, Gradualness and Grammaticalization, 19–44. Amsterdam, Philadelphia: John Benjamins. van der Auwera, Johan. 1998 Conclusion. In: Johan van der Auwera (ed.), Adverbial Constructions in the Languages of Europe, 813–836. Berlin, New York: De Gruyter. Van Valin, Robert. 1990 Semantic parameters of split intransitivity. Language 66(2): 221– 260.

Section 1: (Limits of) Semantic and syntactic gradience

Antonella Sorace, University of Edinburgh

The cognitive complexity of auxiliary selection: from processing to grammaticality judgements However much all things may be so and not so, still there is a more and a less in the nature of things. Aristotle, Metaphysics

1 Split intransitivity and auxiliary selection: from the Unaccusative Hypothesis to now According to the Unaccusative Hypothesis (Perlmutter 1978; Burzio 1986), there are two types of intransitive verbs – unaccusative and unergative – with distinct syntactic properties. The essential insight is that the subject of unaccusative verbs is syntactically comparable to the object of a transitive verb, while the subject of an unergative verb is a true subject. Evidence for the distinction is both syntactic and semantic. For example, in several European languages unaccusative verbs generally select BE as a perfective auxiliary while unergative verbs select HAVE, as shown in (1) and (2): (1) a. Il postino è / *ha arrivato in ritardo The postman is / has arrived late b. Marie est / *a venue à la fête Marie is / has come to the party c. De brief is / *heeft vandaag gekomen The letter is / has today arrived d. Der Zug ist / *hat spät angekommen The train is / has late arrived (2) a. I musicisti hanno / *sono suonati tutto il pomeriggio The musicians have / are played whole the afternoon b. Les ouvriers ont / *sont travaillés toute la nuit The workmen have / are worked whole the night c. De trompettist heeft / *is met bolle wangen geblazen The trumpeter has / is with with all his might blown

24

Antonella Sorace

d. Hans hat / *ist den ganzen Tag gearbeitet Hans has / is the whole day worked Semantically, the subject of unaccusative verbs tends to be a patient or a nonvolitional causer while that of unergative verbs tends to be an agent (Dowty 1991; Van Valin 1990). However, it has proved difficult to fit many verbs unambiguously into one class or the other. On the one hand, there are verbs that do not satisfy unaccusativity diagnostics in consistent ways, both within and across languages; on the other hand, there are verbs that can display either unaccusative or unergative syntax depending on the characteristics of the predicate (see Levin and Rappaport Hovav 1995; Alexiadou et al. 2004; McFadden 2007; van Gelderen et al. 2013 for fuller discussions). One of the main challenges posed by the Unaccusative Hypothesis is therefore to account for the variable behaviour of verbs. Theoretical linguistic research in the last 15 years – expressed in both “projectionist” and “constructional” approaches – has focused on the complex mappings between a lexicalsemantic level of representation and the level of syntactic structure (Levin and Rappaport Hovav 2005; Ramchand 2008). Projectionist approaches enrich the lexical entry of verbs with fine-grained semantic specifications which project to the syntax via a complex system of linking rules. Constructional approaches, on the other hand, assume “bare” lexical entries that are free to project onto enriched syntactic configurations, which in turn determine interpretation (Borer 1994, 2005). However, the projectionist view allows for too little variation because of the deterministic nature of its linking rules, whereas the constructionist view allows too much variation because it lacks a mechanism that rules out impossible mappings. These limitations have been highlighted in particular by work by Sorace and colleagues (see e.g. Sorace 2000, 2004) which has shown that there is systematic variation that cannot be explained by either approach. Instead, her proposal is that intransitive verbs are organized in a hierarchy defined primarily by aspectual notions (telicity/atelicity) and secondarily by the degree of agentivity of the verb. This hierarchy was originally found for auxiliary selection and therefore termed “Auxiliary Selection Hierarchy”; then it was attested for other diagnostic of split intransitivity, which led to the more general term of “Split Intransitivity Hierarchy” (SIH), as in Figure 1. The array of verb classes represented on the SIH reduces to two key factors – telicity and agentivity – whose interaction affects the syntax of split intransitivity and creates gradient satisfaction of morphosyntactic diagnostics of split intransitivity: “telic change” at the core of unaccusativity and “agentive atelic non motional activity” at the core of unergativity. The closer to the core

The cognitive complexity of auxiliary selection

CHANGE OF LOCATION > CHANGE OF STATE > CONTINUATION OF STATE > EXISTENCE OF STATE > UNCONTROLLED PROCESS > CONTROLLED MOTIONAL PROCESS > CONTROLLED NON-MOTIONAL PROCESS >

25

categorically unaccusative

categorically unergative

Fig. 1: The Split Intransitivity Hierarchy (SIH).

a verb is, the more determinate its syntactic status as either unaccusative or unergative, and thus its compatibility with morphosyntactic diagnostics of unaccusativity or unergativity. Sensitivity to contextual or compositional factors also correlates with the distance of a verb from the core: verbs that are stative and non-agentive are the most indeterminate and therefore the most susceptible to alternations and variable syntactic behaviour across languages. What kind of gradience is represented by the SIH? It is important to distinguish gradience from the more general meaning of variation. Variation refers to the existence of linguistic structures that may alternate freely or randomly (albeit within limits); in contrast, gradience refers to alternations that obey tighter constraints and result in degrees of variation (in the sense of graded likelihood to alternate) and graded perception of (un)acceptability. It is gradience – rather than simply variation – that has been the object of investigation in studies on the SIH. Gradience is a property of speakers’ mentally represented grammar because individual speakers agree on intermediate degrees of unacceptability (see e.g. Fanselow et al. 2006 for recent theoretical treatments). In this respect, the gradience embodied by the SIH is also different from Creissels’s (2008) concept of fluid intransitivity: this is defined as “fluctuation” in the behaviour of intransitive verbs leading to “vacillations” in their assignment to the unaccusative or unergative class which are inevitably “exceptions” if a strictly syntactic split is maintained. In contrast, gradience on the SIH, as part of speakers’ linguistic knowledge, is much more systematic and far from being exceptional. Importantly, it affects only certain verbs and coexists with the categorical behaviour of other verbs. Gradience in this sense is typically left unaccounted for by traditional linguistic models of the syntax-lexicon interface. For example, optimality-theoretic accounts (e.g. Legendre’s 2007 work on auxiliary selection) address the issue of variation, but not the phenomenon of gradience. Similarly, projectionist accounts such as Levin and Rappaport Hovav (1995) could not explain the fact that in English verbs of (sound) emission exhibit more variation than verbs of change, or that in Italian durare ‘last’ can take

26

Antonella Sorace

both auxiliary essere ‘be’ and avere ‘have’ but partire ‘leave’ can take only essere. There is now evidence for SIH-style gradience in split intransitivity in more than a dozen typologically diverse languages, including Basque, French, Catalan, Chinese, Croatian, Dutch, German, Irish, Italian, Japanese, Paduan, Sardinian, Spanish, Turkish and, in addition, some sign languages (Sorace, to appear). The SIH has also received support in the literature on diachronic change. Variable verbs in terms of the SIH are diachronically unstable and prone to change, as is well attested in the pan-Romance BE→HAVE shift: change starts from non-core verbs and affects core verbs last (Tuttle 1986 on Italian; Benzing 1931 and Aranovich 2003 on Spanish; Legendre and Knipe 2003 on French; Sankoff and Thibault 1977 on Canadian French; Rohlfs 1969 on Italian; Cennamo 2008 on Old Neapolitan). While effects of the SIH have been found on a variety of manifestations of split intransitivity (e.g. ne-cliticisation in Italian, Sorace 1995; quantifier floating in Japanese, Sorace and Shomura 2000), the original and most detailed demonstrations of the SIH have focused on auxiliary selection. The typological predictions made possible by the SIH can therefore be best illustrated by a comparison of different languages that allow this phenomenon. The SIH predicts that, across languages, telicity is the primary factor, separating BE verbs from HAVE verbs and distinguishing subclasses of BE verbs; agentivity further differentiates among atelic verbs of process, identifying verb subclasses that require HAVE to different degrees (see Sorace 2000 for details). The SIH makes it possible to account for cross-linguistic variation in auxiliary selection systems. Not all languages are predicted to make the same differentiations among verb classes, but core verbs are predicted to select the auxiliary BE or HAVE across all languages, while intermediate verbs are predicted to exhibit crosslinguistic variation: an intermediate verb class could select BE in one language and HAVE in another and exhibit auxiliary alternations within the same language. These predictions have been borne out in several auxiliary-selecting languages (Cennamo and Sorace 2007; Legendre and Sorace 2003, 2007; Sorace 2000; Sorace, to appear). These concepts can be exemplified by a comparison of two auxiliary-selecting languages such as Italian and German. While the SIH has been amply documented for Italian (Sorace 2000, 2004), the literature on German split intransitivity and auxiliary selection has focused on the syntactic bases of the distinction (Grewendorf 1989) and on its semantic bases (Seibert 1993; Kaufmann 1995). Among the researchers working on argument structure and the syntaxsemantics interface, Van Hout, Randall, and Weissenborn (1993), emphasise the centrality of the concept of change for unaccusativity in German (equiva-

The cognitive complexity of auxiliary selection

27

lent terms are Brinkmann’s 1992 “transition” and “locomotion” used by Randall et al. 2004 and Randall 2010), as opposed to the more restricted notion of telic change which has been found to determine unaccusativity in other languages, such as Italian and Dutch. This parametric difference is necessary, in these authors’ view, to account for the fact that Dutch and German select different auxiliaries for verbs denoting displacement without a specific endpoint, as shown in (3): (3) a. Paul und Rita sind stundenlang durch den Saal getanzt. Paul and Rita are for.hours though the room danced ‘Paul and Rita have been dancing around in the room for hours.’ b. Paul en Rita hebben urenlang door de zaal gedanst. Paul and Rita have for.hours though the room danced ‘Paul and Rita have been dancing around in the room for hours.’ Keller and Sorace (2003) set out to assess the validity of the SIH for German by testing (a) auxiliary choice and impersonal passivisation, (b) the extent of the correlation between auxiliary selection and impersonal passivisation in German, i.e., whether the two tests broadly identify the same syntactic classes of verbs and whether they display variation with respect to the same semantic verb classes and (c) the correlation between dialectal variation in auxiliary choice and the position of verbs in the SIH. Based on Magnitude Estimation acceptability judgment data (Bard, Robertson, and Sorace 1996), Keller and Sorace were able to confirm that auxiliary selection in German, as in other languages, is sensitive to telicity and agentivity. Native speakers’ intuitions are most determinate for core verb types (e.g. ankommen ‘arrive’, abreisen ‘depart’ which are strongly preferred with sein; reden ‘talk’, arbeiten ‘work’ which are strongly preferred with haben). Nevertheless, native German intuitions do not differentiate between verbs of change of location and verb of change of state with a telicity-inducing prefix (e.g. verrosten ‘rust’, verwelken ‘wilt’), but exhibit indeterminacy in auxiliary selection with unprefixed indefinite change verbs, which are not inherently specified for telicity (see Sorace 2000 for examples in other languages). As will be seen below, the difference between prefixed and unprefixed change of state verbs is confirmed by experimental data obtained in online tasks. The class of motional process verbs (e.g. swimmen ‘swim’, rennen ‘run’) elicit a strong preference for sein in German, unlike many other languages in which these verbs select HAVE when they are not accompanied by a prepositional phrase indicating the endpoint of the process. Taken together, these results indicate that telicity is a crucial determinant of seinselection, but not the only one: the factor “locomotion” or “spatial transition”

28

Antonella Sorace

also underpins the choice of sein. The factor “transition” by itself (i.e. not specifically spatial) is not sufficient to guarantee the selection of sein, as indicated by the indeterminate behaviour of verbs of indefinite change. Intermediate verbs on the SIH are more variable, as predicted, but do not exhibit precisely the same pattern in German as in other languages. Auxiliary selection is most indeterminate with stative verbs denoting position (e.g. baumeln ‘dangle’, liegen ‘lie’). Verbs of uncontrolled non-motional process (e.g. shaudern ‘shudder’, zittern ‘shiver’) and uncontrolled emission (e.g. rumpeln ‘rumble’, klappern ‘rattle’) show a weaker preference for haben than verbs of controlled, non-motional process – also in line with the SIH. Verbs of continuation of state (e.g. überleben ‘survive’, verharren ‘persist’), however, show a definite preference for haben and no sensitivity to other factors, such as subject agentivity. As Keller and Sorace suggest, it is possible that these verbs are conceptualized as processes rather than continuations of a pre-existing state: the underspecified event structure of these verbs makes them potentially compatible with different conceptualizations.

2 Beyond the Unaccusative Hypothesis: some open questions The SIH is, by itself, a generalization and not a theory. This generalization appears at first sight to suggest that, within their respective classes, some verbs are “more unaccusative” and “more unergative” than others (Legendre, Miyata, and Smolensky 1991). But the unaccusative/unergative split is a binary syntactic distinction and therefore is not compatible with the idea that unaccusativity and unergativity are inherently gradient notions. Does this mean that the Unaccusative Hypothesis should be abandoned after 35 years? The key issue, recently re-proposed by Perlmutter (2010) himself, is whether the relevant phenomena can be accounted for in semantic terms without invoking a syntactic representation of unaccusativity (as in e.g. Bentley and Eythórsson 2003; Bentley 2006). The thesis defended here is that the fundamental intuition underlying the Unaccusative Hypothesis can be maintained (although not the details of the original syntactic analysis – see Alexiadou et al. 2004), but needs to be reconceptualized within a model of the lexicon-syntax interface that explains how a multi-dimensional lexical-semantic level maps onto a binary syntactic level. Depending on the interplay of the lexical semantics of the verb and the aspectual composition of the predicate, some verbs allow only one type of syntactic projection whereas other verbs are compatible with different projections

The cognitive complexity of auxiliary selection

29

to variable degrees. This is the reason why any “syntactocentric” or “semanticocentric” approaches that focus exclusively on either the syntactic or the semantic side of split intransitivity at the exclusion of the other are ultimately bound to provide only a partial picture of this phenomenon. One important limitation of these approaches is the fact that they are either based on purely theoretical argumentations or on corpora and/or offline data. On the one hand, linguistic theories cannot determine exactly when syntactic, lexical and aspectual factors are computed and how they become integrated in the comprehension and production of intransitive verbs appearing in the typical constructions that have served as diagnostics of unaccusativity/unergativity. On the other hand, acceptability judgment data, which have been the main source of evidence for the SIH, do not capture the relative weight of syntactic and semantic factors and their interplay in real-time processing of auxiliaries with intransitive verbs. For example, Keller and Sorace’s study raises some intriguing questions about the role of telicity and agentivity in processing German auxiliary selection that are difficult to address on the basis of their off-line judgmental data. In particular, the difference between telicity inherently encoded in the verb’s argument structure (as in ankommen) and telicity morphologically induced by the presence of a prefix (as in verwelken) is one of compositionality: is one type of telicity more complex than the other? Is compositional telicity computed at a later stage than inherent telicity?

3 Split intransitivity: acquisition, attrition and processing Data from experiments using online measures are potentially more suitable to address these questions, since they do not rely on explicit responses to stimuli on which the speaker has conscious control and therefore provide a more direct picture of implicit knowledge that cannot be consciously manipulated: in other words, they can shed light on the processes that necessarily precede particular acceptability judgments. We will consider some of the more recent experimental evidence supporting the SIH, distinguishing between studies supporting the gradience in verb behaviour and studies supporting the binary distinction between unaccusative and unergative verbs. It is the existence of evidence for both sides – briefly summarized in the next section – that represents a strong argument in favour of modelling their interface.

3.1 Evidence for the syntactic distinction Some of the most telling evidence for the “psychological reality” of the unaccusative/unergative distinction comes from studies of second language acquisi-

30

Antonella Sorace

tion and first language attrition. Developmental studies generally show a split between the syntactic distinction underlying split intransitivity, which is acquired early and remains stable, and the interface conditions determining gradience, which display more variation and instability. Sorace (1993a, 1993b), for example, demonstrated that the linguistic intuitions of non-native Italian speakers initially are most determinate for core verbs and then gradually approximate the SIH, without reaching the determinacy shown by native Italian speakers even at the highest proficiency level. Adult second language learners of languages that do not have overt and consistent morphosyntactic markers of split intransitivity go through a transitional stage in which they introduce these markers in the language (Zobl 1989; Balcom 1997; Hirakawa 2001; Oshita 2001). For example, learners of English from various language backgrounds overextend the passive constructions with core unaccusative verbs: (4) a. My mother was died when I was a baby b. People are fallen in love c. What is happened? Overpassivisation with unaccusative verbs is a strong indication that learners expect to find overt markers of unaccusativity/unergativity in the second language. When these are not found, learners apply markers typically available in other languages (such as auxiliary selection) even if they are not instantiated in either their native language or the target language. Montrul’s (2005) study of native language attrition in second-generation Spanish speakers in the US (“heritage speakers”) shows that these speakers maintain robust knowledge of the syntactic reflexes of unaccusativity in Spanish, since they correctly discriminated syntactically between unaccusative and unergative verbs in contexts requiring postverbal subjects, the absolutive construction and postverbal bare plural subjects. However, these speakers do lose sensitivity to the gradient distinctions along the SIH. Attrition therefore appears to affect the lexicon-syntax interface mappings but not the unaccusative/ unergative syntactic distinction itself. A number of psycholinguistic studies of native language processing offer evidence of the syntactic distinction underlying the Unaccusative Hypothesis. Friedmann et al. (2008) used a cross-modal lexical priming technique, which tests whether or not the subject NP is reactivated after unergatives and unaccusatives verbs during the online processing of a sentence. The experiments revealed that only subjects of unaccusatives reactivate after the verb, but subjects of unergatives do not. The fact that sentences with unaccusative and unergative verbs are processed differently directly supports the Unaccusative

The cognitive complexity of auxiliary selection

31

Hypothesis and the underlying analysis based on the different structural status of the single argument of unaccusative and unergative verbs. Interestingly, some verbs that enter transitive-unaccusative alternations do not show a consistent pattern of trace reactivation, a fact that the authors themselves suggest might be related to their intermediate position on the SIH. The psychological reality of abstract semantic features, such as telicity and agentivity, is addressed in a study of a Semantic Dementia patient by Romagno et al. (2010). This patient showed a dissociation between impaired access to the referential semantic features of verbs (dying, for example, refers to stopping living or existing) and the lexical-semantic features, such as telicity, affecting the syntactic behaviour of verbs, including auxiliary selection. Impairment selectively affected referential semantic features but not abstract lexical semantic features. A body of neurolinguistic studies of aphasia (e.g. Thomson 2003, among others) also supports the reality of the Unaccusative Hypothesis in processing terms. Neurological evidence for the Unaccusative Hypothesis comes from a study by Shetreet, Friedmann, and Hadar (2009). The authors show that the brain distinguishes between unaccusative and unergative verbs, even when they appear in identical structures. Furthermore, different patterns of brain activation were found for syntactic and lexical operations: the inferior frontal gyrus appears to be involved with the execution of the syntactic operation of moving the argument from an object to a subject position, whereas the middle temporal gyrus may be responsible for other lexical operations that are associated with unaccusative verbs in particular languages. In sum, a range of studies offers processing and neurological arguments in support of the syntactic split originally assumed by the Unaccusative Hypothesis. This evidence complements the linguistic arguments for maintaining a syntactic characterization of split intransitivity as a way of accounting for generalizations that unify transitive and intransitive clauses (Levin and Rappaport Hovav 1995; Perlmutter 2010).

3.2 Evidence for gradience Is there real-time processing evidence for the gradient variation of the SIH? Recent studies have begun to provide a perspective on gradience that is complementary to that resulting from earlier acceptability judgment studies. In eye-tracking experiments with native Italian speakers, Bard, FrenckMestre, and Sorace (2010) explored the processing correlates of the SIH by using real-time measures of eye movements in sentence reading. Effects of the

32

Antonella Sorace

SIH were found on second pass reading times, although not on first pass reading times: participants took longer to read sentences with core unaccusative or unergative verbs (as in 5a and 6a) than those with non-core verbs (as in 5b and 6b) when presented with the incorrect auxiliary (i.e. avere in 5 and essere in 6): (5) a. Alla festa il miliardario ha entrato / è entrato da solo nella sala b. Alla festa il miliardario ha rimasto / è rimasto da solo nella sala ‘At the party the millionaire entered/remained alone in the room.’ (6) a. A quella vista il codardo ha urlato / è urlato per lo spavento b. A quella vista il codardo ha trasalito / è trasalito per lo spavento ‘At that sight the coward shouted/jumped in fright.’ The effect was replicated in an eye-tracking experiment by Sorace and Vernice (forthcoming), who found longer reading times in first pass as well as second pass reading measures. In both studies, another signature of the SIH was a “spill-over” effect for non-core verbs, (especially for non-core unergatives) in the words immediately following the past participle, which is interpretable as non-commitment of the processor on auxiliaries with underspecified verbs until the rest of the sentence is encountered. The eyetracking data in Sorace and Vernice reveal another striking difference: auxiliary violations with unergatives trigger more fixations on the subject than auxiliary violations with unaccusatives, which can be interpreted as a sign that the subject of unergatives is more salient for auxiliary selection than the subject of unaccusatives. The eye-tracking data overall suggest that auxiliary selection violations with verbs fully specified for telicity cause more processing disruption than violations with underspecified verbs. Moreover, auxiliary violations with underspecified verbs cause more extended processing disruption than those to core verbs because non-core verbs depend on compositional factors beyond the auxiliary-verb combination. Although at first glance the data do not seem to support a model of auxiliary selection as an operation involving two independent and sequential stages (i.e. the syntactic computation of unaccusativity/unergativity followed by the integration of aspectual and semantic information from the context), the results are open to multiple interpretations, including one that assumes the parallel (late) processing of the syntax and the semantics of split intransitivity (see Bard et al. 2010 for discussion). Event-related brain potentials (ERPs) provide a different and potentially more direct measure of cognitive processing. Because of the high sensitivity and multidimensionality of this measure in combination with the method’s high temporal resolution, ERPs are very well suited to an examination of the

The cognitive complexity of auxiliary selection

33

rapid and complex integration of information in language processing. With respect to language comprehension, research within the ERP paradigm has revealed a number of distinct ERP components that have been functionally classified as arising from semantic or syntactic processes. First, the N400, a centroparietal negativity with a peak latency of approximately 400 ms post-stimulus onset, was described by Kutas and Hillyard (1980) in response to the processing of a semantically anomalous word and has since been regarded as a correlate of lexical-semantic processing. In addition, the amplitude of the N400 is modulated by a variety of lexical-semantic parameters such as word frequency or semantic relatedness (see Kutas and Federmeier 2000 for an overview). Second, the P600, a late positivity with a peak latency of approximately 600 ms after stimulus onset, has been regarded as a correlate of syntactic processing (as for example in response to the processing of a syntactically incongruous word; Osterhout and Holcomb 1992; Hagoort et al. 1993). Although recent research has shown that a strict one-to-one mapping between ERP components and functions cannot be maintained (see Bornkessel-Schlesewsky and Schlesewsky 2008 for a discussion), ERPs not only allow us to distinguish precisely the time course of processing but still provide an insight into qualitative differentiations of processing different information types. The studies briefly reviewed here illustrate the relevance of ERP data, first to the question of split intransitivity itself and second to the question of gradience in auxiliary selection. The first study is Sorace, Filiaci, and Demiral (2013), who used EEG measures of implicit memory to investigate the salience of the subject of intransitive verbs. Salience here is intended as the depth of memory encoding of the argument that is involved in the event represented by the verb. The predictions were that (a) core unergative verbs would lead to a more salient representation of the subject argument because of its agentive nature and its syntactic stability, which would in turn be signalled by better implicit memory performance; (b) the subject of core unergative verbs should be more memorable than the subject of non-core unergative verbs; and (c) graded levels of subject salience would be observed depending on place of the verb on SIH. In the experiment, core and non-core unaccusative and unergative verbs in the present tense were embedded in grammatical sentences: (7) Da secoli un vampiro appare verso la mezzanotte ‘For centuries a vampire appears around midnight.’

core unaccusative

(8) Nel film il pirata viaggia per i sette mari ‘In the film the pirate travels the seven seas.’

core unergative

34

Antonella Sorace

Immediately after reading the sentences, participants saw a picture which represented either the subject of the sentence or a different character and were asked to indicate whether the picture matched the referent of the sentence or not. The measured effects in the EEG signal were the brain responses in the theta frequency band (4–7 Hz) of the human EEG involved in memory performance with respect to load on working memory resources. Higher theta values have been found to correlate with stronger memory retrieval effort (Heine et al. 2006). As predicted, the theta activity for the argument of core unaccusative verbs was higher than the evoked theta activity for the argument of unergative verbs (p < .05). No difference was found with respect to the theta activity for non-core verbs. These preliminary results lend support, first of all, to the very existence of split intransitivity – that is, of a fundamental difference between two types of intransitive verbs. Second, the results support the fundamental assumptions of the SIH: telicity is the primary dimension for unaccusatives whereas subject status is irrelevant for core verbs; moreover, agentivity is primary for unergatives but telicity is irrelevant. In light of the results, this might be related to the fact that when processing unaccusative verbs the attentional focus is shifted from the subject argument to the endpoint of the event. For unergatives, on the other hand, the action denoted by the event is atelic and continuous and the subject argument therefore remains in focus. Third, the results are consistent with previous behavioural results that found violations of auxiliary selection with core unaccusatives to be perceived as stronger than violations of auxiliary selection with core unergative verbs (Bard et al. 2010; Vinson et al. 2007). Speakers need to pay attention to the event when processing unaccusative verbs: the attentional focus is distributed over the event and the auxiliary is a privileged marker of telicity. In contract, when processing unergatives, the prominence and salience of subject status is the main factor, rather than the atelic event. This is reflected in a weaker link between the have auxiliary and unergative verbs than between the be auxiliary and unaccusative verbs. The ERP correlates of German auxiliary selection were studied by Roehm and Sorace (2008) and Roehm, Bornkessel-Schlesewsky, and Sorace (2010, 2012), who found that violations to auxiliary selection with core verbs engendered a biphasic N400-P600 pattern. This pattern was elicited by core unergative verbs (verbs of controlled non-motional process such as reden ‘talk’), core unaccusative verbs (verbs of change of location such as ankommen ‘arrive’), i.e. verbs with an inherent positive or negative lexical specification of the key semantic feature of telicity, and verbs with a telicity-inducing prefix (prefixed change of state verbs, such as verrosten ‘rust’). The N400-P600 is a pattern

The cognitive complexity of auxiliary selection

35

previously associated with syntactic violations that induce a simultaneous syntactic and semantic processing difficulty (Rösler et al. 1993; Ainsworth-Darnell, Shulman, and Boland 1998; Hagoort 2003). Under this interpretation, the N400 indicates a mismatch between the lexical semantic verb type and the auxiliary, while the P600 indicates a global assessment of well-formedness and an integration problem due to the fact that the mismatched past participle does not fit the syntactic structure constructed up to that point. However, the characteristics of the waveforms found by Roehm et al. (2012) were not identical for core and non-core unaccusative verbs in two respects. First, the onset latency of the N400 was delayed for sentences with prefixed verbs in comparison to sentences with inherently telic verbs, which suggests that auxiliary selection is computationally more demanding when it is compositional rather than lexically specified. Second, sentences with unprefixed change of state verbs (e.g. rosten) did not show any differences between sein and haben in grand average ERPs. This apparent indeterminacy of unprefixed verbs raises the question of whether they are flexible because they are syntactically or semantically indeterminate, or whether they have a basic syntactic behaviour but are also compatible with more than one syntactic projection. In order to address this question, Roehm et al. (2012) performed a mixed effect model analysis on the ERP data including behavioural acceptability judgment data that were also collected from participants. These additional analyses for individual participants revealed interesting correlations between the amplitudes of N400 and P600 waves and individual acceptability ratings for habensentences but – crucially – only for unprefixed verbs. Some individuals rated haben with unprefixed change of state verbs as acceptable, other individuals rated it as unacceptable. There was a negative correlation between N400 amplitude and acceptability: smaller N400 for haben correlates with higher acceptability rates for these sentences. The P600 showed a more complex pattern, which suggests two qualitatively distinct late positivities. First, the P600 reflects the categorisation of the sentence as ill-formed, hence it correlates negatively with the acceptability rating for that sentence (as seen before). Second, the P600 also reflects the process of coercing a telic verb into an activity reading in order to render it compatible with the auxiliary haben, hence it correlates positively with the acceptability rating for that sentence, with larger ERP amplitudes in the haben condition for high acceptability ratings. These correlations suggest that some participants apply aspectual coercion in order to render an unprefixed change of state verb compatible with the auxiliary haben: when they do, they give higher acceptability ratings to the sentence and the P600 shows a larger amplitude as a signature of this pragmatic enrichment. We can now revisit unprefixed change of state verbs and the possible causes of their flexibility in German. The absence of an effect is likely to be due to

36

Antonella Sorace

the averaging procedure rather than resulting from the processing system’s general indifference to auxiliary selection with this particular class of verbs. What can be concluded in more general terms? The gradient behaviour of noncore intransitive verbs (e.g. in terms of acceptability judgments) results from a range of more fine-grained processes during online sentence interpretation, which crucially involve compositionality, i.e. the integration of different types of cues. One type of compositional process is triggered by morphologically marked split intransitivity (e.g. telicity given by prefixation in German, by clitic si/se in Italian/Spanish, by suffixation or stem-vowel alternations in Norwegian, etc.), which involves less processing effort and less variation across speakers. A different type of compositionality obtains with morphologically unmarked split intransitivity (e.g. telicity given by adverbials, PPs, etc. or provided by the speaker through aspectual coercion); this involves more processing effort and leaves more scope for individual variation.

4 Linguistic models of split intransitivity: challenges and possible ways forward The challenge for linguistic models of split intransitivity is to make sense of the vast array of types of evidence that has emerged from studies carried out in different research areas and from different theoretical orientations. The most promising models proposed so far are enriched variants of the constructional approach that – unlike the “bare entries” constructional models – attribute a restricted set of aspectual features to lexical entries of verbs, thus constraining their association with syntactic representations, and allow the compositional construction of event structure in the syntax. Elements of this approach can be found in earlier proposals. Rappaport Hovav and Levin (1998), for example, propose Template Augmentation as a lexical operation that builds (“augments”) event structure templates up to other possible templates from less complex to more complex, within a basic inventory of event structure templates (see also McClure 1995). More recently, Ramchand’s “first-phase syntax” (Ramchand 2008; Mateu 2005, 2009; see also Folli and Harley 2006) assumes a set of features (“combinatorial primitives”) with different levels of specification that define a verb and the building of the event structure(s) in which that verb can participate. These models are more flexible than projectionist approaches: instead of exhaustively stating syntactically relevant semantic features at the level of lexical entries, they allow degrees of featural un(der)specification to determine degrees of freedom in the syntactic projection of verbs and thus the extent of their interfacing with com-

The cognitive complexity of auxiliary selection

37

positional semantics and with encyclopaedic knowledge. One of the current limitations of these models, however, is their inability to distinguish between variability in verb behaviour that can be explained structurally or configurationally (e.g. by the licensing of an additional argument, as in the causative alternation the wind broke the window/the window broke) and variability that depends on the integration of semantic or world knowledge (such as auxiliary alternations with verbs like durare ‘last’, or atterrare ‘land’, which prefer BE or HAVE depending on subject agentivity). As a consequence, these models are not well suited to explain gradient effects in acceptability judgments and psycholinguistic measures, such as the ones found by Sorace (2000, 2004) or McKoon and Macfarland (2000). Clearly, more theoretical and experimental research is needed to account for the multiple mappings between the multidimensional lexical-semantic level and the syntactic level of split intransitivity. A useful observation is that this particular lexicon-syntax interface does not seem to be different in principle from other interfaces between syntax and extra-syntactic cognitive systems that have been recently studied. The developmental patterns that have been discovered for the lexicon-syntax interface appears to share properties with those observed at other interfaces, such as the syntax-pragmatics interface (see Sorace 2006; Sorace and Keller 2006 for discussion). For example, the robustness of the unaccusative-unergative distinction in acquisition, attrition and impairment compared to the vulnerability of lexical-semantic gradience on the SIH resembles a parallel split between syntactic properties of null subject languages, which are developmentally stable, and the pragmatic-contextual conditions on the distribution of pronominal forms, which display developmental optionality (Sorace 2011). Once developmental data are taken into account, a theory of learning also becomes necessary in order to account for the acquisition of interface principles in addition to syntactic and lexical knowledge. Following Yang (2002), ideally what is needed is a theory-neutral variational model that incorporates both language-specific and domain-general mechanisms of learning. More generally, what is needed is commitment to a model of competent gradience as the target of learning (Duffield 2003), recognizing that grammatical competence includes knowledge of both grammatical invariance and interface conditions on grammatical realization. A final set of considerations can be made about the nature of the gradience represented by the SIH. We have seen that flexible verbs are not syntactic hybrids between unaccusative and unergative: rather, they allow properties of both categories. This suggests that the gradience on the SIH is produced by the intersection of properties, rather than categories (Aarts 2007). Intersective

38

Antonella Sorace

Categorical unaccusative syntax

CHANGE OF LOCATION > CHANGE OF STATE > CONTINUATION OF STATE > EXISTENCE OF STATE > UNCONTROLLED PROCESS > MOTIONAL PROCESS > NON-MOTIONAL PROCESS Categorical unergative syntax Fig. 2: Intersective gradience on the SIH for Italian.

gradience presupposes subsective gradience within categories and the presence of cut-off points between categories, the location of which may vary across languages and may determine an area of greater variability within individual languages (see also Diedrichsen 2012). For example, in Italian the cutoff point between the unaccusative and the unergative categories is between existence of state verbs and uncontrolled process verbs, as shown in Figure 2. The progressive divergence from the core implies, in behavioural terms, that more compositional effort is required to integrate factors outside the lexical verb itself that affect auxiliary selection.

5 Conclusions Linguistic and psycholinguistic experimental research has considerably expanded our understanding of phenomena related to split intransitivity, including auxiliary selection. It has allowed generalizations about variable verb behaviour that were not possible when the original Unaccusative Hypothesis was proposed. These developments have raised the question of whether the syntactic characterization at the basis of the Unaccusative Hypothesis should be rejected in favour of purely semantic accounts of split intransitivity. All in all, it seems that at the moment this move would not be appropriate in light of the existing evidence. A more promising avenue is to continue to explore the interface between the syntactic and lexical-semantic aspects of split intransitivity, both theoretically and experimentally, and to aim at an explanatory model of how gradient variance and categorical invariance can co-exist in languages and in the human mind.

The cognitive complexity of auxiliary selection

39

References Aarts, Bas. 2007 Syntactic Gradience: The Nature of Grammatical Indeterminacy. Oxford: Oxford University Press. Alexiadou, Artemis, Elena Anagnostopoulou and Martin Everaert (eds.). 2004 The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface. Oxford: Oxford University Press. Aranovich, Raúl. 2003 The semantics of auxiliary selection in Old Spanish. Studies in Language 27: 1–37. Balcom, Patricia. 1997 Why is this happened? Passive morphology and unaccusativity. Second Language Research 13: 1–9. Bard, Ellen Gurman, Daniel Robertson and Antonella Sorace. 1996 Magnitude estimation of linguistic acceptability. Language 71: 32–68. Bard, Ellen Gurman, Cheryl Frenck-Mestre and Antonella Sorace. 2010 Processing auxiliary selection with Italian intransitive verbs. Linguistics 48(2): 325–362. Bentley, Delia. 2006 Split Intransitivity in Italian. Berlin, New York: De Gruyter. Bentley, Delia and Thórhallur Eythórsson. 2003 Auxiliary selection and the semantics of unaccusatives. Lingua 114: 447–471. Benzing, Joseph. 1931 Zur Geschichte von ser als Hilfszeitwort bei den Intransitiven Verben im Spanischen. Zeitschrift für romanische Philologie 51: 385–460. Borer, Hagit. 1994 The projection of arguments. In: Elena Benedicto and Jeffrey Runner (eds.), Functional Projections, 19–47. Amherst: University of Massachusetts. Borer, Hagit. 2005 The Normal Course of Events. Oxford: Oxford University Press. Bornkessel-Schlesewsky, Ina and Matthias Schlesewsky. 2008 An alternative perspective of “semantic P600” effects in language comprehension. Brain Research Reviews 59: 55– 73. Brinkmann, U. 2002 Choice of auxiliary for intransitive verbs of motion: an analysis of an unaccusative diagnostic. Ms., Max Planck Institute for Psycholinguistics, Nijmegen. Burzio, Luigi. 1986 Italian Syntax: A Government-Binding Approach. Foris: Dordrecht. Cennamo, Michaela. 2008 The rise and development of analytic perfects in Italo-Romance. In: Thórhallur Eythórsson (ed.), Grammatical Change and Linguistic Theory: the Rosendal Papers, 115–142. Amsterdam, Philadelphia: John Benjamins. Cennamo, Michaela and Antonella Sorace. 2007 Unaccusativity at the syntax-lexicon interface: evidence from Paduan. In: Raúl Aranovich (ed.), Split Auxiliary Systems. A Cross-linguistic Perspective, 65–100. Amsterdam, Philadelphia: John Benjamins. Creissels, Denis. 2008 Remarks on split intransitivity and fluid intransivity. In: Olivier Bonami and Patricia Cabredo Hofherr (eds.), Empirical Issues in Syntax and Semantics 7, 139–168. Paris: Colloqué de Syntax et Sémantique à Paris. Dowty, David. 1991 Thematic proto-roles and argument selection. Language 67: 547–619. Duffield, Nigel. 2003 Measures of competent gradience. In: Roeland van Hout, Aafke Hulk, Folkert Kuiken and Richard Towell (eds.), The Lexicon=Syntax Interface in Second Language Acquisition, 97–128. Amsterdam, Philadelphia: John Benjamins. Fanselow, Gisbert, Caroline Féry, Ralph Vogel and Matthias Schlesewsky (eds.). 2006 Gradience in Grammar: Generative Perspectives. Oxford: Oxford University Press. Folli, Rafaella and Heidi Harley. 2006 On the licensing of causatives of directed motion: waltzing Matilda all over. Studia Linguistica 60: 121–155. Friedmann, Naama, Gina Taranto, Lewis Shapiro and David Swinney. 2008 The leaf fell (the leaf): the online processing of unaccusatives. Linguistic Inquiry 39: 355–377.

40

Antonella Sorace

van Gelderen, Elly, Michela Cennamo and Jóhanna Barðdal (eds.). 2013 Argument Structure in Flux. Amsterdam, Philadelphia: John Benjamins. Grewendorf, Günther. 1989 Ergativity in German. Dordrecht: Foris. Hagoort, Peter, Colin Brown and Jolanda Groothusen. 1993 The syntactic positive shift (SPS) as an ERP measure of syntactic processing. Language and Cognitive Processes 8: 439–483. Heine, Angela, Sascha Tamm, Markus Hofmann, Rainer Bösel and Arthur M. Jacobs. 2006 Event-related theta activity reflects memory processes in pronoun resolution. Neuroreport 17: 1835–1839. Hirakawa, Makiko. 2001 L2 acquisition of Japanese unaccusative verbs. Studies in Second Language Acquisition 23: 221–245. Kaufmann, Ingrid. 1995 O- and D-predicates: a semantic approach to the unaccusativeunergative distinction. Journal of Semantics 12: 377–427. Keller, Frank and Antonella Sorace. 2003 Gradient auxiliary selection and impersonal passivization in German: an experimental investigation. Journal of Linguistics 39: 57– 108. Kutas, Marta and Steven Hillyard. 1980 Reading senseless sentences: brain potentials reflect semantic incongruity. Science 207: 203–205. Kutas, Marta and Kara Federmeier. 2000 Electrophysiology reveals semantic memory use in language comprehension. Trends in Cognitive Sciences 4: 463–470. Legendre, Géraldine. 2007 On the typology of auxiliary selection. Lingua 117: 1522–1540. Legendre, Géraldine and Hilary Knipe. 2003 From Latin to Modern French: Tracing the evolution of auxiliary selection. Ms., Johns Hopkins University. Legendre, Géraldine, Yoshiro Miyata and Paul Smolensky. 1991 Unifying syntactic and semantic approaches to unaccusativity: a connectionist approach. In: Proceedings of the Seventh Annual Meeting of the Berkeley Linguistics Society, 156–167. Berkeley, CA. Legendre, Géraldine and Antonella Sorace. 2003 Auxiliaires et intransitivite’ en francais et dans les langues romanes. In: Danièle Godard (ed.), Les Langues Romanes; Problemes de la Phrase Simple, 185–234. Paris: Editions du CNRS. Levin, Beth and Malka Rappaport Hovav. 1995 Unaccusativity: at the Syntax-Semantics Interface. Cambridge, MA: MIT Press. Levin, Beth and Malka Rappaport Hovav. 2005 From Lexical Semantics to Argument Realization. Cambridge: Cambridge University Press. Mateu, Jaume. 2005 Arguing our way to the Direct Object Restriction on English resultatives. Journal of Comparative Germanic Linguistics 8: 55–82. Mateu, Jaume. 2009 Gradience and auxiliary selection in Old Catalan and Old Spanish. In: Dins P. Crisma and Giuseppe Longobardi (eds.), Historical Syntax and Linguistic Theory. 176–193. Oxford: Oxford University Press. McClure, William. 1995 Syntactic Projections of the Semantic of Aspect. Tokyo: Hitsujishobo. McFadden, Thomas. 2007 Auxiliary selection. Language and Linguistics Compass 1: 674– 708. McKoon, Gail and Talke Macfarland. 2002 Event templates in the lexical representations of verbs. Cognitive Psychology 45: 1–44. Montrul, Silvina. 2005 Second language acquisition and first language loss in adult early bilinguals: exploring some differences and similarities. Second Language Research 21: 199–249. Oshita, Hiroyuki. 2001 The Unaccusative Trap in second language acquisition. Studies in Second Language Acquisition 23: 279–304.

The cognitive complexity of auxiliary selection

41

Osterhout, Lee and Phillip J. Holcomb. 1992 Event-related brain potentials elicited by syntactic anomaly. Journal of Memory and Language 31: 785–806. Perlmutter, David M. 1978 Impersonal passives and the unaccusative hypothesis. Proceedings of the 4 th Annual Meeting of the Berkeley Linguistics Society, 157–189. Perlmutter, David M. 2010 My path in linguistics. In: Donna B. Gerdts, John C. Moore and Maria Polinsky (eds.), Hypothesis A/Hypothesis B: Linguistic Explorations in Honor of David M. Perlmutter, xvii–xxxvii. Cambridge, MA: MIT Press. Ramchand, Gillian. 2008 Verb Meaning and the Lexicon: A First-Phase Syntax. Cambridge: Cambridge University Press. Randall, Janet. 2010 Linking: the Geometry of Argument Structure. Berlin: Springer. Randall, Janet, Angeliek van Hout, Jiirgen Weissenborn and Harald Baayen. 2004 Acquiring unaccusativity: a cross-linguistic look. In: Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 332–354. Oxford: Oxford University Press. Rappaport Hovav, Malka and Beth Levin. 1998 Building verb meanings. In: Miriam Butt and Wilhelm Geuder (eds.), The Projection of Arguments: Lexical and Compositional Factors, 97–134. Stanford: CSLI Publications. Roehm, Dietmar and Antonella Sorace. 2008 ERP signatures of auxiliary selection in German. Poster presented at the AMLaP conference, Cambridge. Roehm, Dietmar, Antonella Sorace and Ina Bornkessel-Schlesewsky. 2010 The processing of flexible syntax-semantics mappings: a neurophysiological investigation of split intransitivity in German. In: Proceedings of Verb 2010: Interdisciplinary Workshop on Verbs. The Identification and Representation of Verb Features, 59–64. Pisa: Scuola Normale Superiore. Roehm, Dietmar, Antonella Sorace and Ina Bornkessel-Schlesewsky. 2012 Processing flexible form-to-meaning mappings: evidence for enriched composition as opposed to indeterminacy. Language and Cognitive Processes. Rohlfs, Gerhard. 1969 Grammatica storica della lingua Italiana e dei suoi dialetti. Torino: Einaudi. Romagno, Domenica, Costanza Papagno and Alfonso Caramazza. 2010 Evidence from neuropsychology on verb features: The case of a patient with Semantic Dementia. Proceedings of Verb 2010: Interdisciplinary Workshop on Verbs. The Identification and Representation of Verb Features, 333–338. Pisa: Scuola Normale Superiore. Sankoff, Gillian and Pierette Thibault. 1977 L’alternance entre les auxiliaries avoir et être en français parlé à Montréal. Langue Française 34: 81–108. Seibert, Anja. 1993 Intransitive constructions in German and the ergative hypothesis. Working Papers in Linguistics 14. University of Trondheim. Sheetret, Einat, Naama Friedmann and Uri Hadar. 2009 The neural correlates of linguistic distinctions: unaccusative and unergative verbs. Journal of Cognitive Neuroscience 22: 2306–2315. Sorace, Antonella. 1993a Incomplete vs. divergent representations of unaccusativity in nonnative grammars of Italian. Second Language Research 9: 22–47. Sorace, Antonella. 1993b Unaccusativity and auxiliary choice in non-native grammars of Italian and French: asymmetries and predictable indeterminacy. Journal of French Language Studies 3: 71–93. Sorace, Antonella. 1995 Acquiring argument structures in a second language: the unaccusative/unergative distinction. In: Lynn Eubank, Larry Selinker and Mike

42

Antonella Sorace

Sharwood Smith (eds.), The Current State of Interlanguage, 153–175. Amsterdam, Philadelphia: John Benjamins. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76: 859–890. Sorace, Antonella. 2004 Gradience at the lexicon-syntax interface: evidence from auxiliary selection. In: Artemis Alexiadou, Martin Everaert and Elena Anagnostopoulou (eds.), The Unaccusativity Puzzle, 243–268. Oxford: Oxford University Press. Sorace, Antonella. 2006 Gradience and optionality in mature and developing grammars. In: Gisbert Fanselow, Caroline Féry, Matthias Schlesewsky and Ralf Vogel (eds.), Gradience in Grammars: Generative Perspectives, 106–123. Oxford: Oxford University Press. Sorace, Antonella. 2011 Pinning down the concept of ‘interface’ in bilingualism. Linguistic Approaches to Bilingualism 1: 1–33. Sorace, Antonella. (to appear) Gradience in Split Intransitivity. Oxford: Oxford University Press. Sorace, Antonella, Francesca Filiaci and Baris Demiral. 2013 An EEG investigation of salience and aspectual focus in split intransitivity. Internal report, University of Edinburgh (submitted). Sorace, Antonella and Frank Keller. 2005 Gradience in linguistic data. Lingua 115: 1497– 1524. Sorace, Antonella and Shomura, Yoko. 2001 Lexical constraints on the acquisition of split intransitivity: evidence from L2 Japanese. Studies in Second Language Acquisition 23(2): 247–278. Sorace, Antonella and Mirta Vernice. (forthcoming) Gradient auxiliary selection in Italian: eye movement evidence. Ms, University of Edinburgh and University of Milan-Bicocca. Thomson, Cynthia K. 2003 Unaccusative verb production in agrammatic aphasia: The argument structure complexity hypothesis. Journal of Neurolinguistics 16: 151–167. Tuttle, Edward. 1986 The spread of esse as universal auxiliary in central Italo-Romance. Medioevo Romanzo 11: 229–287. Van Valin, Robert. 1990 Semantic parameters of split intransitivity. Language 66: 221–260. Yang, Charles. 2002 Knowledge and Learning in Natural Language. Oxford: Oxford University Press. Zobl, Helmut. 1989 Canonical typological structures and ergativity in English L2 acquisition. In: Susan Gass and Jacquelyn Schachter (eds.), Linguistic Perspectives on Second Language Acquisition, 203–221. Cambridge: Cambridge University Press.

Michele Loporcaro, Universität Zürich

Perfective auxiliation with reflexives in Medieval * Romance: syntactic vs. semantic gradients 1 Introduction Several studies have argued for a semantic account of perfective auxiliary selection and, more generally, of unaccusativity-related features. Sorace (1992, 2000), in particular, has recognized a semantically-based gradience in Romance and beyond. She observes that categorical ‘be’ vs. ‘have’ selection with intransitive verbs occurs at the two poles of the hierarchy in (1), defined in terms of the predicate semantics, while verbs belonging to the intermediate classes may show vacillation: (1) Sorace (2000): Auxiliary selection hierarchy (ASH)

change of location > change of state > continuation of state > existence of state > uncontrolled process > motional process > non-motional process >

categorical BE selection

categorical HAVE selection

For instance, verbs of emission like the one exemplified in (2), denoting an uncontrolled process, may take either auxiliary in Italian: (2) L’eco ha/è risuonato nella caverna ‘The echo resounded in the cave.’

* Previous versions of this research were presented at the Universities of Leiden (May 2011), Roma Tre (October 2011) and Cambridge (November 2012). I thank the audiences there (as well as participants of the Freiburg workshop in June 2012), particularly Adam Ledgeway, as well as two anonymous reviewers for discussion and constructive criticism. I am also grateful to Rachele Delucchi for comments on a previous draft. The usual disclaimers apply. The following abbreviations are used in the examples: ASH = Auxiliary Selection Hierarchy, aux = perfective auxiliary, DO = direct object, E = auxiliary ‘be’, GR = grammatical relation, H = auxiliary ‘have’, It. = Italian, PtP = past participle, RG = Relational Grammar, RRG = Role and Reference Grammar, Sp. = Spanish.

44

Michele Loporcaro

For some of the studies in this line of research, one can speak of “semantic reductionism” (3a), as they try to reduce to semantic properties (telicity, agentivity, etc.) the unergative vs. unaccusative contrast first proposed by Perlmutter (1978) in a different perspective (3b): (3) a. (Semanticist) reductionism: syntactic unaccusativity is an epiphenomenon of semantic properties (telicity, agentivity; e.g. Van Valin 1990; Cennamo 1999); b. Autonomous syntax: unaccusativity is an autonomous syntactic property, correlating with, but not reducible to, semantic properties like telicity, agentivity, etc. (e.g. Rosen 1984; Perlmutter 1989). In this paper, I argue for an approach to auxiliary selection similar to type (3b). I will show that a syntactic representation of the unaccusative vs. unergative contrast is necessary, because in its absence, it becomes impossible to make sense of the data. In particular, I shall demonstrate the existence of a syntactic gradient in Romance, which is orthogonal to and independent from that in (1). Reflexive constructions play a central role in this demonstration: therefore, the paper begins by addressing the uneconomical way in which reflexives are treated, if at all, in semantically-based approaches to auxiliary selection (section 2). It then considers one such treatment in section 3 which focuses instead on reflexives, claiming that “the distribution of the two perfect auxiliaries with reflexive verbs in Old Spanish supports a semantic analysis of split intransitivity, and gives evidence against a syntactic analysis” (Aranovich 2003: 29). In section 4 I present the basics of Perlmutter’s (1978) Unaccusative Hypothesis, and then review in section 5 some of the evidence from empirical domains distinct from auxiliation which show that unaccusativity-related effects require a syntactic representational contrast and cannot be reduced to the semantics. Given these premises, I return in section 6 to Aranovich’s analysis of Old Spanish and the role reflexives play within it. In section 7, I sketch my own account of perfective auxiliation, and show how the different parametric options encountered in auxiliary selection with reflexives across Romance build up to constitute the syntactic gradient. Section 8 brings three Medieval Romance varieties into the picture and shows how auxiliary selection, in reflexives and in general, can be accounted for by means of a parametric syntactic rule which elaborates on the syntactic gradient introduced in section 7. Section 9 presents the conclusions of the previous discussion, pointing to the necessity of a syntactic rule for auxiliary selection in Romance, capitalizing in turn on a syntactic unaccusative vs. unergative representational contrast.

Perfective auxiliation with reflexives in Medieval Romance

45

2 Semantic approaches to auxiliation: the problem with reflexives To exemplify the kind of claims on unaccusativity I will take issue with here, consider the following passage: se il fenomeno [dell’inaccusatività, M.L.] sia sintattico (rifletta cioè la relazione sintattica del soggetto in un livello non finale della rappresentazione), semantico (determinato dal ruolo tematico del soggetto e dall’aspetto lessicale) o sintattico-semantico (determinato semanticamente ma rappresentato sintatticamente […]). In questo lavoro seguiremo l’approccio semantico e sintattico-semantico. (Cennamo 1999: 303) [whether the phenomenon [of unaccusativity] is syntactic (and thus reflects the syntactic relation of the subject in a non-final level of representation), semantic (i.e. determined by the thematic role of the subject and by lexical aspect) or syntactic-semantic (i.e. semantically determined but represented syntactically […]). In this research, we shall follow the semantic and syntacticsemantic approach.]

Much of the research of type (3a), claiming that unaccusativity is syntactically determined and trying to derive the auxiliary selection facts from the semantics, focuses only on intransitive predicates, whereas both transitives and reflectives feature only marginally, if at all. Indeed, such approaches have to postulate separate rules for the auxiliary choice in reflexives, as is the case for Van Valin (1990) on Italian, who first proposes a general rule for auxiliary selection (4a), then later proposes a distinct one that accounts for reflexives (4b): (4) a. Auxiliary selection with intransitive verbs (Van Valin 1990: 233) “Select essere if the LS [= logical structure, M.L.] of the verb contains a state predicate”. b. Auxiliary selection for Italian verbs (Van Valin 1990: 256) “Select avere if the subject is an unmarked actor (with respect to the Actor-Undergoer Hierarchy […]), otherwise essere”. The same applies to Bentley’s (2006) account of French auxiliation: (5) Perfective operator selection in French (Bentley 2006: 72) Select avoir ‘have’ unless a. The construction is marked by se or b. The predicate is a telic intransitive.

46

Michele Loporcaro

The most explicit admission that semantically-based approaches have to postulate several quite disparate rules is to be found in Bentley and Eythórsson (2003: 468), who formulate the following bipartite rule: (6) Perfect formation rule in modern Romance (Bentley and Eythórsson 2003) (i) if V is [+pronominal] > ‘be’ + past participle (ii) a. if P is marked [+Fn] > ‘be’ + past participle b. elsewhere > ‘have’ + past participle Having claimed that “perfect formation involves two rules […] in modern Romance” (Bentley and Eythórsson 2003: 468) – one for reflexives (6i), the other for intransitives (6ii) – they have to posit nonetheless a third distinct rule (select ‘avere’ invariably) for transitives: “It should be stressed that the rule in (22) [i.e. (6) here, M.L.] regards auxiliary selection with all intransitives, but not with transitives” (Bentley and Eythórsson 2003: 461). Of these three rules, (6ii) is the only one which is sensitive to the semantics, since “{Fn} is a subset of {F} including the properties which are relevant for ‘be’ selection in a particular language […]: dynamicity, telicity, stativity” (Bentley and Eythórsson 2003: 460). As we shall see in section 4, a syntactic approach to auxiliation offers a much simpler account.

3 Old Spanish: first part One notable exception to the marginality of reflexives for studies on perfective auxiliary selection of type (3a) is Aranovich (2003). This is an often-quoted study on the gradual replacement of perfective auxiliary ser through haber in the history of Spanish, which does take reflexives into account and even argues that their behaviour with respect to auxiliation is crucial in order to disconfirm a syntactic approach to unaccusativity à la Perlmutter (3b). Aranovich shows that this gradual replacement was driven by semantic factors, since intransitives whose argument is a prototypical patient preserved auxiliary ser longer than other intransitive predicates (Aranovich’s 2003: 11 “Semantic Displacement Hypothesis”). From this gradual displacement, however, Aranovich (2003: 29, 31) concludes that one has to substitute a semantic explanation of Romance auxiliation for the syntactic explanation made available by Perlmutter’s Unaccusative Hypothesis. The selected data in (7) summarize the evidence which brings Aranovich to this conclusion:

Perfective auxiliation with reflexives in Medieval Romance

(7)

a.

time

intransitive verbs

holgar ‘idle about’ ser not later than the 14 th c. morir ‘die’ ser not later than the 17 th c.

b.

47

reflexive verbs

vengarse ‘take revenge’ ser not later than the 13 th c. arrepentirse ‘repent’ ser not later than the 17 th c.

As seen in (7a), the spread of haber affected first verbs like holgar ‘idle about’, agentive and non-telic, and only later affected verbs like morir ‘die’ (non-agentive and telic). In (7b), furthermore, it is shown that the same happened in pronominal verbs too. This, according to Aranovich (2003: 29), “supports a semantic analysis of split intransitivity, and gives evidence against a syntactic analysis”. The argument goes as follows. Considering that “reflexive verbs in Italian and French select the ‘be’ auxiliary is often offered as evidence for the unaccusative analysis of auxiliary selection (Rosen 1981 [1988]; Legendre 1989; Perlmutter 1989; Grimshaw 1990), under the assumption that all reflexive verbs have subjects which are also objects at an underlying level of representation”, Aranovich (2003: 29) feels that his findings disconfirm the Unaccusative Hypothesis for the following reason: “In Old Spanish […], aside from a handful of quasi-reflexive verbs [to be described in section 6, M.L.], I have found no instances of reflexive verbs with the auxiliary ser. The generalization that supports the unaccusative analysis in French and Italian […] does not hold for Spanish, robbing the syntactic analysis of split intransitivity of crucial evidence for Old Spanish”. The reason “[t]he evidence […] about quasi-reflexives selecting ser in Old Spanish also supports the semantic analysis of auxiliary selection” is that the loss of ser with reflexives obeyed the “semantic displacement hypothesis” just as well as intransitives, as shown in (7b): arrepentirse ‘repent’ is more patientive than vengarse ‘take revenge’, and therefore it keeps auxiliary ser longer. There are good reasons for discussing Aranovich’s account of Old Spanish here. Firstly, as I said, this is the only semantically-based treatment of Romance auxiliation that crucially capitalizes on the syntactic behaviour of reflexives, rather than focusing on intransitives only (perhaps devoting just an afterthought to reflexives, or a separate stipulation, as shown in 4–6 above). Secondly, among the studies of type (3a), this is one of the most explicit pleas for semantic reductionism, concluding with the bold claim that the “Old Spanish data may have consequences for a semantic analysis of cross-linguistic variation in split intransitivity, providing the blueprint of an argument to overcome Rosen’s objections against a semantic approach to split intransitivity in general” (Aranovich 2003: 31). Thirdly, it is an influential and much quoted

48

Michele Loporcaro

study among specialists of Spanish, no doubt also because it captures some genuine facts about the gradual replacement of ser through haber in the history not only of Spanish (perhaps not precisely in Old Spanish but slightly later, and in concurrence with other factors, if Rosemeyer 2012 is right), but of IberoRomance, more broadly, as confirmed by the study on Catalan by Batlle (2002). Finally, the argument has some general interest, since it shows that Aranovich’s flawed interpretation of his own data follows from his misrepresenting competing theories and ignoring the comparative Romance perspective and previous studies on other Romance languages. Thus, my refutation is also a case for the necessity of cross-theoretical debate on and a comparative approach to Romance auxiliation.

4 Perfective auxiliation and the Unaccusative Hypothesis The analysis Aranovich attacks is too well-known for me to present in detail here: it is the analysis of perfective auxiliation in Romance – like in other IndoEuropean languages – based on Perlmutter’s Unaccusative Hypothesis, whose basics are summarized in (8)–(10). Perlmutter (1978: 162) posits a bipartition of intransitive predicates, exemplified in (8a–b), with the few initial items from the lists of one-place intransitives provided there: (8) a. b.

unaccusative:

P{2}

burn, fall, drop, sink … (It. bruciare, cadere, affondare …)

unergative:

P{1}

work, play, speak, talk … (It. lavorare, giocare, parlare …)

As shown in the structural representations in (9a–b) – which adopt the formalism of Relational Grammar, the theory in which the Unaccusative Hypothesis was originally cast – the core argument is an underlying direct object (initial 2) for the subclass (8a), while it is an underlying (or initial) subject for the subclass (8b):1

1 Structural representations from (9a–b) follow Davies and Rosen (1988). Grammatical relations are expressed as follows: 1 = subject, 2 = direct object, P = predicate, Cho = chômeur (i.e., “the relation held by a nominal that has been ousted from term status”, Blake 1990: 2). Under Davies and Rosen’s “Predicate Union” theory, chômage also affects predicates, as apparent in (9a–b) where the initial (lexical) predicate is put en chômage by the auxiliary. This technicality is immaterial to our present discussion, which will not assume any familiarity with the framework and introduce all relevant technical notions explicitly.

Perfective auxiliation with reflexives in Medieval Romance

(9)

a.

2

P

1

P

1 P Cho La nave è affondata ‘The boat sank.’

b.

1 1

49

P P

Cho

Gianni ha lavorato ‘John worked.’

Based on the structural contrast in (9), Perlmutter (1989: 81) proposed a simple generalization which accounts for the distribution of auxiliaries essere/avere in all types of clauses in Italian: (10) Perfective auxiliation in Italian (Perlmutter 1989: 81) The perfective auxiliary is essere iff: the final 1 is a 2. Otherwise the perfective auxiliary is avere. This generalization also accounts for the selection of essere in reflexives (11), which was problematic for semanticist accounts, as seen in (3)–(6):2 (11)

1,2 1

P P

direct transitive reflexive

1 P Cho Maria si è lavata ‘Mary washed herself.’ The view of unaccusative and auxiliary selection summarized in (8)–(11) has several theoretical implications. The fundamental one is the one Aranovich would like to dismiss, i.e. the fact that the contrast between unaccusative and unergative predicates has to be expressed in syntactic terms, rather than be reduced to some semantic property. However the contrast is exactly encoded structurally – be it as shown in (9a–b) or some other way – one cannot dispose of it, unlike suggested e.g. by Dowty (1991: 612–613) and Van Valin and La Polla (1997: 257). The latter study

2 In Relational Grammar, reflexiveness is represented as multiattachment of the final subject (i.e. its bearing two grammatical relations) at some previous stratum (cf. Rosen 1981 [1988]). In the Romance languages, multiattachment must be resolved before the final stratum, since no nominal can display both subject and object surface properties at the same time (e.g. a subject precedes the finite verb in the unmarked word order and controls its agreement, while neither is true of a direct object).

50

Michele Loporcaro

is worth commenting on in this context, since it is a general typological study whose discussion of unaccusativity-related properties is doubly tied to our Romance topic: on the one hand, it provides inspiration for Bentley’s (2006) account of Romance auxiliation within the framework of Role and Reference Grammar, mentioned in section 2; on the other, the authors – like Aranovich – dismiss a syntactic view of unaccusativity, and they do so by invoking the results of previous semanticist studies of Italian auxiliary selection within Role and Reference Grammar.

5 On the syntactic nature of unaccusativity Van Valin and La Polla (1997: 255–260) discuss verb agreement in Acehnese (Malayo-Polynesian, Aceh, Sumatra; data from Durie 1985, 1987), which displays an intransitive split, as illustrated in (12): (12) a.

geu-jak (gopnyan) 3-go (3sg) ‘(S)he goes’

b. lôn rhët(-lôn) (1sg) fall(-1sg) ‘I fall’

Acehnese

This can be realized by comparing transitive clauses, exemplified in (13a–b), in which core arguments are expressed through independent and/or clitic pronouns: (13) a.

(gopnyan) geu-mat lôn (3sg) 3-hold 1sg ‘(S)he holds me’

b. (lôn) lôn-mat-geuh (1sg) 1sg-hold-3 ‘I hold him/her’

Acehnese

Among pronominal clitics, the subject is proclitic (geu [gɯ] in 13a, lôn [lon] in 13b), whereas the direct object is enclitic (geu [gɯh] in 13b). Much like in the Romance pro-drop languages, free subject pronouns may be omitted without affecting subject-marking on the predicate, since the proclitic subjective pronoun must be there anyway. On the other hand, just like in Romance, enclitic and free object pronouns are in complementary distribution. As for intransitive clauses, there is a split in core argument cross-referencing on the verb by means of clitic pronominals, since active predicates, whose core argument is semantically an agent (or has an actor macro-role in Role and Reference Grammar terms), get the same proclitic marking as transitive subjects, as shown by (12a) and (13a). Arguments of inactive intransitives, on the other hand (12b), are treated like transitive direct objects, since they cannot

Perfective auxiliation with reflexives in Medieval Romance

51

trigger pronominal proclitic marking on the predicate. This is exactly the kind of intransitive constructions for which Perlmutter’s Unaccusative Hypothesis proposed an underlying representation of the core argument as an initial direct object, assuming unaccusative 2→ 1 advancement. However, Van Valin and La Polla (1997: 257) dismiss such a syntactic analysis for the Acehnese facts: “In order to interpret these facts in terms of grammatical relations, it would be necessary to say that verbs like rhët ‘fall’ have only an object without a subject, or that there is a subject, but is really an object. This is possible, but complicates the theory unnecessarily. The most straightforward account is to say simply that there is one kind of cross-referencing for actors and another for undergoers” (Van Valin and La Polla 1997: 257). Thus, the argument against a syntactic analysis à la Perlmutter is that it would be less economical. By the same reasoning, in the case of Romance auxiliation, one is entitled to dismiss the semantic accounts discussed in section 2 above – including Van Valin’s (1990) – since they are less economical. Romance auxiliation is all the more relevant because in Van Valin and La Polla’s (1997: 662 fn. 3) discussion of Acehnese, the killer argument against a syntactic analysis – and, more broadly, for the dismissal of grammatical relations as a cross-linguistically valid construct – involves two Role and Reference Grammar papers on auxiliary selection in Italian: One could claim that there are multiple levels at which grammatical relations are represented and that at the abstract level the single argument with verbs like rhët is an object in a clause without a subject but that while it is a subject in the overt structure, the verb agrees with it as if it is still an object. This proposal, known as the ‘Unaccusative Hypothesis’, was originally put forward in Perlmutter (1978) and a number of other works; see Van Valin (1990) and Kishimoto (1996) for detailed critiques from the perspective of RRG (Van Valin and La Polla 1997: 662 fn. 3).

Thus, Romance auxiliary selection is, in the authors’ own view, a crucial empirical domain for assessing the correctness of the Unaccusative Hypothesis. By their own line of reasoning, if the Unaccusative Hypothesis was proven indispensable to account for Romance auxiliary selection, it may account for the Acehnese data as well (and for cross-linguistic generalizations on alignment phenomena in general). Consequently, I shall recapitulate in section 5.1 some of the Romance evidence (cf. e.g. Loporcaro 2008) – independent from auxiliary selection – which shows that a syntactic unaccusative vs. unergative contrast is needed and cannot be simply reduced to a semantic contrast.

52

Michele Loporcaro

5.1 Unaccusativity vs. telicity: the evidence from Italian participial clauses Participial clauses tend cross-linguistically to host only telic predicates, as observed for Romance in several studies: (14) Semantic constraint on participial clauses The verb occurring in a participial clause must be telic (e.g. Šabršula 1963; Bertinetto 1986: 266–267; Rosen 1987; Legendre 1987: 9; Van Valin 1990: 239; Hernanz 1991; Dini 1994; etc.) This semantic constraint explains a large set of facts, like the non-occurrence of dependent participles with argumental bare nouns, negation or predicates like possedere ‘possess’, as exemplified in (15) (cf. Dini 1994): (15) a. *Mangiato salame, Gianni riprese il cammino ‘Having eaten salami, Gianni set off again.’ b. *Cadute pietre, la strada è rimasta bloccata ‘(Some) rocks having fallen, the road was blocked.’ c. *Non mangiato (il) salame, Gianni riprese il cammino ‘Not having eaten salami, Gianni set off again.’ d. *Non arrivata Maria, la festa cominciò ‘Maria not having arrived, the party began.’ e. *Posseduta quella villa, Gianni si trasferì ‘Owned that villa, Gianni moved.’ Alongside this semantic constraint, however, it is necessary to recognize independent syntactic conditions. This becomes apparent if one compares participial absolutes (17, below) and dependents, viz. participial clauses with EQUI deletion (i.e., deletion of a subject coreferential with that of the matrix clause, in Perlmutter’s 1984: 308–311 terminology): (16) a. *?(Una volta) lavorato, i contadini non si fecero più vedere ‘Having worked, the farmers did not show up anymore.’ b. (Una volta) vendemmiato, i contadini non si fecero più vedere ‘Having harvested (grapes), the farmers did not show up anymore.’

dependent non-telic

Perfective auxiliation with reflexives in Medieval Romance

c. (Una volta) partiti, i contadini non si fecero più vedere Having left, the farmers did not show up anymore.’

53 telic

d. Raccolta l’uva, i contadini non si fecero più vedere ‘Having picked the grapes, the farmers did not show up anymore.’ In participial dependents, an unergative predicate may occur (16b) provided it is telic, like vendemmiare ‘harvest grapes’, which shows that the ungrammaticality of lavorare in (16a) has a semantic motivation, aptly expressed in the constraint (14).3 On the other hand, in participial absolutes (17), any unergative is ungrammatical, be it telic or non-telic (17a–b):

3 One anonymous reviewer observes that the telicity of vendemmiare “results from an inferable object of the activity (note that in the English gloss ‘harvest grapes’, it is indeed expressed): one can only harvest grapes as long as there are grapes to harvest. This suggests that if we accept that vendemmiare is telic, we also need to characterise it as semantically transitive. Thus, it would fall out of the theory”. The answer to this query is that I am of course aware that there is a long tradition of defining transitivity in semantic terms (e.g. Hopper and Thompson 1980) and/or as a prototypical notion, with reference to several distinct parameters (cf. e.g. Malchukov 2006; Næss 2007, and much other work in the functional-typological approach). Yet, I am following here a different line of research which, within RG, defines transitivity in purely syntactic terms, with reference to propositional strata, whereby a stratum is transitive if and only if it contains both a 1 and a 2 (Perlmutter 1989: 65, with modifications). In this perspective, there is no such thing as a distinction between semantic and syntactic transitivity (cf. the definition of initialization, in fn. 12 below). Note that if such a notion as “semantic transitivity” were adopted, it is unclear what could prevent one from claiming, for instance, that also lavorare is semantically transitive, and that this goes hand in hand with telicity, since “one can only work as long as there is work to be done”. Conversely, under a syntactic view of (in)transitivity, clear diagnostic criteria are available for ascertaining that both lavorare ‘work’ and vendemmiare ‘harvest grapes’ are syntactically intransitive in Italian, but differ in (semantic) telicity, as demonstrated by the standard “in x time” test: (i) a. I contadini hanno vendemmiato in tre ore ‘The farmers have harvested grapes in three hours.’ b. *I contadini hanno lavorato in tre ore ‘The farmers have worked in three hours.’ Note that as soon as lavorare is used within a (syntactically) transitive construction, a definite object telicizes it, as is usually the case with activities: (ii) Hanno lavorato tutto il legno disponibile in tre ore ‘They worked all the available wood in three hours.’

54

Michele Loporcaro

(17) a. *Lavorati/-o i contadini, il vigneto sembrava un campo di battaglia ‘The farmers having worked, the vineyard looked like a battle field.’

absolute

b. *Vendemmiati/-o i contadini, il vigneto sembrava un campo di battaglia ‘The farmers having harvested, the vineyard looked like a battle field.’ c. Partiti i contadini, il vigneto sembrava un campo di battaglia ‘The farmers having left, the vineyard looked like a battle field.’ d. Raccolta l’uva, il vigneto sembrava un campo di battaglia ‘The grapes having been picked, the vineyard looked like a battle field.’ Clearly, telicity alone cannot explain all the data here: it does explain why only telic predicates are fine in participial dependents (16b–d), but the behaviour of participial absolutes, where all unergatives are barred, regardless of their being (a)telic (17a–b), cannot be explained with telicity, nor with other semantic–pragmatic devices like e.g. the Immediate Cause Linking Rule vs. the Directed Change Linking Rule assumed by Rappaport Hovav and Levin (2000) to account for the unergative/unaccusative contrast. Rather, what is needed is a purely syntactic generalization, as formulated in (18) (slightly simplified, from Loporcaro 2003: 240): (18) Participial absolutes in Italian The verb in participial form is accompanied by a nominal which is (i) its P-initial 2 [excludes unergatives and auxiliaries] (ii) the final 1 of the participial [excludes non-passivized transitives] clause. This descriptive result is out of reach for accounts of type (3a) that reduce unaccusativity to telicity and/or non-agentivity.

Perfective auxiliation with reflexives in Medieval Romance

55

5.2 Telic unergatives: a puzzle for semantic approaches to perfective auxiliation Note that the very existence of verbs such as It. vendemmiare ‘harvest grapes’, unergative but telic, shows that reductionist approaches to unaccusativity cannot be right in claiming that “in Italian […] intransitive ‘have’-selecting predicates (‘unergatives’) are not telic” (Bentley and Eythórsson 2003: 461). Admittedly, this is true for most verb lexemes, but crucially not for all of them. There is a statistical correlation, since most syntactically unaccusative predicates are telic and most syntactically unergative predicates are atelic (the gray cells in 19 correspond to the “normal” case). But the fact that all four combinations occur shows that the unaccusative/unergative contrast cannot be reduced to semantics: (19)

semantics

syntax

unaccusative unergative

telic

atelic

morire ‘die’

rimanere ‘stay’

vendemmiare ‘harvest grapes’

lavorare ‘work’

Consider the behaviour of a predicate like Italian traslocare ‘move’, which seems to pose a problem for the Auxiliary Selection Hierarchy (1). This verb, in fact, denotes a telic change of location, and therefore should be a core unaccusative by (1). Indeed, traslocare may show unaccusative syntax, as seen in (20) by the occurrence of past participle agreement (compare the unaccusative participial clause in 16c above). (20) It. traslocare ‘move’: unaccusative syntax (Google search June 21, 2012) a. Quando, una volta traslocata, ho visto la struttura della casa e cioè che al primo piano è situata la clausura delle suore (dove noi non possiamo accedere) ‘When, once having moved, I (1 fsg) saw the structure of the house’ etc. http://www.cestanahoru.org/blog/wp-content/koinonia-eliska.pdf b.

tanto una volta traslocata lavavo sempre tutto! i bicchieri … ‘Anyway, once having moved, I (1 fsg) always washed all glasses’ http://forum.alfemminile.com/forum/f107/__f286319_f107Ot-bicchieri-e-trasloco.html

56

Michele Loporcaro

c.

Enel che al contrario una volta traslocata da Terni, Papigno è finita da anni all’Endesa, sta radicandosi a Perugia ‘Enel [a public service company] which, on the contrary, once having moved from Terni-Papigno ended up years ago at Endesa, is putting down roots in Perugia’ http://quindici.federutility.it/default.aspx?Y0=2006&N0=14

However, one also finds the same verb with unergative syntax, as shown in (21) by the lack of past participle agreement (una volta traslocato) with the plural (unrealized) argument of the participial clause: (21) It. traslocare ‘move’: unergative syntax (Google search June 21, 2012) a. anche E. mi ha detto che è uno degli acquisti che hanno già deciso di fare anche loro, una volta traslocato nella casa nuova ‘Also E. told me that this is one of the things they have already decided to buy, once they will have moved to their new home’ http://gitementali.wordpress.com/2011/09/01/un-simpaticoinsettone/ b.

una volta traslocato creare una tabella di orari per i primi mesi di lavori di casa, spese, etc. ‘Once having moved, create a timetable for the first months of work that the house needs doing on it, purchases, etc.’ http://www.stupidaggini.net/vado-a-vivere-da-solo

As for auxiliation, however, whenever this verb is used in the compound perfect, ‘have’ is selected invariably, as one would not expect, given (1), for a verb denoting a telic change of location:4

4 That a compound perfect is involved in (22a) is guaranteed by the punctual time adverbial ieri ‘yesterday’. This and other similar adverbials, as shown in (22b), are incompatible with ‘be’-selection. This qualification is necessary, since strings like è traslocata do indeed occur: (i) a chi serviranno questi spazi se la moda è già traslocata altrove ‘Who will need all those spaces if fashion has already moved somewhere else?’ (http://ricerca.repubblica.it/repubblica/archivio/repubblica/2007/11/29/zona-tortonaavra-il-marchiodoc-la.html) However, these are examples of stative-resultative constructions, as shown by their co-occurrence with the adverbial già ‘already’.

Perfective auxiliation with reflexives in Medieval Romance

(22) a.

b.

57

la donna che ha traslocato ieri ‘The woman who has moved yesterday.’ (Google search November 27, 2012) no examples of: *è traslocato/-a ieri, *siamo traslocati/-e ieri/un anno fa etc. ‘(He) is moved/(she) is moved yesterday/(we.m/f) are moved yesterday/one year ago etc.’

I do not doubt that this exception can be accommodated somehow, for instance by assuming that telicity for telic change of location verbs such as arrivare ‘arrive’ is “deeper” (in some structural sense) than for traslocare ‘move’.5 But any such account, however elegant, would miss an interesting generalization: not only are unergative telic verbs like traslocare or vendemmiare less numerous, but they also are less likely to resist diachronic change, as shown by the fact that (late) Latin vindemiare ‘harvest grapes’ survives into Italian but not into any other Romance language. Under the view defended here, traslocare or vendemmiare are telic on a par with core unaccusatives like cadere ‘fall’ or arrivare ‘arrive’, but given the correlation in (19), their non-corresponding to the statistically prevailing type represents an adaptive disadvantage, and explains their fading as synthetic verb lexemes in most Romance languages. More generally, although the Auxiliary Selection Hierarchy-based line of research has revealed interesting regularities, it is hard to escape the impression that it has grossly overstated the case for gradience in auxiliary selection, on occasion even by inventing gradualness where it does not exist, at the cost of forcing empirical data into a Procrustean bed. Two such examples are the following: (23) a.

b.

ne hanno venuto tre ‘Three of them have come’ (Conti Jiménez 2005: 1091, 1096, quoting Bentley 2004) sono/?ho rimasto solo ‘I remained alone’ (Legendre and Sorace 2003: 196)

Actually, the selection of ‘have’ in both (23a–b) is utterly ungrammatical in standard Italian, to which the quoted sources ascribe those examples (albeit

5 For instance, something along these lines would be possible in Mateu’s (2009) semantic theory of argument structure, elaborating on Hale and Keyser (2002).

58

Michele Loporcaro

with a question mark, in the second case): needless to say, they can occur in Italo-Romance dialects (or regional substandard varieties of Italian) in which e.g. all unaccusatives select ‘have’ or in which free variation occurs, but of course this is entirely orthogonal to the semantics, whereas Legendre and Sorace (2003: 196) argue that, in standard Italian, ho rimasto solo is predicted to be less ungrammatical than, say, *ho andato ‘I came’ precisely on account of its semantics.6 This is plainly false: any native speaker of standard Italian will confirm that *ho rimasto solo is just as ungrammatical as *ho andato. A further example of “invented” semantically-driven gradualness in auxiliary selection, concerning Old Neapolitan, is discussed in section 8.1.

6 Old Spanish: second part That said, let me revert to Old Spanish auxiliation to see why Aranovich’s strong claims cannot be correct. In what follows, I shall prove him wrong by reanalysing his own data and showing that a syntactic generalization is not only possible but indeed necessary in order to make sense of them. Reflexives, as recalled in section 3, are claimed to provide conclusive evidence against an unaccusativity-based account of perfective auxiliation. More precisely, what would crucially disconfirm such a syntactic account is the fact that auxiliary ser was selected in Old Spanish – not in all reflexive constructions but rather only in those Aranovich calls quasi-reflexives, as exemplified in (24): (24) a.

b.

A Maimino, que se era alçado con tierra de oriente ‘To Maimino, who had rebelled with the Eastern lands’ (Primera crónica general, 13 th c.) Estonçe Rruy Diaz apriessa se fue levantado ‘And then Ruy Diaz had risen in a hurry’ (Mocedades de Rodrigo, 14 th c.)

In all other reflexive (and/or reciprocal) constructions, ser has not occurred since the earliest Spanish texts:7 6 Yet this piece of fake Italian data circulates widely in the literature proposing semantic accounts of Romance auxiliary selection, including e.g. Mateu and Masanell (this volume, 1c). 7 Note that (25b) is actually an instance of reciprocal, not reflexive, construction. Consider, however, that in all Romance clitic se/si-constructions the contrast between reflexives and reciprocals is purely interpretive. As for morphosyntactic properties, including auxiliary selection, they all pattern the same, as shown by the complementary examples in (ia–b):

Perfective auxiliation with reflexives in Medieval Romance

(25) a.

b.

59

y algunos se han ahorcado de desesperados ‘And some of them have hanged themselves out of despair’ (Alonso de Zorita, Relación de los Señores de la Nueva España, 16 th c.) la palabra que entrambos a dos se habian dado ‘The word that each had given to the other’ (Cervantes, La española inglesa, 17 th c.)

The objection is obvious: (24) and (25) are instances of different syntactic constructions. In fact, Aranovich’s (2003: 16) characterization of “syntactic analyses” – ascribing them to “the assumption that all reflexive verbs have the same syntactic representation” and therefore “will show a uniform behaviour with respect to auxiliary selection” – is a misrepresentation that stretches credibility to the limit. Since at least Rosen (1981 [1988]) (referenced in Aranovich’s bibliography), it is clear that reflexive constructions all share the property of having two grammatical relations (later, further constrained as 1,2, since La Fauci 1988) borne by the final subject in one and the same stratum of propositional structure, which by no means implies that all reflexive constructions “have the same syntactic representation”. Indeed, a host of studies (including e.g. La Fauci 1988; Loporcaro 1998: 45–48, 1999, 2001, etc.) have shown that different structural subclasses of reflexive predicates can and must be distinguished by assuming distinct syntactic representations, lest one miss a number of relevant generalizations. Thus, the fact that ser occurs only in (24), not in (25), can be explained by invoking a syntactic condition, pace Aranovich. To see how, it is necessary to introduce another kind of gradualness – syntactic rather than semantic in nature.

(i) a. no se hubieran destruido los unos a los otros ‘They would not have destroyed each other’ (Alonso de Zorita, Relación de los Señores de la Nueva España, 16 th c.) b. y me he comido su pan ‘And I have eaten up his bread’ (Baltasar Gracián, El comulgatorio, 17 th c.) (ia) is a reciprocal (as opposed to the direct transitive reflexive in 25a), while (ib) is a dyadic reflexive (as opposed to the reciprocal in 25b). Yet, nothing changes as far as auxiliary selection is concerned.

60

Michele Loporcaro

7 The syntactic gradient In Loporcaro (2007, 2011), I have elaborated on the binary contrast in (9) above (unaccusative vs. unergative), showing that (9a–b) are actually two poles of a scale, starting from the observation that reflexive constructions sometimes pattern uniformly with unaccusatives (say, in standard Italian), but sometimes part ways, so that one subset goes with unaccusatives and the complementary one with unergatives. Alongside direct transitive reflexives (cf. 11 above), more types of reflexive constructions should be distinguished. The crucial one for my reanalysis of the Old Spanish data is that of retroherent unaccusatives (in Rosen’s 1981 [1988], 1982 terminology), corresponding to Aranovich’s “quasi-reflexives”:8 (26)

2 2,1 1 1 Maria Ruy Diaz

P P P P si è se fue

Cho alzata levantado

retroherent unaccusatives

‘Mary got up’ ‘Ruy Diaz had risen’ (24b)

There are then two structural subtypes of indirect reflexives: viz. (27a), monadic indirect reflexives (i.e. reflexive constructions which involve only one nominal core argument, like in unaccusatives 9a); and (27b), indirect transitive (dyadic) reflexives (i.e. reflexives in which two distinct nominals bear the subject and the direct object relations initially, like in plain transitive constructions): (27) a.

1,3 1,2 1

P P P

b.

1 P Cho Maria si è risposta ‘Mary answered herself’

1,3 1,2 1

P P P

2 Cho Cho

indirect reflexives

1 P Cho Cho Maria si è lavata le mani ‘Mary washed her hands’

Once these different classes of reflexives are distinguished, one can move on to classify their behaviour with respect to auxiliation (and other unaccusativi8 As seen in the representation in (26), Romance si/se-predicates of this kind are argumentally just like monovalent unaccusatives such as It. partire/Sp. partir ‘leave’. The occurrence of the reflexive clitic is determined by a purely syntactic process, i.e. the resolution of 2,1 multiattachment (cf. fn. 2) between the second and the third stratum.

61

Perfective auxiliation with reflexives in Medieval Romance

ty-related properties) across Romance. This has been done for past participle agreement in Loporcaro (1998) and for perfective auxiliation in Loporcaro (2007, 2011, 2014), resulting in an implicational scale (here exemplified with auxiliation) in which unaccusatives and unergatives occupy the two extremes:9 (28)

inactive unacc.

active reflexive retroherent

a. b. c. d. e. f.

Italian Log. Sardinian Old Romanesco Old Florentine Leccese Spanish

dir. trans.

E E E E E

indir. unerg.

indir. trans.

trans./ unerg.

H H H H H H

This is an implicational scale, since no Romance variety described so far seems to show options that are arranged discontinuously along it.10 For instance, standard Italian (28a) (like French) has auxiliary ‘be’ in plain unaccusatives and all reflexives, whereas Logudorese Sardinian (28b) minimally differs in selecting ‘have’ in dyadic reflexives, as exemplified in (29e): (29) a. maria ɛs Mary is

paltiːða left.fsg

Logudorese Sardinian unaccusative

b. maria z ɛst arrɛnɛɣaːda Mary refl is gotten-angry.fsg c. maria z ɛl bestiːða Mary refl is dressed.fsg d. maria z ɛr Mary refl is

rispɔsta answered.fsg

retroherent dir. trans. reflexive indir. unerg. refl. aux E

9 The headings on the first line (active/inactive) are motivated by the circumstance that, as already stressed by Perlmutter (1978), the unergative/unaccusative contrast corresponds to what in linguistic typology is called active/inactive alignment. 10 At least, this seems to hold in all systems described so far in which auxiliary selection is not subject to massive free variation, whereas in transitory systems in which variation is pervasive, violations of the implications in (28) can be observed, as discussed at the end of section 8.2 below.

62

Michele Loporcaro

aux H e. maria z a ssamunaːðu zal maːnɔs aux H indir. trans. refl. Mary refl has washed.msg the hands f. maria a Mary has

mmaniɣaːðu eaten.msg

(za minɛstra) (the soup)

unergative/ transitive

The scale is motivated deductively, on structural grounds (cf. fn. 13 below), but it is also intuitively plausible.11 Thus, among the constructions displaying a reflexive clitic on the predicate, retroherent unaccusatives (26) are closest to plain unaccusatives, given that both have just an argumental DO relation. Consequently, it is to be expected that if reflexives do not pattern all the same, retroherent unaccusatives will be most likely to behave like plain unaccusatives. Next come direct transitive reflexives (29c) where, as seen in (11) above, the final subject is also an argumental direct object, namely the one initialized by the transitive predicate.12 Thus, if reflexives differ with regard to some syntactic property and the rule is sensitive to initial objecthood, then plain transitive reflexives will pattern together with unaccusatives, while if it is sensitive to initial subjecthood, they will pattern together with the following constructions along the scale, viz. (29d–g). One step further, indirect unergative reflexives (29d) share with the preceding constructions the property of being monadic, given that just one noun phrase, the final subject, is initialized as argument by the predicate also in another grammatical relation, which, however, is not the direct object relation. In being monadic, they differ with respect to the next class (29e, dyadic reflexives), where in addition to the final subject there is a distinct nominal that bears the direct object relation argumentally. Thus, if some syntactic property defines two complementary subsets of reflexive constructions, one will expect dyadic reflexives to pattern rather like transitives (29f) with which they share the property of having a subject which is distinct from the initial direct object.13 11 It has also been adopted recently within other frameworks such as Role and Reference Grammar (Bentley 2010) and Minimalism (Ledgeway 2012: 321). 12 Initialization (Dubinsky 1985) is the attribution of grammatical relation and semantic role by a predicate to its argument(s). 13 As shown in Loporcaro (2011: 82), the options in (28a–e) can be formalized by stating different structural requirements on the agreement controller. In order of increasing restrictiveness, these are the following: (i) a. the final 1 is a 2 b. the final 1 is the only nuclear-term argument c. the final 1 is an initial 2

= (28a) = (28b) = (28c)

Perfective auxiliation with reflexives in Medieval Romance

63

This syntactic gradient is orthogonal to the semantic one in (1). It is in principle possible that both may be relevant to the synchronic description of Romance auxiliary selection systems as well as to the illustration of diachronic change. Which one of them is indeed relevant is an empirical issue and, as I will show, the empirical evidence demonstrates that (28) always is and that semantic constraints play a more marginal and subordinate role for auxiliary selection.14

8 Perfective auxiliation in Romance: diachronic variation along the syntactic gradient The scale in (28) has proven instrumental in classifying different auxiliation options found across space and time in the Romance languages and dialects, and Old Spanish is no exception. Actually, it closely resembles Old Florentine (28d), except that Old Spanish texts already show from the outset that the system is in a transitory state, which eventually results in (28f). In Old Florentine, as reflected in Dante’s texts, auxiliary avere is selected with all reflexives, including direct transitive ones (30c), but with the exception of retroherent unaccusatives (30b):15 (30) a. Fuggito è ogni augel che ’l caldo segue ‘All birds that follow the heat have fled’ (Rime c 27)

d. the final 1 is an initial 2, not a P-initial 1 e. the final 1 is an initial 2, not multiattached

unaccusative

= (28d) = (28e)

The notion “P-initial x” denotes that the nominal at issue bears the grammatical relation x in the first stratum of the P(redicate) sector of the predicate P. The P-sector is defined in turn as the set of strata in which a given predicate bears the P-relation (cf. Davies and Rosen 1988: 57). 14 One anonymous reviewer comments: “Unfortunately, this descriptive statement is not explicit enough”. He/She may have written this before reading the following section, where I show that in Old Florentine, Old Sicilian and Old Neapolitan the syntactic scale (29) is indispensable for modelling diachronic change in auxiliary selection, whereas no conclusive evidence for the relevance of the semantically-based hierarchy in (1) has been adduced yet. (Cf. in particular section 8.2 and fns. 15–16 for a rebuttal of putative evidence in support of such relevance for Old Neapolitan.) 15 Cf. La Fauci (2004) and Loporcaro (2004: 57–61) for a more detailed discussion of auxiliary selection in Old Florentine.

64

Michele Loporcaro

b. io mi sarei brusciato ‘I would have burnt myself’ (If xvi 49)

retroherent aux E

c. la donna che […] ci s’hae mostrata ‘The woman that showed herself to us’ (Vn xxxviii 3) d. Quand’io m’ebbi dintorno alquanto visto ‘After I looked around me for a while’ (If xxxii 40) e. poscia che tanti/speculi fatti s’ha ‘After he has made (for himself) so many mirrors’ (Pd xxix 143–144) f.

aux H direct trans. refl.

indirect unerg. refl.

indirect trans. refl.

Ma i Provenzai che fecer contra lui/non hanno riso (Pd vi 130–131) ‘But Provençals, who slandered him, did not laugh’

unergative

As shown in (26), this is the very same subclass of reflexives which Aranovich calls “quasi-reflexives”, i.e. the only one in which auxiliary ser occurs in Old Spanish (although already in competition with haber), as seen in (24) above. Now, as evident from what has been said in section 7, this is a syntactically defined class, which makes a syntactic analysis of Old Spanish auxiliation possible. This can be shown starting from thirteen-century Florentine. The rule accounting for the observed facts is simpler here, since Dante’s corpus represents a steady state with no significant variation. It can be stated as in (31) (from La Fauci 2004: 252, with the slight modifications proposed in Loporcaro 2011: 77): (31) Perfective auxiliation in Dante’s Florentine The perfective auxiliary is essere iff the final 1: (i) is a 2; (ii) is not a P-initial 1. Otherwise the perfective auxiliary is avere. The difference lies in the fact that the Old Spanish system, ever since the earliest texts, already appears to be undergoing a transition from stage (32a) to stage (32b), with haber fully generalized:

65

Perfective auxiliation with reflexives in Medieval Romance

(32) Perfective auxiliation in the diachrony of Castilian stage (a): Aux ser iff the final 1: > stage (b): Aux haber. (i) is a 2; (ii) is not a P-initial 1. Aux haber elsewhere. In this transition, semantic factors like those spotted by Aranovich did play a role, so that among both plain and retroherent unaccusatives, loss of auxiliary ser occurred earlier for those predicates that were from the outset semantically closer to the semantic type that prevails in unergatives (non-telic, with an agentive argument). But the semantic factors at play are clearly subordinate to the syntax, in that their scope is defined syntactically in the first place, as shown in (32a): in fact, all predicates that do not satisfy those syntactic conditions select only haber from the outset. In refuting Aranovich’s analysis in Loporcaro (2011: 79–81), I hinted at the fact that two other well-studied medieval Italo-Romance varieties actually provide a closer match to Old Spanish in that they too show variation in perfective auxiliation with unaccusatives. In what follows, I shall elaborate on this point and assess whether a) a syntactic rule can be formulated for those varieties and whether b) the predicate’s semantics plays a role in auxiliary choice.

8.1 Old Neapolitan Formentin (2001) has shown that in 14th- and 15th-century Neapolitan texts, plain unaccusatives overwhelmingly select essere (with very few exceptions: 1.8 %), whereas ‘have’ occurs in 45 % of retroherent unaccusatives (cf. the absolute figures in 33i) and is selected categorically elsewhere: only (33i), mirroring the medieval situation, reports absolute figures; (33ii–iii) concern two further diachronic stages, to be addressed at the end of this section. (33) Old Neapolitan a. b. c. d. e. f. g.

unaccusatives retroherent unacc. direct transitive refl. indirect unerg. refl. dyadic reflexives unergatives transitives

(i) 14th‒15th c.

(ii) 16th‒19th c.

(iii) 20th‒21st c.

E

H

E

E

552 17 3 0 2 0 0

10 14 19 2 49 61 all

E E (H) E (H) no data (E) H H H

H

H

E/H E/H E/H E/H H H H

66

Michele Loporcaro

Note that the figures in (33i) do not include, for unaccusatives, examples like the following (from a late 14th-century text), in which the verb is in the pluperfect subjunctive or past conditional: (34) a.

b.

si li Grieci no l’avessero andato in succurso ‘Had the Greek not come to his aid’ (De Blasi 1986: 72.30) nde averria insuto acqua in quantitate ‘Plenty of water would have gotten out of there’ (De Blasi 1986: 184.13)

The reason for this exclusion is that auxiliary choice with unaccusative verbs in just those tenses/moods has been shown to be in free variation (cf. Formentin 2001: 98–99; Ledgeway 2003: 609).16 This is relevant for my study, given that auxiliated verb forms in the pluperfect subjunctive or past conditional feature in most of the examples adduced by Cennamo (1999: 322–325) in order to substantiate her claim that in Old Neapolitan auxiliary ‘have’ spread to noncore unaccusatives in compliance with the Auxiliary Selection Hierarchy (1). This claim has been refuted conclusively by Formentin’s (2001: 98–99) systematic study of the Old Neapolitan corpus: la variazione tra essere e avere […] non è una caratteristica degli inaccusativi “centrali” […] l’alternanza dell’ausiliare che si osserva in averria socciesso […] e tutte quelle cose che erano soccese […] ha le stesse ragioni dell’alternanza che si osserva in si li Grieci no l’avessero andato in succurso e cortesemente lo ademandao perché era andato; e tali ragioni appaiono di rilievo squisitamente morfosintattico, non già lessicale o semantico (Formentin 2001: 98–99). [variation between ‘have’ and ‘be’ […] is not a characteristic of “central” unaccusatives […] The alternance in auxiliary that is observed in averria socciesso ‘it would have happened’ […] and tutte quelle cose che erano soccese ‘all those things that had happened’ […] has the same reasons as the alternance that is observed in (34a) [here, M.L.] and in cortesemente lo ademandao perché era andato ‘courteously did he ask him

16 Cennamo (2002: 198) refuses to take stock of this straight fact, claiming that “È difficile stabilire se avere si diffonda inizialmente in alcuni contesti modali e temporali (congiuntivo trapassato e condizionale passato) veicolanti la modalità irreale” [It is hard to establish whether ‘have’ spreads initially in some modal and temporal contexts (pluperfect subjunctive and past conditional) which vehiculate irrealis modality]. In fact, quite to the contrary, as Ledgeway (2009: 602) puts it, “Difficile sarebbe […] non riconoscere in tale sistematicità la spia di un modello di alternanze condizionate da considerazioni modali di tipo reale/irreale” [One cannot fail […] to acknowledge in such systematicity the unmistakable signs of a pattern of alternation conditioned by a realis/irrealis modal opposition]. Ledgeway (2009: 613) also shows that since the 16th century the modal constraint on auxiliary selection no longer seems to apply in Neapolitan.

Perfective auxiliation with reflexives in Medieval Romance

67

why he had gone’; and those reasons appear to be of a purely morphosyntactic, rather than lexical or semantic, nature.]

As it happens, Cennamo’s (1999) account manages to present Old Neapolitan data as though the Auxiliary Selection Hierarchy did constrain perfective auxiliation, only thanks to lack of control of a relevant variable (the modal constraint). As soon as one does control for it, it turns out that there are no examples of auxiliary ‘have’ with “central” unaccusatives in Old Neapolitan, which would support the claim.17 Note that the modal constraint on auxiliation Formentin (2001) revealed for Old Neapolitan receives comparative support from Old Spanish (Stolova 2006) and Old Sicilian (to be discussed in section 8.2, example 37).18 Having ascertained that the Auxiliary Selection Hierarchy (1) played no role in the spread of ‘have’ with plain unaccusatives in Old Neapolitan, one might still wonder whether the same was true of retroherent unaccusatives, where free variation is an indisputable fact (with 17 occurrences of ‘be’ vs. 14 of ‘have’ in 33ib). Adopting Aranovich’s (2003) grid and procedure, the following results are obtained: (35) Perfective auxiliary with Old Neapolitan retroherent predicates

(telic) change of location (Aranovich 2003: “directed motion”, e.g. salirse ‘go out’)

essere

avere

irese ‘go’, partirese ‘depart’, ponerese (in via) ‘hit the road’, sollevarese ‘rise’, tirarese ‘head’

adunarese ‘gather’, arroccharese (irr.) ‘take refuge in a fortress’, dellongarese ‘depart’, fugirese (irr.) ‘flee’, moverese (irr.) ‘move’, partirese (irr.) ‘depart’

17 Faulty readings of the textual evidence (e.g. transitive ànno partuto ‘they have shared’ interpreted erroneously as unaccusative ‘they have departed’) account for the rest of Cennamo’s putative evidence, as shown by Formentin (2001: 98–99 fn. 41) (cf. the discussion in Loporcaro 2011: 81 fn. 31). 18 One anonymous reviewer observes that “the paper is not explicit on why the “irrealis effect” favours a syntactic approach to auxiliary selection”. To clarify, let me emphasize that what I intend to show here is that lack of control for irrealis modality was instrumental in paving the way for (demonstrably wrong) claims to the relevance of the Auxiliary Selection Hierarchy (1) for the change in auxiliary selection in the history of Neapolitan. Once this independent factor is controlled for, it emerges clearly, as shown in (33i–iii), that both the synchronic distribution of auxiliaries in Old Neapolitan and diachronic change in the subsequent stages were constrained by the syntactic scale (28), not by the semantics.

68

Michele Loporcaro

change of state

adonarese ‘realize’

adonarese (irr.) ‘realize’, compirese ‘come true, be over’, farese (bella) ‘beautify oneself’, imbullarese ‘wrap oneself up’, spezarese ‘break’

continuation/ existence of state

trovarese ‘find oneself’

fissicarese ‘remain’ (irr.), romanirese ‘stay’, trovarese (irr.) ‘find oneself’

uncontrolled process

allegrarese ‘rejoice’, mutarese (irr.) ‘change’, reposarese ‘relax’, contentarese ‘content oneself with’, maravigliarese ‘be astonished’, penczarese ‘think’

non-motional process

accompagniarese, aconciarese1 ‘get set’, portarese ‘behave’, restrengerese ‘restrain oneself’

adastarese ‘hurry up’, delectarese ‘take delight of’, faticarese ‘get tired’, fidarese (irr.) ‘trust’, inclinarese ‘be inclined’ (irr.), solazarese ‘amuse oneself’

volitional achievement (Aranovich)

acconciarese2 ‘agree with’, chiavarese ‘lock oneself’, rebellarese ‘rebel’

inpaciarese ‘meddle in’, potrestarese ‘protest’

The abbreviation “irr(ealis)” means that ‘have’ occurs only in the pluperfect subjunctive and/or past conditional with that given verb. Once those cases are factored out, it is still clear that ‘have’ and ‘be’ are distributed over all semantic classes, including the two extremes of the Auxiliary Selection Hierarchy. One might argue, however, that the Neapolitan situation is more complex than the Castilian one, since its auxiliation system did not eventually evolve towards the generalization of ‘have’, as shown by the two subsequent dia-

Perfective auxiliation with reflexives in Medieval Romance

69

chronic stages in (33ii–iii).19 However, there is another well-investigated southern Italo-Romance variety which yields an even closer match to Old Spanish: Old Sicilian.

8.2 Old Sicilian The history of Sicilian is more similar to that of Spanish in that the modern dialect only uses ‘have’ as a perfective auxiliary with all kinds of predicates (e.g. tò suoru ha nisciutu ‘your sister has gone out’, Loporcaro 1998: 161). The change away from the Proto-Romance double auxiliary system was already near to completion by the time of the earliest extant texts, as shown by La Fauci’s (1992) study of a 14th- and 15th-century corpus. The results are summarized in (36), where quantification is added only to illustrate variation, but is omitted whenever auxiliary choice is categorical in the corpus: (36) a. b. c. d. e. f. g.

unaccusatives retroherent unacc. dir. trans. refl. indir. unerg. dyadic reflexives unergatives transitives

aux E

aux H

%H

319 13 never never never never never

57 72 always always always always always

15.1 84.7 100.0 100.0 100.0 100.0 100.0

Old Sicilian

(La Fauci 1992)

The situation closely parallels the Old Spanish one in that ‘have’ is selected categorically not only with transitives/unergatives (36f–g) but also with all reflexives except retroherent unaccusatives (36b). In the latter, as well as in plain unaccusatives (36a), free variation is observed instead, as shown by the following examples:

19 The data for 16 th‒17 th century Neapolitan in (33ii–iii) are drawn from Vecchio (2006) and Ledgeway (2009: 612), whose results diverge a bit. In Ledgeway’s (2009: 612–613) corpus, ‘have’ is categorical with dyadic reflexives (33e) and occurs in just a handful of cases with unaccusatives (33a), while free variation is found with retroherents and monadic reflexives (33b–c). Vecchio (2006: 70–75), on the other hand, reports a slightly fuzzier picture, with only scarse occurrences of ‘have’ in (33b–c) and of ‘be’ in (33e): this is why the corresponding abbreviations appear in brackets in some cells in (33ii). As for the modern stage, Ledgeway (2000: 228, 2009: 618–620) and Vecchio (2006: 75) report free variation in all unaccusatives and pronominal verb constructions, except for dyadic reflexives, where ‘have’ is selected categorically as in transitives and unergatives.

70

Michele Loporcaro

(37) a.

unu homu chi havia andatu a lu lectu ‘A man that had approached the bed’ (di Girolamo 1982: 48.10)

b.

lu quali era andatu ad visitari lu corpu di lu beatu Iheronimu ‘(The one) that had gone to visit the body of beatus Jerome’ (di Girolamo 1982: 59.2)

Also parallel to Old Spanish (cf. Stolova 2006) is the relevance of the modal constraint already seen at work in Old Neapolitan. This is shown by the fact that the pluperfect subjunctive and past conditional verb forms have a higher incidence in the subset of ‘have’-auxiliated examples (38i) than in the complementary one of ‘be’-auxiliated ones (38ii): (38) Old Sicilian

(i)

a. unaccusatives b. retroherent

aux E

irrealis (%)

319 13

19 (4.1%) 1 (7.7%)

(ii)

aux H

irrealis (%)

57 72

11 (19.3%) 10 (13.9%)

Finally, (39) shows (with just some selected examples) that in Old Sicilian, too, predicates located at all steps of Sorace’s Auxiliary Selection Hierarchy occur with both auxiliaries. Contrary to Neapolitan, I present the results for plain unaccusatives here:20

20 Retroherent unaccusatives offer much less ground for comparison, since there are only 13 occurrences of auxiliary ‘be’ in the corpus. Among these, however, one finds clear instances of verbs which are not “core unaccusatives” semantically, such as confessarisi ‘confess’ (relig.), or guardarisi ‘guard (against)’, as shown in the following examples (from La Fauci 1992: 66): (i) a. tostu ti sia confessatu ‘Do confess soon!’ (Cusimano 1951–52: w.70) b. di quilli di ki illu si est guardatu ‘Of them against whom he has guarded’ (Bruni 1973: 70.14) Conversely, auxiliary ‘have’ occurs also with “core unaccusatives” (verbs denoting telic change of location) such as tornari ‘return’: (ii) in lu primu locu undi si avia tornatu ‘In the first place where it had returned’ (di Girolamo 1982: 42.13)

Perfective auxiliation with reflexives in Medieval Romance

71

(39) Perfective auxiliary with Old Sicilian unaccusatives essiri

aviri

(telic) change of location

andari ‘go’, cadiri ‘fall’, cascari ‘fall’, dixindiri ‘go down’, exiri ‘get out’, fugiri ‘flee’, intrari ‘enter’, muntari ‘go up, rise’, naxiri ‘be born’, passari ‘pass through/by’, tornari ‘return’, veniri ‘come’

andari ‘go’, cadiri ‘fall’, cascari ‘fall’, xindiri ‘go down’, exiri ‘get out’, fugiri ‘flee’, intrari ‘enter’, muntari (irr.) ‘go up, rise’, naxiri ‘be born’, passari ‘pass through/by’, turnari ‘return’, veniri ‘come’

change of state

accadiri ‘happen’, adveniri ‘happen’, intraveniri ‘happen’, moriri ‘die’, resursitari ‘rise (from the dead)’, trapassari ‘pass away’

accadiri ‘happen’, adveniri ‘happen’, intraveniri ‘happen’, moriri ‘die’, resussitari ‘rise (from the dead)’, trapassari ‘pass away’

continuation/ existence of state uncontrolled process

campari ‘live’, plachiri ‘please’

campari ‘live’, plachiri ‘please’

crixiri ‘grow’

crixiri ‘grow’

To be sure, it cannot be excluded that finer-grained quantification (by number of occurrences for verb lexemes as well as by date of text) may show some semantically-driven skewing of the auxiliary data, and thus reveal some semantic effects on the spread of ‘have’ of the type detected by Aranovich for Old Spanish. What my quantifications certainly (and above all La Fauci’s 1992 study) do show, is that the gradual spread of aviri ‘have’ in Old Sicilian was sensitive to the syntactic gradient (28), illustrated above in section 7, the same that proved relevant to describe auxiliary selection in Old Spanish, Old Florentine and Old Neapolitan: in Old Sicilian, selection of ‘have’ is categorical in (36c–g), whereas plain unaccusatives in (36a) show only 15 % occurrences of ‘have’ and, in between, retroherent unaccusatives (36b) are already close (84.7 %) to being annexed to the province of categorical ‘have’ selection. Summing up, it is clear from the above that both the Old Neapolitan and the Old Sicilian data demonstrate the relevance of the syntactic scale in (28) for modelling diachronic change in auxiliary selection. In both cases, variation in auxiliary selection over transitional stages has been shown to be sensitive to the syntactic factors mirrored in (28).

72

Michele Loporcaro

As hinted in fn. 10 above, this need not be universally the case, since systems displaying free variation occasionally do reveal violations of the implications in (28). Thus, for instance, in the variety of Acadian French studied in Balcom (2005: 86–87), avoir/être occur variably with the same unaccusative verbs which select être in standard French, whilst with retroherent unaccusatives (s’étendre ‘spread’, se briser ‘break’, se promener ‘stroll’, etc.) her informants use être categorically.21 Consider however that – as Balcom herself observes – other studies on Acadian French report uniform selection of avoir with all unaccusative predicates, included those with clitic se (cf. Péronnet 1991; King and Nadasdi 2001). Note, on the other hand, that as for plain unaccusatives, Balcom (2005: 86) reports the highest rates of selection of être with arriver ‘arrive’ (85 %) and mourir ‘die’ (52 %), whereas other change of location and change of state verbs display much lower rates, thus showing that the subclasses in Sorace’s hierarchy (1) are not crucial for the description of this variable behaviour. Consider also that with rester ‘stay’ (continuation of state), être is selected in 46 % of the occurrences, thus unexpectedly (given that hierarchy) outranking change of location verbs like descendre ‘go down’ (20 %), monter ‘climb’ (18 %) or tomber ‘fall’ (36 %). Obviously, verb semantics is only a partial predictor of variation in auxiliary selection in this variety.

9 Conclusion Concluding one of the most influential semanticist reanalyses of unaccusativity, Dowty (1991: 612–613) wrote: “To be sure, the extensively-argued advancement analyses, such as those of Rosen 1984, Burzio 1986, and others for Italian demand to be answered in detail – either to argue that each unaccusativity phenomenon is semantic or to provide a plausible monostratal alternative for any grammatical unaccusativity – to follow through on this suggestion”. The suggestion was that unaccusativity phenomena can – and indeed should – be explained “without invoking ‘grammatical object’ in the description of intransitives at all” (Dowty 1991: 612), i.e. dispensing with the Unaccusative Hypothesis. Two decades later, the situation has not changed substantially. Despite the bulk of work which has pursued this goal, some of which I have discussed

21 This can be formalized, within the framework adopted here, by assuming a condition [+multiattached] (cf. fn. 2) to be still enforced categorically in the auxiliary selection rule for this variety, whereas the basic condition seen in (10) above (final 1 is a 2) is only optionally enforced.

Perfective auxiliation with reflexives in Medieval Romance

73

here, semanticist reanalyses of Romance perfective auxiliation cannot match the descriptive economy and empirical accuracy of those proposed in the line of research inaugurated by Perlmutter’s (1978, 1989) seminal work on the Unaccusative Hypothesis. In fairness, one must recognize that semantic factors have been shown to play a role in auxiliary selection in both synchrony and diachrony. Synchronically, they play a role in the way the predicate valency of an intransitive verb is defined in the lexicon: thus, if e.g. verbs of emission like It. squillare ‘ring’ show variable auxiliation, this must mean that the lexical entry for such verbs provides for both an unergative and an unaccusative syntactic construction.22 That this kind of variation does not occur with telic cadere ‘fall’ is predicted by the Auxiliary Selection Hierarchy in (1), and the contrast between such classes of verbs represents a bona fide illustration of the relevance of semantic factors. Diachronically, semantic factors have been shown to play a role in transitory stages such as those studied by Aranovich for Old and Classical Spanish.23 However, this is not to say that those factors must either necessarily play a role in diachronic change or, worse, may “giv[e] evidence against a syntactic analysis” (Aranovich 2003: 29). The diachronic evidence reviewed here, in fact, shows that semantic factors have been demonstrated to be irrelevant for perfective auxiliation in Old Neapolitan (section 8.1) and that for Old Sicilian the relevance of the syntactic auxiliary selection gradient is an ascertained fact (section 8.2), while sensitivity to the semantics does not seem to be observed. The foregoing discussion has also shown that claims to the superordinate, or even exclusive, role of semantic conditions on auxiliary selection have been based on a mix of the following ingredients: a) blatant misrepresentation of

22 Sorace (2000: 833), discussing analyses of this kind, claims that this “burdens the lexicon with a proliferation of multiple entries that make it unlearnable”. I am not aware of any conclusive demonstration (computational or other) that this should indeed be the case. In recent work in theoretical syntax, several different formalizations of this kind of unaccusative/unergative alternations have been proposed, all of them entailing a complication of both syntactic structure and the lexicon. Cf. e.g., most recently, Acedo-Matellán and Mateu (2013: 231–233) and the earlier references quoted there, who analyse John danced into the room as an instance of “syntactic compounding of a manner root √dance with a phonologically null event (i.e. GO)”, also termed a “phonologically null unaccusative verb”. 23 A similar case has been made for Romanian by Dragomirescu and Nicolae (2009: 19), who see a last remnant of a binary perfective auxiliation rule in the occurrence of a fi ‘be’ with the participle of unaccusative verbs denoting change of location/state to form a periphrasis which is usually interpreted as a stative resultative periphrasis involving a copula (cf. e.g. Sandfeld and Olsen 1936: 316, 23; Avram 1994: 509; Loporcaro et al. 2004: 23) but which they analyse as a compound (resultative) perfect.

74

Michele Loporcaro

syntactic unaccusativity-based analyses, as exemplified by Aranovich’s (2003: 16) misleading statement that, under syntactic analyses, “all reflexive verbs have the same syntactic representation” (section 6); b) ignorance of the relevant literature, as exemplified by Cennamo’s (2008: 126) assertion that auxiliary selection in “early Italian vernaculars […] still awaits investigation”; c) disregard for selected portions of the available empirical evidence, as exemplified by Bentley and Eythórsson’s (2003: 461) statement that “In Italian […] intransitive ‘have’-selecting predicates (‘unergatives’) are not telic” (section 5.1); and, last but not least, d) primary data-twisting, as exemplified by Legendre and Sorace’s (2003: 196) contention that (ungrammatical) *ho rimasto solo is marginally acceptable in standard Italian (section 5.2). An unbiased analytical procedure which is exempt from the above flaws must lead to the conclusion that a purely syntactic unaccusative vs. unergative contrast, and the syntactic gradient based upon it in (28), play a decisive role in perfective auxiliary selection in most documented stages of most Romance languages.

References Acedo-Matellán, Víctor and Jaume Mateu. 2013 Satellite-framed Latin vs. verb-framed Romance: A syntactic approach. Probus 25: 227–265. Aranovich, Raúl. 2003 The semantics of auxiliary selection in Old Spanish. Studies in Language 27: 1–37. Avram, Larisa. 1994 Auxiliary configurations in English and Romanian. Revue roumaine de linguistique 5–6: 493–510. Balcom, Patricia. 2005 Le paramètre de l’inaccusativité chez les Acadiennes et les Acadiens du Nouveau-Brunswick. In: Patrice Brasseur and Anika Falkert (eds.), Français d’Amérique: approches morphosyntaxiques, 83–93. Paris: l’Harmattan. Batlle, Mar. 2002 L’expressió dels temps compostos en la veu mitjana i la passiva pronominal. El procés de substitució de l’auxiliar ‘ésser’ per ‘haver’. Barcelona: Institut d’Estudis Catalans i Publicacions de l’Abadia de Montserrat. Bentley, Delia. 2004 ‘Ne’-cliticisation and split-intransitivity. Journal of Linguistics 40: 219– 262. Bentley, Delia. 2006 Split Intransitivity in Italian. Berlin, New York: De Gruyter. Bentley, Delia. 2010 Principles of subject markedness in Romance. Archivio glottologico italiano 95: 152–189. Bentley, Delia and Thórhallur Eythórsson. 2003 Auxiliary selection and the semantics of unaccusativity. Lingua 114: 447–471. Bertinetto, Pier Marco. 1986 Tempo, aspetto e azione nel verbo italiano. Il sistema dell’indicativo. Florence: Accademia della Crusca. Blake, Barry J. 1990 Relational Grammar. London: Routledge. Bruni, Francesco (ed.). 1973 Libru di li vitii et di li virtuti. 3 vols. Palermo: Centro di studi filologici e linguistici siciliani.

Perfective auxiliation with reflexives in Medieval Romance

75

Burzio, Luigi. 1986 Italian Syntax. Dordrecht: Reidel. Cennamo, Michela. 1999 Inaccusatività tardo-latina e suoi riflessi in testi italiani antichi centro-meridionali. Zeitschrift für romanische Philologie 115: 300–331. Cennamo, Michela. 2002 La selezione degli ausiliari perfettivi in napoletano antico: fenomeno sintattico o sintattico-semantico. Archivio glottologico italiano 87: 175–222. Cennamo, Michela. 2008 The rise and development of analytic perfects in Italo-Romance. In: Thórhallur Eythórsson (ed.), Grammatical Change and Linguistic Theory: The Rosendal Papers, 115–142. Amsterdam, Philadelphia: John Benjamins. Conti Jiménez, Carmen. 2005 Existentials sentences with preposed locative phrases and postverbal determinerless subjects in Spanish. Linguistics 44: 1079–1104. Cusimano, Giuseppe (ed.). 1951–52 Poesie siciliane dei secoli XIV e XV. 2 vols. Palermo: Centro di studi filologici e linguistici siciliani. Davies, William and Carol Rosen. 1988 Unions as multi-predicate clauses. Language 64: 52–88. De Blasi, Nicola (ed.). 1986 “Libro de la destructione de Troya”. Volgarizzamento napoletano trecentesco da Guido delle Colonne. Rome: Bonacci. di Girolamo, Costanzo (ed.). 1982 Libru di lu transitu et vita di misser sanctu Iheronimu. Palermo: Centro di studi filologici e linguistici siciliani. Dini, Luca. 1994 Aspectual constraints on Italian absolute phrases. Quaderni del Laboratorio di Linguistica della Scuola Normale Superiore di Pisa 8: 52–87. Dowty, David. 1991 Thematic proto-roles and argument selection. Language 67: 547–619. Dragomirescu, Adina and Alexandru Nicolae. 2009 Relics of auxiliary selection in Romanian. Paper given at Going Romance 2009, Nice, December 3–5, 2009. Dubinsky, Stanley. 1985 Union constructions in Japanese: a unified analysis of -sase and -rare. Ph.D. dissertation, Cornell University. Durie, Mark. 1985 A Grammar of Acehnese on the Basis of a Dialect of North Aceh. Dordrecht, Cinnaminson: Foris. Durie, Mark. 1987 Grammatical relations in Acehnese. Studies in Language 11: 365–399. Formentin, Vittorio. 2001 L’ausiliazione perfettiva in antico napoletano. Archivio glottologico italiano 86: 79–117. Grimshaw, Jane. 1990 Argument Structure. Cambridge, M.A.: The MIT Press. Hale, Ken and Samuel Jay Keyser. 2002 Prolegomenon to a Theory of Argument Structure. Cambridge, M.A.: MIT Press. Hernanz, M. Lluïsa. 1991 Spanish absolute constructions and aspect. Catalan Working Papers in Linguistics 1: 75–128. Hopper, Paul J. and Sandra A. Thompson. 1980 Transitivity in grammar and discourse. Language 56: 251–299. King, Ruth and Terry Nadasdi. 2005 Deux auxiliaires qui voulaient mourir en français acadien. In: Patrice Brasseur and Anika Falkert (eds.), Français d’Amérique: approches morphosyntaxiques, 103–111. Paris: l’Harmattan. Kishimoto, Hideki. 1996 Split intransitivity in Japanese and the unaccusative hypothesis. Language 72: 248–286. La Fauci, Nunzio. 1988 Oggetti e soggetti nella formazione della morfosintassi romanza. Pisa: Giardini. [English translation 1994 Objects and Subjects in the Formation of Romance Morphosyntax. Bloomington, Indiana: IULC.] La Fauci, Nunzio. 1992 Capitoli di morfosintassi siciliana antica: tassonomia dei costrutti medi e ausiliari perfettivi. In: Studi linguistici e filologici offerti a Girolamo Caracausi, 185–220. Palermo: CSFLS. [Also in La Fauci (2000: 41–73)].

76

Michele Loporcaro

La Fauci, Nunzio. 2000 Forme romanze della funzione predicativa. Pisa: ETS. La Fauci, Nunzio. 2004 Armonia differenziale dell’ausiliazione perfettiva nel volgare di Dante. In: Maurizio Dardano and Gianluca Frenguelli (eds.), SintAnt. La sintassi dell’italiano antico. Atti del Convegno internazionale di studi (Università “Roma Tre”, 18–21 settembre 2002), 237–252. Rome: Aracne. Ledgeway, Adam. 2000 A Comparative Syntax of the Dialects of Southern Italy: a Minimalist Approach. Oxford, Boston: Blackwell. Ledgeway, Adam. 2003 L’estensione dell’ausiliare perfettivo avere nell’antico napoletano: intransitività scissa condizionata da fattori modali. Archivio glottologico italiano 88: 29–71. Ledgeway, Adam. 2009 Grammatica diacronica del napoletano (Beihefte zur Zeitschrift für romanische Philologie 350). Tübingen: Niemeyer. Ledgeway, Adam. 2012 From Latin to Romance. Morphosyntactic Typology and Change. Oxford: Oxford University Press. Legendre, Géraldine. 1987 Topics in French syntax. Ph.D. dissertation, University of California, San Diego. Legendre, Géraldine. 1989 Unaccusativity in French. Lingua 79: 95–164. Legendre, Géraldine and Antonella Sorace. 2003 Auxiliaires et intransitivité en français et dans les langues romanes. In: Danièle Godard (ed.), Les langues romanes. Problèmes de la phrase simple, 185–233. Paris: Editions du CNRS. Loporcaro, Michele. 1998 Sintassi comparata dell’accordo participiale romanzo. Turin: Rosenberg & Sellier. Loporcaro, Michele. 1999 L’ausiliazione perfettiva nelle parlate di Zagarolo e di Colonna e lo studio della sintassi dei dialetti mediani. Contributi di Filologia dell’Italia Mediana 13: 203–226. Loporcaro, Michele. 2001 La selezione dell’ausiliare nei dialetti italiani: dati e teorie. In: Federico Albano Leoni, Rosanna Sornicola, Eleonora Stenta Krosbakken and Carolina Stromboli (eds.), Dati empirici e teorie linguistiche. Atti del XXXIII Congresso Internazionale di Studi della Società di Linguistica Italiana (SIL XXXIII). Napoli, 28–30 ottobre 1999, 455–476. Rome: Bulzoni. Loporcaro, Michele. 2003 The Unaccusative Hypothesis and participial absolutes in Italian. Perlmutter’s generalization revised. Rivista di Linguistica/Italian Journal of Linguistics 15: 199–263. Loporcaro, Michele. 2007 On triple auxiliation in Romance. Linguistics 45: 173–222. Loporcaro, Michele. 2008 L’allineamento attivo-inattivo e il rapporto fra lessico e morfosintassi. In: Emanuela Cresti (ed.), Prospettive nello studio del lessico italiano. Atti del IX Congresso della Società Internazionale di Linguistica e Filologia Italiana (SILFI 2006). Firenze, 14–17 giugno 2006. Volume 1, 311–320. Florence: Firenze University Press. Loporcaro, Michele. 2011 A euroversal in a global perspective: auxiliation and alignment. In: Peter Siemund (ed.), Linguistic Universals and Language Variation, 55–91. Berlin, New York: De Gruyter. Loporcaro, Michele. 2014 Perfective auxiliation in Italo-Romance: the complementarity of historical and modern cross-dialectal evidence. In: Adam Ledgeway, Paola Benincà and Nigel Vincent (eds.), Diachrony and Dialects: Grammatical Change in the Dialects of Italy, 48–70. Oxford: Oxford University Press. Loporcaro, Michele, Lorenza Pescia and Maria Ana Ramos. 2004 Costrutti dipendenti participiali e participi doppi in portoghese. Revue de linguistique romane 78: 15–46.

Perfective auxiliation with reflexives in Medieval Romance

77

Malchukov, Andrej. 2006 Transitivity parameters and transitivity alternations: constraining co-variation. In: Leonid Kulikov, Andrej Malchukov and Peter De Swart (eds.), Case, Valency and Transitivity, 175–190. Amsterdam, Philadelphia: John Benjamins. Mateu, Jaume. 2009 Gradience and auxiliary selection in Old Catalan and Old Spanish. In: Paola Crisma and Giuseppe Longobardi (eds.), Historical Syntax and Linguistic Theory, 176–193. Oxford: Oxford University Press. Mateu, Jaume and Mar Masanell. (this volume) A constructional approach to auxiliary selection: evidence from existential constructions. Næss, Åshild. 2007 Prototypical Transitivity. Amsterdam, Philadelphia: John Benjamins. Perlmutter, David M. 1978 Impersonal passives and the Unaccusative Hypothesis. Berkeley Linguistics Society 4: 157–189. Perlmutter, David M. 1984 Working 1s and inversion in Italian, Japanese and Quechua. In: David M. Perlmutter and Carol Rosen (eds.), Studies in Relational Grammar 2, 292–330. Chicago, London: Chicago University Press. Perlmutter, David M. 1989 Multiattachment and the Unaccusative Hypothesis: The perfect auxiliary in Italian. Probus 1: 63–119. Péronnet, Louise. 1991 Système des modalités verbales dans le parler acadien du sudest du Nouveau-Brunswick. Journal of the Atlantic Provinces Linguistic Association 13: 85– 98. Rappaport Hovav, Malka and Beth Levin. 2000 Classifying single argument verbs. In: Peter Coopmans, Martin Everaert and Jane Grimshaw (eds.), Lexical Specification and Insertion, 269–304. Amsterdam, Philadelphia: John Benjamins. Rosemeyer, Malte. 2012 How to measure replacement: auxiliary selection in Old Spanish bibles. Folia Linguistica Historica. 33(1): 135–174. Rosen, Carol. 1981 [1988] The relational structure of reflexive clauses. New York: Garland [Ph.D. dissertation, Harvard University 1981]. Rosen, Carol. 1982 The unaccusative hypothesis and the ‘inherent clitic’ phenomenon in Italian. Chicago Linguistics Society 18: 530–541. Rosen, Carol. 1984 The interface between semantic roles and initial grammatical relations. In: David M. Perlmutter and Carol Rosen (eds.), Studies in Relational Grammar 2, 38–77. Chicago, London: Chicago University Press. Rosen, Carol. 1987 Star means bad: a syntactic divertimento for Italianists. Italica 64: 443– 476. Šabršula, Jan. 1963 La signification des verbes français et les problèmes d’aspect (étude comparative: langue française et langues slaves). Beiträge zur Romanischen Philologie 1: 166–179. Sandfeld, Kristian and Hedwig Olsen. 1936 Syntaxe roumaine. vol. 1. Emploi des mots et flexion. Paris: Droz. Sorace, Antonella. 1992 Lexical conditions on syntactic knowledge: Auxiliary selection in native and non-native grammars of Italian. Ph.D. dissertation, University of Edinburgh. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76: 859–890. Stolova, Natalya I. 2006 Split intransitivity in Old Spanish: irrealis and negation factors. Revue roumaine de linguistique 51: 301–320. Van Valin, Robert. 1990 Semantic parameters of split intransitivity. Language 66: 221–260. Van Valin, Robert and Randy La Polla. 1997 Syntax. Cambridge: Cambridge University Press. Vecchio, Paola. 2006 L’ausiliazione perfettiva in napoletano. Studio di sintassi diacronica. Revue de linguistique romane 70: 53–94.

Pierre-Don Giancarli, Université de Poitiers

Auxiliary selection with intransitive and reflexive verbs: the limits of gradience and scalarity, 1 followed by a proposal 1 Introduction The Auxiliary Selection Hierarchy (ASH, Sorace 2000, also called Split Intransitivity Hierarchy, Sorace 2006, 2011) is today’s most elaborate gradient representation dedicated to the auxiliary selection of BE or HAVE2 within intransitive verbs. Nevertheless, I wish to emphasise the limits of gradience before proposing a binary and therefore non-gradient and even non-scalar representation of auxiliaries including both intransitive and reflexive verbs. This is not the first binary attempt in the literature: according to the Unaccusative Hypothesis (UH), unaccusatives select BE and unergatives select HAVE.3 The UH was initially proposed in Relational Grammar in Perlmutter (1978) and was adapted into the Government-Binding theory elaborated by Burzio (1981, 1986). According to the UH there are two types of intransitives that differ syntactically: the single argument (and surface subject) of an unergative is an external argument equivalent to the subject of a transitive verb, while the single argument (and surface subject) of an unaccusative is an internal argument equivalent to the direct object of a transitive verb promoted to subject position. It also claims that the distinction is systematically related to the semantic characteristics of these verbs: unergativity correlates with agentivity and unaccusativity correlates with patienthood (Perlmutter 1978; Van Valin 1990). However, the UH had several flaws – for example, mismatches across unaccusativity diagnostic tests, difficulty in explaining the differences between the patienthood of the subject of an intransitive inaccusative and that of the subject of a transitive verb in the passive, inconsistencies in the alignment between the syntactic and semantic properties of split intransitivity, some

1 I am grateful to the two anonymous reviewers for their comments. Any error or inaccuracy is of course my own. 2 Our spelling in upper case letters is meant to abstractly cover all the forms that auxiliaries can take in the various languages examined here, for example HAVE stands for FF avoir, Acadian aouèr and Corsican avè, all written in lower case letters. 3 At least in Perlmutter 1978’s and 1980’s versions.

80

Pierre-Don Giancarli

verbs being unable to satisfy unaccusativity diagnostics in consistent ways while other verbs display either unaccusative or unergative syntax, etc. (see Rosen 1984; Levin and Rappaport Hovav 1995; Alexiadou et al. 2004; McFadden 2007; Giancarli 2011). We wish to offer a different binary approach. This article is structured as follows. After drawing a distinction between scalarity and gradience, and basing myself explicitely on ASH publications and quotations from Sorace in particular, I argue in part 1 that gradience (not as an objective situation but as a possible representation of split intransitivity) is not an adequate representation of auxiliary selection, due to the problems raised by gradience as the particular representation developed by the ASH and due to a certain degree of mismatch with the data. Part 2 shows the problems posed by scalarity in general, drawing on authentic extracts from French, Acadian and Corsican4 corpora5. Part 3 provides a non-scalar global representation of auxiliaries in Corsican, including not only intransitive (non reflexive) verbs, which form a well-documented pattern, but also reflexive ones, which have received little attention in that respect.

2 Gradience as the particular representation developed by the ASH 2.1 Gradience for intransitive verbs In terms of representation, I distinguish scalarity from gradience: scalarity refers to a scale made up of elements that are quantitatively different but qualitatively identical. It is a quantitative-only scale. Gradience is a particular kind of scale that makes relevant use of prototypicality (cf. Aarts 2007: 87, 107, 241). It refers to a scale built on one (or more) prototypical occurrence(s), consisting of occurrences that are both quantitatively and qualitatively different: some

4 French here refers to the standard French spoken in France, FF for short. Traditional Acadian is a variety of French spoken in Eastern Canada both open to anglicisms and conservative regarding archaisms dating back to pre-classic French. Corsican is a Roman language with a pre-Roman substratum spoken in Corsica and northern Sardinia. 5 The Acadian corpus Péronnet “1985” from the University of Moncton is made up of recordings of French-speaking informants living in south-east New Brunswick, Canada. The Corsican corpus is composed mainly of novels and stories, either translations from French into Corsican, which will be useful when a comparison between the French and Corsican perfects is in order, or original texts in Corsican; see details in the Corpus section before the References.

Auxiliary selection with intransitive and reflexive verbs

81

belong to the core(s) and are closer to the prototype(s), others belong to the periphery/ies. It is a quantitative and qualitative scale. The ASH is all the more complex as it is not mono-dimensional but bidimensional, since it involves not a single class of elements but two classes of elements and an additional intermediate zone. It is gradience between two key categories (unaccusatives and unergatives6) that converge on each other7, insofar as some verbs on the scale between the two ends display some characteristics of both categories (Aarts 2007: 55, 97, 124). The verbs situated at both ends are presented as prototypical of one or the other auxiliary8; they are central verbs (if one wishes to put it that way), while those in the middle are, paradoxically, peripheral. In such a representation, the two parts of the scale are separated by an intermediate zone, a sort of zero-zone from which a part of the scale starts in one direction and another part in the opposite direction, which corresponds to the ASH underspecified verbs. A verb reaches its lowest level (of one and the other properties) in the middle of the scale, and reaches its highest levels at the ends, the maximum level of unaccusativity at one end of the scale and the maximum level of unergativity at the other end. Does this apply to the unergative/unaccusative pair? In addition to some mismatches with the data, at least five difficulties can occur here:

6 One reviewer requires quotations to substantiate this point: “The systematic differences within the syntactic classes of unaccusative and unergative verbs may be captured by a hierarchy” (Sorace 2004: 255); “Verbs at the BE end of the ASH are core unaccusatives […]; verbs at the HAVE end are core unergatives” (Keller and Sorace 2003: 60), “There exist gradient dimensions or hierarchies which distinguish core unaccusative and unergative monadic verbs from progressively more peripheral verbs” (Legendre and Sorace 2003: 6); “The closer to the core a verb is, the more determinate its syntactic status as either unaccusative or unergative” (Sorace 2011: 69). 7 I leave aside the question as to whether unaccusativity and unergativity, i.e. a dichotomy between verbs that take an internal argument and verbs that take an external argument, correspond to gradable properties. Proponents of the ASH have had to justify themselves on this point on many occasions (“The Split Intransitivity Hierarchy substantiates the intuition that, within their respective classes, some verbes are more unaccusative and more unergative than others […] Crucially however this does not mean that unaccusativity or unergativity are inherently gradient notions” Sorace 2006: 110). I shall admit that, even if they are not gradable in themselves, they are amenable to scalar or gradient interpretations. 8 HAVE for central unergative core verbs and peripheral verbs closer to the unergative core, BE for central unaccusative core verbs and peripheral verbs closer to the unaccusative core (Sorace 2004: 260–263).

82

Pierre-Don Giancarli

2.1.1 Agentivity and telicity on the same scale? The ASH scale reduces unergativity to agentivity, and unaccusativity to telicity9, with BE-verbs at the telic end, HAVE-verbs at the agentive one10 and verbs that are neither telic nor agentive in between,11 so that because agentivity and telicity are not pointing in opposite directions the question arises as to how they can be made to appear at the two opposite ends of one and the same scale. Moreover, let us remember that some verbs can be telic and agentive at the same time: if one looks at the ASH category n°1 (change of location), i.e. the verbs considered the most telic, like FF arriver (arrive), partir (leave), venir (come), revenir (come back) (Sorace 2000: 256), old Spanish huir (run away) and escapar (escape) (Legendre 2007), do they not bear an agentive component? One could easily argue that they do (cf. also Ruwet 1988: 385).

2.1.2 Two asymmetric properties and mismatches with the data The second problem is that the roles played by the two properties that are supposed to stand in opposition are not equal but asymmetric: one is defined as the main factor, namely telicity.12 Because telicity is what distinguishes one end of the scale from the other end 13, it covers the whole length of the scale,

9 One reviewer requires quotations to substantiate this point: “I verbi intransitivi si collocano su un gradiente (Split Intransitivity Hierarchy) definito da due fattori: la telicità è l’agentività” [intransitive verbs are positioned on a gradient defined by two factors: telicity and agentivity] (Sorace’s abstract in Sorace 2011: 67); “The array of verb classes represented on the Split Intransitivity Hierarchy reduces to two key factors: telicity and agentivity” (Sorace 2011: 69); “Verbs at the BE end of the ASH are core unaccusatives and denote telic change; verbs at the HAVE end are core unergatives and denote agentive activity” (Keller and Sorace 2003: 60); “Core verbs (those at the extremes of the hierarchy) are inherently specified for telicity and agentivity” (Sorace 2004: 265); “Telic change appears to be the core (prototypical) feature of unaccusative semantics” (Bentley and Eyrthorsson 2003: 453). Unaccusative verbs, consequently the verbs selecting BE since unaccusativity is supposed to be in connection with the choice of auxiliary, would be telic; see also Sorace 2000: 884. 10 “The two key notions are telic change, which strongly correlates with BE, and agentive unaffecting process, which strongly correlates with HAVE” (Sorace 2000: 861–862). 11 “Intermediate verbs, which are neither telic nor agentive […]” (Sorace 2004: 265). 12 “In languages that have auxiliary selection, telicity is the main factor” (Sorace and Keller 2005: 17). The importance of telicity as a criterion is a widely shared position, even outside and before the ASH, cf. Van Valin 1990; Tenny 1992; Borer 1994, etc. 13 “The primary distinction separating the unaccusatives [at one end of the hierarchy] from the unergatives [at the other end of the hierarchy] is aspectual telicity” (Gurman Bard, FrenckMestre, and Sorace 2010: 329); “Telicity is the main factor that separates verbs which select E [BE] from verbs which select A [HAVE] (Legendre 2007: 13); “Telicity is the primary factor that

Auxiliary selection with intransitive and reflexive verbs

83

from a high degree of telicity at the top (the unaccusative end) to a low degree of telicity at the bottom (the unergative end).14 The other property (agentivity) is deemed less important, with a scope that does not cover the whole scale but is limited to the category of HAVE-verbs, i.e. the unergative portion of the scale15, i.e. just a portion of the scale. So the question is this: can one establish a scale on the basis of two key properties, one of which covers the whole scale, and the other only a part of it? Moreover, a lot of data cannot be accommodated by the ASH, not to say that they are at odds with it: the ASH predicts that, across languages, telicity is the primary factor separating BE-verbs from HAVE-verbs (Sorace 2011). Let us consider Corsican data, representing about 350 BE-selecting intransitives according to our countings, and French data, 23 BE-selecting intransitives according to Benveniste’s countings.16 Telicity plays no role with regard to French verbs such as rester or demeurer (‘stay’). Legendre and Sorace noticed the paradoxical atelicity of these verbs (“de manière inattendue, quelques verbes atéliques dénotant l’absence de changement sélectionnent être”, Legendre and Sorace 2003: 213), but this did not result in their questioning telicity as their main factor. Moreover, regardless of the context, telicity can only apply to verbs such as monter (‘go up’) or descendre (‘go down’) because it is understood in a very broad sense that denotes delimited events with an endpoint, such as arriver (‘arrive’) (Sorace 2004: 246), as well as others expressing a direction without an endpoint,

separates BE-verbs from HAVE-verbs” (Sorace 2004: 265); “Across languages, telicity is the primary factor, separating BE verbs from HAVE verbs (Sorace 2011: 71). 14 Telicity as well as agentivity are regarded by the ASH as “gradient notions” (Sorace 2000: 882). 15 The ASH publications on this subject all agree that the criterion of agentivity is only relevant for the HAVE-portion of the scale (Sorace 2011: 71: “Across languages […] agentivity further differentiates among atelic verbs of process”; Sorace 2004: 265: “Agentivity is a secondary factor that differentiates among HAVE-verbs”; Gurman Bard, Frenck-Mestre, and Sorace 2010: 329: “A secondary distinction, agentivity, separates core unergatives from peripheral unergatives”, except one ASH publication that on the contrary argues that agentivity is only relevant for the BE-portion of the scale: “agentivity is a secondary factor that further distinguishes among BE verbs” (Keller and Sorace 2003: 88). 16 Entrer (‘get in’), sortir (‘get out’), naître (‘be born’), éclore (‘hatch’) [doubtful], mourir (‘die’), décéder (‘die’), partir (‘leave’), tomber (‘fall’), échoir (‘fall due’) [rare], aller (‘go’), venir (‘come’), survenir (‘appear’, ‘occur’), advenir (‘happen’), devenir (‘become’), provenir (‘come from’), parvenir (‘reach’, ‘manage’), intervenir (‘intervene’, ‘take place’), monter (‘go up’), descendre (‘go down’), rester (‘stay’), accourir (‘run up’, ‘rush’), arriver (‘arrive’, ‘succeed’, ‘happen’), retourner (‘go back’) (Benveniste 1974: 181). Demeurer (‘stay’), paraître (‘appear’) and passer (‘pass’, ‘become’) could be added.

84

Pierre-Don Giancarli

such as monter or descendre (Sorace 2000: 865). In Legendre and Sorace (2003: 212) these verbs are directly added to the list of exceptions. Any exception to the relatively short list of French verbs is all the more problematic for the ASH, since French BE-selecting verbs are supposed to identify a subset of intransitives at the BE-end of the hierarchy, i.e. the most unaccusative and telic part of the ASH gradient (Legendre and Sorace 2003: 225). The same argument could be used for Corsican firmà / ristà / rimane (‘stay’), cullà / salle (‘go up’), scende / discende / scalà / falà / tramuntà (‘go down’), continua and perdurà (‘go on’). However, in Corsican, some verbs of activity such as girà (‘go round’, ‘move’) or sfilà (‘parade’) also select BE, as well as some verbs of state (see section 2.1.4). These verbs also constitute a challenge for any attempt to characterise the selection of auxiliary BE in terms of telicity, ASH’s pimary factor separating BE-verbs from HAVE-verbs, while they can be directly accounted for by our approach; see 4.1.4 and 4.1.2.

2.1.3 Positioning of the 0-point The third problem, related to the previous question of asymmetric properties, is that this situation makes the positioning of the 0-point of the graduation difficult to understand: agentivity, limited to the unergative portion of the scale, has its highest degree at the bottom of the scale and its zero-point in the middle, whereas telicity has its highest degree at the top of the scale and its zero-point at the bottom.17 The question is: is the zero-zone in the middle of the scale, or at one of its ends?

2.1.4 Fourth and fifth problem: the consequence this has on the verbs in the middle of the hierarchy and mismatches with the data If agentivity only applies to a portion of the scale, starting from a so-called zero-point in the middle and reaching its highest level at the bottom end, then the most gradient verbs (those directly open to variation since they admit competition between the two auxiliaries and which, in Italian for example, are situated in the middle of the scale, see Legendre and Sorace 2003: 197 or Sorace

17 In passing, this element contradicts what the reader might have previously understood regarding the intermediate zone, i.e. that the two mutually exclusive parts of the scale were separated by an intermediate zone standing for a zero-zone, the starting point for two opposite orientations.

Auxiliary selection with intransitive and reflexive verbs

85

2004: 260) are least taken into account by the gradient representation developed by the ASH, since they are underspecified.18 Because they are both the least unaccusative and the least unergative, the most gradient verbs escape the two polar criteria. Moreover, if telicity applies to the whole length of the scale, then the middle of the hierarchy is occupied by state verbs, which display neither an average degree of telicity nor a varying degree of telicity according to the verbs under consideration, since by definition they are all atelic with a zero-degree of telicity. This is a surprise, considering one is not looking at the bottom of the scale where the lowest degree of telicity can be found, but at the middle, which means that the verbs under examination are not situated at the expected place.19 Moreover, Corsican state verbs, far from being indeterminate, do not exhibit variation: most select HAVE categorically but some select BE in an equally categorical manner20. They cannot be explained with the ASH but they can be with our model, see 4.1.4.

18 Cross-linguistically, state verbs have an “underspecified aspectual structure” (Sorace 2000: 883); “Stative verbs tend to be variable across languages” (Sorace 2004: 249); “Intermediate verbs show variation” (Sorace 2004: 258), “Intermediate verbs are predicted to exhibit crosslinguistic variation” (Sorace 2011: 71); “Intermediate verbs, which are neither telic nor agentive, are the most variable and the least determinate” (Sorace 2004: 265); “Verbs that are stative and non-agentive are the most indeterminate and therefore the most susceptible to syntactic alternations and variation across languages” (Sorace 2011: 70). 19 Verbs of category n°4 are not the only ones to be not at the expected place: those belonging to category n°6 (motional controlled processes) should clearly side with HAVE since they are almost at the very end of the HAVE part of the scale (6 out of 7). Yet German selects sein (BE) with a number of category 6 verbs, and so do Dutch and Italian (Legendre and Sorace 2003: 198). Conversely, but based on the same logic, verbs belonging to category n°2 (change of state) should clearly side with BE since they are almost at the very end of the BE part of the scale. Yet these verbs are not completely unresponsive to HAVE and some of them even accept it very well (Sorace 2000: 865; Legendre and Sorace 2003: 195–196; Sorace 2004: 259). Legendre and Sorace note that these verbs show “des alternances régulières [regularly alternate]”: can one say of a verb that regularly alternates between BE and HAVE that it is stable, deeply rooted on the BE side? Also in German the relative order of the verbs on the scale is not the one expected either (Keller and Sorace 2003: 102). This difficulty was noticed, and the answer given was that “continuation of pre-existing state” verbs are not “continuation of pre-existing state” verbs, but “processes” (Keller and Sorace 2003: 102). 20 Vive / campà (‘live’), esiste (‘exist’), occorre / bisugnà / vulè (‘be necessary’), ghjuvà (‘be good’), calzà / cunvene (‘suit’), cunfà (‘be good’), scunvene (‘be unsuitable’), garbà / piace (‘please’), spiace / dispiace (‘displease’), durà (‘last’), custà (‘cost’), valè (‘be worth’), bastà (‘be sufficient’), mancà (‘be insufficient’), etc.

86

Pierre-Don Giancarli

2.2 Gradience for reflexive verbs? When dealing with auxiliary selection, one deals with intransitive verbs, and in doing so, one neglects – if not forgets – reflexives. The ASH does not take them into account, possibly because the languages under study do not display any auxiliary variation in their reflexive form (BE in Italian, BE in French …) and certainly because the ASH is limited to intransitive (monovalent) verbs, while reflexive verbs are dyadic (bivalent) (Cennamo and Sorace 2006: 88). Nevertheless, an absence of auxiliary variation with the reflexive verbs of a language should not make the issue of its auxiliary selection less deserving of interest. Why should the selection of BE receive an explanation only when the verbs are monovalent and not bivalent? No reason is provided by the ASH.21 In fact, telicity, the main factor used by the ASH with regard to intransitives, has already been tested for reflexives and cannot be sustained for the latter. The ASH scholars who for a while supported telicity with intransitives compared with telic BE-selecting reflexives on the one hand and non-telic non-BEselecting non-reflexives on the other hand (in pairs such as French ramollir/se ramollir [‘get soft’] or Italian asciugare/asciugarsi [‘dry oneself’]) were forced to give up (Legendre and Sorace 2003: 214–216), since it was evidently in contradiction with the facts: s’éterniser (‘drag on’), s’attarder (‘linger’), se maintenir (‘keep steady’), se pavaner (‘strut about’), se regarder (‘look at each other’), etc. are not telic, but BE-selecting. Reflexives cannot belong to the domain of the ASH, since it cannot account for them. Some scholars say that it is “a constraint specific to reflexives” but do not specify its nature (Legendre 2007: note 13). One can conclude that, even facing a language with what would be considered gradience by the ASH in terms of auxiliary selection for reflexive verbs, the ASH would use their bivalence (but see note 21) as a reason for excluding them. Even if it did include them, it would not have the adequate means to do so, which means that the ASH’s view on auxiliary selection is at best bound to remain partial and ignore a whole domain of auxiliary selection.

21 Moreover, the position of the ASH concerning the valence of reflexives used to justify their exclusion from the field of study is not easy to comprehend: in order to consider reflexives as bivalent (Cennamo and Sorace 2006: 88), the ASH must consider the clitic as an argument and avoid saying that the clitic is a valency-reducing marker, since this would make reflexives monovalent verbs (either unaccusatives or unergatives). However, the ASH explicitly sides with those who claim that the clitic is a valency-reducing marker (Legendre and Sorace 2003: 216). Then again, if the ASH considers reflexive verbs monovalent, nothing should prevent them from being treated alongside monovalent non-reflexive verbs and from passing the test of telicity, ASH’s primary and most reliable factor.

Auxiliary selection with intransitive and reflexive verbs

87

I think reflexive verbs are concerned with the question of auxiliary selection, and that the explanation must be of the same nature as that put forward for intransitive verbs. I side with those who say that reflexives are fundamentally bivalent, but I consider that this is no obstacle for integrating reflexives into a global explanation (see Giancarli 2011: 156–250). The study of a language such as Corsican, which admits a variation as far as auxiliary selection with reflexive verbs is concerned, can be of great help (see 4.2 below).

3 Scalarity in general as an adequate/inadequate factor to account for auxiliary selection 3.1 Certain syntactic phenomena in context escape any kind of prediction made from a scalar representation I shall take the example of Corsican, where avè (HAVE) can govern a BE-verb when this verb is either elided or moved away.

3.1.1 The BE-verb is elided after a servile verb In Corsican (as in Italian) there are a number of words that share particular characteristics, i.e. the servile verbs pudè/vulè/duvè (‘can/want/must’), to which I wish to add turnà (‘start again’) and cessà/stancià (‘stop’), in other words three modal verbs and three aspectual ones. They belong to a closed list of oxytonic verbs that share the particularity of being transparent: they can be preceded by either auxiliary. What counts is the chosen verb and the auxiliary that would have been selected if the verb had been conjugated directly. HAVE can be used with a BE-verb (and such a verb would be positioned as BE-selecting on a gradient scale) when the latter is elided after a servile verb.22 In case of ellipsis, BE is replaced by HAVE. Let us consider the verbs vultà (‘come back’), campà (‘live’) and andà (‘go’). BE is normally their only possible auxiliary. Yet the following extracts selected HAVE: (1) De Soto era campatu più chè ciò ch’ iddu De Soto be.pst.ipfv.3sg live.ptcp.m.sg more than what that he avaria duvutu, era andatu al dilà have.cond.3sg must.ptcp.m.sg be.pst.ipfv.3sg go.ptcp.m.sg beyond

22 This applies in a low register, because BE can be maintained in a higher register.

88

Pierre-Don Giancarli

di ciò ch’ iddu avaria pussutu. of what that he have.cond.3sg can.ptcp.m.sg ‘De Soto had lived longer than he should have, he had gone beyond what he could have.’ (SGP) (2) Ùn sò quanti, chì si ne eranu Not know.prs.1sg how many who clit from there be.pst.ipfv.3pl andati à fà furtuna induv’ ellu ci era da go.ptcp.m.pl to make fortune where it there be.pst.ipfv.3sg of chì fà, si ne pentinu è dissenu, chì, s’ what make clit of that regret. pst.pfv.3pl and say. pst.pfv.3pl that if elli avianu pussutu, eranu vultati. they have.pst.ipfv.3pl can.ptcp.m.sg be.pst.ipfv.3pl return.ptcp.m.pl ‘I don’t know how many, who had gone away to make their fortunes where there was the opportunity, wished they hadn’t, and said that, if they could have, they would have returned home.’ (AFA) (1) is a comparative and (2) is a conditional. Both are elliptical structures, corresponding respectively to: (1′) De Soto era campatu più chè ciò ch’ iddu De Soto be.pst.ipfv.3sg live.ptcp.m.sg more than what that he avaria duvutu campà, era andatu al dilà di ciò ch’ have.cond.3sg go.ptcp.m.sg beyond of what that he what that iddu avaria pussutu andà. he have.cond.3sg can.ptcp.m.sg go ‘De Soto had lived longer than he should have lived, he had gone beyond what he could have gone.’ (2′) […] s’ elli avianu pussutu vultà, eranu […] if they have.pst.ipfv.3pl can.ptcp.m.sg return, be.pst.ipfv.3pl vultati. return.ptcp.m.pl ‘if they could have returned home, they would have returned home.’ In these extracts vultà/campà/andà select BE, however HAVE is also selected: avianu pussutu, avaria duvutu and avaria pussutu. Why? Because they are

Auxiliary selection with intransitive and reflexive verbs

89

elliptical forms. With the servile verbs pudè/duvè/vulè the expected auxiliary is the one normally taken by the verb that follows, i.e. vultà/campà/andà, thus BE (for vulè, this is only the case if the verb is used in the sense of ‘must’). The ellipsis of the BE-selecting verbs suppresses the requirement for the expected auxiliary (BE), and HAVE becomes selectable. HAVE is chosen, while a scalar representation would have predicted the selection of BE for these verbs. The BE-verb does not necessarily have to be elided. Any kind of syntactic structure that moves the BE-selecting verb away will pave the way for the selection of HAVE, as I will now demonstrate.

3.1.2 The BE-verb is moved away from the servile verb and auxiliary This can be done by means of a relative clause, a cleft-sentence, or an apposition. I will limit myself to the relative clause by way of example: (3) Petru era pussutu vultà in stu paesi. Peter be.pst.ipfv.3sg Can.ptcp.m.sg return in this village ‘Peter could have gone back to this village.’ (4) Stu paesi, chì vultacci Petru avia pussutu, This village that go back-there Peter have.pst.ipfv.3sg can.ptcp.m.sg si chjama Pila-Canali. clit call.prs.3sg Pila-Canali. ‘This village, that Peter could have gone back to, is called Pila-Canale.’ Because vultà is a BE-selecting verb, era (‘was’) is selected in front of the servile verb pudè. However, (4) accepts HAVE because the verb was moved away from the servile verb and from the auxiliary governing it. The reasoning would be the same with a simple inversion: (3′) Avia pussutu Petru vultà in stu paesi. have.pst.ipfv.3sg can.ptcp.m.sg Peter return in this village ‘Peter could have gone back to this village.’ What is common to the two phenomena in 3.1 is that 1) when a BE-verb is elided or moved away that verb loses its capacity to determine the choice of auxiliary. Nevertheless, an auxiliary is necessary. HAVE is then resorted to as a default auxiliary. This shows how vulnerable esse is compared to avè, even in a language that counts more than 350 BE-selecting intransitives. 2) HAVE is

90

Pierre-Don Giancarli

chosen while a gradient or scalar representation would have forecast the selection of BE.

3.2 The problem of a cross-linguistic scale To set out a wide variety of languages on a single scale, one must either take for granted in accordance with the Null Hypothesis that all the compound forms (all perfects, all pluperfects, etc.) of the languages under consideration are equivalent to each other, or disregard the fact that they are not, whether they be Romance languages or Germanic languages or any other. The problem is that they are not equivalent to each other, and this has to be taken into account.23

3.2.1 All the compound forms (perfects, etc.) of different languages are not equivalent to each other All perfects must not be put on the same footing, and this criticism is essential. This claim may be shared by many, but the consequences must also be accepted. For example, French passé composé and Corsican passatu quancianu are two compound forms with an auxiliary in the present, both superficially like the English present perfect. However, they should not be carelessly equated with each other because their uses are different; they do not always select the same auxiliary, and I shall go so far as saying that their natures are synchronically different. The French passé composé can be a temporal anteriority marker (its most common use nowadays: hier il a mangé un gateau ‘yesterday he has eaten/he ate a cake’) or a present perfective, i.e. an aspect marker (its initial use in French: il a bien mange ‘he has eaten well/he has had a good meal’). The temporal passé composé is aoristic (like French passé simple il mangea, the only difference being the plane of utterance: passé composé pertains to the plane of discourse and not the plane of historical narrative). It is a past tense that gives a whole view on the event from the outside (unlike French imparfait). The aspectual passé composé on the other hand is not aoristic, and as a

23 The ASH is exposed to this difficulty, and there are only two ways to minimise it: remove certain languages from the scale or, as the ASH does it, set up rules that apply to some languages but not to others (addition of the criterion “internal motion” in Legendre 2007, or “locomotion” in Keller and Sorace 2003). In both cases the cross-linguistic scope of the cross-linguistic scale is reduced.

Auxiliary selection with intransitive and reflexive verbs

91

compound form it is perfective, as opposed to passé simple and imparfait which are simple forms. It capitalises on the resulting state following the change of state, and it is a present form. French and Corsican choices look identical, i.e. an auxiliary in the present followed by a past participle when the examples chosen are present perfective aspectual perfects, as in (5), but let us look at what happens with temporal variants: (5) Maintenant qu’ils sont arrivés, c’est une autre now that they be.prs.3pl arrived it be.prs.3SG a other angoisse. anxiety ‘Now they have arrived, it is another cause for anxiety.’ Avà ch’elli sò ghjunti, è un’ antra angoscia. now that they be.prs.3pl arrived be.prs.3SG a other anxiety ‘Now they have arrived, it is another cause for anxiety.’ (CDL) (6) Tu es bien arrivé ce soir avec les croisés ? You be.prs.2sg well arrived tonight with the crusaders? ‘You did arrive tonight with the crusaders?’ Tù, ghjunghji sta sera incù i cruciati, ùn hè ? You, arrive.pst.pfv.2sg tonight with the crusaders, not is? ‘You arrived tonight with the crusaders, didn’t you?’ (H) (7) C’ est peut-être le même qui rôdait déjà quand mon papa It is perhaps the same that loitered already when my father a disparu. have.prs.3sg disappear.ptcp.m.sg ‘It may be the same one [shark] that was already loitering when my father disappeared.’ Ch’ ellu ùn sia listessu pesciu cane chè quand’ ellu That he not be.prs.sbjv.3sg the same shark as when he s’ anneiò u mo poveru babbu? clit drown.pst.pfv.3sg my poor father? ‘Could it be the same shark as when my poor father drowned?’ (P)

92

Pierre-Don Giancarli

Examples like (6) and (7) show how different the French “perfect” and the Corsican “perfect” are. They do not match. The French “perfect” can keep its auxiliarised form, while Corsican has to move to a non-auxiliarised aorist form (this is particularly the case for educated and older speakers), since the passatu quancianu is not suitable for narrative contexts or with explicit temporal dates. The difference is that the French “perfect” usually has a temporal past value and sometimes an aspectual perfective one, whereas the Corsican “perfect”, even if engaged in an aoristic drift, is basically an aspect. One will easily admit that two semantically different forms cannot appear side by side on the same scale and be identically treated. Treating them identically nonetheless is one of the main reasons for the difficulties experienced by the ASH, since telicity, its primary factor on the gradient whatever the language, faces difficulty in French (cf. 2.1.2), is suited neither to German24, nor Corsican (cf. 2.1.2 and 2.1.4) nor Acadian (see 3.2.2), and is challenged by Italian and Dutch data (see note 19), just to mention a few languages. But any scalar representation of auxiliaries in a cross-linguistic perspective would meet difficulties, since positioning the compound forms of different languages on one and the same scale forces one to see them from the same perspective and reduces the differences between them to a quantitative difference. As Mcfadden and Alexiadou (2010) state25, the perfect is not a stable and consistent cross-linguistic category, and “two languages will differ in their auxiliary selection at least in part because of the differences in the makeup of their perfects” (390). Taking these differences into account in my analysis of French of France, English, Acadian and Corsican led me in Giancarli (2011) to a different explanation of auxiliary selection than that of the ASH – not a single quantitative explanation but qualitatively different explanations for each language considered. I shall limit myself here to giving an overview of auxiliary selection in Acadian before providing a non-scalar representation of auxiliaries in Corsican.

3.2.2 The example of Acadian French: a language whose “perfect’s” double nature is directly correlated with auxiliary selection A model that would only take into account the morphology of the French perfect and ignore its double nature, as would any scale, would hardly be in a 24 The ASH itself had to give up telicity as its number-one criterion for German (Keller and Sorace 2003). 25 These authors build on the difference between the perfect of result and all the others (in particular the experiential perfect, but including also the perfect of result) in Earlier English to explain the choice of be or have, respectively (420–421).

Auxiliary selection with intransitive and reflexive verbs

93

position to incorporate a language whose choice of auxiliary would rely on the French “perfect’s” double nature. Such a language exists, since in Acadian one single intransitive verb will be (using the unaccusative/unergative terminology) unaccusative when in the aspectual perfect and unergative when in the temporal “perfect”, and will therefore select the BE or HAVE auxiliary on that basis. At least ten verbs can undoubtedly select BE in Acadian: aller, arriver, descendre, monter, mourir, partir, passer, rentrer (entrer is not so common), sortir, venir/menir.26 I shall take partir (‘leave’) for illustrative purposes. The same speaker27 uses the same verb partir in (8) and in (9): (8) J’ ai sauvé une princesse, pis i dit, je sus obligé d’ aller avec I saved a princess then he says I am obliged to go with sous peine de mort, i dit, ielle est partie asteur, j’ on pain of death he says she be.prs.3sg leave.ptcp.f.sg now I ai été trahi have been betrayed ‘She has left now.’ (9) On avait pas soupé quand qu’ on a parti, We had ‘nt had dinner when we have.prs.3sg leave.ptcp.m.sg pis c’ est la faim qui nous a fessé then it is hunger that us struck ‘We hadn’t had dinner when we left, then we were hunger stricken.’ In (9) the “perfect” is temporal and aoristic: it marks the event as past, located as disconnected from the origin situation, in a temporal adverbial clause introduced by the conjunction quand que (‘when that/when’). Because the “perfect” is of a temporal nature, auxiliary selection is made in favour of HAVE: quand qu’on a parti. In (8), on the other hand, the perfect is of an aspectual nature and, as a compound form, it refers to a perfective value: the link with the situation of uttering is demonstrated by the presence of the deictic adverb asteure (‘now’). Ielle est partie asteur is about a third person’s absence from the enunciative space. You could add an aspectual quantifier like depuis trois semaines (‘for three weeks’) to underline this point: ielle est partie depuis trois semaines,

26 Respectively: ‘go’, ‘arrive/succeed/happen’, ‘go down’, ‘go up’, ‘die’, ‘leave’, ‘pass/become’, ‘go in/come back’, ‘get out’, ‘come’. 27 This speaker is one of seven informants recorded while telling a tale and living in southeast New Brunswick, Canada.

94

Pierre-Don Giancarli

which is synonymous with ielle est absente depuis trois semaines: because partir is discrete28, the temporal adjunct depuis trois semaines does not quantify the verb but the state resulting from the process, i.e. the fact of not being there. Because this perfect is aspectual, auxiliary selection is made in favour of BE. FF and Acadian French have the same opposition between two “perfects”: a temporal anteriority “perfect” and an aspectual present perfective perfect. However, in FF the aspectual perfect is compatible with HAVE as well as with BE, as was shown by extract (5), while in Acadian the aspectual perfect only selects BE. And while in FF the temporal “perfect” is compatible with HAVE and BE, as was shown by extracts (6) and (7), in Acadian the temporal “perfect” only selects HAVE. In FF this dichotomy between the temporal “perfect” and the aspectual perfect transcends the difference between BE and HAVE, so that the explanation for auxiliary variation must be sought elsewhere29. It must not be sought elsewhere in Acadian, however, which builds on the double nature of its “perfect” and establishes a correlation between the kind of “perfect” and the choice of the auxiliary. Listing Acadian verbs on a variation scale looking only at the verb itself would not account for a dimension of variation based not on the verb but on a pair of values attached to the compound form.

4 A non-scalar representation of Corsican auxiliaries Corsican selects HAVE with its non-reflexive verbs when they are bivalent and BE with at least 350 of its monovalents. In the reflexive domain, certain types of reflexive verbs, unlike French or Italian, admit either BE or HAVE. Here are examples of an intransitive with HAVE (10), one with BE (11), two examples of variation with an intransitive (1 compared to 4, 12 compared to 13) and two with a reflexive (14 and 15): (10) Sin’ à avale aghju travagliatu per mantene à Babbu. Until to now have.prs.1sg work.ptcp.m.sg to support to Father ‘Up to now, I have worked to help my father.’ (P)

28 We speak of a discrete process when it can be individuated as an occurrence that manifests a quantitative delimitation of the notion, a dense process when neither the quantitative nor qualitative dimension is prevalent; while compact applies to what is indivisible in the strictest sense and open at both ends (Culioli 1991–92: 12). 29 See Giancarli 2011: 94 ff.

Auxiliary selection with intransitive and reflexive verbs

95

(11) Elli ci sò sfilati à millaie sott’ à l’ Them there be.prs.3pl parade.ptcp.m.pl by thousands under to the ochji. eyes ‘They paraded by the thousands before our eyes.’ (ABM) (1)

De Soto era campatu più chè ciò ch’ iddu De Soto be.pst.ipfv.3sg live.ptcp.m.sg more than what that he avaria duvutu, era andatu al dilà have.cond.3sg must.ptcp.m.sg be.pst.ipfv.3sg go.ptcp.m.sg beyond di ciò ch’ iddu avaria pussutu. of what that he have.cond.3sg can.ptcp.m.sg ‘De Soto had lived longer than he should have, he had gone beyond what he could have.’ (SGP)

(4)

Stu paesi, chì vultacci Petru avia This village that go back-there Peter have.pst.ipfv.3sg pussutu, si chjama Pila-Canali. can.ptcp.m.sg clit call.prs.3sg Pila-Canali. ‘This village, that Peter could have gone back to, is called Pila-Canale.’

(12) Bulà, aghju bulatu un pocu inghjilocu è a Fly have.prs.1sg fly.ptcp.m.sg a little everywhere and geography giograffia, hè virità, m’ hè ghjuvata assai. it is truth to me be.prs.3sg be useful.ptcp.f.sg much ‘I have flown a little everywhere, and geography, that’s correct, served me a lot.’ (PP) (13) Hè vulatu à pezzi. be.prs.3sg fly.ptcp.m.sg to pieces ‘It was shattered to pieces.’ (14) Microcosimu u paese ? Ben intesu, a dritta è a manca, i Microcosm the village? Well heard the right and the left the neri è i rossi, dui partiti chì si sò black and the red two parties that clit be.prs.3pl

96

Pierre-Don Giancarli

spartuti una cintunara d’ elettori, cuntendu i share.ptcp.m.pl a hundred or so of voters counting the braganati. multicoloured ‘Is the village a microcosm? Of course, the right and the left, the black and the red, two parties that shared among each other about a hundred voters, including the multicoloured ones.’ (DF) (15) – “Alora, spartaremmu i vacchi, avali !” – “Seti vo ! Seti – Then share.fut.1pl the cows now – You are you You are vo !” S’ hani spartutu i vacchi, dici alora: you clit have.prs.3pl share.ptcp.m.sg the cows he says then: “Qualissi voli ?” Which you want? ‘– “Then we’ll share the cows, now!” – “Up to you! Up to you!” They shared the cows, then he says “Which ones do you want?”’ (CS) Each of these illustrative extracts will be given an explanation in this article.

4.1 Intransitive verbs 4.1.1 Definitions of Source and Goal I call Source and Goal thematic macro-roles, encompassing the roles of agent and patient. I argue that in Corsican, HAVE is selected when the subject corresponds to a Source, an acting role, while BE is selected when the subject corresponds to both a Source and a Goal, acting and acted upon, so it satisfies two thematic roles. This applies to simple (non-reflexive) BE-selecting verbs, whose subject has this dual status intrinsically, and to BE-selecting reflexive ones, whose subject satisfies this dual status either intrinsically or (most of the time) obtains it by means of a co-reference. What comes under the terms Source and Goal? Source and Goal are macroroles. Each macro-role is accompanied by a distinction between several levels, but a hierarchy between those levels, though easy to draw, is not necessary,30 30 This is unlike Van Valin’s macro-roles “Actor” and “Undergoer” (Van Valin 1990; Van Valin and Lapolla 1997) which are organised in a relevant gradient hierarchy. Secondly, theirs are two discrete categories, to which each argument does or does not belong, thus excluding double membership, Thirdly our macro-roles are also chosen to comply as nearly as possible with

Auxiliary selection with intransitive and reflexive verbs

97

since the selection of HAVE requires the activation of only one macro-role (Source) and the selection of BE is justified as long as the two macro-roles are involved, whatever the particular roles within the macro-roles. It is of no importance whether the subject is positioned at the top or at the bottom of a Goal hierarchy – what counts is that it is recognised as bearing any role encompassed within that macro-role. A scalar view, and a fortiori a gradient one with one (or more) core(s) and one (or more) periphery/ies, is consequently irrelevant. The Source has the primitive properties of an agent rather than a patient, of a cause rather than a consequence, but Source is not confined to agent, and Goal is not reducible to patient: a Source is the instigator of a process for which it is responsible. It encompasses the semantic roles of agent, instrument or stimulus source of a perception or knowledge necessarily pre-existing this perception or knowledge. A Goal is an affected argument in the sense that it undergoes a change of state, with at least one of its properties being the case before the change of state but not after. It can be a locative, i.e. an inanimate argument affected by a state; an experiencer, i.e. an animate argument affected by a state; a theme, i.e. a subject argument whose localisation is predicated by a state verb; a patient affected by a change of state; a beneficiary; or a recipient. Source and Goal will be useful to us in the following way: with Corsican simple (non-reflexive) verbs, avè (HAVE) will be selected when the subject has the Source macro-role, and esse (BE) when it is both Source and Goal. Three categories will be distinguished: process verbs with no variation, process verbs with variation and state verbs.

4.1.2 Discrete and dense process verbs with no variation: when the subject is an agent-Source, and a patient-Goal This has to do with two cases: firstly when the subject is an agent-Source, and a patient-Goal, which will concern discrete and dense process verbs. Secondly, when the subject is an agent-Source and a locative-Goal, which will concern weather verbs (dense). I shall only present the first one here; for an illustration of the weather verbs see Giancarli (2011: 123 ff.).

the semantic realisations of the arguments actually found in the range of both intransitive and reflexive constructions.

98

Pierre-Don Giancarli

When this dual-status argument is both an agent-Source and a patientGoal, it triggers a process that affects an argument that happens to be itself. It is an agent-patient affected by the resulting state; it initiates a process but undergoes it at the same time. This view encompasses not only the subclass of verbs that some would call telic and that select BE both in Corsican and in their French equivalents,31 but also verbs of movement.32 Corsican verbs of movement select BE like the verbs of change of place because they are not based on the criterion of telicity but on the dual status agent-patient, and this requirement is met: (11) Elli ci sò sfilati à millaie sott’ à l’ Them there be.prs.3pl parade.ptcp.m.pl by thousands under to the ochji. eyes ‘They paraded by the thousands before our eyes.’ (ABM) (16) Sò corsu à mettemi à u pianò. be.prs.1sg run.ptcp.m.sg to put myself to the piano ‘I hurried to get to my piano.’ (BC) This reminds one of the middle voice, by which the subject “effectue en s’affectant” [effects while being affected], “effectue quelque chose qui s’accomplit en lui” [achieves something which is being achieved in him] (Benveniste 1974: 173).

4.1.3 Discrete and dense process verbs that admit either one or the other auxiliary Some verbs, less than a dozen according to my countings, admit either one or the other auxiliary and select BE 1) when the subject is both a Source and an inanimate Goal, 2) when it is both a Source and an animate non-volitional

31 Nasce/naître (‘be born’), more/mourir (‘die’), diventà/devenir (‘become’), etc. 32 Corre (‘run’), girà (‘move’, ‘circulate’), sfilà (‘parade’), etc.

Auxiliary selection with intransitive and reflexive verbs

99

Goal, or 3) when it is both a Source and an animate non-controller Goal. I shall provide a few illustrations of 1) here: 1) Corre, saltà, vulà (‘run’, ‘jump’, ‘fly’, respectively): the subject is both a Source and an animate agentive Goal (HAVE) / the subject is both a Source and an inanimate Goal (BE). Corre, saltà, vulà can select either BE or HAVE: BE when their subject is both a Source and a Goal. They select HAVE when the subject is animate and agentive, and BE when it is inanimate. Because the subject of a BE-selecting verb involves an undergoing dimension, there is a distribution on the following basis: the verb takes BE when the subject is inanimate (and therefore unagentive), which makes the Goal component prevalent, and complementarily HAVE is selected when the subject is animate and agentive, which makes the Source component prevalent. Hence the following contrast between (12) and (13) with the verb vulà: (12) Bulà, aghju bulatu un pocu inghjilocu è a Fly have.prs.1sg fly.ptcp.m.sg a little everywhere and geography giograffia, hè virità, m’ hè ghjuvata assai. it is truth to me be.prs.3sg be useful.ptcp.f.sg much ‘I have flown a little everywhere, and geography, that’s correct, served me a lot.’ (PP) (13) Hè vulatu à pezzi. It be.prs.3sg fly.ptcp.m.sg to pieces ‘It was shattered to pieces.’ Vulà takes HAVE in (12) with a first-person animate subject and has its literal sense (here ‘fly in the air aboard a plane’), but it takes BE when its subject is inanimate, with a figurative sense, in (13) ‘to be shattered’, not ‘fly’ but ‘fly into pieces’.

4.1.4 Compact verbs I will now try to show why Corsican compact verbs can select BE, namely when the subject is both a stimulus-Source and a patient-Goal – to be more precise, a Goal because it undergoes an assessment/evaluation. A certain number of Corsican BE verbs – inexplicable by telicity since, as states, they present no

100

Pierre-Don Giancarli

change of state, change of place, or movement – can find an explanation through the Source-Goal duality.33 Here are a few examples from our corpus: (17) Sò campatu assai cù e parsone maiò. be.prs.1sg live.ptcp.m.sg much with the people big ‘I have lived a great deal among grown-ups.’ (PP) (18) St’ arburu, vai puru, da nulla ci hè This tree definitely of nothing to us be.prs.3sg ghjuvatu. be good.ptcp.m.sg ‘This tree has definitely been of no use to us.’ (EAG) (19) Mi saria piaciutu à principià ista storia cum’ elle Me be.cond.3sg please.ptcp.m.sg to begin this story as they principianu e fole. Mi saria piaciutu à dì: begin the fairy tales Me be.cond.3sg please.ptcp.m.sg to say: ‘It would have pleased me to begin this story in the manner fairy tales begin. It would have pleased me to say:’ (PP) They are state verbs, intransitive and unpassivizable. As state verbs, their subject is non-agentive by definition , so it cannot be viewed as partly patient and partly agent, thus excluding any interpretation in terms of middle voice. This will not stop us from claiming that their subject has a dual status – both Source as a stimulus (source of the perception/feeling/cognition of an experiencer) and Goal as a patient, insofar as it undergoes an evaluation. For example with ghjuvà (‘be useful’) the subject is Source of the perception of the object qualified as good, and patient of an evaluation which makes it qualified as such.

33 The verbs concerned are: bastà (‘be sufficient’), bisugnà (‘be necessary’), calzà (‘be suitable’), campà (‘live’), cunfà (‘be fit’), cunvene (‘be convenient’), custà (‘cost’), dispiace (‘displease’), durà (‘last’), esiste (‘exist’), esse (‘be’), garbà (‘be enjoyable’), ghjuvà (‘be good’, ‘be useful’), mancà (‘be lacking’), occorre (‘be convenient’), parè (‘appear’), piace (‘please’), scunvene (‘not be convenient’), spiace (‘displease’), stà (‘be’, ‘stay’), vive (‘live’), valè (‘be worth’), vulè (‘be necessary’), sembrà (‘seem’).

Auxiliary selection with intransitive and reflexive verbs

–

–

–

–

101

When I say “evaluation/assessment”, the term is four-fold: It can be a root modality (Culioli’s modality n°434), enabling the speaker to assess how autonomous the subject is. The verbs concerned are occorre and bisugnà. The very common verb vulè also means ‘be necessary’ when BE is selected, but it experiences an interesting variation (cf. Giancarli 2011: 140). It can take the form of a qualitative evaluation by the speaker, a value judgment on the grammatical subject passed with a positive or negative valuation (modality n°3). The verbs concerned are ghjuvà, calzà, cunfà, cunvene, scunvene, garbà, piace, spiace, dispiace. It can be a quantative evaluation of the subject by the speaker, in the case of durà, custà and valè which are all indicative of how big/tall/long/expensive the referent of the subject is – in other words its quantative dimension. Finally, it can be a superimposition of the last two evaluations: firstly with bastà and mancà with which the speaker passes judgment on a quantity assigned to the subject. Secondly with parè and sembrà which, on top of a qualitative evaluation, involve a quantative evaluation through the epistemic modality they carry (modality n°2). They enable the speaker to modulate the strength of his/her assertion by giving it a degree of assertability that quantifies how probable the relationship between the subject and the predicate is.

As for the verbs of existence (esse, stà, vive, esiste, campà, to which parè and sembrà can be added), they lend themselves to a quantative or qualitative evaluation according to the context (see Giancarli 2011: 134 ff.).

4.2 Reflexive verbs 4.2.1 The types of reflexive verbs that admit variation, either one or the other auxiliary 4.2.1.1 Presentation Corsican reflexives are composed of a clitic personal pronoun (clit), pre-verbal in the finite forms and post-verbal the rest of the time. It is identical to the 34 Type 4 modality concerns the relation between the grammatical subject and the predicate, type 3 modality is a qualifying modality concerned with judgments about the content of the predicative relation, type 2 modality establishes a relation between the enunciator and the propositional content and assesses the (im)probability of the validation of the relation (Culioli 1990 and 1999).

102

Pierre-Don Giancarli

non-reflexive accusative pronoun in all the persons (mi/ti/ci/vi) except the third person in the plural and singular, where it has the special form si. They are not limited to BE as is the case in French or Italian; they can select either BE or HAVE. Ten kinds of reflexive verbs can be distinguished in Corsican, bivalent or monovalent: Case 1: true reflexive (20) Petru si hè lavatu. Peter clit be.prs.3sg wash.ptcp.m.sg ‘Peter has had a wash.’ Case 2: true reflexive of inalienable possession (21) Petru si hè Lavatu i capelli. Peter clit be.prs.3sg wash.ptcp.m.sg the hair (21′) Petru si hà lavatu i capelli. Peter clit have.prs.3sg wash.ptcp.m.sg the hair ‘Peter washed his hair.’ Case 3a: ethic reflexive of alienable possession with a bivalent verb (22) Petru si hè cumpratu una vittura. Peter clit be.prs.3sg buy.ptcp.m.sg a car (22′) Petru si hà cumpratu una vittura. Peter clit have.prs.3sg buy.ptcp.m.sg a car ‘Peter bought a car.’ Case 3b: ethic reflexive with a monovalent verb (23) Petru si hè fermu quì. Peter clit be.prs.3sg stay.ptcp.m.sg here ‘Peter stayed here.’ Case 4a: reciprocal reflexive without an object (24) Petru è Paulu si sò azzuffati. Peter and Paul clit be.prs.3pl fight.ptcp.m.pl ‘Peter and Paul fought.’ Case 4b: reciprocal reflexive with an object (25) I zitelli si sò spartuti e cubi. The children clit be.prs.3pl share.ptcp.m.pl the marbles

Auxiliary selection with intransitive and reflexive verbs

103

(25′) I zitelli s’ anu spartutu e cubi. The children clit have.prs.3pl share.ptcp.m.sg the marbles ‘The children shared the marbles.’ Case 5: inherent reflexive (26) Una turista taliana si hè svanita. A tourist Italian clit be.prs.3sg faint.ptcp.f.sg ‘An Italian tourist has fainted.’ Case 6: medio-passive reflexive (27) Petru si hè avanzatu. Peter clit be.prs.3sg move forward.ptcp.m.sg ‘Peter moved forward.’ Case 7a: reflexive with a passive meaning with a bivalent verb (28) E ghjembe si sò tronche. The branches clit be.prs.3pl break.ptcp.f.pl ‘The branches have broken.’ Case 7b: with a monovalent verb, where a distinction must be made between 29 and 30: (29) Si hè / hà travagliatu. clit be.prs.3sg / have.prs.3sg work.ptcp.m.sg ‘One has worked.’ (30) Si hè andati. clit be.prs.3sg leave.ptcp.m.pl ‘One has left.’ However, it is not just any kind of reflexive that accepts variation: of the ten listed, six must select BE (cases 1, 3b, 4a, 5, 6, 7a), and four need not select BE exclusively and can therefore select HAVE as well (cases 2, 3a, 4b, and possibly some monovalents of 7b).

35 The reflexive e ghjembe si sò tronche should not be confused with the (non-reflexive) anticausative (e ghjembe hanu troncu, literally ‘the branches have broken’). They differ in terms of productivity, relationship with a transitive counterpart and, most importantly, presence/absence of a cause: if an anticausative presents a process occurring without a cause (Reinhart 2002; Härtl 2003; Talmy 2003; Reinhart and Siloni 2005; Kallulli 2006, etc.), e ghjembe hanu troncu (‘the branches have broken’) is an anticausative, but Corsican reflexive e ghjembe si sò tronche is not, since a secondary Source is present; see 4.2.2.

104

Pierre-Don Giancarli

4.2.1.2 Reasons for the variation with reflexive verbs: valency and orientation. Subject-orientation (BE) / object-orientation (HAVE) What do BE-selectable reflexive verbs have in common? If case 7b were disregarded, the answer would be: bivalence and the explicit presence of an object. Or perhaps one part of 7b can be included in the explanation, the one illustrated by si hè/hà travagliatu (one has worked). Of course there is neither bivalence nor an object here, since the verbs are monovalent (they are unergatives); however, they can be attributed the same functioning as the bivalent forms 2– 3a-4b, thanks to the internal object. Unlike unaccusatives, whose monovalence is undisputable, unergatives globalise the process by integrating features that could be rendered by complements, the complement is “incorporated” (Hale and Keyser 2002). Herslund (1996) even claims that unergatives are bivalents in the guise of monovalents. They have an object, but instead of externalising it in the form of a NP after the verb, it is lexicalised inside the verb root; we usually call this an internal object. Travaglià can be glossed as ‘do some work’, i.e. ‘produce some work through working’, parlà by ‘produce some speech through speaking’, and so on. Showing the deep similarities that monovalent unergatives have with bivalents is a way of justifying the possible selection of HAVE with unergatives in 7b. I claim that the series of reflexives that has an object (internal object included) is also the one open to the selection of HAVE, while the series without an object is restricted to BE. The choice of auxiliary thus has something to do with valency. Unfortunately, the criterion of valency shows its limits, since the presence of an object argument makes the selection of HAVE possible but not compulsory due to the fact that BE remains selectable (remember 21, 22, 25 and 29). We are now facing another auxiliary variation, but with an equal number of arguments. To understand this variation within the Corsican reflexive construction, it is necessary to turn to a complementary explanation, namely the orientation. The criterion of orientation (or in more general terms topicalisation36) helps one understand how the selection is made under certain conditions. In bivalent constructions, the auxiliaries are the markers of different orientations: BE towards the subject, HAVE towards the object. The sentence will be oriented towards one of the arguments, and will say something about the state it is in.

36 This refers to the phenomenon in which some element is singled out as the topic: “An entity, E, is the topic of a sentence, S, iff in using S the speaker intends to increase the addressee’s knowledge about, request information about, or otherwise get the addressee to act with respect to E” (Gundel 1988: 210).

Auxiliary selection with intransitive and reflexive verbs

105

This criterion is relevant only when the speaker faces a choice in terms of auxiliary, i.e. with reflexives including an object, which I referred to as cases 2, 3a and 4b above. Here are a couple of examples of reciprocal reflexives with an object (case 4b) showing variation of the same verb sparte (‘share’), conjugated with BE in 14 and HAVE in 15: (14) Microcosimu u paese ? Ben intesu, a dritta è a manca, i Microcosm the village? Well heard the right and the left the neri è i rossi, dui partiti chì si sò black and the red two parties that clit be.prs.3pl spartuti una cintunara d’ elettori, cuntendu i share.ptcp.m.pl a hundred or so of voters counting the braganati. multicoloured ‘Is the village a microcosm? Of course, the right and the left, the black and the red, two parties that shared among each other about a hundred voters, including the multicoloured ones.’ (DF) (15) – “Alora, spartaremmu i vacchi, avali !” – “Seti vo ! Seti – Then share.fut.1pl the cows now – You are you You are vo !” S’ hani spartutu i vacchi, dici alora: you clit have.prs.3pl share.ptcp.m.sg the cows he says then: “Qualissi voli ?” Which you want ‘– “Then we’ll share the cows, now!” – “Up to you! Up to you!” They shared the cows, then he says “Which ones do you want?”’ (CS) In (14), sparte is associated with BE, which highlights the prevailing role of the political parties on the island, while the electors are presented as disposable and insignificant, mere puppets in the hands of the elected officials that enable them to become powerful local leaders. HAVE (dui partiti chì s’anu spartutu una cintunara d’elettori) would have given the electors an autonomous dimension out of the subject’s control, inadequate in the local context. In (15), three brothers cannot fairly share the family cows, after one of them cheated the others with the corn and wine. The cows are the third object to be shared on the list, and everyone is very careful about the last share. I vacchi in s’hani spartutu i vacchi is the main element of the sentence around which the discourse is centred, and it is also the topic of the question Qualissi

106

Pierre-Don Giancarli

voli? (‘which ones do you want?’). Conversely, the choice of BE would have been subject-centred, and would have presented the collective subject as a block, as if they were acting by common consent in full agreement without making the cows (grammatical object) an issue. As for si hè travagliatu/si hà travagliatu (the unergative of case 7b), the acceptability of HAVE is understandable on the basis of what was said previously in terms of valency. Unergatives internalise a bivalent relationship inside the verb root through an internal object, so they do have an object. The idea of an internal object is useful in terms of valency when it brings the monovalent unergatives back to bivalents. But within the context of a choice between BE and HAVE based on orientation it is useless, since I cannot claim that HAVE is still the marker of an object-orientation, considering there is no autonomous object in itself.

4.2.2 Si is an argument (Source) and the subject has a dual status: Goal + Source I claim that the pronoun si is an argument (a deficient argument but nevertheless an argument 37) to which the thematic role of Source is assigned. Apart from one exception (the monovalent unergative), the clitic pronoun faces a subject endowed with a dual status: Goal + Source.38 The dual status of the subject is the reason for the selection of BE with reflexive Corsican verbs39, except in one case: the reflexive with a passive meaning with a monovalent unergative. This position needs justifying, as there is no consensus on the matter. For some scholars, the clitic is not an argument, but a valency-reducing marker that suppresses an argument and leads to a monovalent construction (which I call View 1). According of the first version of View 1, si absorbs the internal argument, which leaves the subject as the sole remaining argument, an external argument (Source), and the verb is a monovalent unergative – the

37 Si (with both monovalent and bivalent reflexives) cannot be conjoined with a lexical subject, it is not quantifiable, not qualifiable, not suffixable, not cleavable, hardly dislocable, and not stressable. It must be in contact with the verb/auxiliary, and to the left of the verb/auxiliary in finite assertive forms. 38 This dual status is obtained either by means of a co-reference (most of the time), or the subject has it intrinsically (see Giancarli 2011: chapter 4 for a more detailed discussion). 39 This applies at least as a primary reason; remember that Corsican reflexives accept auxiliary variation: they select BE only if in the first stage the subject has a dual status, and if in the second stage the reflexive is bivalent + subject-oriented, since complementarily HAVE is an object-orientation marker.

Auxiliary selection with intransitive and reflexive verbs

107

opposite of a passive construction. Reflexives would be valency-reducing markers by means of absorption of the Goal argument, making the verb intransitive. This is View 1a, supported by Alexiadou and Anagnostopoulou (2004) and Reinhart and Siloni (2005). This view is not sustainable. For example, it is at odds with the predictable choice of auxiliary: if the only argument left is the Source of a verb considered unergative, it is the auxiliary HAVE that should govern the reflexive (unless reflexives are a separate case completely disconnected from non-reflexive intransitives and any hope of a global explanation is abandoned), but BE is the only possible auxiliary in Italian or French, and either HAVE or BE is selected in Corsican. Moreover, this view obviously does not apply to a language such as Corsican whose reflexives accept monovalents (cf. 23 Petru si hè fermu quì ‘Petru is stayed here’) and are therefore not derivable from a bivalent structure. This capacity rules out any approach in terms of absorption, whether an absorption of the Goal argument or of the Source argument 40 since there is no transitive to be turned into an intransitive. According to other scholars, si suppresses the external argument, which leaves the subject as the sole remaining argument, an internal argument, and the verb is a monovalent unaccusative, similar to a passive construction. This is View 1b, supported by Burzio (1986), Legendre and Sorace (2003) and Cennamo and Sorace (2006). This view is not sustainable either – firstly, because languages such as Corsican have reflexives that accept monovalents and secondly, because in Corsican, Italian or French (to name but a few), if the clitic was used when the Source is not present, one would also expect it to appear when the passive is lacking a Source, i.e. an agent, but this never happens: (31) E ghjembe sò state tronche. The branches be.prs.3pl be.pass.ptcp.f.pl break.ptcp.f.pl I rami sono stati rotti. The branches be.prs.3pl be.pass.ptcp.m.pl break.ptcp.m.pl Les branches ont été cassées. The branches have.prs.3pl be.pass.ptcp.f.pl break.ptcp.f.pl ‘The branches have been broken.’

40 See Giancarli (2011: 183–185) for reasons arguing against View 1a in other languages too, even when their reflexive construction is limited to bivalent verbs.

108

Pierre-Don Giancarli

(31″) *E ghjembe si sò state tronche. *The branches clit be.prs.3pl be.pass.ptcp.f.pl break.ptcp.f.pl * I rami si sono stati rotti. *The branches clit be.prs.3pl be.pass.ptcp.m.pl break.ptcp.m.pl *Les branches s’ ont été cassées. *The branches clit have.prs.3pl be.pass.ptcp.m.pl break.ptcp.f.pl The Corsican, Italian and French sequences above are ungrammatical because the clitic pronoun itself already instanciates a Source argument. Another position holds that the clitic is an argument (which I call View 2) placed within the valency of the verb, whether monovalent or bivalent. According to one version of View 2, which I call View 2a, it is an internal argument, while the subject would be the external argument. This is supported by Rizzi (1986) for Italian and Dobrovie-Sorin (1998) for Roman languages in general. Alternatively, it is an external argument, while the subject would be the internal argument. This is View 2b, supported by Kayne (1988) and Embick (2004) for Roman languages, and which is the closest to View 2c I shall put forward. I share with View 2b the idea that si is a Source argument. Because si is a Source argument, the sentence containing it is not compatible with the presence of an agent (Source), even when the reflexive construction sounds similar to a passive, while a passive is compatible with an agent: (28) E ghjembe si sò tronche. The branches clit be.prs.3pl break.ptcp.f.pl ‘The branches have broken (themselves).’ (28′) *E ghjembe si sò tronche da Petru / da u *The branches clit be.prs.3pl break.ptcp.f.pl by Peter / by the ventu. wind *‘The branches have broken by Peter / by the wind.’ (31) E ghjembe sò state tronche. The branches be.prs.3pl be.pass.ptcp.f.pl break.ptcp.f.pl ‘The branches have been broken.’ (31″) E ghjembe sò state tronche da Petru/ The branches be.prs.3pl be.pass.ptcp.f.pl break.ptcp.f.pl by Peter/

Auxiliary selection with intransitive and reflexive verbs

109

da u ventu. by the wind ‘The branches have been broken by Peter / by the wind.’ (20) Petru si hè lavatu. Peter clit be.prs.3sg wash.ptcp.m.sg ‘Peter has had a wash.’ (20′) *Petru si hè lavatu da Petru. *Peter clit be.prs.3sg wash.ptcp.m.sg by Peter *‘Peter has had a wash by Peter.’ (32) Petru hè lavatu. Peter be.prs.3sg wash.ptcp.m.sg ‘Peter is washed.’ (32′) Petru hè lavatu da Petru. Peter be.prs.3sg wash.ptcp.m.sg by Peter ‘Peter is washed by Peter.’ (24) Petru è Paulu si sò azzuffati. Peter and Paul clit be.prs.3pl fight.ptcp.m.pl ‘Peter and Paul fought.’ (24′) *Petru è Paulu si sò azzuffati da Ghjuvanni. *Peter and Paul clit be.prs.3pl fight.ptcp.m.pl by John *‘Peter and Paul fought by John.’ (33) Petru è Paulu sò stati azzuffati. Peter and Paul be.prs.3pl be.pass.ptcp.m.pl fight.ptcp.m.pl ‘Peter and Paul have been beaten.’ (33′) Petru è Paulu sò stati azzuffati da Peter and Paul be.prs.3pl be.pass.ptcp.m.pl fight.ptcp.m.pl by Ghjuvanni. John ‘Peter and Paul have been beaten by John.’ Si is an argument to which the thematic role of Source is assigned (as in View 2b), but the subject is not a Goal argument. I disagree with Views 2a and 2b on

110

Pierre-Don Giancarli

the three following issues and for the following reaons: 1) within the reflexive with a passive meaning a distinction must be made between two levels of Source. 2) Si and the subject are co-referential. It is because si is a Source argument and because si and the subject are co-referential that the subject is not only a Goal argument. Hence my third point: 3) I see si as a Source argument (a deficient argument), but that does not mean the subject is a Goal argument, since the subject has a double status of Goal + Source, due to the co-reference. This double status is the reason for the selection of BE in the Corsican reflexive construction (provided the criteria of valency and orientation do not modify this choice). The subject of a passive construction is a Goal argument, but it is because the subject of a reflexive construction is not simply a Goal argument (unlike View 2b) that the clitic is unacceptable in a passive: (31) E ghjembe sò state tronche. The branches be.prs.3pl be.pass.ptcp.f.pl break.ptcp.f.pl ‘The branches have been broken.’ (31′) *E ghjembe si sò state tronche. *The branches clit be.prs.3pl be.pass.ptcp.f.pl break.ptcp.f.pl In a passive, the subject must be a Goal argument only. Our position can be called View 1c and summarised in the following way: si is a Source argument while the subject (Goal argument) is assigned the complementary thematic role by means of a co-referential relationship with the clitic pronoun41 and therefore a dual status of Goal + Source, hence the selection of BE. I shall provide a few illustrations with commentaries: True reflexive (case 1): (20) Petru si hè lavatu. Peter clit be.prs.3sg wash.ptcp.m.sg ‘Peter has had a wash.’ The clitic pronoun and the subject are co-referential. The clitic is an animate Source argument (Source in the sense of an agent), and the subject is both a Goal argument (Goal in the sense of a patient) intrinsically + a Source argument by means of its co-reference with the clitic. There are two arguments (the

41 For the sake of simplicity I leave out the Corsican variant si hè troncu ghjembe (see Giancarli 2011: 225 ff.).

Auxiliary selection with intransitive and reflexive verbs

111

subject and the clitic) corresponding to one actor. Petru si hè lavatu can be glossed by ‘si has washed Petru’, with a locating relation of identification between the subject and the clitic: Petru (the Goal argument) is washed but is also the washer, and the dual status of the subject (Goal + Source) is the reason for the selection of BE. Ethic reflexive with a monovalent verb (case 3b): (23) Petru si hè fermu quì. Peter clit be.prs.3sg stay.ptcp.m.sg here ‘Peter stayed here.’ There are two arguments (the subject and the clitic pronoun) corresponding to one actor. The clitic and the subject are co-referential. The clitic is a Source argument (Source in the sense of an agent with ghjucà ‘play’, ciuttà ‘dive’, firmà ‘stay’, or a stimulus with esse and stà), and the subject Petru is both a Goal argument (Goal in the sense of beneficiary) intrinsically + a Source argument by means of its co-reference with the clitic. Petru si hè fermu quì can be glossed as ‘Si has stayed for the benefit/to the detriment of Petru’, with a locating relation of identification between Petru and si. The dual status of the subject (Goal + Source) is the reason for the selection of BE. Reciprocal reflexive without an object (case 4a): (24) Petru è Paulu si sò azzuffati. Peter and Paul clit be.prs.3pl fight.ptcp.m.pl ‘Peter and Paul fought.’ The same kind of reasoning as above could be used, but the subject already has a dual status from the start because of the process itself, intrinsically. The dual status of the subject (Goal + Source) is the reason for the selection of BE. Reflexive with a passive meaning, first with a bivalent verb (case 7a): (28) E ghjembe si sò tronche. The branches clit be.prs.3pl break.ptcp.f.pl ‘The branches have broken.’ E ghjembe si sò tronche is close to the passive e ghjembe sò state tronche. However, as I showed above, this gloss is not reliable since a passive makes it possible to express an agent, i.e. a Source argument, whereas a reflexive with a so-called passive meaning is not compatible with the presence of an agent; cf. (28/28′) and (31/31′). The reason, which also confirms my invariant analysis

112

Pierre-Don Giancarli

of the reflexive construction, is that the reflexive construction already contains a Source argument (an exclusively Source argument) marked by si, and therefore cannot take a second one. The difference between the reflexive with a passive meaning and the passive also concerns the subject: it is a Goal argument (an exclusively Goal argument) in a passive, but in a reflexive with a passive meaning like (28c), the subject is not a Goal argument, or more exactly it is not only a Goal argument, since it has a dual status. I would like to elaborate slightly on the dual status of the reflexive with a passive meaning. For this reflexive, I have to make a distinction between two levels of Source: a Primary Source, and a Secondary Source. There is a Primary Source, which is outside the sentence. It is this source that cannot be made explicit by means of an agent introduced by da (‘by’) in the sentence (cf. 28′). As for si, it is a Source argument, but a Secondary one. And the subject is coreferential with it. “Secondary Source” means that the subject does not merely undergo the process (a process whose instigator is an outside argument, i.e. the Primary Source), but is also partly responsible for it. This is because it possesses the necessary qualities for the realisation of the process, and these qualities are involved in the realisation of the change of state. For example when one says (28) e ghjembe si sò tronche the main cause of the breaking of the branches is indeed taken on by a Primary Source, for instance Petru or u ventu, but as a second resort there is a secondary cause of the breaking which is the branches themselves, embodied by si, with which e ghjembe is co-referent. Si embodies the Source dimension obtained by the grammatical subject through the coreference. The branches essentially broke because there was too much pressure on them, but even if the branches did not directly break the branches, e ghjembe si sò tronche means that the breaking of the branches is partly ascribable to the branches, whose solidity was not sufficient. Using Levin and Rappaport Hovav’s terminology with bare intransitives42, one could almost claim that truncassi within a pronominal construction is both an external and internal causation verb, except that here the responsibility is only partial: this Corsican construction has the capacity of ranking the responsibilities between a Primary external Source and a Secondary internal one. An unspecified Primary Source

42 “With an intransitive verb describing an internally caused eventuality, some property inherent to the argument of the verb is responsible for bringing about the eventuality […]. Unlike internally caused verbs, externally caused verbs by their very nature imply the existence of an ‘external cause’ with immediate control over bringing about the eventuality described by the verb: an agent, an instrument, a natural force, or a circumstance” (Levin and Rappaport Hovav 1995: 91–92).

Auxiliary selection with intransitive and reflexive verbs

113

(Petru or u ventu) acts upon ghjembe, which is a Goal argument but also the Secondary Source, because the language takes into account the existence of a Secondary Source that shows some agentivity or rather some responsibility on the part of the subject. This is in line with the interpretation of (28), is coherent with the argument and themactic status I gave of si, and in fine the dual status of the subject (Goal + Source through the co-reference with the Source argument) is the reason for the selection of BE, as it was with reflexive forms other than the passive reflexive and with non-reflexive intransitives. This is also the difference between reflexive (28c) E ghjembe si sò tronche and non-reflexive E ghjembe hanu troncu: (28) E ghjembe si sò tronche. The branches clit be.prs.3pl break.ptcp.f.pl ‘The branches have broken (themselves).’ (34) E ghjembe hanu troncu. The branches have.prs.3pl break.ptcp.m.sg ‘The branches have broken.’ The reflexive construction selects BE because the subject has a dual status, whereas the non-reflexive construction selects HAVE because the subject does not have a dual status and is not a Secondary Source. It is not a Secondary Source because it bears no responsibility for the realisation of the process. Three forms can be compared from the active voice (35): (35) X hà troncu e ghjembe X have.prs.3sg break.ptcp.m.sg the branches ‘X has broken the branches.’

(active meaning and active morphology)

(34) E ghjembe hanu troncu The branches have.prs.3pl break.ptcp.m.sg (*da X) (*by X) ‘The branches have broken (*by X).’

(passive meaning and active morphology)

(31) E ghjembe sò state The branches be.prs.3pl be.pass.ptcp.f.pl tronche (*da X) break.ptcp.f.pl (*by X) ‘The branches have been broken (*by X).’

(passive meaning and passive morphology)

114

Pierre-Don Giancarli

(28) E ghjembe si sò The branches clit be.prs.3pl tronche (*da X) break.ptcp.f.pl (*by X) ‘The branches have broken (*by X).’

(partly passive meaning and reflexive form)

(34) e ghjembe hanu troncu is an example of a reversible verb: it has a passive meaning and active morphology, and it does not licence an external argument because there is no Source. Unlike (28), (34) is a lexical phenomenon limited to certain verbs (for the kinds of verbs possible, see for example Heidinger and Schäfer 2008: 138 ff. or Schäfer 2008: 187 ff.). There is no Source, whether Primary or Secondary, and because of this absence of cause they are called “anticausatives”. Any derivational relationship with a bivalent construction in the active voice is artificial, whether the causative alternant is derived from the anticausative via causativisation (Pesetsky 1995, etc.) or the anticausative lacks an implicit external argument due to a process of detransitivisation (Reinhart 2002, etc.), precisely because with an anticausative the process is presented as spontaneous and agentless (Haspelmath 1993: 90), it expresses an “autonomous event”, “occurring in and of itself (Talmy 2003: 472), without implying that it has a Source. In (31), e ghjembe sò state tronche (da X), there is a passive meaning and passive morphology: there is one single Source, outside the sentence and explicitable in a da-phrase, and (when this Source is explicit) two arguments and two actors. It is a fundamentally bivalent construction, and a place must be reserved for the agent even when it is not mentioned. Finally, in (28) e ghjembe si sò tronche (*da X), which has a partly passive meaning and is a reflexive form, I shall argue there are two Sources: two coreferential arguments and therefore one actor. Two levels of Source with one Primary Source outside the sentence and not mentionable and a Secondary Source (embodied by the clitic pronoun with which the Goal subject is co-referential) responsible for the realisation of the process due to its qualities. The subject of reflexive e ghjembe si sò tronche, but not of non-reflexive e ghjembe hanu troncu, has a dual status. This is why the former, but not the latter, selects BE. Finally, we turn to monovalent unaccusative and unergative verbs: at first glance, the monovalent si travaglia and the monovalent si va could not be more similar, at least in simple (non-reflexive) forms. But as soon as one considers their corresponding compound forms (29) and (30), the difference is clear in the final vowel:

Auxiliary selection with intransitive and reflexive verbs

115

(29) Si hè travagliatu. clit be.prs.3sg work.ptcp.m.sg ‘One has worked.’ (30) Si hè andati. clit be.prs.3sg leave.ptcp.m.pl ‘One has left.’ Within the group of monovalents, two cases must therefore be distinguished, corresponding to what the UH calls unaccusatives and unergatives. Unaccusatives accept a BE reflexive form and a past participle in the plural, while unergatives accept a BE reflexive form (but HAVE is grammatical, for instance si hà travagliatu) but a past participle in the singular. This is why we say si hè travagliatu-masc-sing and not *si hè travagliati-masc-pl. The number cannot vary, nor can the gender (*si hè travagliata-fem-sing, *si hè travagliate-fem-pl). Let us first take a closer look at unaccusatives, like si hè andati: since, in the compound forms, the past participle agrees in the plural while the auxiliary agrees in the singular, the past participle has to agree with a subject which must be plural, too. The question is: where is this subject? I contend that the past participle agrees with si, i.e. si is a subject 43. This why a ghjente (‘people’) can be substituted for it as a subject: si hè andati = a ghjente sò andati. To be more precise, si is a subject that is grammatically singular but semantically plural: grammatically singular to take into account the auxiliary, which is singular, and semantically plural to take into account the past participle, which (with an unaccusative) is in the plural.44 Why is it semantically plural? Because it refers to an indeterminate human animate (the quantifier tutti (‘all’) could be added, obviously in the masculine plural: si ci hè tutti andati ‘we all left’), roughly synonymous with a ghjente (‘people’) or omu in Corsican, or with French on (on y est tous allés). Note a ghjente and si cannot be used together, since they have the same function. An unaccusative reflexive as in si hè andati selects BE, not because the subject (si), which is intrinsically a Goal-argument, has a double status by means of a co-reference with a Source Argument (this would describe the situa-

43 This subject faces a number of constraints listed in footnote 37, all of which also affect the French pronoun on, a subject pronoun. 44 It is a partial syllepsis, since the singular agreement only bears on the auxiliary. This is exactly the situation of the French pronoun on, which takes an auxiliary in the singular but can take a past participle in the plural: On est venus trop tôt ‘One is-SG come-PL too early’. (EAG)

116

Pierre-Don Giancarli

tion of the bivalent reflexive with a passive meaning), but because of the verb itself, qualified as unaccusative by the UH. It does not matter whether the subject is si; what counts is the verb, and the analysis will depend on the verb. For instance si hè andati will be analysed like i nostri crunachisti sò andati: (36) I nostri crunachisti sò andati in Moulinserre é só vultati the our reporters are gone in Moulinserre and are returned cù ste maghjine di felicità (BC) with these images of happiness ‘Our reporters are gone in Moulinsart and they are returned with these images of happiness.’ ‘Our reporters have gone to Moulinsart and they have come back with these images of happiness.’ The subject (si, just like i nostri crunachisti) is an agent-Source and a patientGoal because andà is a verb expressing a middle voice. Consequently the subject has a double status, and the auxiliary BE is selected. Now let us turn to si hè travagliatu, whose auxiliary selection is the only one that escapes all possible explanations within our framework. In si hè travagliatu, si is a subject, just like it is in si hè andati. More precisely, it refers to an indeterminate human animate and is plural. In this respect, there is no difference between the si of unaccusatives and the si of unergatives. With a verb such as travaglià, which selects HAVE in the non-reflexive form (unlike andà in si hè andati), one would not have expected si hè travagliatu but si hà travagliatu (and note that si hà travagliatu is grammatical). One would conclude that si hè andati, with an unaccusative, selects BE, while si hà travagliatu, with an unergative, selects HAVE. But this is not the case. Why does an unergative like travaglià, logically HAVE-selecting in a non-reflexive form, usually select BE in a reflexive form? The selection of BE is not due to a passive: si hè travagliatu is not a passive because travaglià is monovalent while the Corsican passive is limited to bivalent verbs. It is not due to a double status of the subject by means of the verb: with a so-called unergative, the subject si is only a Source. Consequently, no double status can be considered the explanation of the selection of BE, unlike unaccusatives. The explanation I provided for unaccusatives is not extendable to unergatives. It is not due to a double status of the subject by means of a coreference: the explanation I provided for bivalent verbs is not extendable to unergatives (it is not extendable to monovalent verbs in general) since si is not balanced with an intrinsic Goal-argument that would have established a coreference relationship between them. It is not due to the reflexive form in itself

Auxiliary selection with intransitive and reflexive verbs

117

either because in Corsican this form does not require the selection of BE (unlike French or Italian): si hè travagliatu coexists alongside si hà travagliatu. Si hè travagliatu, the form commonly used, is the one least expected: the past participle has no agreement while in the two types of monovalents the subject si refers to an indeterminate human in the plural and BE is selected – in other words, an auxiliary usually going with an agreement. Why is there no agreement? Is it because si is not a subject? We know it is. Is it because the subject si is not a plural? We know it is. The question that arises is then to know why the past participle agrees with si in the case of andà (unaccusative) but not in the case of travaglià (unergative). Our answer to this question is also our answer to the question of why a so-called unergative like travaglià, quite logically HAVE-selecting in a non-reflexive form, can select BE in a reflexive form. To make it short, our answer is: because the BE in si hè travagliatu is not a real BE; it is a BE with the properties of auxiliary HAVE. I think the lack of agreement in si hè travagliatu is revealing because, due to this lack of agreement, si hè travagliatu behaves as if its auxiliary were avè (HAVE), precisely the auxiliary you would expect with an unergative. Nevertheless, the monovalent unergative verb in Corsican is an exception among all the Corsican reflexive constructions. We leave this question open pending further research.

5 Conclusion We started by voicing several criticisms: 1) the problems raised by the ASHS’s gradient representation built on prototypical occurrences, as evident from both the mismatches with some data and the constitution of the gradient itself, 2) the problems raised without prototypical occurrences by scalarity in general because certain syntactic phenomena in context escape any kind of scalar representation and because the nature of the compound form of a specific language is a relevant factor that cannot be ignored and makes cross-linguistic scales problematic, and 3) the unjustified exclusion of reflexive verbs from the field of study of auxiliary selection. This led us to propose a binary vision of auxiliaries that takes into account both simple intransitive verbs and reflexive verbs, one different from the Unaccusative Hypothesis. This vision, not so much quantitative in terms of thresholds but qualitative, aims at giving a specific explanation for each language under consideration regarding auxiliary selection. With regard to Corsican, the following global binary explanation is provided: HAVE is selected when the subject corresponds to a Source macro-role, while BE is selected (in the case of simple BE-verbs) or selectable (in the case of reflexive BE-verbs) when the

118

Pierre-Don Giancarli

subject corresponds to both Source and Goal macro-roles, i.e. it satisfies two thematic roles. The subject of simple BE-verbs has this dual status intrinsically, the subject of reflexive BE-verbs satisfies this dual status either intrinsically or obtains it mostly by means of a co-reference with the clitic pronoun treated not as a valency-reducing marker but as an argument. Reflexive verbs give rise to a second binary distinction: BE is actually selected when, in the presence of a bivalent verb and an explicit object 45, the subject is topicalised, while HAVE is selected when the object is topicalised.

References Aarts, Bas. 2007 Syntactic Gradience: the Nature of Grammatical Indeterminacy. Oxford: Oxford University Press. Alexiadou, Artemis and Elena Anagnostopoulou. 2004 Voice morphology in the causativeinchoative alternation. In: Artemis Alexiadou, Martin Everaert and Elena Anagnostopoulou (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 114–136. Oxford: Oxford University Press. Bentley, Delia and Eyrthorsson Thorhallur. 2003 Auxiliary selection and the semantics of unaccusativity. Lingua 114: 447–471. Benveniste, Émile. 1974 Problèmes de linguistique générale. Paris: Gallimard. Borer, Hagit. 1994 The projection of arguments. In: Elena Benedicto and Jeffrey Runner (eds.), Functional Projections, 19–47. (University of Massachusetts Occasional Papers 17.) Amherst: Graduate Linguistic Student Association. Burzio, Luigi. 1981 Intransitive verbs and Italian auxiliaries. Ph.D. dissertation, MIT. Burzio, Luigi. 1986 Italian Syntax, a Government-Binding Approach. Dordrecht: Reidel. Cennamo, Michela and Antonella Sorace. 2006 Auxiliary selection and split intransitivity in Paduan. In: Raúl Aranovich (ed.), Split Auxiliary Systems: a Cross-Linguistic Perspective, 65–99. (Typological Studies in Language 69.) Amsterdam, Philadelphia: John Benjamins. Culioli, Antoine. 1990 Pour une linguistique de l’énonciation. Tome 1. Paris: Ophrys. Culioli, Antoine. 1991–92 Structuration d’une notion et typologie lexicale. BULAG 17: 7–12. Culioli, Antoine. 1999 Pour une linguistique de l’énonciation. Tome 2. Paris: Ophrys. Dobrovie-Sorin, Carmen. 1998 Impersonal SE constructions in Romance and the passivization of unergatives. Linguistic Inquiry 29: 399–437. Embick, David. 2004 Unaccusative syntax and verbal alternations. In: Artemis Alexiadou, Martin Everaert and Elena Anagnostopoulou (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 137–158. Oxford: Oxford University Press. Giancarli, Pierre-Don. 2011 Les Auxiliaires ÊTRE et AVOIR : étude comparée corse, français, acadien et anglais. Rennes: Presses Universitaires de Rennes. Gundel, Jeanette. 1988 Universals of topic-comment structure. In: Michael Hammond, Edith Moravcsik and Jessica Wirth (eds.), Studies in Linguistic Typology, 209–239. Amsterdam, Philadelphia: John Benjamins.

45 One exception: BE is also selectable with monovalent unergatives.

Auxiliary selection with intransitive and reflexive verbs

119

Gurman Bard, Ellen, Cheryl Frenck-Mestre and Antonella Sorace. 2010 Processing auxiliary selection with Italian intransitive verbs. Linguistics 48: 325–361. Hale, Kenneth and Samuel Keyser. 2002 Prolegomenon to a Theory of Argument Structure. Cambridge, Mass.: MIT Press. Härtl, Holden. 2003 Conceptual and grammatical characteristics of argument alternations. The case of decausative verbs. Linguistics 41: 883–916. Heidinger, Steffen and Florian Schäfer. 2008 The French reflexive passive and anticausative. In: Benjamin Fagard, Sophie Prévost, Bernard Combettes and Olivier Bertrand (eds.), Évolutions en français, Études de linguistique diachronique, 135–152. Bern: Peter Lang. Herslund, Michael. 1996 Det franske sprog [The French Language], Kapitel III, Valens og transitivitet. Forelobig version. Kallulli, Dalina. 2006 On unaccusative morphology and argument realization. Unpublished manuscript, University of Vienna. Kayne, Richard S. 1988 Romance SE / SI. GLOW Newsletter 20: 33. Keller, Frank and Antonella Sorace. 2003 Gradient auxiliary selection in German. Journal of Linguistics 39: 57–108. Legendre, Géraldine. 2007 On the typology of auxiliary selection. Lingua 117: 1522–1540. Legendre, Géraldine and Antonella Sorace. 2003 Auxiliaires et intransitivité en français et dans les langues romanes. Les Langues Romanes, Sciences du Langage, CNRS Editions: 185–233. English version at http://www.cog.jhu.edu/faculty/legendre/index.htm. Levin, Beth and Malka Rappaport Hovav. 1995 Unaccusativity: At the Syntax-Semantics Interface. Cambridge, Mass.: MIT Press. McFadden, Thomas. 2007 Auxiliary selection. Language and Linguistics Compass 1: 674– 708. Perlmutter, David. 1978 Impersonal passives and the Inaccusative Hypothesis. In: Jeri Jaeger and Anthony Woodbury (eds.), Proceedings of the Fourth Annual Meeting of the Berkeley Linguistic Society, 157–189. University of California at Berkeley. Perlmutter, David. 1980 Relational grammar. In: Edith Moravcsik and Jessica Wirth (eds.), Syntax and Semantics 13: Current Approaches to Syntax, 195–229. New York: Academic Press. Pesetsky, David. 1995 Zero Syntax. Cambridge, Mass.: MIT Press. Reinhart, Tanya. 2002 The theta system: An overview. Theoretical Linguistics 28: 229–290. Reinhart, Tanya and Tal Siloni. 2005 The lexicon-syntax parameter. Linguistic Inquiry 36: 389–436. Rizzi, Luigi. 1986 On chain formation. In: Hagit Borer (ed.), Syntax and Semantics. vol. 19. The Syntax of Pronominal Clitics, 65–95. New York: Academic Press. Rosen, Carol. 1984 The interface between semantic roles and initial grammatical relations. In: David Perlmutter and Carol Rosen (eds.), Studies in Relational Grammar 2: 38–77. Chicago: University of Chicago Press. Ruwet, Nicolas. 1988 Les verbes météorologiques et l’hypothèse inaccusative. In: Claire Blanche-Benveniste, André Chervel and Maurice Gross (eds.), Grammaire et Histoire de la Grammaire. Hommage à la mémoire de Jean Stéfanini, 383–402. Université de Provence. Schäfer, Florian. 2008 Middles as anticausatives. In: Emily Efner and Martin Walkow (eds.), Proceedings of NELS 37: 183–196. Amherst: Graduate Linguistic Student Association. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76: 859–890.

120

Pierre-Don Giancarli

Sorace, Antonella. 2004 Gradience at the lexicon-syntax interface. In: Artemis Alexiadou, Martin Everaert and Elena Anagnostopoulou (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 243–268. Oxford: Oxfored University Press. Sorace, Antonella. 2006 Gradedness and optionality in mature and developing grammars. In: Gisbert Fanselow, Caroline Fery, Matthias Schlesewsky and Ralph Vogel (eds.), Gradience in Grammars, 106–123. Oxford: Oxford University Press. Sorace, Antonella. 2011 Gradience in split intransitivity: the end of the unaccusativity hypothesis? Archivio Glottologico Italiano 96: 67–86. Sorace, Antonella and Frank Keller. 2005 Gradience in linguistic data. Lingua 115: 1497– 1524. Talmy, Leonard. 2003 Toward a Cognitive Semantics. Cambridge, Mass.: MIT Press. Tenny, Carol. 1992 The aspectual interface hypothesis. In: Ivan Sag and Anna Szabolsci (eds.), Lexical Matters, 1–27. Stanford: Center for the Study of Language and Information. Van Valin, Robert. 1990 Semantic parameters of split intransitivity. Language 66: 221–260. Van Valin, Robert and Randy Lapolla. 1997 Syntax: Structure, Meaning and Function. Cambridge: Cambridge University Press.

Corpus Acadian: Corpus Péronnet “85”, a corpus of traditional Acadian consisting of 75,000 words recorded in southeast New Brunswick, Canada. French and Corsican: The French and Corsican corpora, which amount to 405,327 and 353,212 words respectively, are composed of the following texts.46 Antoni, Petru. 2000 Detti è fatti. La Marge. (DF) Beckett, Samuel. [1948] 1988 En Attendant Godot. Les Éditions de Minuit. Corsican translation by D. Geronimi 1985 Intantu. In: Rigiru 23 Éditions Cyrnos et Méditerranée. (EAG) Biancarelli, Marcu. 2001 San Ghjuvanni in Patmos. Albiana. (SGP) Collodi, Carlo. [1883] 2002 Les aventures de Pinocchio. Corsican translation by G. Fusina 2001 Pinochju. Éditions Stamperia Sammarcelli. (P) Daudet, Alphonse. [1873] 2001 Contes du lundi. Gallimard. Corsican translation by M. Ceccaldi 1980 Fole di u luni. Klincksieck. (CDL) Delmon-Casanova, Jean-Luc. 2002 Heresia. DCL Éditions. Corsican translation by S. Casta. (H) Giacomo-Marcellesi, Mathée. 1989 Contra Salvatica. Edisud. (CS) Hergé. 1963 Les bijoux de la Castafiore. Casterman. Corsican translation by F.-M. Perfettini 1994 I ghjuvelli di a Castafiore. Casterman. (BC) Saint-Exupery, Antoine de. [1943] 1990 Le Petit Prince. Gallimard. Corsican translation by S. Casta 1990 U Principellu. Akenaton & Squadra di u Finusellu. (PP) Thiers, Jacques. 1990 A Funtana d’Altea, Albiana. (AFA) Thiers, Jacques. 1996 A barca di a Madonna. Albiana. (ABM)

46 The corresponding abbreviations are given in brackets.

Section 2: Between constructional variation and auxiliary selection

Artemis Alexiadou, Universität Stuttgart

On the irrealis effect on auxiliary selection

1

1 Introduction Across languages, it has been argued that auxiliary selection is sensitive to modality: (1) If a language had a choice between HAVE vs. BE as a perfect auxiliary, in modal contexts HAVE replaced BE; the switch is unidirectional and is from BE to HAVE. This effect has been labelled in the literature the irrealis effect on auxiliary selection (see Shannon 1995), and has been reported to hold for (the history of) Germanic (e.g. English, Scandinavian, Middle Dutch and Middle Low German) and Romance (e.g. Spanish, Old French, Old Neapolitan). The phenomenon has been described as relating to split intransitivity, and is (often) accompanied by a switch to HAVE as the perfect auxiliary in other contexts as well. Two questions arise and will be dealt with in this contribution. First, is this really an issue about BE vs. HAVE or a more general issue concerning competition between two “perfect” forms and how these fit the semantics of modal/ irrealis contexts? Second, does this switch lead to auxiliary selection loss? My answer to the first question will be that it is not about HAVE replacing BE; it is more about “competition” between two forms used in relationship to what we call the perfect. These could be BE vs. HAVE or analytic vs. synthetic, etc., as illustrated in (2a–b). Pattern A characterises languages such as English and Swedish, while evidence for Pattern B comes from the history of the Greek perfect: (2) a. Pattern A: HAVE vs. BE b. Pattern B: Synthetic vs. Analytic (BE + periphrasis)

English/Swedish Earlier Greek

At the core of the switch in (1) is the grammaticalisation path of the perfect (Harris 1984; Bybee and Dahl 1989; Elsness 1997 among others). These authors 1 I am indebted to two anonymous reviewers and the editors of this volume for their comments and suggestions. My research was supported by a Deutsche Forschungsgemeinschaft grant to the project B6 “Underspecification in voice systems and the syntax-morphology interface” of the Collaborative Research Center 732 “Incremental Specification in Context” at the Universität Stuttgart.

124

Artemis Alexiadou

have established that there are three stages in perfect development, illustrated in (3): (3) Stage I: resultative Stage II: anterior/experiential Stage III: simple past It is a well-established fact that perfects across languages begin as resultatives, stage (I). On the basis of the results in McFadden and Alexiadou (2010), I argue that the irrealis effect is observed when one of the variants in (2) remains a resultative construction and the other one grammaticalises into an experiential perfect. If both variants of A and B become experiential perfects, then most likely the one will not replace the other. If only one of them does, then this form will win the competition. This is in line with the view that what we call the perfect is neither a primitive of the theory nor a category that has a single precise definition in terms of such primitives. As McFadden and Alexiadou (2010) stress, following Iatridou, Anagnostopoulou, and Pancheva (2003), the perfect is not a simple category with a universal definition and consistent properties; rather, perfect is a cover term for a wide range of complex constructions that share a similar makeup. What unifies perfects is that they involve multiple pieces related to syntacticosemantic levels of tense, aspect and potentially even Aktionsart, and convey either explicit or implicit anteriority. What distinguishes among the various types of perfect is the presence or absence of specific pieces, the precise specification and interpretation of those pieces, and how the syntacticosemantic pieces map onto morphosyntactic ones. Turning to the second question, my answer to this is that the grammaticalisation path can also help us explain why loss of auxiliary selection, where BE was replaced by HAVE in all contexts, is observed in some but not all the languages that have been argued to show an irrealis effect, e.g. English vs. German. Crucially, if only HAVE perfects grammaticalise into experiential perfects in a language, then these will become the only form available to express the perfect in this language. Evidence for this comes from the development of the German perfect, as described in Gillmann (2011). The paper is structured as follows. In sections 2 and 3, I document the irrealis effect for a number of languages, namely English, Swedish, German, Dutch and Spanish. In section 4, I show that this is not merely a matter of BE vs. HAVE. Evidence will come from the behaviour of perfects in Earlier Greek. In section 5, I discuss the formal analysis of the perfect and in section 6, I turn to the German perfect whose development will help us understand why auxilia-

On the irrealis effect on auxiliary selection

125

ry selection loss took place in some languages but not in others. Section 7 summarises and concludes.

2 The irrealis effect on auxiliary selection in English

2 ,3

2.1 Summary of the change Earlier English had periphrastic constructions consisting of the past participle of the main verb plus an auxiliary have or be, as in (4), a pattern similar to the one found in Present-Day German, Dutch, French and Italian. (4) a. I am come as ƺe bade me. ‘I have come as you asked me.’ (Mirk,75.2015)

(be periphrasis)

b. he haþe foghten wyth þe fend. ‘He has fought with the enemy.’ (Mirk,116.3171)

(have periphrasis)

2 All data are taken from McFadden and Alexiadou (2010). Following the practice in this paper, Earlier English is used as a cover term for Old English (OE, to 1150 CE), Middle English (ME, from 1150 to 1500) and Early Modern English (EModE, from 1500 to 1710), and English as it is currently spoken is referred to as Present-Day English (PDE). McFadden and Alexiadou point out that the changes in the perfect system were completed as recently as 1900, thus it would not be sufficiently restrictive to speak of Modern English (which could go back as far as 1500) or even Late Modern English (which would include the 18 th and 19 th centuries). The data discussed in McFadden and Alexiadou come from the York-Toronto-Helsinki Parsed Corpus of Old English Prose (Taylor, Warner, Pintzuk, and Beths 2003), the Penn-Helsinki Parsed Corpus of Middle English, 2 nd edition (Kroch and Taylor 1999) and the Penn-Helsinki Parsed Corpus of Early Modern English (Kroch, Santorini, and Delfs 2005), and their dating follows that of the corpora. The final line in each example gives an abbreviated label for the source text followed by the sentence ID from the corpus file. Complete information on the texts from which examples have been taken is given in McFadden and Alexiadou (2010). 3 As is well known, the present perfect in English has four main uses (see Iatridou, Anagnostopoulou, and Pancheva 2003 for discussion and references): a) the universal perfect conveys the meaning that the predicate holds throughout some interval stretching from a certain point in the past up to the present; b) the experiential perfect asserts that the subject has a certain experience and can be formed from an underlying eventuality of any Aktionsart, unlike the universal perfect that requires an unbounded eventuality; c) the perfect of result which is said to be possible only with telic predicates and only for as long as the effect of the underlying eventuality holds. It infers that the target state holds at the moment of utterance; d) the perfect of recent past that reports an eventuality that has just happened.

126

Artemis Alexiadou

However, as is well known, English subsequently diverges from these other languages in losing the version with BE. In Present-Day English (PDE), have is used to form perfects with all verbs. (5) a. I have/*am come as you asked me. b. He has fought with the enemy. As with other Indo-European languages, both the periphrasis with have and that with be had their pre-OE origins in stative resultatives – the former with transitives, the latter with intransitives. (6) a. I have my bags packed. (i.e., ‘I have my bags in a state of having been packed.’) b. My bags are arrived. (i.e., ‘My bags are in the state of having arrived.’) In ME, around 1350, have began to replace be with unaccusatives. This process was gradual, involving a period of variation lasting several hundred years, where the relevant verbs could appear with either have or be. The former appeared earliest and most consistently in modal and irrealis contexts, past and infinitive perfects, and clauses with iterative or durative semantics. From this point onward, be lost more and more ground to have, finally disappearing by the end of the nineteenth century. McFadden and Alexiadou (2010), based on an analysis of data covering the OE, ME, and EModE periods, showed that: a) The periphrasis with be – unlike that with have – did not change significantly from OE to EModE. It remained a stative resultative construction with a restricted interpretation and correspondingly restricted use. b) The have periphrasis, on the other hand, developed by the end of the ME period the full range of interpretations characteristic of the PDE perfect. The change in the have periphrasis from a resultative to a more general perfect came around 1350. At this point, have started showing up with verbs that had previously only appeared with be—but it was not replacing be in the perfectof-result contexts with these verbs. Rather, it spread into experiential perfect contexts that had been the domain of the morphosyntactic simple past in OE and early ME. The curious restrictions on the use of be in late ME and EModE can all be reduced to its perfect-of-result interpretation and resulting incompatibility with contexts that would require an experiential perfect.

On the irrealis effect on auxiliary selection

127

2.2 Restriction on the be-periphrasis The strongest restriction on the be periphrasis in Earlier English comes from modal and/or irrealis contexts. More precisely: the relevant contexts are past counterfactuals, see (7): (7) a. and if they had come sooner, they could haue holpen them. ‘And if they had come sooner, they could have helped them.’ (Giff,G3V.246) b. he had never come to himself … if he had not met with this allay. ‘He would never have come to himself … if he had not met with this distraction.’ (Behn,189.165) c. I am satisfy’d … else I had not come to Town at all. ‘I am satisfied … otherwise I wouldn’t have come to town at all.’ (Vanbr,32.10,11) d. And he … will wish he had with the poore peoples children gon barefoot. ‘And he … will wish he had gone barefoot with the poor people’s children.’ (Locke,35.46) The effect of such past counterfactuals on the choice of auxiliaries in ME and EModE is nearly categorical. Even verbs like come, which always took be in OE and early ME, are forced to take have in these contexts.

Tab. 1: Middle English auxiliary frequency by modality (suitably restricted).

Past counterfactuals All other intransitives

be

have

% be

3 535

59 68

4.8 88.7

Tab. 2: Early Modern English auxiliary frequency by modality (suitably restricted).

Past counterfactuals All other intransitives

be

have

% be

6 984

99 365

5.7 72.9

128

Artemis Alexiadou

This shows that the periphrasis with be was distributed differently than that with have in late ME and EModE. The reason for this is that be and have periphrases differed semantically in terms of their temporal-aspectual interpretations. As argued in detail in McFadden and Alexiadou (2010), the be periphrasis is similar to the German stative passive. I come back to this in section 6. (8) summarises the distribution of auxiliaries in late ME and EModE periphrases: (8) The distribution of auxiliaries in late ME and EModE periphrases Be only forms perfects of result where the result state holds of the subject. Have appears in all experiential perfects and in perfects of result where the result state holds of something other than the subject. Returning to the contexts where the be periphrasis is restricted, they are precisely those where a perfect-of-result interpretation would be ruled out or strongly dispreferred in favour of an experiential perfect. Duratives (for example) are atelic, describing a process rather than its result. (9) þei han gon all aboute the cytee. they have gone all about the city ‘They have walked all around the city.’ (ManTr,117.2859) An atelic eventuality does not yield a good result state, so these are clearly existential perfects, predicted to use have; see footnote 3. A similar path has been observed in other languages as well, and I will briefly go over some cases here. For each language discussed, first the irrealis effect will be described, and then a comment will be offered as to whether or not auxiliary loss can be observed in this language in order to establish a possible relationship between the two facts, i.e. irrealis effect and HAVE replacing BE.

3 Beyond English 3.1 Other Germanic languages 3.1.1 Old Swedish Larsson (2009) points out that in Swedish texts before the 17 th century, a majority of the examples of unaccusatives embedded under HAVE are used in counterfactuals. A suggestive example is offered in (10):

On the irrealis effect on auxiliary selection

129

(10) Hade ock iagh, tå dödt medh them, / Tå hade ey skeedt, had also I then died with them / then had not happened thet wij nu seem. that we now see ‘If also I had died with them then, it would not have happened, what we now see.’ (Petri Tobie *1493: 404; from Johannisson 1945: 117) Larsson argues that actually the perfect with BE is perfect-like, not a “real” perfect. Her description suggests that it is actually quite similar to the perfect in Earlier English. As in English, in Swedish HAVE developed into a perfect auxiliary. I will not discuss Swedish further here. I assume, however, that its development is very similar to that of English.

3.1.2 Middle Dutch Shannon (1995, citing Kern 1912) notes that there is a strong tendency for unaccusative verbs to appear under HAVE in irrealis contexts, as shown in (11): (11) a. ne hadde sijn spere niet ghebroken, Pollidamas ne hadde nemmer gesproken. ‘If his spear had not broken, Pollidamas would never have spoken.’ b. daer brac dat pert inden ijs, so dat hi bi na verdrenckt hadde. ‘Then the horse broke into the ice, so that he almost would have drowned.’ c. daerom, seyd Jhesus, ist ghesciet, want waer mijn rijc van hier beneden, mijn dienres hadder voer ghestreden dat mi dit niet en had gesciet. [270] ‘Therefore, said Jesus, it has happened, because were my kingdom from here below, my servants would have fought for it.’ d. veel luden sijn ghevallen … die niet ghevallen souden hebben dan … ‘Many people have fallen … who would not have fallen but …’ As is well known, HAVE did not replace BE in Dutch. I will come back to this issue in section 6.

3.1.3 Middle Low German Shannon (1995, citing Magnusson 1939) observes that the same influence of irrealis contexts on perfect auxiliary selection can be found in Middle Low German, see (12) and (13):

130

Artemis Alexiadou

(12) he … wolde ghegan hebben to hilghen steden. ‘He … wanted to have gone to holy places.’ (13) dat ik gerne tho Lübeke hedde ghekomen. ‘That I would like to have come to Lübeck.’ In German, as in Dutch, HAVE did not develop into the perfect auxiliary. Gillmann (2011) reports that the German BE perfects, similar to German HAVE perfects, arose out a resultative construction, a pattern familiar from the description of English. The question that arises is why the BE construction grammaticalised into an experiential perfect in German, but not in English. I will discuss this issue in some detail in section 6.

3.2 Romance: Spanish 4 Mateu (2009) notes that Old Spanish had auxiliary selection of the familiar type, i.e. BE occurs with unaccusatives while HAVE occurs with unergatives (14): (14) a. El Rey le dixo que antes allí avía descansado con mucho placer The king him told that before there had rested with much pleasure b. (…) digo que su Rey es muerto no en la batalla (…) sino en su cama, say-I that their king is died not in the battle (…) but in his bed de enfermedad natural from illness natural ‘I say that their king has not died in battle, but in his bed because of natural illness.’ Stolova (2006) reports that HAVE seems to have replaced BE in irrealis contexts: Tab. 3: (her Table 1) BE and HAVE with intransitive compound pasts. Total number of tokens

Number of tokens with BE

Number of tokens with HAVE

Ratio of BE-based to HAVE-based

882

584

298

1.96 (roughly 2)

4 Stolova, citing Nordhal (1977) and Togeby (1974), describes a similar situation for Old French. However, as an anonymous reviewer correctly points out, these sources discuss the behaviour of a single verb, namely aller ‘go’, and it is not clear to what extent one can draw a more general conclusion out of this. Thus I leave French out of my discussion here.

On the irrealis effect on auxiliary selection

131

Tab. 4: (her Table 2) BE and HAVE with intransitive compound pasts marked as [+irrealis] or [+negation]. Total number of tokens

Number of tokens with BE

Number of tokens with HAVE

Ratio of BE-based to HAVE-based

153

77

76

1.01 (roughly 1)

Stolova states explicitly that the comparison between the ratio of the BE-based constructions to the HAVE-based ones from her Table 1 (i.e., roughly 2) and the ratio of the BE-based constructions to the HAVE-based ones from her Table 2 (i.e., roughly 1) indicates that the rate at which the intransitive compound pasts marked as [+irrealis] or [+negation] take the auxiliary HAVE is almost twice as high as the average. This finding suggests that modality should be considered a factor when it comes to the Medieval Spanish split intransitivity. As is well known, in Spanish HAVE replaced BE. The literature on Spanish that I consulted, namely Mateu (2009) and his sources, does not discuss whether BE perfects had a resultative interpretation or not, but since this a general tendency I assume that they did. Sánchez-Marco (2012) studied texts from the 12 th to the 20 th century that contain more than 40 million words, and presents evidence that the extension of haber as the only perfect auxiliary took place already in the earliest centuries in a gradual and orderly way, in that it spreads from some types of predicates to others and it is constrained by certain semantic properties. In particular, the replacement of ser by haber takes place first through dynamic atelic eventualities like caminar ‘walk’, llorar ‘cry’ or correr ‘run’. This pattern of development is reminiscent of what we have seen in the previous section for English, where dynamic atelic predicates take HAVE and not BE. I will not discuss the Spanish situation in any further detail; see Rosemeyer (this volume) for some discussion on the conservation of BE across text genres.

3.3 Summary To summarise, across languages, we seem to have support for (1): (1) If a language had a choice between HAVE vs. BE as a perfect auxiliary, in modal contexts HAVE replaced BE; the switch is unidirectional and is from BE to HAVE. In some languages, but not in others, this was followed by auxiliary selection loss. This leads two questions. First, is a matter of BE vs. HAVE or a more

132

Artemis Alexiadou

general fact about the perfect forms in competition? Second, why is it the case that only in some languages the one pattern gets lost? The hypothesis I will pursue is that it is not about BE vs. HAVE, but about resultative perfects. In particular, it seems to me that there are two things going on. First, if a language has two variants for perfect formation, replacement in irrealis contexts takes place if the variant that is replaced is a stative resultative; this is independent of the HAVE vs. BE issue. This is in my view the first step in explaining the behaviour of perfects across languages. In a second phase, and if one of the variants grammaticalises into an experiential perfect but the other one does not, loss of the one variant as a structure than can be used to express perfect semantics is possible. The evidence for this is presented in next section and comes from the history of the Greek perfect. My discussion below builds on and owes a lot to Bentein’s (2012) discussion.

4 Evidence for Step 1: it is about how the pieces combine There are two forms of perfect formation in earlier stages of Greek (beginning with Homeric Greek, see below): a synthetic and a periphrastic one. Both started out as resultatives, but then both extended to environments that are characterised as experiential perfects in English. This has been observed for the synthetic perfect in Gerö and von Stechow (2003) and references therein. Bentein (2012) argues that a similar development can be observed for the analytic perfect as well. It is precisely this development that shows the complex relationship between irrealis contexts and resultative perfects.

4.1 Notes on the formation of the perfect in the history of Greek Let us briefly consider some historical facts about perfect formation in the history of Greek. In Archaic/Homeric Greek (8 th century BC), we find both a synthetic and an analytic form. The synthetic form is built as follows: verb roots reduplicate in the perfect by taking the first consonant plus epsilon as a prefix. Verbs that begin with vowels, or some consonant clusters, add only the epsilon. Examples of reduplication are given in (15) from McCall (1999). The verbal form then inflects for person and number: (15) Reduplication for 1 st person base: λοιπ- [loip-] ‘leave’, perfect is λελοιπα [leloipa] base: αγ- [ag-] ‘lead’, perfect is εαγμαι [eaŋmai] or ηγμαι [e:ŋmai] with vowel coalescence

On the irrealis effect on auxiliary selection

133

The analytic variant was formed on the basis of the auxiliary BE and the perfect participle, which bears a medio-passive ending, the affix -men-. The resultative is typically associated with the passive voice. The participle inflects for number, person and gender, and also contains reduplication. (16) mala gar pepnymenos esti. very for breathe-ppp. is ‘For he is very wise indeed.’ McCall (1999) observes that in Classical Greek (5 th century BC) there was an alternation between the synthetic and the periphrastic form that was phonologically conditioned, i.e. in the beginning the periphrastic form occurred in thirdperson plural only, and then spread to other persons and to the singular. A summary of the distribution of the periphrastic perfect across persons and numbers is offered in Table 5 from Bentein (2012: 16, 17): Tab. 5: Distribution of the periphrastic perfect (person ⁄ number). Author

Dates

Total

1SG

1PL

2SG

2PL

3SG

3PL

Aeschylus Sophocles Herodotus Euripides Thucydides Aristophanes Lysias

c. 525 ⁄ 4–456 ⁄ 5 BC c. 496 ⁄ 5–406 BC c. 485–424 BC c. 485 ⁄ 0–406 BC c. 460 ⁄ 55–c. 400 BC c. 460–386 BC ?459 ⁄ 8–c. 380 BC

16 14 71 17 31 20 33

0 2 0 1 0 1 2

1 0 0 2 0 0 1

1 0 0 2 0 0 1

0 0 0 0 0 1 6

12 8 29 7 8 12 6

2 4 42 5 23 6 16

As Bentein notes, the person⁄number endings of the medio-passive synthetic perfect and pluperfect are used in the other tenses as well, but there they are attached to the verb stem with the help of an intervening vowel (e and o in the present⁄imperfect/future and a in the aorist). What is important is the fact that in the 3PL of consonant-final root verbs such as trepo, the endings -ntai/-nto became -atai⁄-ato (see Table 6). This is generally assumed to be the regular result of the vocalisation of the nasal phoneme in interconsonantal contexts (Smyth 1984 [1920]: 16, 155), so what would have resulted in a consonantal accumulation is avoided (*tetrapntai ). Jannaris (1897: 91) actually claimed that Greek speakers had an aversion towards such clusters, and Smyth (1984) offers further discussion for other euphonic changes that affected the trepo paradigm. By the end of the 5 th century, the use of the periphrasis had become the only possibility with consonant final roots.

134

Artemis Alexiadou

Tab. 6: Medio-passive perfect paradigms of the verbs timao and trepo.

1SG 2SG 3SG 1PL 2PL 3PL

timao

trepo

tetimemai tetimesai tetimetai tetimemetha tetimesthe tetimentai

tetrammai tetrapsai tetraptai tetrammetha tetraphthe tetraphatai

4.2 The development of BE + participle in the history of Greek As Bentein (2012) correctly notes, most work on the history of the perfect has focused on the synthetic perfect, where the consensus is that the resultative use of the perfect is more or less the only meaning of this tense in Homeric Greek. During the classical period, the Greek synthetic perfect often yields a resultative reading. However, as Gerö and von Stechow (2003) show after closer inspection of the classical texts, one finds practically all the different readings that are commonly associated with e.g. the English perfect. What Gerö and von Stechow show is that the Greek synthetic perfect has the readings of the Present-Day English perfect. Bentein (2012) demonstrates that the same development is observed for the BE + participle periphrasis. In particular, he first shows that in Archaic/Homeric Greek the BE construction is undoubtedly resultative. During the Classical Greek period (5 th century BC), however, the BE periphrasis spreads, and while most of the examples receive a resultative interpretation, examples can be found where the construction has a meaning similar to the experiential perfect: it is found with atelic predicates and can be accompanied by agentive phrases, as shown in (17): (17) engelthe autoi hoti Megara aphest eke kai it: was: reported to: him that Megara it: has: revolted and Peloponn esioi mellousin esbalein es t en Attik Peloponnesians they: are: about: to to: invade into the Attica en kai hoi phrouroi Ath enaion diephtharmenoi eisin hupo Megareon and the garrison of: Athenians slaughtered they: are by Megarians ‘it was reported to him that Megara had likewise revolted, that the Peloponnesians were on the point of invading Attica, and that the Athenian garrison had been slaughtered by the Megarians.’ (Thuc. 1.114.1)

135

On the irrealis effect on auxiliary selection

During the 4 th century BC, the use of the BE periphrasis becomes more general, with both telic and atelic verbs. Interestingly, the periphrasis is more regularly used in the subjunctive and especially optative mood, contrary to what seems to be the case with the synthetic perfect (see Table 7): Tab. 7: Distribution of the periphrastic perfect (mood). Author

Dates

Homer

8 th cent. BC?

Aeschylus Sophocles Herodotus Euripides Thucydides Aristophanes

c. 525 ⁄ 4–456 ⁄5 BC c. 496 ⁄ 5–406 BC c. 485–424 BC c. 485 ⁄ 0–406 BC c. 460 ⁄ 55–c. 400 BC c. 460–386 BC

Lysias Isocrates Xenophon Plato Isaeus Aeschines Demosthenes Aristotle

?459 ⁄ 8–c. 380 BC 436–338 BC c. 430–354 BC c. 429–347 BC c. 420–340s BC c. 397–c. 322 BC 384–322 BC 384–322 BC

Total

IMP

IND

INF

OPT

PART SUBJ

38

2

26

6

4

0

0

16 14 71 17 31 20

0 0 0 0 0 0

16 11 56 17 27 18

0 0 0 0 0 0

0 2 7 0 4 0

0 0 0 0 0 0

0 1 8 0 0 2

33 39 138 131 27 9 120 56

0 0 0 4 0 0 0 4

30 32 85 60 21 7 73 37

1 0 2 2 0 0 17 0

2 4 29 47 5 2 27 8

0 0 0 0 0 0 0 0

0 3 22 18 1 0 19 7

Key. IMP = imperative; IND = indicative; INF = infinitive; OPT = optative; PART = participle; SUBJ = subjunctive. (Bentein 2012: 30).

It is not clear whether an aoristic character became available, and the evidence cited is inconclusive. Importantly, however, the synthetic perfect got lost some time between Classical Greek and late Koine Greek (Koine Greek dates between 300BC and 300AD), so that forms such as léluka ‘I have untied’ became obsolete relatively early on in the post-Classical period. The BE + participle periphrasis is the only true perfect form through the medieval period (see also Gerö and von Stechow 2003). A new periphrastic perfect form emerges during the medieval/Byzantine Greek period (roughly between 300 AD and 1453 AD). This developed out of an earlier have future/conditional tense that took an aorist infinitive as its complemennt (i.e. HAVE + aorist infinitive competes with BE + Participle). The completion of the grammaticalisation of this form takes place towards the 19 th century (see Horrocks 1997). Nowadays, Standard Modern Greek uses only the HAVE perfect to express the meaning of the experiential perfect, and there is a synthetic past tense with

136

Artemis Alexiadou

distinct semantics. The BE periphrasis seems to be mostly resultative again. This, however, is not the case across Greek dialects and we find two patterns of variation. As Ralli (2006) notes, in the Tsakonian dialect all tenses are built on the basis of a BE periphrasis, while in Cypriot Greek and Greek dialects spoken in Salento, it is the simple past tense that extended to perfect semantics, and all (HAVE + BE) periphrastic forms are resultative. To summarise, the evidence provided by the Greek development of the perfect suggests that the irrealis effect is observed with forms that can grammaticalise into experiential perfects, and is independent of the actual form/pattern used. This answers the first question raised in this paper. In the next section, I turn to the formal analysis of the perfect. Here, I will contrast the syntax and semantics of experiential perfect to that of the resultative construction.

5 A formal analysis of the perfect 5.1 Structural details of the experiential perfect I will follow McFadden and Alexiadou (2010) and assume that the following building blocks are present to yield experiential perfect semantics. As shown in (18), there is a functional head Perf below T (see, e.g., Iatridou, Anagnostopoulou, and Pancheva 2003). This is the source of the anteriority semantics. (18)

This Perf head has a clear semantic contribution. Following von Stechow (1999), McFadden and Alexiadou (2010) assume a version of the Extended Now (XN) Theory. Rather than expressing a simple relation between the reference and event times, the perfect introduces its own interval – the XN – which has the reference time as its right edge and extends to some contextually specified

On the irrealis effect on auxiliary selection

137

time anterior to the reference time. The eventuality is then situated within the XN, with temporal and quantificational adverbials determining exactly where. The Perf head is what is responsible for creating an XN interval extending into the past and is spelled out as the perfect auxiliary. The T head situates the ending point of that interval relative to the speech time and is spelled out as the finite tense marking. The ending point of the XN is thus analogous to the Reichenbachian reference time. The most important point is that the periphrasis contains an element (which we identify as Perf) expressing anteriority to the reference time. The anteriority arises here from the fact that the perfect XN extends into the past from its T-anchored ending point.

5.2 Structural details of resultatives The structure of (stative) resultative constructions is different (see, e.g., Kratzer 2000; Anagnostopoulou 2003; Embick 2004; Alexiadou and Anagnostopoulou 2008). (19)

In (19), the auxiliary be is nothing more nor less than the normal copula that appears with predicate adjectives and nouns. The AspR head (from Embick 2004; R for resultative) is spelled out as the participial morphology and produces a state that is the result of an event (see von Stechow 1998; Kratzer 2000, for proposals on the semantic details). Because of this, it needs as its complement an eventuality that can produce a result state, a requirement that iteratives, duratives and atelic predicates in general do not meet. Hence, they are not eligible to appear in this structure and do not show up in the be periphrasis unless they are somehow coerced to yield a result state (potentially by the addition of some adverbial element that supplies its own result state). Transitives and unergatives are ruled out by an incompatibility between AspR and the voice head that introduces external arguments. The output of AspR is stative, and thus the wrong input for agentive voice. Voice cannot ap-

138

Artemis Alexiadou

pear above AspR. Similarly, the output of voice (an event) is the wrong input for AspR (which wants an event plus a state), so a structure with AspR above voice is equally impossible.

5.3 Explaining the counterfactual effect In languages like English counterfactuality is encoded by finite past tense morphology, which McFadden and Alexiadou (2010) assume spells out material in T. Crucially, a clause that is formally a simple past can in principle be interpreted as past or as counterfactual – but not as both. In order to get a past counterfactual meaning, some additional morphosyntactic material is necessary. What we get is formally a pluperfect, as in (20b). (20) a. If I owned a car, I would drive to Vegas. b. If I had owned a car, I would have driven to Vegas. In such instances, the T material spelled out as past morphology on the auxiliary supplies the counterfactuality, while the perfect morphosyntax yields the “pastness”. This works because the have periphrasis has an interpretation that, although distinct from the simple past, does involve anteriority. What we refer to as “past counterfactuals” are thus more properly counterfactuals of anterior perfects. The structure for the resultative structure, however, has no source for “pastness” or anteriority beyond a past T. But on the assumption that T only occurs once per clause, we get either past or counterfactual, not both. If the latter, then in strictly compositional fashion the structure produces the counterfactual of a result state reading that we saw. The above discussion and structures lead to the conclusion that if resultatives and experiential perfects have distinct structures, and distinct semantics, (19) is crucially “incompatible” with counterfactual contexts, and only (18) is licit. Thus in counterfactual environments, languages will use the one pattern that yields experiential perfect semantics, whatever form (2a vs. 2b) this may have. If both patterns grammaticalise into experiential perfects, then we do not expect an irrealis effect on the choice of a particular perfect form, as is the case in German. I turn to this issue in the next section.

6 The German perfect From a synchronic perspective, we can argue that the structure in (18) is adopted for both HAVE and BE perfects in Modern German (see e.g. Rathert 2003).

On the irrealis effect on auxiliary selection

139

The question that arises is why HAVE did not replace BE in German, since at least during some stage in its history, documented in section 3, it seemed to be behaving similar to English. The answer to this is, as suggested above, that this did not happen as both periphrases shifted to experiential perfects. As Sapp (2011) shows in detail, in Early New High German mood plays no role and both HAVE and BE perfects are licit in these contexts (see Table 8): Tab. 8: Analysis of auxiliary choice in EMHG by mood, Sapp (2011: 41). Factor group

Factor

sein

haben

Factor weight

mood p = 0.074 Total

indicative subjunctive

48 (77.4 %) 17 (94.4 %) 65 (81.2 %)

14 (22.6 %) 1 (5.6 %) 15 (18.8 %)

0.553 0.324

The above picture is consistent with an analysis according to which the BE perfect in German was always a perfect similar to the HAVE perfect. In German, it is the adjectival passive that shows the restrictions observed with BE perfects in English, i.e. it is out in e.g. durative contexts. As acknowledged in the literature, it is this construction that has the semantics of the perfect of result. This is illustrated in (21) below, where we see that the German stative passive is out in durative contexts that happily co-occur with experiential (HAVE and BE) perfects: (21) Duratives a. Seitdem hat Erosion die Festung immer weiter zerstört. since has erosion the fort ever further destroyed (haben perfect) ‘Since then, erosion has destroyed the fort more and more.’ b. Seitdem ist die Festung immer weiter verfallen. since is the fort ever further decayed ‘Since then, the fort has decayed more and more.’

(sein perfect)

c. ?*Seitdem ist die Festung immer weiter zerstört. since is the fort ever further destroyed (stative passive) Intended: ‘Since then, the fort has been destroyed more and more.’ (data from McFadden and Alexiadou 2010) The most revealing parallel comes, however, from the interpretation of counterfactuals. As McFadden and Alexiadou (2010) showed, if we take either German perfect and put its auxiliary in the past subjunctive, we get a past counterfactual, conveying that a proposition was contrary to fact at a time in the past,

140

Artemis Alexiadou

as in (22a) and (22b). When we take a stative passive and put its auxiliary in the past subjunctive, on the other hand, what we get is a contrary-to-fact present state that is the result of a past event: a present counterfactual. (22) Counterfactuals a. Wenn er gearbeitet hätte … if he worked had.sbj ‘If he had worked …’ b. Wenn er angekommen wäre … if he arrived were.sbj ‘If he had arrived … ’ c. Wenn er geheilt wäre … if he healed were.sbj ‘If he were (in the state of having been) healed … ’ (data from McFadden and Alexiadou 2010) A stative resultative construction is impossible in precisely the contexts where the Earlier English be periphrasis did not appear. General perfects, however, work just fine in these contexts, whether with HAVE or with BE. Gillmann (2011) shows convincingly that in German BE perfects have been extended to atelic verbs, a state of affairs that did not take place in English. Her discussion substantiates the point that in German both HAVE and BE perfects grammaticalised into experiential perfects, and a similar development also took place in Dutch. Dutch, as she points out, went even a step further and can now form perfects of even transitive verbs with zijn, something that is not possible in German. Gillmann also shows that in German HAVE and BE perfects are similar semantically, while a semantic distinction exists between BE perfects and stative passives. In fact, in modern German, it is rather uncontroversial that BE perfects and resultatives are distinct semantically, and I briefly revisit here some of the evidence put forth in favour of this distinction, based on Löbner (2002). One difference concerns the distribution of event-related adverbials. With resultatives, although the situation expressed in (23) logically arises from her having gone away some time before, this event is not a component of the situation referred to. Accordingly, it cannot be further qualified with adverbials that relate to the event itself; these select the perfect reading(s): (23) Sie ist gestern / eilig / heimlich verreist. she is yesterday / in a hurry / secretly gone away ‘She has gone away/went away yesterday/in a hurry/secretly.’ (perfect only; Löbner 2002)

On the irrealis effect on auxiliary selection

141

On the other hand, the resultative can be modified by adverbials not possible in the perfect reading, for example solange clauses or bis adverbials, as in (24), Löbner’s (7): (24) Sie ist bis nächste Woche verreist. she is until next week gone away ‘She is gone away until next week.’ (Resultative only; Löbner 2002) I am aware of two explanations as to why HAVE did not replace BE in German. The first one is given in Shannon (1995), who claims that HAVE did not grammaticalise as an irrealis marker due to the strength of subjunctive morphology. However, as Sapp correctly notes, this cannot extend to Dutch, which lost subjunctive morphology and presumably it does not generalise across Romance, in the sense that Romance languages lost auxiliary selection but did not lose subjunctive morphology. A second reason is offered in Sapp (2011). According to him, a more plausible explanation for the maintenance of the split auxiliary system in Dutch, German, but also French and Italian is the much higher frequency of the perfect tenses in those languages. In spoken French, German and Italian, the present perfect is used instead of the preterite. Dutch still maintains the preterite, but the present perfect is frequently used with preterite meaning. This would suggest that speakers are aware of the difference between BE + participle as a perfect/past vs. BE + participle as a stative adjectival construction. The two have very different semantics. In view of Gillmann’s results, however, it seems to me that a most plausible explanation is one according to which HAVE did not replace BE, as both types of constructions grammaticalised into experiential perfects. While the German situation provides evidence for my hypothesis that a replacement takes place only if one of the forms shifts in interpretation, the question that arises is why some forms change but others not. In other words, why did BE periphrases in English not shift? While at the moment I cannot offer a conclusive answer to this question, one could speculate that this development is related to the fact that BE did not spread to other verb classes as was the case in German and thus created an asymmetry, as suggested in Lipson (1999). Lipson claims that having the morphological resources to distinguish the two senses of the perfect in only one verb class was too cumbersome. As she points out, the meaning of the one entails the other: I am come entails I have come: any situation in which I am come is felicitously uttered also supports the statement I have come. Therefore, she concludes, if one or the other

142

Artemis Alexiadou

interpretation had to survive, it makes sense that it would be the more broadly applicable. In addition, as MacFadden and Alexiadou (2010) point out, the counterfactual effect is a product of a fairly early stage in the grammaticalisation of the BE perfect. In English, the counterfactual effect results because the ME and EModE BE perfect remains at least fairly close to its resultative origins and does not develop true past semantics the way the HAVE perfect does. Other European languages such as German whose BE perfects have the same historical source went through a similar stage, but they crucially diverged in their subsequent development. The issue awaits further research.

7 Conclusions In this paper, I addressed two questions concerning the switch from BE to HAVE. These were: 1) Is this switch a matter of BE vs. HAVE or a more general issue concerning competition between two “perfect” forms and how these fit the semantics of modal/irrealis contexts? 2) Does this switch lead to auxiliary loss? My answer to the first question was that it is not about HAVE replacing BE, but it is more about “competition” between two forms used in relationship to what we call the perfect. These could be BE vs. HAVE, analytic vs. synthetic, etc. In the case of HAVE vs. BE competion, the hypothesis is that HAVE replaces BE if HAVE perfects grammaticalise into experiential perfects, while BE ones do not. My answer to the second question was that this is actually related to the grammaticalisation path of perfects across languages. Specifically, if only HAVE perfects grammaticalise into experiential perfects in a language, they will become the only form available to express the perfect in this language.

References Alexiadou, Artemis and Elena Anagnostopoulou. 2008 Structuring participles. In: Charles B. Chang and Hannah J. Haynie (eds.), Proceedings of West Coast Conference on Formal Linguistics 26, 33–41. Somerville, MA: Cascadilla Proceedings Project. Anagnostopoulou, Elena. 2003 Participles and voice. In: Artemis Alexiadou, Monika Rathert and Arnim von Stechow (eds.), Perfect Explorations, 1–36. Berlin, New York: De Gruyter. Bentein, Klaas. 2012 The periphrastic perfect in Ancient Greek: a diachronic mental space analysis. Transactions of the Philological Society 110: 1–41. Bybee, Joan and Östen Dahl. 1989 The creation of tense and aspect systems in the languages of the world. Studies in Language 13: 51–103. Elsness, Johan. 1997 The Perfect and the Preterite in Contemporary and Earlier English. Berlin, New York: De Gruyter.

On the irrealis effect on auxiliary selection

143

Embick, David. 2004 On the structure of resultative participles in English. Linguistic Inquiry 35: 355–392. Gerö, Eva-Carin and Arnim von Stechow. 2003 Tense in time: the Greek perfect. In: Regine Eckardt (ed.), Words in Time: Diachronic Semantics from Different Points of View, 251– 294. Berlin, New York: De Gruyter. Gillmann, Melanie. 2011 Die Grammatikalisierung des sein-Perfekts. Eine korpuslinguistische Untersuchung zur Hilfsverbselektion der Bewegungsverben im Deutschen. Beiträge zur Geschichte der Deutschen Sprache und Literatur 133: 203–234. Harris, John. 1984 Syntactic variation and dialect divergence. Journal of Linguistics 20: 303–327. Horrocks, Geoffrey. 1997 Greek: a History of the Language and its Speakers. London: Longman. Iatridou, Sabine, Elena Anagnostopoulou and Roumiana Pancheva. 2003 Observations about the form and meaning of the Perfect. In: Artemis Alexiadou, Monika Rathert and Arnim von Stechow (eds.), Perfect Explorations, 153–204. Berlin, New York: De Gruyter. Jannaris, Antonius. 1987 An Historical Greek Grammar, Chiefly of the Attic Dialect. London: Macmillan. Kratzer, Angelika. 1996 Severing the external argument from its verb. In: Johan Rooryck and Laurie Zaring (eds.), Phrase Structure and the Lexicon, 109–137. Dordrecht: Kluwer. Kratzer, Angelika. 2000 Building statives. Proceedings of BLS 26: 385–399. Larsson, Ida. 2009 Participles in time: the development of the Perfect tense in Swedish. Ph.D. dissertation, Göteborg University. Lipson, Mimi. 1999 The loss of auxiliary selection in English. University of Pennsylvania Working Papers in Linguistics 6.2: 49–61. Löbner, Sebastian. 2002 Is the German Perfekt a perfect Perfect? In: Ingrid Kaufmann, Barbara Stiebels (eds.), More than Words: a Festschrift for Dieter Wunderlich, 369–391. Berlin: Akademie Verlag. Mateu, Jaume. 2009 Gradience and auxiliary selection in Old Catalan and Old Spanish. In: Paola Crisma and Giuseppe Longobardi (eds.), Historical Syntax and Linguistic Theory, 176–194. Oxford: Oxford University Press. McCall, Betsy. 1999 On the rise of periphrasis in the Greek Perfect medio-passive. Indiana University, LING L760 Diachronic Syntax. McFadden, Thomas and Artemis Alexiadou. 2010 Perfects, resultatives and auxiliaries in Earlier English. Linguistic Inquiry 41: 389–425. Nordahl, Helge. 1977 Assez avec alé: estre et avoir comme auxiliaires du verbe aler en ancien français. Revue Romane 12: 54–67. Ralli, Angeliki. 2006 Syntactic and morpho-syntactic phenomena in Modern Greek dialects: the state of the art. Journal of Greek Linguistics 7: 121–159. Rathert, Monika. 2003 Universal-existential-puzzles. In: Artemis Alexiadou, Monika Rathert and Arnim von Stechow (eds.), Perfect Explorations, 363–380. Berlin, New York: De Gruyter. Rosemeyer, Malte. (this volume) Entrenchment and discourse traditions in Spanish auxiliary selection. Rothstein, Björn. 2008 The Perfect Time Span. Amsterdam, Philadelphia: John Benjamins. Sanchéz-Marco, Cristina. 2012 The gradualness of change in auxiliary selection: Evidence from Spanish. Paper presented at the Workshop on Auxiliary selection: gradience and gradualness. Freiburg University, June 22–23, 2012.

144

Artemis Alexiadou

Sapp, Christopher. 2011 Auxiliary selection in the Early New High German tenses. Groninger Arbeiten zur germanistischen Linguistik 53: 29–43. Shannon, Thomas. F. 1995 Toward a cognitive explanation of perfect auxiliary variation: some modal and aspectual effects in the history of Germanic. American Journal of Germanic Linguistics & Literatures 7: 129–163. Smyth, Herbert Weir. 1984 Greek Grammar. Boston: Harvard University Press. Stolova, Natalya. 2006 Split intransitivity in Old Spanish: irrealis and negation effects. Revue roumaine de linguistique LI(2): 301–320. Togeby, Knud. 1974 Précis Historique de Grammaire Française. København: Akademisk Forlag. von Stechow, Arnim. 1998 German participles II in distributed Morphology. Unpublished manuscript, University of Tübingen. von Stechow, Arnim. 1999 Semantics and syntax of the perfect and future constructions. Lectures at the LOT Summer School, Potsdam.

Ida Larsson, Department of Linguistics and Scandinavian Studies, University of Oslo

The HAVE/BE alternation in Scandinavian – perfects, resultatives and unaccusative structure 1 Introduction All of the Scandinavian languages have an alternation between HAVE and BE in constructions with an active participle; see the Swedish and Danish examples in (1) and (2) below, with BE + participle of an unaccusative verb and HAVE + participle of an unergative verb.1 This alternation is well known in e.g. German, Dutch, French and Italian, but not in Present-Day English. (1) a. Han är hemkommen. he be.prs home-come.ptcp.c.sg ‘He has come home.’

(Swedish)

b. Hon har sovit länge. she have.prs sleep.sup long ‘She has slept for a long time.’ (2) a. Han er kommet. he be.prs come.ptcp.n.sg ‘He has come.’

(Danish)

b. Hon har sovet længe. she have.prs sleep.ptcp.n.sg long ‘She has slept for a long time.’ The variation is however not of the same kind in all of the languages. For instance, Icelandic, Norwegian and Swedish differ from Danish by also allowing HAVE with unaccusative verbs; cf. (3) and (4). (3) Han har redan kommit hem. he have.prs already come.sup home ‘He has already come home.’

(Swedish)

1 Capitalised BE and HAVE are used to cover the forms of be and have in English, vara and ha in Swedish and the equivalents in the other Germanic languages. Modern Swedish makes a morphological distinction between perfect and passive/resultative participles. The former is traditionally referred to as the supine, and I gloss it as sup. Modern Mainland Scandinavian common gender is glossed as c.

146

Ida Larsson

(4) *Han har kommet he have.prs come.ptcp.n.sg

(Danish)

As we will see further below, only Danish has auxiliary selection in perfects, like German and Dutch.2 In Icelandic, Norwegian and Swedish, perfects are always formed with HAVE, and the construction with BE is a resultative construction. (This means that the translation of the Danish example in 2a as an English present perfect is more accurate than the translation of the Swedish example in 1a.) In the following, I review the alternation between HAVE and BE in Scandinavian with respect to examples like (1) and (2), taking into account both present-day and historical data (see Larsson 2014 for an overview of HAVE/BE in Scandinavian dialects).3 I discuss what properties distinguish the resultative construction with BE from a perfect, and what factors determine the choice between HAVE-perfect and BE-resultative at different times and in the different languages. The first part of the paper establishes a distinction between perfects and resultatives, and argues that a BE-perfect is missing both in older and modern Icelandic, Norwegian and Swedish. I will suggest that differences between the languages with respect to the choice between HAVE and BE relate to variation in the aspectual contribution of the participle used in the resultative. Section 2 gives an overview of the diagnostics that distinguish perfects from resultatives and introduces a distinction between two kinds of resultatives (target states and resultant states) with partly different properties. I analyse the construction with BE in Icelandic, Norwegian and Swedish as a construction with a tenseless participial clause, while in Danish, BE forms a perfect tense together with a tensed participle. In section 3, I investigate the distribution of the constructions in older Icelandic and Swedish, which show partly different patterns with respect to BE, despite the fact that neither language has a BE-perfect.4 The second part of the paper is concerned with the properties of unaccusative verbs and the Auxiliary Selection Hierarchy (ASH; Sorace 2000). In section

2 The situation in Faroese is unclear. According to Thráinsson et al. (2004), Faroese patterns with Icelandic, but dialect data suggest that it might sometimes rather pattern with Danish (see Larsson 2014). It is well known that Danish has some influence on Faroese. 3 The discussion is to a large extent based on parts of Larsson (2009a), where the details of the investigated material are given. 4 I use the term older Swedish to refer to both Old Swedish (c. 1200–1500) and Early Modern Swedish (c. 1500–1700).

147

The HAVE/BE alternation in Scandinavian

4, I consider variation and change in the distribution of HAVE-perfects relative to BE-resultatives and show that the pattern of variation and change is the same in Swedish, Icelandic and Norwegian as in auxiliary selection languages, and that it follows the Auxiliary Selection Hierarchy. Section 5 relates the variation to the structure of unaccusatives. Section 6 concludes the paper with a short discussion of gradience, gradualness and auxiliary selection.

2 Perfects and resultatives – tense and aspect In this section, I first give a brief overview of diagnostics that distinguish perfects from resultatives and then introduce a distinction between two types of resultatives. The distinctions are taken to relate to the aspectual and temporal composition of the participial clause. I illustrate the constructions with modern Icelandic and Swedish data, but the same distinctions can be made for Norwegian. As we will see in section 3 below, the same distinctions can also account for the patterns in older Scandinavian.

2.1 Distinguishing resultatives from perfects As argued by Larsson (2009a, 2012) and McFadden and Alexiadou (2010), perfects (with HAVE) and resultatives (with BE) behave differently in a number of contexts. Firstly, perfects, but generally not resultatives, allow an experiential reading, i.e. a reading where the target state of the participial event does not hold at the reference time; cf. (5a) and (5b) (but see further below). In languages like Icelandic and Swedish, HAVE is the only option in experiential contexts; see (5c). In Icelandic, BE is the unmarked option in resultative contexts with unaccusative verbs, and HAVE is somewhat degraded; see (5d).5 (5) a. Hún er farin til Kína (núna) she be.prs.3sg go.ptcp.f.sg.nom to China now ‘She has gone to China (now).’

(Icelandic)

b. *Hún er farin til Kína þrisvar sinnum she be.prs.3sg go.ptcp.f.sg.nom to China three times ‘She has gone to China three times.’ 5 For a discussion of the distribution of different perfect-like constructions in Icelandic, see Larsson (2009b) and references there.

148

Ida Larsson

c. Hún hefur farið til Kína þrisvar sinnum. she have.prs.3sg go.ptcp.n.sg to China three times ‘She has gone to China three times.’ d. ? Hún hefur farið til Kína núna she have.prs.3sg go.ptcp.n.sg to China now ‘She has gone to China now.’ Secondly, past perfects can have a past counterfactual reading, but past resultatives cannot. In Icelandic, past tense BE + active participle necessarily has a present tense reading; see (6). Since Danish, unlike Icelandic, has a perfect tense with BE, BE is possible in counterfactuals like (7). (6) a. *Ef hann væri kominn í gær, (Icelandic) if he be.pst.subj.3sg come.ptcp.m.sg.nom yesterday þá … then ‘If he had come yesterday, then …’ b. Ef hann væri kominn núna, þá… if he be.pst.subj.3sg come.ptcp.m.sg.nom now then (7) Jeg kunne godt have ønsket mig at han (Danish) I could.refl well have.inf wish.ptcp.n.sg refl that he var kommet til mødet i går. be.pst come.ptcp.n.sg to meeting.def.n.sg yesterday ‘I would have wished that he had come to the meeting yesterday.’ Following McFadden and Alexiadou (2010), and based on Iatridou’s (2000) account of counterfactuals, we can assume that this difference between past perfects and past resultatives is due to the fact that past perfects involve two layers of past morphology, whereas past tense resultatives only have one layer (see Larsson 2009a: 157–168 for discussion). In fact, it is well known that past time adverbials are ambiguous in past perfects, but they are generally not ambiguous in past tense resultatives; see (8). This difference is straightforwardly accounted for if we assume that the past perfect has two layers of past morphology, and the past resultative only one.6

6 See Larsson (2009a: 79–95) for a discussion of the restriction on past adverbials in the present perfect.

149

The HAVE/BE alternation in Scandinavian

(8)

a. Han hade rest hem klockan fem. (Swedish) he have.pst travel.sup home clock.def.c.sg five ‘He had gone home at five o’clock.’ i. The event of going home took place before five o’clock. ii. The event of going home took place at five o’clock. b. Han var hemrest klockan fem. he be.pst home-travel.ptcp.c.sg clock.def.c.sg five ‘The event of going home took place before five o’clock.’ Not: ‘The event of going home took place at five o’clock.’

Furthermore, perfects and resultatives differ with respect to stativity. It has often been assumed that perfects are stative (see e.g. Katz 2003), but they do not behave like other statives in e.g. pseudoclefts and with respect to manner adverbials. Resultatives, on the other hand, are clearly stative and therefore disallowed in pseudoclefts and with manner adverbials; see the Present-Day Swedish examples in (9) and (10). (9)

a. *Vad som hände var att han var what that happen.pst be.pst that he be.pst hemkommen home-come.ptcp.c.sg

(Swedish)

b. Vad som har hänt är att han har what that have.prs happen.sup be.prs that he have.prs kommit hem come.sup home ‘What has happened is that he has come home.’ (10) a. Han är hemrest (*snabbt) med flyg. (Swedish) he be.prs home-travel.ptcp.c.sg quickly with plane ‘He has gone home by airplane.’ b. Han har rest hem snabbt med flyg. he have.prs travel.sup home quickly by plane ‘He has gone home quickly by airplane.’ We can conclude that the difference between perfects and resultative constructions relates to tense morphology and stativity. Perfects involve a non-finite past tense (cf. Julien 2001), whereas resultative constructions express perfective or resultative aspect (see further below), and have a stative participle. In

150

Ida Larsson

this way, we can account for the differences with respect to counterfactuals, adverbial modification and pseudoclefts. In the next section, I make a further distinction among the resultatives, and propose that some, but not all, resultative participles carry aspectual meaning. I take Kratzer’s (2000) distinction among stative passives as a starting point.

2.2 Two types of resultatives – with and without aspect Using the terminology of Parsons (1990), Kratzer (2000) distinguishes between two different kinds of stative passives: passives which (like perfects) express a resultant state that necessarily holds forever, and target state passives which denote the, in principle, reversible target state specified by the participial predicate: (11) a. Resultant state: “For every event e that culminates, there is a corresponding state that holds forever after. […] If Mary eats lunch, then there is a state that holds forever after: the state of Mary’s having eaten lunch.” b. Target state: “It is important not to identify the Resultant-state of an event with its ‘target’ state. If I throw a ball onto the roof, the target state of this event is the ball’s being on the roof, a state that may or may not last for a long time” (Parsons 1990: 234–235; see also Kratzer 2000). Only telic verbs can form target state participles, but not all of them do (see further Larsson 2009a, 2012 and references cited there). Resultant states are not lexically restricted in this way; according to Parsons, they follow from every event that culminates (or terminates). As noted by Kratzer (2000), only target state passives can be modified by the adverb fortfarande ‘still’:7 (12) a. Staden är fortfarande förstörd. city.def.c.sg is.prs still destroy.ptcp.c.sg ‘The city is still destroyed.’

(Swedish; target state passive)

7 Like German, both older and modern Swedish have stative passives with BE and eventive passives with a verb meaning ‘become’ + participle (in addition to morphological passives).

The HAVE/BE alternation in Scandinavian

b. Huset är (*fortfarande) house.def.n.sg is.prs still ommålat. repaint.ptcp.n.sg ‘The house has been repainted.’

151

(Swedish; resultant state passive)

In this respect, resultant state passives resemble perfects: (13) De har (*fortfarande) förstört they have.prs still destroy.sup staden. city.def.c.sg ‘They have destroyed the city.’

(Swedish; perfect)

Only participles that disallow fortfarande (resultant state participles) allow clause-final frequency adverbs in Present-Day Swedish: (14) a. *Staden är förstörd ofta. city.def.c.sg be.prs destroy.ptcp.c.sg often b. Huset är ommålat ofta. house.def.n.sg be.prs repaint.ptcp.n.sg often ‘The house has been repainted often.’

(Swedish; target state passive) (Swedish; resultant state passive)

Again, resultant state passives resemble perfects: (15) De har förstört staden ofta. they have.prs destroy.sup city.def.c.sg often ‘They have destroyed the city often.’

(Swedish; perfect)

This means that resultant state passives can have an experiential reading in Present-Day Swedish, as well as a resultative reading. Compare (16a) where the result of taking away dinner still holds (resultative reading) at the speech time, and the experiential example in (16b) where the result of putting dinner on the table no longer holds (i.e. dinner is not on the table at the speech time). (16) a. Middagen är (*fortfarande) bortplockad. dinner.def.c.sg be.prs still away-take.ptcp.c.sg ‘Dinner has been taken away.’

152

Ida Larsson

b. Middagen är både framdukad och dinner.def.c.sg be.prs both put.on.the.table.ptcp.c.sg and bortplockad. (Du är sen.) away-taken.ptcp.c.sg you be.prs late ‘Dinner has both been put on the table and taken away. (You are late.)’ Importantly, the same distinction between target states and resultant states can be made for active participles with BE; cf. (17). (17) a. Han är fortfarande försvunnen. he be.prs still disappear.ptcp.c.sg ‘He is still missing.’

(Target state)

b. Deltagarna är (*fortfarande) (Resultant state) participant.def.pl be.prs still hitresta från Stockholm. here-travel.ptcp.pl from Stockholm ‘The participants have come here from Stockholm.’ Again, the resultant state construction can have an experiential reading, as well as a resultative reading; see (18). Examples with an experiential reading, however, are often somewhat marked in Present-Day Swedish. (18) a. Ansökan är inkommen och application.def.c.sg be.prs in-come.ptcp.c.sg and ivägskickad. (Så allt är under kontroll.) away-sent.ptcp.c.sg so everything be.prs under control ‘The application has come in and has been sent away. (So everything is under control.)’ b. Frida är både hemkommen och utgången. Frida be.prs both home-come.ptcp.c.sg and out-go.ptcp.c.sg (Nu vet jag inte var hon är.) now know.prs I not where she be.prs ‘Frida has both come home and gone out. (Now I don’t know where she is.)’ Parsons (1990) introduces resultant states to account for the semantics of the perfect, and the resultant state construction with BE resembles a perfect in several ways. It is well known that perfects too can have both resultative and

The HAVE/BE alternation in Scandinavian

153

experiential readings. In fact, many of the Swedish resultant state passives are best translated with English HAVE-perfects, whether the reading is resultative or not. The distinction between resultative and experiential readings can be assumed to relate to the aspectual composition of the participle (see Iatridou, Anagnostopoulou, and Izvorski 2001; Pancheva 2003; Larsson 2009a) in perfects with HAVE, as well as in the construction with BE. In Larsson (2009a, 2012), I suggest that while target state participles lack grammatical aspect, resultant state participles can express perfective or resultative aspect in Swedish. However, unlike HAVE, BE only combines with a tenseless complement. As we have seen, this accounts for differences with respect to adverbial modification and counterfactuals. Examples like (18) are taken to involve a perfective, but tenseless participial clause. They are therefore not true perfects, but pattern with stative passives. As we saw above, neither Swedish (nor Icelandic) has a perfect with BE. In the following, I continue to refer to the Icelandic/Norwegian/Swedish construction with BE as a resultative, despite the fact that it is not necessarily restricted to resultative readings but sometimes also has an experiential reading. In other words, I refer to examples like (18) as involving a BE-resultative.

3 Distribution of HAVE and BE in older Swedish and Icelandic Although perfects and resultatives differ with respect to temporal and aspectual properties, they have similar semantics, and it can be difficult to classify individual examples as one or the other. The perfect has historically developed from a resultative, and the semantic and morphological similarity is clearly a prerequisite for this development (see Larsson 2009a for extensive discussion). While in certain contexts a HAVE-perfect is required (e.g. in past counterfactuals), many contexts allow for both perfects and resultatives. In fact, throughout the history of the Scandinavian languages (and other Germanic languages as well), we can note an alternation between resultatives and perfects where the choice between the two is made in different ways at different times and in different languages. In this section, I investigate this alternation in the history of Scandinavian, focussing on Icelandic and Swedish. Norwegian largely patterns with Icelandic, and up until the 17th century Danish patterns with Swedish. In section 3.1, we look briefly at the use of HAVE with different groups of verbs in Old Icelandic and Old Swedish. We will see that there is reason to account for the alternation between HAVE and BE in the historical records with

154

Ida Larsson

the same distinction between HAVE-perfects and BE-resultatives as we have observed in the modern languages. Section 3.2 points to a difference between Icelandic and older Swedish with respect to resultatives with BE: resultant state participles with perfective aspect (and an experiential reading) are more generally available in older Swedish than in Icelandic. Section 3.3 gives a summary.

3.1 Early HAVE-perfects HAVE-perfects occur with all types of verbs in the oldest Scandinavian sources; see Tables 1 and 2.8 The fact that unaccusative verbs occur in HAVE-perfects suggests that Old Scandinavian did not have auxiliary selection like modern Danish and German. However, we can note that examples with unaccusatives embedded under HAVE are rare, and that the number of examples is lower in Old Swedish (1 % in all) than in Old Icelandic (7 %). Unaccusative verbs are generally used with BE, as in (19).

Tab. 1: Verbs in the complement of HAVE in four Old Norse texts (Larsson 2009a: 131). Text

Unaccusative

Unergative

-sk-forms

Transitive

ALL

The Poetic Edda Grágás Íslendingabók (c. 1130) Landnámabók (12th century) TOTAL

7 ( 5 %) 4 ( 4 %) 4 (14 %) 7 (12 %) 22 ( 7 %)

21 (14 %) 12 (13 %) 11 (38 %) 17 (29 %) 61 (19 %)

1 (1 %) 2 (2 %) 2 (7 %) 3 (5 %) 8 (2 %)

119 (80 %) 76 (81 %) 12 (41 %) 31 (53 %) 238 (72 %)

148 94 29 58 329

Tab. 2: Verbs in the complement of HAVE in four Old Swedish texts (Larsson 2009a: 131). Text

Unaccusative Unergative

Transitive

ALL

Äldre Västgötalagen (c. 1220) Upplandslagen (1297) Gutalagen (c. 1300) Ett fornsvenskt legendarium (c. 1300) TOTAL

0 (0 %) 2 (1 %) 0 (0 %) 3 (3 %) 5 (1 %)

36 (86 %) 181 (87 %) 35 (88 %) 83 (82 %) 335 (86 %)

42 207 40 101 390

6 (14 %) 24 (12 %) 5 (13 %) 15 (15 %) 50 (13 %)

8 Icelandic has a mediopassive with the suffix -sk, which is kept separate from unergatives and transitives.

The HAVE/BE alternation in Scandinavian

155

(19) Han […] spordhe hwi iak var komin hære he ask.pst.3sg why I be.pst.1sg come.ptcp.m.sg.nom here ‘He asked why I had come here.’ (Old Swedish; Ivan c. 1303: 462) In both languages, examples of HAVE-perfects of unaccusative verbs generally have an experiential reading, as in (20a) below, or a counterfactual reading, as in (20b). (20) a. þvíat hvern hefi ec heim um komit for every have.prs.1sg I world prt come.ptcp.n.sg ‘for I have come into every world’ (Old Icelandic; Edda Vm.43) b. hafþe þin kona hær comit. have.pst.3sg your.f.sg.nom wife.f.sg.nom here come.ptcp.n.sg aldre skulde hon hæþan coma: liuande: never would she away come.inf living ‘if your wife had come here, she would never get away alive’ (Old Swedish; Leg c. 1300: 19) Importantly, there are no clear-cut examples of past counterfactuals with BE in the older Scandinavian records, and the alternation between HAVE and BE with unaccusatives is to a large extent accounted for if we distinguish counterfactual contexts. In Johannisson’s Old Swedish material, 185 (52 %) of the total 355 examples of HAVE with participles of unaccusative verbs involve counterfactuals (1945: 97). Compare the counterfactual with HAVE in (21a) and the similar, but factual, example with BE in (21b). (21) a. Jach hade gerne och faaret medh tyl ider I have.pst.sg gladly also go.ptcp.n.sg with to you ‘I would also gladly have gone with them to you.’ (Early Modern Swedish; Siggesson a. 1500: 15) b. j samme dage fych iach Knwtz j Gylle scriffuilse. in same day get.pst.sg I Knut.gen in Gylle letter athy ware faren aff Westeraars til Stocholm. that-you be.pst.subj go.ptcp.m.sg from Västerås to Stockholm ‘The same day, I got the letter from Knut in Gylle that you were gone from Västerås to Stockholm.’ (Early Modern Swedish; Siggesson a. 1500: 15)

156

Ida Larsson

In the material in Larsson (2009a), there is not a single clear-cut example of BE in a past counterfactual context. In Johannisson’s study, there are altogether nine examples of counterfactuals with BE, but they all have a present counterfactual reading. According to McFadden and Alexiadou (2010), this is also the pattern in older English, and as noted, perfects and resultatives behave differently with respect to counterfactuals also in the present-day languages. The fact that BE appears to have been categorically banned in past counterfactuals in the historical material is clearly a strong argument to treat the construction with BE as a resultative in older Scandinavian and older English, as in the modern languages. I therefore conclude that older Swedish and older Icelandic lacked a BEperfect, just like the modern languages. As noted, there are still contexts where there is a choice between a HAVE-perfect and a BE-resultative, and there is some (limited) variation in the oldest sources. HAVE sometimes occurs in contexts where BE would be possible, as in (22). (22) The […] sagdo at the haffdo tith komit alt they say.pst.3pl that they have.pst.3pl there come.ptcp.n.sg all affbrat too-soon ‘They said that they had come there too soon.’ (Old Swedish; Erikskrönikan 14th c.: 1632 ff.) In younger texts, HAVE and BE sometimes seem to be used interchangeably in Swedish; cf. (23a) with an almost identical example in (23b). (23) a. Så seer man här tilgått wara so see.pres.sg one here about-go.ptcp.n.sg be.inf ‘one sees that things have happened in this way here’ (Early Modern Swedish; Petri *1493: 48) b. Kan man wel merkia huru thå haffuer can.prs.sg one well notice.inf how then have.pres.sg tilgått about-go.ptcp.n.sg ‘One can well notice how things then have happened’ (Early Modern Swedish; Petri *1493: 31) There is also considerable variation between texts. In some texts, there are no examples of unaccusatives with HAVE, while in other texts over 40 % of the examples contain HAVE (see Johannisson 1945: 135–136, Table 25). To some

The HAVE/BE alternation in Scandinavian

157

extent, the variation is due to contextual factors (e.g. the frequency of counterfactuals). However, we can also view the variation, and what seems to be a gradual increase in the frequency of HAVE, as part of the establishment of the perfect tense in Swedish. The perfect is an innovation in the Old Germanic languages, and it is therefore expected that the frequency increases as the perfect is gradually established in the speech community. In Swedish (as in English), the establishment of the HAVE-perfect is at the expense of the BE-resultative (cf. Larsson 2009a; McFadden and Alexiadou 2010). We noted above that unaccusatives occurred more often in perfects with HAVE in Old Icelandic than in Old Swedish. We could possibly take this difference between Old Icelandic and Old Swedish to relate either to contextual differences (text types, etc.) or perhaps to differences in the time of the establishment of the HAVE-perfect. It turns out, however, that in Old Swedish, BE is used in contexts where (Old) Icelandic requires HAVE. Next, we will take a closer look at the different types of examples.

3.2 A difference between older Swedish and Icelandic The difference between Old Swedish and Old Icelandic suggested by the data in Table 1 and Table 2 above, is not simply a difference in the frequency of experiential perfects or counterfactuals. Even regarding only experiential contexts, BE is more common in older Swedish than in both Old and Present-Day Icelandic. In iterative contexts, Johannisson (1945: 103 fn. 2) notes 56 examples with HAVE and 60 with BE in older Swedish. A couple of experiential examples with BE are given in (24). (24) a. Thet är offta skeedt, at […] it.n.sg be.prs.sg often happen.ptcp.n.sg that ‘It has often happened that …’ (Early Modern Swedish; Petri *1493: 3) b. Vbbo […] bygde Vbsal / Ther altidh är Ubbo build.pst.sg Uppsala there always be.prs.sg skedt Kongars waal happen.ptcp.n.sg king.pl.gen election ‘Ubbo built Uppsala where kings always have been elected.’ (Early Modern Swedish; Messenius I *1579: 4) In Johannisson’s Old Norse material, there is only one example with BE and 25 examples with HAVE together with adverbs of frequency or iteration (1945: 36).

158

Ida Larsson

In other words, Old Icelandic and Old Norwegian seem to be very similar to modern Icelandic, where adverbs of iteration and frequency are excluded in the resultative with BE, as we saw in example (5b) above (see also Jónsson 1992; Thráinsson 2007: 12). Old Norse is also similar to modern Icelandic in dispreferring the HAVE-perfect in resultative contexts (cf. 5d above). Based on the higher number of examples with BE in experiential contexts, Johannisson concludes that BE is a temporal auxiliary in older Swedish, but not in Old Norse, Present-Day Icelandic and Present-Day Swedish. However, this leaves us without an explanation for the fact that BE never occurs in past counterfactuals. Moreover, even if BE sometimes occurs in experiential contexts, HAVE is more common in experiential than in resultative contexts. Finally, the active construction with BE patterns with stative passives throughout the history of Swedish. In older Swedish, stative passives are sometimes used where Present-Day Swedish requires a perfect. Kirri (1975) gives the examples in (25): (25) a. Nær domen war affsagd, kom when verdict.def.m.sg be.pst.sg off-speak.ptcp.m.sg come.pst.sg Margaretha […] fram for rætten Margaretha forward before court.def.m.sg ‘When the verdict had been spoken, Margaretha came before the court.’ (Early Modern Swedish; Petri tänkebok *1493: 261; from Kirri 1975: 37) b. Nær thetta læset war sade hwar when this.n.sg read.ptcp.n.sg be.pst.sg say.pst.sg each sitt ther til poss.refl there to ‘When this had been read, everyone gave his/her comment.’ (Early Modern Swedish; Petri tänkebok *1493: 313; from Kirri 1975: 37) From examples like these, Kirri concludes that BE + a passive participle sometimes had the reading of a perfect in older Swedish. This would clearly make older Swedish rather exceptional. We saw in section 2.2 above that BE-resultatives sometimes have an experiential reading in modern Swedish, and that they are sometimes compatible with frequency adverbs. If we assume that there are two different kinds of resultant state participles also in older Scandinavian, we have a straightforward way of accounting for the difference between older Swedish and Icelandic: in Icelandic, but not in Swedish, resultant state participles necessarily have resultative aspect, and frequency adverbs are therefore excluded (see Larsson

The HAVE/BE alternation in Scandinavian

159

2009a, 2012 for discussion). In Present-Day Swedish, examples with a perfective reading are often marked, and a perfect with HAVE is generally preferred. In 15th- and 16th-century Swedish, on the other hand, these kind of examples appear to have been a more common phenomenon, both with passive and active participles.

3.3 Intermediate summary We have seen that the construction with BE is less restricted in older Swedish than in (older) Icelandic, but more restricted than perfects with BE in e.g. Danish and German. Importantly, the construction with BE never occurs in past counterfactuals (unless embedded under a perfect). In fact, it seems to pattern with stative passives, rather than with perfects, throughout the history of Swedish. I conclude that neither older Swedish nor Icelandic had a BE-perfect, but that the resultant state construction with BE could have perfective aspect in older Swedish, but was restricted to resultative aspect in older Icelandic (as in Present-Day Icelandic). I suggested above that the construction with BE + active participle in experiential contexts can be understood as involving a participle with perfective aspect. Unlike a perfect participle, the participle is tenseless and stative. The difference between target states and resultant states, I assume is a question of both presence/absence of aspect and the structure of the verb phrase (cf. e.g. Alexiadou and Anagnostopoulou 2008; Larsson 2009a). This gives us the following (partial) structures of the BE-resultatives and the HAVE-perfect (I return to the structure of the verb phrase below): (26) a. BE + target state participle: … BE [VoiceP Voice [PartP –en [VP … b. BE + resultant state participles (resultative or perfective aspect): … BE [AspP Asp [VoiceP Voice [PartP –en [VP … (27)

HAVE + perfect participle (past tense): … HAVE … [TP T [AspP Asp [VoiceP Voice [PartP –en [VP …

Viewed thus, what we can observe as apparently gradual differences in the distribution of HAVE and BE between the older Scandinavian languages, or between older and modern Swedish, is a consequence of the availability of three different structures with similar semantics. The preference for one or the other construction varies depending on time and language, as well as context

160

Ida Larsson

and real-world knowledge. Perfective resultant state participles are more restricted in Icelandic than in older Swedish, and tensed participles are used in Present-Day Swedish, where the older languages would have a tenseless participial structure. In older Scandinavian, as in modern Icelandic, BE is preferred over HAVE in contexts where there is a choice between the two. We have seen that Icelandic and Swedish are not auxiliary selection languages: perfects are always formed with the auxiliary HAVE, and BE is restricted to resultatives (with varying aspectual properties). At the same time, BEresultatives in several ways resemble the BE-perfects we find in languages like modern Danish. In particular, BE-perfects (in languages like Danish) and BEresultatives (in Icelandic and Swedish) occur with the same groups of verbs. In the next two sections, we investigate which unaccusative verbs occur with BE, and examine their structure. Section 4 introduces the Auxiliary Selection Hierarchy and shows that the patterns of variation and change parallel those found in auxiliary selection languages. In section 5, I suggest that the variability relates directly to event structure and the lexical specifications of verbs, and I briefly sketch a structural account of eventive unaccusative verbs, using the verb phrase structure from Ramchand (2008).

4 Variation and change According to Johannisson (1945), Swedish and Danish largely behave the same until the 17 th century (cf. Falk and Torp 1900). Johannisson observes that HAVE was more common with unaccusative verbs in older than in Present-Day Danish, and that HAVE was particularly common in counterfactuals (1945: 219). In Present-Day Danish, BE is the only possibility with unaccusative verbs, and it is possible also in past counterfactuals.9 Thus, what started out as an extended use of BE with a perfective, resultant state participle, led to reanalysis and the emergence of a BE-perfect. This change seems to have taken place in Danish around the 17 th century. Swedish took a different turn: rather than developing a BE-perfect, the use of HAVE-perfects was extended to resultative contexts, and the construction with BE subsequently became more restricted. In this section, we look briefly at how this change proceeded. First, we look at patterns of variation in the present-day Scandinavian languages. (For the full discussion, see Larsson 2009a, chapter 7.)

9 Larsson (2009a, 2012) suggests that the development of HAVE-perfects in Germanic starts with an extended use of resultant state participles, and this also seems to be the case with BEperfects in Danish. Additional support comes from Present-Day Icelandic (see Larsson 2009b).

The HAVE/BE alternation in Scandinavian

161

4.1 Unaccusatives with BE and the Auxiliary Selection Hierarchy It is well known that the diagnostics for unaccusativity often pick out a subset of the unaccusatives, and that the subset can vary depending on language and diagnostic (cf. e.g. Levin and Rappaport Hovav 1995: 20). The cross-linguistic variation is, however, not unrestricted but rather follows certain patterns, as shown with respect to auxiliary selection by the investigations by e.g. Sorace (2000, 2004), Legendre and Sorace (2003) and Cennamo and Sorace (2007). The results are summarised in the Auxiliary Selection Hierarchy in (28) below (from Sorace 2000: 863); the feature specifications are from Legendre and Sorace (2003) (cf. Legendre 2007a: 1525). The verbs that are higher on the hierarchy most consistently (both cross-linguistically and within a language) form perfects with BE, and the verbs at the bottom most consistently form perfects with HAVE. The verbs above the line occur with BE in Scandinavian. (28) The Auxiliary Selection Hierarchy BE Change of location (arrive, come)

Change of state

(appear, die)

(grow, wilt)

Continuation of pre-existing state Existence of state Uncontrolled process (non-motional) Controlled process (motional)

+telic; +/−agentive; +directed change; +motional displacement +telic; −agentive; +directed change; −motional displacement −telic; −agentive; +directed change; −motional displacement

(last)

+stative

(be, exist) (sweat, tremble)

+stative −telic; −agentive; −directed change; −motional displacement −telic; +agentive; +directed change; +motional displacement −telic; +agentive; −directed change; −motional displacement

(swim, run)

Controlled process (work) (non-motional) HAVE (adapted from Sorace 2000: 863)

162

Ida Larsson

In Danish, the verbs that form perfects with BE include verbs that express telic change of location (‘come’, ‘arrive’), telic change of state (‘become’, ‘wake up’), and gradual change of state (‘grow’); see (29). Verbs that form perfects with HAVE are stative verbs (including BE), and (generally) all groups of atelic verbs; see (30). (29) a. Han er/*har kommet. he be.prs/have.prs come.ptcp.n.sg ‘He has come.’

(Danish)

b. Der er/*har sket noget. there be.prs/have.prs happen.ptcp.n.sg something ‘Something has happened.’ c. Han er/*har vokset meget. he be.prs/have.prs grow.ptcp.n.sg much ‘He has grown much.’ (30) a. Jeg *er/har boet i Stockholm. I be.prs/have.prs live.ptcp.n.sg in Stockholm ‘I have lived in Stockholm.’

(Danish)

b. Jeg *er/har arbejdet med artiklen. I be.prs/have.prs work.ptcp.n.sg with paper.def.c.sg ‘I have worked with the paper.’ c. Han *er/har sovet længe. she be.prs/have.prs sleep.ptcp.n.sg long ‘He has slept for a long time.’ The same groups of verbs that form perfects with BE in Danish have active participles with BE in Icelandic and older Swedish. Older Swedish examples of change of location verbs and change of state verbs are given in (31). (31) a. Han […] spordhe hwi iak var he ask.pst.3sg why I be.pst.1sg komin hære come.ptcp.m.sg.nom here ‘He asked why I had come here.’ (Old Swedish; Ivan c. 1303: 462)

(Older Swedish)

b. Herra Iwan ær ater vordhin stark Sir Ivan be.prs.1sg again become.ptcp.m.sg.nom strong ‘Sir Ivan has become strong again.’ (Old Swedish; Ivan c. 1303: 2509)

163

The HAVE/BE alternation in Scandinavian

c. är en qist vuxin ur thet ena be.prs.sg a.m.sg twig grow.ptcp.m.sg out.of the.n.sg one.def trädh tree.def.n.sg ‘has a twig grown out of the first tree’ (Early Modern Swedish; Bureus I *1568: 200) In Present-Day Swedish, a smaller group of verbs occur in the active resultative with BE. Neither variable behaviour verbs without a particle (but with a telic PP), nor gradual or punctual change of state verbs are possible with BE in Present-Day Swedish; see (32). (32) a. *Han är sprungen till grannen. he be.prs run.ptcp.c.sg to neighbour.def.c.sg

(Swedish)

b. *Han är drunknad. he be.prs drown.ptcp.c.sg c. *Han är bliven misstänksam. he be.prs become.ptcp.c.sg suspicious d. *Bomben är exploderad. bomb.def.c.sg be.prs explode.ptcp.c.sg e. *Priset är ökat. price.def.n.sg be.prs increase.ptcp.n.sg The only verbs that consistently have active participles in the complement of BE in Present-Day Swedish are inherently telic change of location verbs, and variable behaviour verbs with an incorporated particle (cf. Teleman, Hellberg, and Andersson 1999/2: 606): (33) a. De är redan anlända. they be.prs already arrive.ptcp.pl ‘They have already arrived.’

(Swedish)

b. Båten är redan ankommen till bryggan. boat.def.c.sg be.prs already arrive.ptcp.c.sg to wharf.def.c.sg ‘The boat has already arrived.’ c. Han är nyligen återvänd från England. he be.prs recently return.ptcp.c.sg from England ‘He has recently returned from England.’

164

Ida Larsson

d. Frida är hitflyttad från Stockholm. Frida be.prs here-move.ptcp.c.sg from Stockholm ‘Frida has moved here from Stockholm.’ Present-Day Icelandic behaves very much like Old Icelandic: all groups of unaccusatives are still possible with BE; see (34). There is, however, some variation with respect to variable behaviour verbs and gradual change of state verbs; cf. (35a) and (35b) (and see e.g. Yamaguchi and Pétursson 2003). With variable behaviour verbs, grammaticality can depend on the presence/absence of a particle; see (36). (34) a. Ég er alveg komin. I be.prs.1sg just come.ptcp.f.sg.nom ‘I have just come.’

(Icelandic)

b. Þetta er orðið fínt. this.n.sg be.prs.3sg become.n.sg fine.n.sg ‘This has become fine.’ c. Snjórinn er nýlega bráðnaður. snow.def.m.sg.nom be.prs recently melt.ptcp.m.sg.nom ‘The snow has recently melted.’ (35) a. Grasið er gulnað. grass.def.n.sg be.prs.3sg yellow.ptcp.n.sg ‘The grass has turned yellow.’

(Icelandic)

b. *Það er blánað. it.n.sg be.prs.3sg turn.blue.ptcp.n.sg (36) Jón er genginn út/*frá John.m.sg.nom be.prs.3sg go.ptcp.m.sg.nom out/from Selfossi. Selfoss.m.sg.dat

(Icelandic)

In Norwegian, there is considerable variation with regard to the preference for HAVE-perfects or BE-resultatives in resultative contexts, partly depending on the individual verb. The restrictions on BE are however very similar to Icelandic; examples are given in (37). Change of location verbs and change of state verbs are generally possible with BE, but there is variation with respect to variable behaviour verbs (in contexts without a particle).

The HAVE/BE alternation in Scandinavian

(37) a. Hun er dratt til Stockholm. she be.prs go.ptcp.n.sg to Stockholm ‘She has gone to Stockholm.’

165

(Norwegian)

b. Det er blitt kaldere. it be.prs become.ptcp.n.sg cold.comp ‘It has become colder.’ c. Vinden er økt wind.def.m.sg be.prs increase.ptcp.n.sg ‘The wind has increased.’ d. *Han er løpt til naboen he be.prs run.ptcp.n.sg to neighbour.def.m.sg The variation found in Present-Day Icelandic, Norwegian and Swedish follows the pattern expected from the Auxiliary Selection Hierarchy. The verbs at the top of the hierarchy are the only ones possible with BE in Present-Day Swedish, and in Icelandic and Norwegian, there is more variation with verbs that are lower down on the hierarchy. Note that this is despite the fact that the hierarchy is based on auxiliary selection languages – as we have seen, neither Icelandic, nor Norwegian, nor Swedish has auxiliary selection. Only Swedish shows a clear diachronic change in the verbs that have BEresultatives; in Norwegian and Icelandic (and Danish), the same verbs that occurred with BE in earlier stages are still possible with BE (modulo individual and dialectal variation). In the next section, we look at the loss of resultatives with BE in the history of Swedish. We will find additional support for the hierarchy.

4.2 Loss of BE in Swedish In the 17 th century, there is a shift in the distribution of BE-resultatives in Swedish, and the frequency of BE + a participle of an unaccusative verb drops (relative HAVE-perfects); see Table 3. In 17th-century Swedish, examples with BE and adverbs of iteration disappear, and BE is (again) restricted mainly to strictly resultative contexts. Moreover, the frequency of HAVE + unaccusative verb in resultative perfects increases and from the 17 th century onwards, examples like (38) become common in Swedish.

166

Ida Larsson

Tab. 3: The frequency of BE (relative HAVE) with unaccusatives in older Swedish. Past counterfactual contexts are not included (From Larsson 2009a: 248.).

Period I

Period II

Period III

Author (or text)

Date

# BE/tot

% BE

Bureus I + II Gyllenhielm Rosenhane Horn Columbus I, Mål-roo Columbus II, Ordeskötsel Börk I + II Philomela Gyllenborg TOTAL

*1568 *1574 *1611 *1629 *1642 *1642 *c. 1660 1668 *1679

29/ 36 43/ 45 78/ 89 36/ 57 23/ 39 19/ 43 11/ 54 12/ 37 19/ 51 270/451

81 % 96 % 88 % 63 % 59 % 44 % 20 % 32 % 37 % 60 %

(38) a. När en af hans trogneste konungzmänn hade when one of his faithful.superl kingsman.pl have.pst aflijdit decease.ptcp.n.sg ‘when one of his most faithful kingsmen had deceased’ (Early Modern Swedish; Columbus I *1642: 29) b. att dels ha redan flytt that some have.prs.pl already flee.ptcp.n.sg ‘that some have already fled’ (Early Modern Swedish; Börk I *c. 1660: 295) To some extent, the change is simply a shift in frequency; perfects with HAVE + an unaccusative verb can already be attested in Old Swedish. The loss of BE can be understood as the end of the rather slow establishment of the HAVEperfect. However, the change cannot only be described in terms of shifts in frequency or a change in the choice between two equally available constructions, the HAVE-perfect and the BE-resultative. We have already seen that older and modern Swedish differ with respect to which verbs are possible with BE. In the historical change of the language, different groups of verbs are affected in different ways; see Table 4. In the earliest period, variable behaviour verbs (in contexts without a particle) already occurred with BE less often (72 %) than did change of location verbs (92 %) and punctual change of state verbs (85 %). When conservative writers show variation between HAVE and BE, it tends to be with variable behaviour verbs or with change of state verbs; cf. (39a) and (39b).

The HAVE/BE alternation in Scandinavian

167

Tab. 4: The frequency of BE with different kinds of events.10 Past counterfactual contexts are not included (From Larsson 2009a: 262).

Period I Period II Period III TOTAL

Gradual change of state

Punctual change of state

Change of location

Variable behaviour

# BE

% BE

# BE

% BE

# BE

% BE

# BE

% BE

4/ 4 1/ 3 2/ 7 7/14

(10 %) (33 %) (29 %) 50 %

47/ 55 26/ 52 23/ 90 96/197

85 % 50 % 26 % 49 %

86/ 93 49/ 65 17/ 28 152/186

92 % 75 % 61 % 82 %

13/18 2/15 0/17 15/50

72 % 13 % 0% 30 %

(39) a. Jost Cursel […] och andra lifländare mera som vore aff Jost Cursel and other Livonian.pl more who be.pst.pl of godh villia medh redne good will with ride.ptcp.pl ‘Jost Cursel […] and several other Livonians who had ridden along out of free will’ (Early Modern Swedish; Gyllenhielm *1574: 327) b. haffver herr Nils Bielke […] rididt till herttigen have.prs.sg sir Nils Bielke ride.ptcp.n.sg to duke.def.m.sg ‘has Sir Nils Bielke ridden to the duke’ (Early Modern Swedish; Gyllenhielm *1574: 280) In the most modern texts in the material, variable behaviour verbs never occur with BE in the absence of a particle. Change of location verbs, on the other hand, still occur with BE in more than half of the cases in the third period in Table 4.

10 Following Ramchand and Svenonius (2002), I assume that particles can introduce a target state, and treat verb-particle combinations like rida ut ‘ride out’ together with inherently telic change of location verbs like anlända ‘arrive’. They also behave the same with respect to BE. I distinguish between examples that express a gradual change of state (e.g. brinna ‘burn’ and växa ‘grow’) and examples that express telic/punctual change of state (bli ‘become’, ske ‘happen’ and somna ‘fall asleep’). The latter group also involves constructions with verbs that otherwise express a gradual change of state in constructions with a particle (as in e.g. brinna upp ‘burn up’). Moreover, I distinguish variable behaviour verbs (e.g. resa ‘travel’) that do not occur in a construction with a particle from inherently telic change of location verbs (e.g. anlända ‘arrive’) which I group with variable behaviour verbs with particle. Change of location verbs that are not inherently telic and occur without particle (but with a PP) are referred to as variable behaviour verbs.

168

Ida Larsson

The loss of BE thus appears to involve a drop in the frequency of BE in all cases, but a greater drop in some contexts than in others, followed by the complete loss of BE with certain kinds of events. The change can be described as in (40), where the presence/absence of a target state in the verb phrase is taken to be a relevant factor, together with agentive/causative semantics. (40) Loss of BE in Swedish Change of location/controlled processes with target state (arrive, come) Change of state with target state (appear, happen) Gradual change of state Change of location without target state

+target state; +/−agentive +target state; −agentive +telic; −target state; −agentive +telic; −target state, +/−agentive

Thus, the pattern of change in Swedish is exactly what we expect given the variation in the modern Scandinavian languages, and the Auxiliary Selection Hierarchy.

4.3 A known pattern – but not auxiliary selection We have now seen that both synchronic and diachronic data from Scandinavian lend additional support to the Auxiliary Selection Hierarchy. The only group of verbs that is systematically possible in the complement of BE in Present-Day Swedish are the change of location verbs; they are at the top of the hierarchy. In Norwegian and Icelandic, there is more variation in the possibility of BE with gradual than with punctual change of state verbs, as expected from the hierarchy. However, two things should be noted. Firstly, although we have so far referred to variation and change with respect to the possibility of BE depending on verb types, it is clear that is not only a question of lexical properties of verbs (as the hierarchy might suggest). Although motional and non-motional processes are low on the hierarchy, they are possible with BE in Present-Day Swedish, when they incorporate a particle that introduces a target state: (41) a. Han är precis färdigsimmad. he be.prs just ready-swim.ptcp.c.sg ‘He has just finished with the swimming.’

The HAVE/BE alternation in Scandinavian

169

b. Han är precis färdigdansad. he is.prs just ready-dance.ptcp.c.sg ‘He has just finished with the dancing.’ Secondly, the hierarchy describes the likelihood of BE or HAVE in perfects: the lower on the hierarchy, the more consistently a verb forms a perfect with HAVE in languages like Danish, German or Italian, and the higher on the hierarchy, the more likely it is that BE is the only option. As we have seen, Icelandic, Norwegian and Swedish do not have auxiliary selection in perfects. Instead, HAVE is possible with all groups of verbs, and it is the only option in perfects (e.g. in past counterfactual contexts). In resultatives, on the other hand, BE is always the only possibility with intransitive verbs. Intransitive verbs that do not have active participles in resultatives with BE, do not have resultatives with HAVE. In other words, there is nothing corresponding to auxiliary selection in resultatives. The fact that the same patterns of variation and change can be found in both perfects and resultatives calls for an explanation. In the next section, I consider the varying structures of unaccusative verbs and point to properties that resultatives and perfects with unaccusative verbs arguably have in common.

5 Unaccusative structures In this section, I first give a structural account of the different groups of verbs in the Auxiliary Selection Hierarchy, using Ramchand’s (2008) verb phrase structure. In section 5.2, I suggest a way of unifying the different groups of unaccusative verbs. I also propose that the differences in variability are due to how specific the verbs’ lexical entries are.

5.1 Verb phrase structure and active participles with BE Ramchand (2008) argues that the verb phrase can be decomposed syntactically into three parts corresponding to three subeventualities of an event: initP represents the initial or causing state, procP the process or transition part of the event and resP the target state. The structure of a full verb phrase is given in (42). The projections are always ordered as in (42), but they do not all need to be present, and they are only present when they are associated with lexical material.

170

Ida Larsson

(42) Structure of the verb phrase:

(from Ramchand 2008) Ramchand assumes that verbs have combinations of category features that determine what structure they associate with. Arguments can have a combination of theta-roles, and, importantly, the verb sometimes specifies what the combinations must be (below this is represented by co-indexing of category features). The structures for process transitives like läsa ‘read’ and resultative transitives like förstöra ‘destroy’ are given in (43). Both läsa and förstöra take a subject which initiates a process, but in the case of förstöra, the object DP (and not the subject) undergoes the process and ends up in the target state. (43) a. Lexical entry for läsa ‘read’: [initi, proci] [initP x läsa [procP x läsa [y]]] b. Lexical entry for förstöra ‘destroy’: [init, proci, resi ] [initP x förstöra [procP y förstöra [resP y förstöra]]] (cf. Ramchand 2008a: 64–65) We now have a way to account structurally for the difference between older and Present-Day Swedish, as well as for the differences between Swedish, Norwegian and Icelandic. In this framework, the groups of verbs in the auxiliary selection hierarchy receive slightly different feature specifications and are associated with different structures; see (44). Punctual change of state verbs like hända ‘happen’ and gradual change of state verbs like växa ‘grow’ differ with regard to the res feature, but neither of them has initP in their structure. They can therefore typically be transitivised in English, since English presumably has null morphology that lexicalises init.

The HAVE/BE alternation in Scandinavian

(44) Loss of BE in Swedish Change of location/controlled processes with target state (arrive, come) Change of state with target state (appear, happen) Gradual change of state Change of location without target state

171

initi, proci, resi proci, resi proci initi, proci

We have seen that telic change of location verbs like anlända ‘arrive’ are the verbs with the most consistent unaccusative behaviour. These verbs express that their sole argument causes itself to move and end up in a different place/ state. In other words, change of location verbs involve a full verb phrase and take a subject which is the Initiator, Undergoer and Resultee. Unlike verbs like grow, verbs like arrive are lexically specified as intransitive and cannot be transitivised; cf. (45). Gradual change of state verbs like växa ‘grow’ have less specific lexical entries and also show variable behaviour. (45) a. *He arrived the boat. b. *He travelled her to Stockholm. c. She grows tomatoes. Let us now consider motional or non-motional controlled processes like simma ‘swim’ and dansa ‘dance’, which are at the bottom of the Auxiliary Selection Hierarchy and which are generally regarded as unergative verbs. Both swim and dance are eventive and take agentive subjects, and in the absence of e.g. a telic PP or particle, they are atelic. We can therefore assume that they are both lexically specified with init and proc features, and that they both take a subject which is the Initiator as well as the Undergoer (cf. Ramchand 2008a: 71 ff.). Neither of them is lexically specified for a target state. (46) a. Lexical entry for simma ‘swim’: [initi, proci] b. Lexical entry for dansa ‘dance’: [initi, proci] The difference between simma or dansa and telic change of location verbs like anlända does not lie in the presence/absence of an initP or an external argument (Initiator), but in the complement of proc: anlända involves resP, whereas proc in the structure of simma and dansa takes a Rheme or Path DP complement. However, it is presumably not specified in the lexical entry of these verbs what the complement of proc must be. This explains why, depending on con-

172

Ida Larsson

text, these verbs can display unaccusative behaviour. As we saw in example (41) above, both simma and dansa can have active participles in the resultative with BE in Present-Day Swedish when they incorporate a particle. In this respect, they behave just like variable behaviour verbs such as flyga ‘fly’ or springa ‘run’. Note also that dance can express change of location, and that swim does not need to: (47) a. Han simmade fjärilsim (men rörde sig inte alls). he swim.pst butterfly-swim but move.pst refl not at.all ‘He did the butterfly stroke (but didn’t move at all).’ b. Han dansade hela vägen hem. he dance.pst all way.def.c.sg home ‘He danced all the way home.’ In Present-Day Swedish, there are two requirements on active participles in the complement of BE. First, either the verb itself or an element (a particle) incorporated into the participle has to carry a res feature; a telic PP is not enough for BE to be possible with variable behaviour verbs: (48) a. *Han är sprungen till affären he be.prs run.ptcp.c.sg to store.def.c.sg ‘He has run to the store.’ b. Han är hitsprungen. he be.prs here-run.ptcp.c.sg ‘He has run here.’ However, not all inherently telic verbs or particle verbs are possible with BE; see (49). It seems that an init feature is also required in Present-Day Swedish. The difference between unergatives like work and unaccusatives like arrive can therefore not be understood in terms of (non-)agentivity or absence of an external argument. (49) *Den är uppbrunnen. it.c.sg be.prs up-burn.ptcp.c.sg ‘It has burnt up.’ The verbs that occur with BE most consistently (in perfects or resultatives) are verbs that have an init feature and that express agentivity or causation. This is clearly not how we usually distinguish unaccusatives. Unaccusative verbs are

The HAVE/BE alternation in Scandinavian

173

generally taken to be non-agentive and have a subject which is an underlying object (see e.g. Burzio 1986). In the next section, I look closer at the arguments for including initP in the structure of some unaccusatives, and I briefly sketch a way of unifying the different groups of unaccusative verbs that comes close the standard account but allows for variation.11

5.2 A note on unaccusativity and variability With respect to change of location verbs, it is quite clear that they involve an initP, and that the sole argument is the Initiator. These verbs have active participles in the complement of BE in Present-Day Swedish, but they also occur in impersonal passives: see (50). Change of state verbs like intransitive grow or wilt cannot be passivised; see (51). (50) a. Det skulle resas till Stockholm och gås på there would travel.inf.pass to Stockholm and go.inf.pass on teater. theatre ‘People would travel to Stockholm and go to the theatre.’ b. Det ankoms och avrestes hela dagen. there arrive.pst.pass and off-travel.pst.pass whole day.def.c.sg ‘People were arriving and departing all day.’ c. Sedan anländes det till Arlanda och stods i then arrived.pst.pass there to Arlanda and stand.pst.pass in kö i timmar. line for hour.pl ‘Then there was the arrival to Arlanda and the standing in line for hours.’ (51) a. *Det växtes i trädgårdarna. there grow.pst.pass in garden.def.pl b. *Det vissnas i vasen. there wilt.prs.pass in vase.def.c.sg

11 I disregard the structure of stative verbs. They typically do not have participles in resultatives or other tenseless contexts, and in Danish they do not have perfects with BE.

174

Ida Larsson

As pointed out by Zaenen (1988) and Thráinsson (2007), verbs tend to require an agentive reading in impersonal passives (see also Barðdal and Mólnar 2000: 129 and Hoekstra and Mulder 1990 for Icelandic and Dutch, respectively). It is also standardly assumed that passivisation requires an external argument (an Initiator). Since other verbal passives necessarily have a transitive structure (with one implicit and one explicit argument), verbs like arrive, which are obligatorily intransitive, are restricted to impersonal passives. The presence of an initP makes impersonal passivisation possible, but it does not necessarily make the verb unergative. It is clear that verbs like arrive have an initP also in the resultative construction with BE. Agent-oriented adverbials are compatible with BE throughout the history of Swedish period; consider (52).12 In the 17th century, BE is even somewhat more common in clauses with agent-oriented adverbials, than overall (Larsson 2009a: 257). (52) Owe Rammel war upsåtligen och friwilligt inflytt i Owe Rammel be.pst.sg intentionally and voluntarily in-flee.ptcp in Christianstad Christianstad ‘Owe Rammel had intentionally and voluntarily fled into Christianstad.’ (Early Modern Swedish; Spegel *1645: 57) I conclude that unaccusativity cannot be characterised by the absence of an external argument (Initiator) (cf. Zaenen 1988). The question, then, is what we mean when we say that verbs like arrive, drown or grow are unaccusative, or that some verbs have more consistent unaccusative behaviour than others. On the account sketched here, a particular verb is not marked as unaccusative in its lexical entry (i.e. there is no grammatical feature ±unacc or a vunacc); a verb like grow is compatible both with unaccusative and transitive structures, and a verb like arrive behaves like a typical unaccusative with respect to certain diagnostics (auxiliary selection) and like an unergative according to others (impersonal passivisation). If we account for the variable behaviour of verbs like grow by saying that the root can occur in both unaccusative and transitive/unergative structures, we are either forced to say that this is also the case for verbs like arrive, or to assume that only one

12 The participle inflytt in (52) looks like a non-agreeing neuter singular form. However, Spegel often has devoicing of final /d/, and the masculine singular and neuter singular will therefore sometimes be identical.

The HAVE/BE alternation in Scandinavian

175

of the diagnostics (participle formation and passivisation) distinguishes unaccusatives from unergatives. Given the well-known mismatches between different diagnostics for unaccusativity and the cross-linguistic variation, it seems clear that no single factor can be invoked to cover all cases. It is well known that different diagnostics can be sensitive to different properties; impersonal passivisation requires an external argument (an Initiator), whereas only verbs that lack an init feature partake in the causative-inchoative alternation (at least in the absence of morphological marking). For the possibility of active participles with BE in PresentDay Swedish, we have seen that both resP and initP are relevant. However, the requirements vary depending on language and time. At the same time, the group of verbs that form active participles with BE or in e.g. reduced relatives, are rather stable, cross-linguistically speaking, and even in Present-Day Swedish, verbs like växa ‘grow’ and drunka ‘drown’ (unlike e.g. arbeta ‘work’) have active participles in attributive position, although they are not possible with BE. With the present account, this remains to be explained. We would clearly lose a generalisation about verb structure if we were only able to refer to individual groups of verbs (e.g. telic change of location verbs), or individual features (e.g. res) depending on diagnostics, rather than to the general distinction between unaccusatives and unergatives. Now, it is important that in the account sketched above, no theta-criterion is assumed. This means that an argument can be both an external argument (Initiator) and an internal argument (i.e. base-generated lower down in the verb phrase). In fact, all groups of verbs under discussion have a subject which is base-generated in an internal argument position. This is true also for change of location verbs where the subject is the Undergoer and Resultee, as well as the Initiator. In this respect, the Unaccusative Hypothesis is preserved in the present system. We can assume that gradual change of state verbs involve an adjectival or scalar structure embedded under proc. This is a rather standard analysis of deadjectival verbs like gulna ‘turn yellow’, and the analysis can be extended straightforwardly to verbs like vakna ‘awaken’ and drunkna ‘drown’ which also have the inchoative suffix -na. For verbs like växa ‘grow’ we have to assume an abstract scale, which is defined relative to the argument which is promoted to subject in the intransitive use. This would give us a structure for unaccusatives as in (53), where XP is either resP or a phrase introducing a scale. Formulated like this, all unaccusative verbs under discussion have a subject generated in a predicational structure in the complement of proc. This account comes very close to suggestions by e.g. Moro (1997), Hoekstra (1999) and Folli and Harley (2006).

176

Ida Larsson

(53) Unaccusative participial structure: [PartP DPi –en … [procP DPi proc … [XP DPi X We expect the nature of the complement of proc to affect the temporal contour of the process (telicity), but on the present account, unaccusative structures are neither defined in terms of telicity, nor (non-)agentivity. With an account along these lines, we have a way of structurally identifying the superset of unaccusative verbs, while still allowing individual diagnostics to pick out structurally defined subsets. Diagnostics can, for instance, be sensitive to the exact nature of XP, and, as we have seen, also to other properties of the subject and the verb phrase. A couple of questions relating to the observed variation still remain. First, more needs to be said in order to explain the fact that verbs like arrive have a more stable behaviour with respect to auxiliary selection and BE-resultatives than do verbs like break and grow. Secondly, we need to explain why this same pattern can be found in both perfects and resultatives. With respect to the difference in variability, this follows directly from the way we formulated the lexical specifications of the verbs. The verbs with the most stable behaviour have the most specific lexical entry: change of location verbs always denote complex events with both initP, procP and resP, and they are always associated with the full verb phrase structure (cf. Sorace 2000 for a similar idea). Telic change of state verbs have less lexical specification and a somewhat more variable behaviour. Finally, verbs like grow have little lexical specification, can occur in both transitive and intransitive structures, and in fact sometimes combine with a DP other than the subject, which measures out the process (cf. grow a centimeter with *arrive a kilometer). With respect to the fact that the same pattern of inter- and intra-language variation and change can be observed for both resultatives and perfects, it seems reasonable to view this as a consequence of the properties of the participle. I have taken the participial clauses to have partly different structures in the different constructions: perfects involve a tensed participial clause, resultant state constructions a tenseless but aspectual participle, whereas target state participles lack both tense and aspect. At the same time, I have (implicitly) assumed that participial morphology spells out a participial head below Tense, Aspect and the functional head responsible for passive voice (see the structures given in 26, 27 and 53 above), and this arguably shares properties across constructions (see Larsson and Svenonius 2013 for discussion). We know that participle formation is sensitive to unaccusativity, also outside perfects and resultatives with BE.

The HAVE/BE alternation in Scandinavian

177

6 Conclusion: gradualness, gradience and auxiliary selection In this paper, I have investigated different types of variation with respect to the alternation between HAVE and BE in the Scandinavian languages. First, we considered the alternation between HAVE-perfects and BE-resultatives in Old Icelandic and Old Swedish, and noted variations in the aspectual composition of the resultative participle. Secondly, we looked at the varying restrictions on the unaccusatives that occur in resultatives with BE. In this section, I conclude the paper by relating the variation to the questions of gradualness and gradience, and I briefly comment on the difference between the auxiliary selection and the alternation between HAVE and BE in Scandinavian.

6.1 Variation, change, gradualness and (non-)gradience Throughout the history of Scandinavian, we can observe gradual shifts in the distribution of HAVE-perfects and BE-resultatives, and the preference for resultatives with BE over perfects with HAVE, or for perfects over resultatives, varies depending on language and time. In Present-Day Swedish, the perfect with HAVE is generally used in both experiential and resultative contexts, whereas in Icelandic and in Old Scandinavian, the resultative construction with BE + active participle of an unaccusative verb is often preferred over a perfect in resultative contexts. Moreover, older Swedish sometimes had a perfective resultant state participle where Icelandic and Present-Day Swedish would have a perfect. It is, however, never a question of complete blocking: both BE and HAVE are possible with (groups of) unaccusative verbs, depending on the context. In both Swedish and Icelandic, there is a small, but clear, difference in meaning between the HAVE-perfect and the resultative construction with BE, and as we have seen there are contexts where only the former is allowed. On the present account, the differences between the constructions with BE and the perfect with HAVE are non-gradient. The difference between (Old) Swedish and (Old) Icelandic can also be given a non-gradient, structural explanation: in Icelandic, perfective resultant state participles do not seem to be possible. The differences between the languages are in other words not solely a question of speakers making different choices between equally possible constructions. In addition to the general variation in the distribution of perfects and resultatives, we have noted variation in the groups of verbs that have active participles with BE, and we have seen that variation and change in Scandinavian follows the pattern expected according to the Auxiliary Selection Hierarchy,

178

Ida Larsson

although in Scandinavian, the alternation between HAVE and BE should not be understood in terms of auxiliary selection. Again, the account that I have sketched does not make any direct reference to gradience, but assumes that the different groups of verbs in the auxiliary selection hierarchy have different structural requirements (tied to their semantics). In other words, I take the difference between older and Present-Day Swedish, or between Present-Day Swedish and Icelandic or Norwegian to be structural and, in principle, nongradient. In Present-Day Swedish, the subject of the participle needs to both the Initiator and Resultee in the resultant state construction with BE, whereas in Norwegian and Icelandic, the requirement is that the subject is generated in predicational structure in the complement of proc (but as we have seen, there can be further restrictions on the type of complement). However, by tying the restrictions to the lexical specifications of verbs, we expect different verbs to behave in different ways, and it is clear that also the encyclopaedic content of the lexical items and our real-world knowledge can be significant for the possibility of BE and the choice between HAVE and BE. This is also the case for stative passives. The fact that (54a) is perfectly natural in Present-Day Swedish, whereas (54b) is odd, does not necessarily depend on structural differences between the verbs rapa ‘burp’ and sparka ‘kick’ but rather relates to the fact that we can more easily provide a context for (54a). (54) a. Babyn är rapad. baby.def.c.sg be.prs burp.ptcp.c.sg ‘The baby has been burped.’ b. # Däcken är sparkade tires.def.pl be.prs kick.ptcp.pl In other words, what we observe to be gradual differences between speakers (texts, languages) depends partly on individual choices (tied to the conventions of the speech community and real world knowledge) and partly on grammatical differences between languages. In the historical process of language change, the two are interrelated. When the perfect with HAVE + participle of unaccusative verbs is generalised in 17th-century Swedish, the resultative construction with BE consequently becomes more restricted. In a sense, the development begins as soon as the perfect with HAVE emerges, i.e. already in the oldest Scandinavian sources. At the initial stage of the development, HAVEperfects are infrequent with all groups of verbs, but perfects with HAVE + unaccusative verbs occur as early as in Old Scandinavian, as we have seen.

The HAVE/BE alternation in Scandinavian

179

6.2 HAVE/BE in Scandinavian vs. auxiliary selection I have argued that the alternation between HAVE and BE in the Scandinavian languages, apart from Danish, should not be understood as auxiliary selection. As I pointed out in section 4.3 above, there is no correspondence to auxiliary selection in resultatives: the unaccusatives that are disallowed in resultatives with BE do not form resultatives with HAVE. However, there is a resultative construction with possessive HAVE and a passive participle (see Larsson 2009a, 2012 for discussion): (55) a. Vi har fortfarande väskorna packade. we have.prs still bags.def.pl pack.ptcp.pl ‘We still have the bags packed.’ b. Vi har (*fortfarande) blommorna we have.prs still flowers.def.pl vattnade. water.ptcp.pl ‘We have the flowers watered.’

(Target state)

(Resultant state)

Following Harley (2002), we can assume that possessive HAVE decomposes into BE + a preposition that introduces possessive semantics and the possessor DP (cf. Kayne 1993). A schematic structure of possessive HAVE + resultant state participle is given in (56). (56) Possessive HAVE + resultant state participle: [DPi BE [PP DPi Pposs [AspP DPj Asp [VoiceP Voice [PartP DPj –en [initP Temporal HAVE, on the other hand, is clearly a raising verb, and it lacks possessive semantics. We can therefore assume that temporal HAVE does in fact not involve a possessive prepositional element. In Larsson (2009a: 361–370), I suggest that temporal HAVE is an alternative spell-out of the copula in languages like Danish and German, i.e. languages with auxiliary selection.13 Unlike the HAVE/BE alternation in Swedish and Icelandic, auxiliary selection is 13 Cf. Hoekstra (1994) who argues that possessive HAVE and temporal HAVE differ with regard to the presence/absence of an element X in the complement; X and not HAVE is responsible for theta-assignment, and it is present with possessive HAVE and not with temporal HAVE. In Larsson (2012b), I argue that the prepositional element is present in perfects in languages like Swedish, which lack auxiliary selection, but that it then takes temporal arguments and makes a semantic contribution. In this way, I account for the restrictions on positional past tense adverbials in the present perfect in Swedish.

180

Ida Larsson

then basically a morphological agreement phenomenon (cf. Bentley and Eythórsson 2003). Notably, auxiliary selection, but not the possibility of BE in resultatives, can depend on tense, mood, person and number, and it does not always correlate with the structure of the verb phrase (see McFadden 2007 for an overview). In the Procidano dialect of Italian, for instance, the present perfect has HAVE, whereas the pluperfect has BE: (57) a. jé nun hó/*so’ sturiéto, ma … I not have.prs.1sg/be.prs.1sg study.ptcp.m but ‘I haven’t studied, but …’ b. a primma lenza nen lu fovo canisciuto at first look non him be.pst.1sg know.ptcp.m ‘at first I had not recognized him’ (Ledgeway 2000: 202) At the same time, the restrictions on resultatives with BE in Swedish have clear parallels in auxiliary selection languages. As we have seen, variation and change in Scandinavian resultatives follows the pattern expected from the auxiliary selection hierarchy. In other words, we want an account of resultatives with BE that captures the similarities between resultatives and perfects, as well as the differences. We can assume that the copular verb BE is the same across the constructions, and that the copula is also present in possessive and perfect structures with HAVE. Moreover, I have assumed that despite differences relating to stativity, aspect and tense, participial morphology spells out the same structure across the constructions (but I have left the properties of the participial head unspecified). Lastly, I have tied the fact that some verbs show more variable behaviour than others to differences in lexical specification: some verbs do not express complex events (in the absence of additional morphology) and simply have a less specified lexical entry than other verbs. The different groups of unaccusative verbs are, however, unified by having a subject which is base-generated in the complement of proc.

References Alexiadou, Artemis and Elena Anagnostopoulou. 2008 Structuring participles. In: Charles B. Chang and Hannah J. Haynie (eds.), Proceedings of the 26 th West Coast Conference on Formal Linguistics, 33–41. Somerville, MA: Cascadilla Proceedings Project. Barðdal, Jóhanna and Valéria Mólnar. 2000 Passive in Icelandic – Compared to Mainland Scandinavian. Working Papers in Scandinavian Syntax 65: 109–146. Bentley, Delia and Thórhallur Eythórsson. 2003 Auxiliary selection and the semantics of unaccusativity. Lingua 114: 447–471.

The HAVE/BE alternation in Scandinavian

181

Burzio, Luigi. 1986 Italian Syntax: a Government-Binding Approach. Dordrecht: Reidel. Carey, Kathleen. 1994 The grammaticalization of the perfect in Old English. An account based on pragmatics and metaphor. In: William Pagliuca (ed.), Perspectives on Grammaticalization, 103–117. Amsterdam, Philadelphia: John Benjamins. Cennamo, Michela and Antonella Sorace. 2007 Auxiliary selection and split intransitivity in Paduan. In: Raúl Aranovich (ed.), Split Auxiliary Systems: a Cross-Linguistic Perspective, 65–99. Amsterdam, Philadelphia: John Benjamins. Falk, Hjalmar and Alf Torp. 1900 Dansk-norskens syntax i historisk fremstilling [DanishNorwegian Syntax from a Historical Perspective]. Kristiania: H. Aschehoug & Co. Folli, Raffaella and Heidi Harley. 2006 On the licensing of causatives of directed motion: Waltzing Matilda all over. Studia Linguistica 60: 121–155. Harley, Heidi. 2002 Possession and the double object construction. Linguistic Variation Yearbook 2: 29–68. Hoekstra, Teun. 1999 Auxiliary selection in Dutch. Natural Language and Linguistic Theory 17: 67–84. Hoekstra, Teun and René Mulder. 1990 Unergatives as copular verbs: locational and existential predication. The Linguistic Review 17: 1–79. Iatridou, Sabine. 2000 The grammatical ingredients of counterfactuality. Linguistic Inquiry 31: 231–270. Iatridou, Sabine, Elena Anagnostopoulou and Roumyana Izvorski. 2001 Observations about the form and meaning of the perfect. In: Michael Kenstowicz (ed.), Ken Hale: a Life in Language, 189–238. Cambridge, Mass.: MIT Press. Julien, Marit. 2001 The syntax of complex tenses. The Linguistic Review 18: 125–167. Johannisson, Ture. 1945 Hava och vara som tempusbildande hjälpverb i de nordiska språken [Have and be as temporal auxiliaries in the Nordic languages]. Lund: Gleerups. Katz, Graham. 2003 On the stativity of the English perfect. In: Artemis Alexiadou, Monika Rathert and Arnim von Stechow (eds.), Perfect Explorations, 205–234. Berlin, New York: De Gruyter. Kayne, Richard S. 1993 Toward a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. Kirri, Arto. 1975 Studier över passivkonstruktioner i nysvenskt skriftspråk [Studies over Passive Constructions in Early Modern and Modern Swedish Written Language]. Helsingfors: Helsingin yliopiston monistuspalvelu. Kratzer, Angelika. 2000 Building statives. In: Lisa J. Conathan, Jeff Good, Darya Karitskaya, Alyssa B. Wulf and Alan C. L. Yu (eds.), Proceedings of the 26th Annual Meeting of the Berkeley Linguistics Society, 385–399. Berkeley: University of California, Berkeley Linguistics Society. Larsson, Ida. 2009a Participles in time. The development of the perfect tense in Swedish. Gothenburg: University of Gothenburg. Larsson, Ida. 2009b Becoming perfect. Observations about Icelandic vera wbúinn að. Íslenskt mál 30: 53–92. Larsson, Ida. 2012 On the development of the perfect (participle). In: Ans van Kemenade and Nynke de Haas (eds.), Historical Linguistics 2009: Selected Papers from the 19 th International Conference of Historical Linguistics, 191–210. Amsterdam, Philadelphia: John Benjamins. Larsson, Ida. 2014 HAVE/BE + participle of an unaccusative verb. Nordic Atlas of Language Structures (NALS) Journal Vol. 1: 381–395, http://www.tekstlab.uio.no/nals.

182

Ida Larsson

Larsson, Ida and Peter Svenonius. 2013 English and Scandinavian participles and the syntax-morphology interface. Paper presented at the 25 th Scandinavian Conference of Linguistics in Reykjavík, 12–15 May 2013. Ledgeway, Adam. 2000 A Comparative Syntax of the Dialects of Southern Italy: A Minimalist Approach. Oxford, Boston: Blackwell. Legendre, Geraldine and Antonella Sorace. 2003 Split intransitivity in French: an optimality-theoretic perspective. In: Danièle Godard (ed.), Les langues romanes : problèmes de la phrase simple, 185–234. Paris: CNRS Editions. McFadden, Thomas. 2007 Auxiliary selection. Language and Linguistics Compass 1: 674– 708. McFadden, Thomas and Artemis Alexiadou. 2006 Pieces of the be perfect in German and Older English. In: Donald Baumer, David Montero and Michael Scanlon (eds.), Proceedings of the 25 th West Coast Conference on Formal Linguistics, 270–278. Somerville, MA: Cascadilla Proceedings Project. McFadden, Thomas and Artemis Alexiadou. 2010 Perfects, resultatives, and auxiliaries in earlier English. Linguistic Inquiry 41(3): 389–425. Moro, Andrea. 1997 The Raising of Predicates. Cambridge: Cambridge University Press. Pancheva, Roumyana. 2003 The aspectual makeup of perfect participles and the interpretations of the Perfect. In: Artemis Alexiadou, Monika Rathert and Arnim von Stechow (eds.), Perfect Explorations, 277–308. Berlin, New York: De Gruyter. Parsons, Terence. 1990 Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, Mass: MIT Press. Ramchand, Gillian. 2008a Verb meaning and the Lexicon. A First-Phase Syntax. Cambridge: Cambridge University Press. Ramchand, Gillian and Peter Svenonius. 2002 The lexical syntax and lexical semantics of the verb-particle construction. In: Line Mikkelsen and Christopher Potts (eds.), Proceedings of the 21 st West Coast Conference on Formal Linguistics, 387–400. Somerville, Mass: Cascadilla Press. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76: 859–890. Sorace, Antonella. 2004 Gradience at the lexicon–syntax interface: evidence from auxiliary selection and implications for unaccusativity. In: Artemis Alexiadou, Elena Anagnostopoulo and Martin Everaert (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 243–268. Oxford: Oxford University Press. Teleman, Ulf, Staffan Hellberg and Erik Andersson. 1999 Svenska Akademiens grammatik 1–4 [The Swedish Academy Grammar 1–4]. Stockholm: Norstedts Ordbok. Thráinsson, Höskuldur, Hjalmar P. Petersen, Jógvan Í Lon Jacobsen and Zakaris Svabo Hansen. 2004 Faroese, An Overview and Reference Grammar. Tórshavn: Føroya Fró∂skaparfelag. Thráinsson, Höskuldur. 2007 The Syntax of Icelandic. Cambridge: Cambridge University Press. Vikner, Sten and Rex A. Sprouse. 1988 Have/Be-selection as an A-chain membership requirement. Working Papers in Scandinavian Syntax 38. Yamaguchi, Toshiko and Magnús Pétursson. 2003 The speaker and the perfective auxiliaries hafa and vera in Icelandic. Language Sciences 25: 331–352. Zaenen, Annie. 1988 Unaccusative verbs in Dutch and the syntax-semantics interface. CSLI Report 123. Stanford: CSLI Publications.

Jaume Mateu and Mar Massanell i Messalles, Universitat Autònoma de Barcelona

A constructional approach to auxiliary selection: * evidence from existential constructions 1 Introduction In Mateu (2009) the following descriptive generalisation was pursued: the intransitive verbs that exhibit gradient auxiliary selection preference in Contemporary Italian (Sorace 2000, 2004) typically coincide with the ones that earlier lost the BE auxiliary in both Old Catalan (Batlle 2002) and Old Spanish (Aranovich 2003).1 Drawing on data from these sources,2 Mateu (2009) pointed out that it cannot be a mere coincidence that in both Old Catalan and Old Spanish verbs of existence and appearance3 were among the first to admit the HAVE auxiliary, while the rest of the unaccusative verbs were more reluctant to accept it. Interestingly, as pointed out by Sorace (2000, 2004), intransitive verbs of existence and appearance show gradience in Italian (e.g., see her examples

* We want to thank the editors and the anonymous reviewers for constructive comments and helpful suggestions. All errors are ours. The first author acknowledges the funding of grants FFI2011-23356 (Spanish Ministerio de Economía y Competitividad) and 2014 SGR-1013 (Generalitat de Catalunya). The second author acknowledges the funding of the grant 2014 SGR-1328 (Generalitat de Catalunya). 1 E.g., cf. Aranovich (2003: 5–6): “A quick glance at the verbs (…) reveals that the degree of affectedness of the subject is a factor in the displacement of ser by haber as the perfect auxiliary. At one end of the continuum are the subjects of stative verbs of existence and appearance like quedar ‘remain’. The subjects of these verbs do not suffer any changes in state or location, hence they are not affected in any way by the event. This is the first class to lose its ability to select ser. At the opposite end are subjects of verbs of directed motion and verbs of change of state. These subjects are affected since they are in a new location or state as a consequence of the event. These classes are the last ones for which haber displaces ser as the perfect auxiliary of choice. In between these two extremes are verbs of manner of motion like correr ‘run’, and dynamic verbs of existence and appearance like desaparecer ‘disappear’. (…) The chronology of split auxiliary selection in Spanish, then, falls under the generalization that the less affected the subject, the earlier a verb lost its ability to select auxiliary ser”. 2 Unfortunately, Sorace’s (2000) important work on gradience in auxiliary selection with intransitive verbs was not mentioned by Batlle (2002) nor by Aranovich (2003). 3 Levin and Rappaport Hovav (1995: 282) include remain, endure, come, and exist within this class. Come is also classified within the class of “verbs of inherently directed motion” (1995: 281).

184

Jaume Mateu and Mar Massanell i Messalles

in 1c–d). In contrast, core or prototypical unaccusatives like (1a–b) do not show any gradient variation at all. (1) a. Gianni è / *ha arrivato. Gianni is / has arrived

(Italian)

b. Gianni è / *ha morto. Gianni is / has died c. La guerra è durata / ?ha durato a lungo. the war is lasted / has lasted for long d. I dinosauri sono esistiti / ??hanno esistito 65 milioni di anni the dinosaurs are existed / have existed 65 millions of years fa. ago (2) shows Sorace’s (2000: 863) famous Auxiliary Selection Hierarchy (ASH), which is nicely paraphrased in (3) by Keller and Sorace (2003: 60–61). In particular, it is worth pointing out that core verbs like the ones in (1a–b) are referred to as “those on which native grammaticality judgments are maximally consistent”. In contrast, non-core or intermediate verbs like the ones in (1c–d) “exhibit gradient auxiliary selection preference”. (2) change of location change of state continuation of a pre-existing state existence of state uncontrolled process controlled process (motional) controlled process (nonmotional)

selects BE – least variation

selects HAVE – least variation

(3) Verbs at the BE end of the ASH are core unaccusatives and denote telic change; verbs at the HAVE end are core unergatives and denote agentive activity in which the subject is unaffected. Intermediate verbs between the two extremes incorporate telicity and agentivity to lesser degrees, and tend to have a less specified (basically stative) event structure […]. Core verbs are those on which native grammaticality judgments are maximally consistent, and are acquired early by both first and second language learners. In contrast, intermediate verbs are subject to crosslinguistic differences and exhibit gradient auxiliary selection preferences (Keller and Sorace 2003: 60–61).

A constructional approach to auxiliary selection

185

The main goal of the present chapter is to deal with some apparent counterexamples to Sorace’s (2000, 2004) ASH: for example, in Old Catalan (OC) socalled core unaccusative verbs like venir ‘come’ are found in an existential argument structure where HAVE becomes the typical auxiliary. We will see that OC verbs like venir ‘come’ can be associated with two different syntactico-semantic structures: a subject of result structure, which selects BE quite systematically, and an existential structure, which typically selects HAVE. For instance, consider the following relevant example from OC where the same verb venir ‘come’ appears to be integrated into these two structures: (3) E ab veritat se pot dir que, de tots quants grans senyors and with truth refl can.3sg say that, of all those great lords hi són venguts, no n’í ha vengut negú qui ab loc.cl. are come, no part.cl.-loc.cl. has come none who with tan gentil orde sia vengut so gentle order be.subj.3sg. come ‘And one can truthfully say that, among all those great lords who came, no one has come who came so gently.’ (1490 [1460], Martorell, Tirant, p. 279, l. 24) The structure of this chapter is as follows. In section 2 we show that Sorace’s (2000, 2004) ASH, which has been interpreted in terms of lexical-semantic classes of verbs (cf. 2), can be recast in terms of syntacto-semantic constructions. In particular, the different structural properties associated with subject of result constructions, which select BE quite systematically in Old Catalan, and existential constructions, which typically select HAVE in this language, will be shown to favour a constructional approach to auxiliary selection. In section 3 we revise synchronic evidence from Italian existential constructions that also favours a constructional approach to auxiliary selection. In particular, evidence for this approach comes from the fact that so-called “core” unergative verbs like It. lavorare ‘work’ and cantare ‘sing’ do not select HAVE when they are integrated into existential constructions, which can be taken as a sign of their unaccusativity. Section 4 contains some concluding remarks. In particular, we conclude that (i) unaccusativity is not a property of verbs but rather of constructions (see also Moro 1997 or Rigau 1997, among others, as well as Gillmann, this volume) and (ii) auxiliary selection is not a matter of verbs but rather of constructions (see also McFadden and Alexiadou 2010 and Rosemeyer 2014: chapter 6, among others).

186

Jaume Mateu and Mar Massanell i Messalles

2 Some diachronic evidence for a constructional approach to auxiliary selection The results obtained from our study of auxiliary selection in Old Catalan (OC) using the CICA corpus (Massanell i Messalles and Mateu, in prep.)4 show that constructions with verbs like venir ‘come’ select BE when a definite theme is involved (see 5) and select HAVE in examples that contain an indefinite one (cf. 6).5 Notice, for instance, the contrast between the examples in (5c) and (6b), which are taken from the same text. (5) a. Quant la barcha fo venguda, … when the boat was come (1275–1299, Desclot, Crònica II, p. 100, l. 15) b. E quant En Volusià fo vengut a·N Pilat, … and when det. Volusianus was come to Pilate … (1275–1299, Vides, p. 366, l. 28)

4 The recent CICA corpus (cf. http://www.cica.cat) has been created with the intention of helping the authors of the Gramàtica del Català Antic [Grammar of Old Catalan], which will include contributions by eighty scholars. At the moment the research presented in this chapter was carried out, the CICA corpus contained more than 400 texts, from the 11 th to the 18 th century. For the relevance of CICA to the topic of perfect auxiliaries in Old Catalan, see Massanell i Messalles (2009). 5 HAVE is also typically selected when venir ‘come’ takes a sentential argument, be it an infinitival clause (ia) or a finite clause introduced by que ‘that’ (ib–c). This notwithstanding, see the example in (ii) for a conservative use of BE. (i) a.

Allí estigérem lo dimarts ý partírem lo dimecres, ý à·ns there be.past.1pl the Tuesday and left.past.1pl the Wednesday, and has-usdat vengut en camí pasar per la Membrilla (…). come on path pass through the Membrilla (1525–1542, Liori i Requesens, Epistolaris, c. 38, l. 18)

b.

À·m vengut a conte que ella fos per a tenir càrec del chic (…). has-medat come to mind that she was for to have charge of.the boy (1525–1542, Liori i Requesens, Epistolaris, c. 110, l. 40)

c.

ha vingut a sa notícia que los jurats de la dita vila (…) has come to his notice that the juries of the mentioned village (1585, Corts, p. 36, l. 33)

(ii) ara és vengut a nostra sabuda, que vós havets fet (…) now is come to our knowledge that you have made (1378–1399, Mostassaf VIII, p. 232, l. 29)

A constructional approach to auxiliary selection

187

c. de part lur és vengut lo mal from part their is come the wrong (1397–1398, Mostassaf I, p. 158, l. 31) d. E baste a tu aquest dan tan gran que per tu me and suffices to you this damage so great that for you to.me és vengut is come (1442–1458 [1425–1448], Curial, p. 321, l. 2) e. La malaltia ere venguda per peccat the illness was come for sin (1445–1449 [1410–1415], Sant Vicent, Sermons IV, p. 154, l. 8) f. La marquesa de Agilar és ja venguda the marchioness of Agilar is already come (1525–1542, Liori i Requesens, Epistolaris, c. 65, l. 52) g. uns hòmens de Marthorell que eren venguts a negociar ab some men from Martorell who were come to negotiate with mi me (1525–1542, Liori i Requesens, Epistolaris, c. 177, l. 10) (6) a. per lo qual trencament los à vengut gran dan a by the which break themdat has come great damage to lur molí their mill (1263–1265, Alcoi, f. 10r, l. 10) b. ne negun mal (…) no ha vengut de part lur nor none wrong not has come from part their (1397–1398, Mostassaf I, p. 156, l. 20) c. dubtava que a sa filla no hagués vengut algun mal doubted.3sg that to his daughter not had come some wrong soptosament suddenly (1490 [1460], Martorell, Tirant, p. 1379, l. 29) d. E dix a sa mare congoxa li havia venguda and said to her mother anguish dat.cl. had come (c. 1499, París, f. 8v, l. 20)

188

Jaume Mateu and Mar Massanell i Messalles

e. requerós àn vengut alguns messengers have.3pl come some (1525–1542, Liori i Requesens, Epistolaris, c. 139, l. 14) f. me ha vingut ara a la memòria una altra cosa sobre to.medat has come now to the memory an other thing about assò this (1557, Despuig, Col·loquis, p. 176, l. 5) Similarly, the relevant contrasts shown in (7) and (8), which are drawn from Batlle (2002), lead to the descriptive generalisation that BE is associated with a construction that involves a definite theme (e.g., cf. les noves ‘the news’; la dessa Pal·las ‘the goddess Palas’), whereas HAVE is associated with one that involves an indefinite theme (e.g., cf. nova ‘new(s)’; grans llams y foch ‘great lightning and fire’). (7) a. A 14 de yuliol, per les noves que heren vingudes que los at 14 of july, by the news that were come.pp.fem.pl that the tortosins … Tortosians (1572–1600, F. Despalau) b. Vuy, que contam a 3 de desembre, ha vingut nova com today, that count.we at 3 of December, has come new(s) how don Alonso … Mr. Alonso (1572–1600, F. Despalau) (8) a. Ayxí com los romans tenien la imatge de la dessa like as the Romans had the image of the goddess que deyan que era cayguda del cel en los murs who said that was fallen from.the sky on the walls y aprés portada en Roma, … and after brought in Rome (XVI–XVII, Ponç d’Icard)

Pal·las, Palas, de Troya of Troya

b. Encara aquest no havia acabat quant aplegà altre portant yet this not had finished when gathered another bringing nova com grans llams y foch havia caygut del cel, … news how great lightning and fire had fallen from.the sky, […] (XVI, Llibre de Job)

A constructional approach to auxiliary selection

189

Subjects of result constructions like (5a–g), (7a) or (8a), which contain a definite DP subject, select BE. It could not be otherwise since they appear to behave as copular resultative constructions.6 In contrast, existential constructions, which contain an indefinite theme argument, tend to select HAVE (cf. 6a–f, 7b or 8b). There is a plausible explanation for the correlation between definiteness and the type of the construction involved: existential constructions are typically used to introduce a referent, which consequently is indefinite. However, a conservative use of BE in existential constructions is not excluded, as can be shown in (9). (9)

a. és venguda provisió e inhibició del senyor rey que is come provision and inhibition of.the Mr. king that alcú … someone (1311–1324, Epistolari Ia, c. 41, l. 14) b. fo gran mal vengut e gran destrucció was great wrong come and great destruction (1346, Gestes, p. 111, l. 8)

The following data in (10) also exemplify the above-mentioned contrast. The more frequent pattern of existential constructions is the one in (10a): HAVE is typically selected when an indefinite argument (i.e., coreu ‘mail’) appears. In contrast, the use of BE in (10b) can be regarded as conservative. (10) a. De l’enperador no à vengut coreu from the emperor not has come mail (1525–1542, Liori i Requesens, Epistolaris, c. 109, l. 56) b. no era vengut coreu not was come mail (1525–1542, Liori i Requesens, Epistolaris, c. 190, l. 10) Similarly, consider the following examples with the verb arribar ‘arrive’, drawn from Batlle (2002):

6 For the systematicity of BE-selection in subject of result constructions in Old English, see also McFadden and Alexiadou (2010).

190

Jaume Mateu and Mar Massanell i Messalles

(11) a. Vui ha arribat correu de Sa Majestat, que’l deixà today has arrived mail from Her Majesty, which.himac left molt bo. very well (Epistolari d’Estefania de Requesens, p. 161; XVI c.) b. Vui és arribat correu de Barcelona. today is arrived mail from Barcelona (Epistolari d’Estefania de Requesens, p. 134; XVI c.) An important remark is in order here, which can be shown to militate against pure lexical-semantic approaches to auxiliary selection that do not take morphosyntactic factors into account. Indeed, from a lexical-semantic perspective, the verb venir ‘come’ in (5e) (cf. La malaltia ere venguda ‘The illness came’) or in (7a) (cf. Les noves heren vingudes ‘The news came’) can be lexically classified as “verb of appearance” (see Levin and Rappaport Hovav 1995: 282). However, from our present constructional perspective, these examples do not involve an existential construction:7 as pointed out above, we want to distinguish the subject of result constructions in (5c,d,e) and (7a), where BE is systematically selected, from those existential constructions where HAVE is typically selected (cf. 6a–d and 7b). In our constructional approach, BE-HAVE alternations of OC verbs like venir ‘come’ (e.g., cf. 5–6)8 can be accounted for by allowing them to be compat-

7 The same holds for the following examples, which are subject of result constructions, in spite of involving “appearance” of lo mal ‘the harm’ in (iiia), lo dampnatge ‘the damage’ in (iiib), and la compunctió ‘the compunction’ in (iiic). Cf. also the examples in (5c,d,e). (iii) a.

(…) lo mal que li fós vengut per les dites metzines the harm that himdat was come for the mentioned poisons (1374–1377, Procés criminal, f. 31v, l. 16)

b.

(…) per assò los és vengut lo dampnatge for this themdat is come the damage (1275–1299, Vides, p. 169, l. 11)

c.

(…) e quan és venguda la compunctió and when is come the compunction (1445–1449 [1410–1415], Sant Vicent, Sermons IV, p. 75, l. 12)

8 These BE-HAVE alternations are not, of course, only typical of venir ‘come’. For example, the verb romandre (later, restar) ‘remain, rest’ can also be found in both constructions: e.g., cf. the data in (iv) and (v). Cf. the similar contrast in (vi), taken from Batlle (2002). If it is true that the change from BE to HAVE began with existential constructions, stative verbs like ro-

191

A constructional approach to auxiliary selection

ible with or integrated into different argument structures. In particular, the construction BE is systematically selected as a copular structure that contains a participle predicated of a subject of result (as noted, it is then not surprising that a definite or specific subject is found; cf. the data in 12a from Contemporary Catalan). In contrast, the construction where HAVE is typically selected is an existential construction whose theme argument is indefinite. (12) a. *(Els) nois estan preocupats./ the boys are.perf.3pl worried.pp.masc.sg Estan preocupats *(els) nois.

(Catalan)

b. *(Els) nois són feliços. / Són feliços *(els) nois. the boys are happy mandre or restar should be prone to appearing in existential constructions like the ones in (v), according to Aranovich’s (2003) findings. (iv) a.

a Jesuchrist són romases les plagues que rebé en la sua Passió to Jesus-Christ are remained the wounds that received in the his Passion (1445–1449, Sant Vicent, Sermons vi, p. 27, l. 7)

b.

ab aquella pocha gent que restada li era isqué de la ciutat with that few people that remained dat.cl was exited from the city (1490 [1460], Martorell, Tirant, p. 104, l. 20)

c.

lo regne és restat a la filla del dit rey the realm is rested to the daughter of.the mentioned king (1475–1499, Sumari i, p. 103, l. 37)

(v) a.

car gran riquesa li havia romasa de son pare since great richness dat.cl. had remained from his father (1374–1399 [1290], Llull, Santa Maria, p. 188, l. 31)

b.

no li ha romàs un pa not dat.cl. has remained some bread (1400, Epistolari ie, c. 52, l. 36)

c.

No penses tu que m’ aja restada gens de sanch en la mia persona not think you that medat has remained nothing of blood in the my person (1490 [1460], Martorell, Tirant, p. 663, l. 34)

(vi) a.

b.

(…) e ls rayms qui eren romases fossen maduratz the grapes that were remained were ripened (…) no·m ha res romàs de dubte not.medat has anything remained of doubt (XIV–XV) [Diàlegs]

192

Jaume Mateu and Mar Massanell i Messalles

As expected, the replacement of BE by HAVE did not start in a copular construction where the resulting state of a definite subject is asserted (e.g., cf. Engl. The knights are gone). It is clear that alternations like the one exemplified in (13a–b) are predicted to be found in the latest stages of the replacement of BE by HAVE, i.e., examples like (13a) are the most reluctant ones to accept HAVE. (13) a. Els cavalers són venguts. the knights are come.pp.masc.pl b. Els cavalers han vengut. the knights have come The relevant change started where a syntacto-semantic meaning of existence was involved (cf. also Batlle 2002 and Aranovich 2003, i.a.). As noted, an intermediate or “peripheral” (i.e., non-core) status is assigned to the appearance/ existence meaning in Sorace’s (2000, 2004) ASH (see 2). In her framework the diachronic changes are expected to take place from the periphery of ASH to its core and not the other way around (see also Cennamo 2008; Massanell and Mateu, in prep.; and Mateu 2005, 2009 for relevant discussion). Consider the existential constructions in (14) (cf. Engl. There came {two/*the} knights from Spain), which are quite frequent in OC and can be relevant to test our constructional proposal: (14) a. A la emperadriu ha venguts ·II· cavalers d’Espaya to the emperess has come.pp.masc.pl two knights from Spain (1275–1299, Desclot, Crònica II, p. 53, l. 24) b. nos fes saber que avia venguts ·V· usdat made.3sg know that had.3sg come.pp.masc.pl five cavallers genets en Oriola knights riders in Oriola (1327–1335, Alfons III, d. 53, l. 11) c. que galees hi havia vengudes moltes that galleys there had.3sg come.pp.fem.pl many (1343, Jaume I, Fets, f. 114v, l. 7) d. e d’altres hòmens que hi havia venguts and of.others men who there had.3sg come.pp.masc.pl (1343, Jaume I, Fets, f. 115r, l. 4)

A constructional approach to auxiliary selection

193

e. hi ac venguts més de ·C· míllia hòmens there has come.pp.masc.pl more than hundred thousand men a cavall on horseback (1352, Muntaner, Crònica, f. 32rb, l. 26) f. et aquí atrobam que havia vengudes ·X· galeas […] and there find.1pl that had.3sg come.pp.fem.pl ten galleys (1352, Muntaner, Crònica, f. 128ra, l. 5) g. e n’ agra molt més venguts si no and part.cl. would.have.3sg much more come.pp.masc.pl if no fos la guerra was the war (1352, Muntaner, Crònica, f. 161vb, l. 4) The existential constructions in (14) have syntactic and semantic properties that make them similar to the ones in (15), which are found in Contemporary Catalan (cf. Fr. Il y a and Sp. Hay; for different analyses of this existential construction, see Hoekstra and Mulder 1990; Moro 1997; Rigau 1997, among others; for a diachronic account of this construction in Old Catalan, see Ramos 2000). As can be seen in both (14) and (15), there is no agreement between the verb haver ‘have’ and the indefinite theme argument, which only agrees with the past participle.9

9 Compare the following data in (vii), (viii), and (ix), which are all taken from the same text. The existential construction in (vii) is similar to the ones exemplified in (14), the constructions in (viii) contain the existential verb haver ‘have’ (cf. Contemporary Catalan haver-hi; see Rigau (1997), i.a.) and, finally, the subject of result constructions in (ix) is similar to the ones exemplified in (5). (vii)

tantes de géns avia vengudes a Murcia many of people had.3sg come.pp.fem.pl to Murcia (1275–1299, Desclot, Crònica III, p. 7, l. 25)

(viii) a.

b.

so és la vila de Fraga, hon à grans géns e bons garrers that is the village of Fraga, where has great people and good warriors ‘This is the village of Fraga, where there are great people and good warriors.’ (1275–1299, Desclot, Crònica II, p. 12, l. 2) en aquesta vila ha moltes géns d’armes a cavayl e a peu in this village has many people of army on horseback and on foot (1275–1299, Desclot, Crònica III, p. 39, l. 13)

194

Jaume Mateu and Mar Massanell i Messalles

(15) a. Hi ha cavallers vinguts de tot arreu. loc.cl has knights come.pp.masc.pl from all places

(Contemp. Catalan)

b. Hi ha cavallers arribats de tot arreu. loc.cl has knights arrive.pp.masc.pl from all places As pointed out above, it is clear that the relevant replacement of BE by HAVE did not start with the subject of result constructions like (13a), where BE selection is quite systematic until the latest stages. Rather, the change started within the domain of existential constructions. It seems plausible, then, that existential constructions like the ones in (14a–b) could compete with the ones in (16a– b), where the auxiliary BE would be conservative (recall that the preferred auxiliary in those existential constructions that have an indefinite argument turns out to be HAVE). After an intermediate stage like the one in (14), the last stage turns out to be the one in (17), where it is the auxiliary and not the participle that agrees with the subject. The pattern in (17) is in fact the one that is found in Contemporary Catalan, which lacks auxiliary selection: see (18a). Interestingly, in the northwestern variety of Catalan the theme argument does not agree with the verb whenever it is indefinite (see Rigau 1997, i.a.). Only definite subjects do: cf. (19b–c). (16) a. A la emperadriu són venguts II· cavalers d’Espaya to the emperess are come.pp.masc.pl two knights from Spain b. eren venguts ·V· cavallers genets en in were come.pp.masc.pl five knights riders Oriola Oriola

(ix)

c.

e ha-hi molt valens géns d’armes and has-there many valiant people of army (1275–1299, Desclot, Crònica III, p. 93, l. 6)

a.

que cavalers crestians e géns del rey d’Aragó s’eren that knights Christian and people of.the king of Aragon refl.were venguts atendar en aquel […] come.pp.masc.pl attend in this (1275–1299, Desclot, Crònica II, p. 140, l. 9)

b.

totes les géns de València foren vengudes a Algesira all the people of Valentia were come.pp.fem.pl to Algeciras (1275–1999, Desclot, Crònica III, p. 156, l. 4)

c.

aqueles géns eren aquí vengudes per lo príncep those people were here come.pp.fem.pl for the prince (1275–1299, Desclot, Crònica IV, p. 40, l. 20)

A constructional approach to auxiliary selection

195

(17) a. A la emperadriu han vengut ·II· cavalers d’Espaya to the emperess have come two knights from Spain b. havien vengut ·V· cavallers genets en Oriola had.3pl come five knights riders in Oriola (18) a. Han vingut / arribat dos cavallers. have.3pl come / arrived two knights b. Ha vingut / arribat dos cavallers. has come / arrived two knights (19) a. Han vingut {dos/els/Ø} cavallers. have.3pl come.pp {two/the/Ø} knights

(Central Catalan)

(Northwestern Catalan)

(Central Catalan)

b. Ha vingut {dos/*els/Ø} cavallers. has come.pp {two/the/Ø} knights

(Northwestern Catalan)

c. Han vingut els cavallers. have.3pl come.pp the knights

(Northwestern Catalan)

Apparent exceptions to the above-mentioned descriptive generalisation with respect to OC venir ‘come’ (i.e., HAVE is not found when the subject is definite) can involve factors like restructuring (cf. 20) and irrealis modality (cf. 21). It seems plausible, then, to argue that the auxiliary HAVE in (20a–b) is motivated by the transitive nature of the verb in infinitival form (cf. aucir ‘kill’ or veer ‘see’). Notice, for example, that the clitic climbing shown in (20b) can be taken as clear evidence for restructuring.10

10 Restructuring was not an obligatory phenomenon in OC, as shown by the examples in (x), where BE is selected: (x) a.

b.

On con totz aquels qui eren vengutz per él a vezer lo where with all those who were.3pl come.pp.pl to him to see himac plorassen cry.past.subj.3pl (1275–1299, Vides, p. 159, l. 4) él li dix que demoni era, e que era vengut per ela a decebra he dat.cl says that evil was, and that was come to her to deceive (1275–1299, Vides, p. 272, l. 29)

196

Jaume Mateu and Mar Massanell i Messalles

(20) a. Nós èm vengutz per tu a aucir we have.1pl come.pp.pl to you to kill ‘We have come to kill you.’ (1275–1299, Vides, p. 103, l. 24) b. Tots los barons d’Alamanya l’ agren vengut veer all the knights of Germany himac have.3pl come see ‘All the knights of Germany have come to see him.’ (1275–1299, Desclot, Crònica II, p. 61, l. 18) Other apparent counterexamples like the ones in (21) can be motivated by factors related to modality. In particular, HAVE has been shown to be heavily favoured over BE in irrealis clauses in Old Neapolitan (Ledgeway 2003; see also Cennamo 2002) and several older Germanic languages (cf. Shannon 1995; McFadden 2007; and McFadden and Alexiadou 2010, among others). The counterfactual clauses in (21) convey that the considered proposition was contrary to fact at a particular time in the past. The auxiliary HAVE is selected here, since no real resulting state is actually involved in (21). These examples do not involve subject of result constructions. (21) a. si d’ aquí a aquel dia no havíets vengut 11 if from here to that day no had.2sg come (1268, Illes XIII-27, p. 126, l. 3) b. si vós, sényer, vos agésets vengut if you, lord, you had.subj.2sg come (1375–1376, Sereneta, Cartes II, c. 15, l. 24)

c.

que tots los sarraÿns de les montanyes de Prades that all the Saracens from the mountains of Prades tota la tera eren venguts assetgar la ciutat all the land were.3pl come.pp.pl besiege the city (1275–1299, Desclot, Crònica II, p. 82, l. 22)

e de Ciurana e de and from Ciurana and from de Barcelona of Barcelona

11 Here is the full context where the auxiliary HAVE in the counterfactual construction can be compared with the auxiliary BE in the previous clause: (xi) vos deïm e us manam […] que vós d’assí you.dat say.1pl and you.dat order.1pl that you from.here vinent […] siats vengut danant nós […] sabent coming be.subj.2sg come.pp.sg before us knowing aquel dia no havíets vengut nós enantarem contra vós. that day not had.2sg come we would.go.1pl against you

a Sent Andreu to Saint Andrew que si d’aquí that if from.here

primer next a to

A constructional approach to auxiliary selection

197

All in all, it seems that the following descriptive generalisation holds for Old Catalan: subject of result constructions involve BE-selection, whereas the rest (existential constructions included) involve HAVE-selection. In particular, we have shown that in Old Catalan alleged core unaccusative verbs like venir ‘come’ can enter into an existential construction where HAVE is typically selected.12 We have claimed that this fact can be accounted for in constructional approaches in a more appropriate way than in non-constructional ones. According to Sorace (2000, 2004), core unaccusative verbs like It. venire ‘come’ or arrivare ‘arrive’ are predicted to select only BE in languages in which auxiliary selection is an active phenomenon (e.g., cf. It. *Ha arrivato posta vs. È arrivata posta ‘There arrived mail’). Nevertheless, by looking at some relevant data from Old Catalan we have concluded that, as far the auxiliary selection is concerned, there are not core verbs but rather core constructions. If we consider core constructions as the last to be affected by the relevant grammatical change, it seems plausible to take subject of result constructions (e.g., cf. OC.

12 The case of the verb venir ‘come’ is not unique. Core unaccusative verbs like entrar ‘enter’, anar ‘go’, arribar ‘arrive’, eixir ‘exit’, and passar ‘pass’ can also be found in existential constructions like the ones exemplified in (xii): (xii)

a.

sarraÿns a cavayl e a peu avia entrats en la vayl Saracens on horseback and on foot had.3sg enter.pp.masc.pl in the valley d’Albayda of Albaida (1275–1299, Desclot, Crònica III, p. 16, l. 9)

b.

car massa n’ à anats en pena perdurable since many part.cl has go.pp.masc.pl in pity lasting (1374–1399 [1290], Llull, Santa Maria, p. 252, l. 17

c.

en lo port de Contestina havia arribades sis naus de genovesos in the port of Contestian had.2sg arrive.pp.fem.pl six ships of Genoesian (1490 [1460], Martorell, Tirant, p. 1337, l. 18)

d.

ha exides de Mònech IIII galeas de Mònech per dampnificar les has exit.pp.fem.pl from Monaco three galleys of Monaco to damage the vostres géns your people (1343–1349, Reintegració ii, c. 387, l. 20)

e.

E dixeren-nos que hòmens havia passats tres tro a la and told.3pl-us that men had.3sg pass.pp.masc.pl three up to the barbacana barbican (1343, Jaume I, Fets, f. 114r, l. 21)

198

Jaume Mateu and Mar Massanell i Messalles

El cavaller és/*ha arribat ‘lit. The knight is/*has arrived’) as the “core” ones, whereas existential constructions (e.g., cf. OC Ha/és arribat coreu, lit. ‘Has/is arrived mail’, i.e., ‘There arrived mail’; cf. 11) can be regarded as “non-core” and thus predicted to be the first affected by the relevant grammatical change. Indeed, it is important to realise that this fact is coherent with Keller and Sorace’s (2003: 60) conception of “intermediacy” in (3) iff we replace “verbs” by “constructions”: i.e., cf. “intermediate [constructions] between the two extremes incorporate telicity and agentivity to lesser degrees, and tend to have a less specified (basically stative) event structure”. As predicted, existential constructions in Old Catalan, like the ones in which HAVE is typically selected, are atelic, less agentive and more stative than subject of result constructions.

3 Some synchronic evidence for a constructional approach to argument structure In this section we put forward synchronic evidence that shows that in Romance languages such as Contemporary Catalan and Italian, core unergative verbs like sing or work can enter into an unaccusative existential construction like the one in (22c–d), a fact that appears to be unexpected from a non-constructional perspective.13 After reviewing some interesting data from Catalan that favour a constructional approach to argument structure, we will discuss their Italian counterparts by focussing on the puzzle they pose to theories of auxiliary selection. (22) a. Els joves canten. the boys sing.3pl

(Catalan)

b. *Canten joves. sing.3pl boys

13 Recall that from the non-constructional perspective core unergative verbs like sing or work are not expected to be “variable behaviour verbs” (vs. cf. manner-of-motion verbs like dance or verbs of sound like rumble, which are not analysed as core unergatives by Sorace (2000, 2004): see her distinction in (2) between “controlled processes-motional” (e.g., dance) and “controlled processes-nonmotional” (e.g., sing or work). Verbs of sound (e.g., rumble) would fall under the intermediate class of “uncontrolled processes” in 2). See also Levin and Rappaport Hovav (1995, 2005) and Bentley and Thórhallur (2003), for relevant discussion.

A constructional approach to auxiliary selection

199

c. (En aquesta coral) hi canten joves. in this choir loc.cl sing.3pl boys ‘There are boys singing (in this choir).’ d. (En aquesta coral), (de joves) n’hi canten molts. in this choir of boys part.cl.-loc.cl sing.3pl many ‘There are many boys singing (in this choir).’ As is well-known (e.g., see Burzio 1986), bare plurals like joves ‘boys’ cannot act as postverbal subjects of unergatives (see 22b). In contrast, these NPs can function as postverbal subjects of unaccusatives (e.g., Cat. Vénen joves lit. ‘Come boys’, i.e., there come boys) or as direct objects of transitive verbs (e.g., Cat. Les drogues maten joves ‘Drugs kill boys’). Following Torrego (1989), Rigau (1997) and Mateu and Rigau (2002) argue that the unaccusativity of the existential construction in (22c) or (22d) is possible due to the presence of the obligatory locative marker hi ‘there’.14 Such syntactic flexibility or “elasticity” is not expected according to a lexicalist account à la Levin and Rappaport Hovav (1995), who, following Lonzi (1985), argue that some Italian examples similar to (22c) and (22d) are not unaccusative. Before showing that their conclusion for Italian is not correct, let us deal with the analysis of the Catalan data, which will be useful to show that these existential constructions are unaccusative. To start with, notice that our assumption that the existential construction in (22c) is unaccusative can directly account for the contrast in (23a) and (23b): there are two direct internal arguments that appear to compete for the same syntactic position in the unaccusative construction in (23a): cançons ‘songs’ and joves ‘boys’. In contrast, this problem does not arise in (23b) since this construction

14 Similarly, as is well-known (see Levin and Rappaport Hovav 1995; Hoekstra 1998; or Sorace 2000, inter alia), the presence of a directional PP like into the hall is crucial in the explanation of why the constructions in (xiiib) and (xivb) are unaccusative (e.g., cf. BE-selection in xivb). The German data in (xiv) are adapted from Sorace (2000: 876; ex. 40). (xiii) a. b.

They danced (in the hall). They danced into the hall.

(xiv) a.

Hans und Rita haben (im Saal) getanzt. (German) Hand and Rita have in the hall danced ‘Hans and Rita danced (in the hall).’

b.

Hans und Rita sind *(in den Saal) getanzt. Hans and Rita are into the hall danced ‘Hans and Rita danced into the hall.’

200

Jaume Mateu and Mar Massanell i Messalles

is unergative (i.e., transitive with a cognate object; see Hale and Keyser [2002]): the partitive NP joves ‘boys’ is an internal argument in (23a), whereas the definite DP els joves ‘the boys’ is a postverbal external argument in (23b). (23) a. *En aquesta coral hi canten cançons joves in this choir loc.cl sing.3pl songs boys ‘There are boys singing songs in this choir.’

(Central Catalan)

b. En aquesta coral hi canten cançons els joves. in this choir loc.cl. sing.3pl song the boys ‘The boys sing songs in this choir.’ As pointed out by Rigau (1997) and Mateu and Rigau (2002), an even clearer piece of evidence for the unaccusativity of the existential constructions in (22c) and (22d) can be found in the northwestern variety of Catalan, where there is no agreement between the indefinite argument joves ‘boys’ and the verb: see (24b). Indeed, the lack of agreement in (24b) would be unexpected if the bare plural NP were the subject/external argument of an unergative verb/construction (cf. the unaccusativity of the example in 25b). (24) a. Els joves canten. the boys sing.3pl

(NW Catalan)

b. (En aquesta coral) hi canta joves. in this choir loc.cl sings boys c. (En aquesta coral), (de joves) n’hi canta molts. in this choir of boys part.cl.-loc.cl sings many.pl (25) a. Vénen (els) joves. come.3pl (the) boys.

(Central Catalan)

b. Ve joves. comes boys

(NW Catalan)

c. *Ve els joves. comes the boys

(NW Catalan)

d. Vénen els joves. come.3pl the boys

(NW Catalan)

As noted, the unaccusative construction is possible if it is licensed by a locative element like hi ‘there’ (cf. [26a] and [24b]). Moreover, the same contrast shown

A constructional approach to auxiliary selection

201

in the Central Catalan data in (23) also applies to its northwestern variety: cf. (26b) and (26c). (26) a. *Canta joves. sing.3sg boys

(NW Catalan)

b. *En aquesta coral hi canta cançons joves in this choir loc.cl sings songs boys c. En aquesta coral hi canten cançons els joves. in this choir loc.cl sing.3pl songs the boys Drawing on Hale and Keyser’s (1993) configurational theory of argument structure, Mateu and Rigau (2002) claim that the syntactic analysis of the agentive unergative structure is the one in (27a), whereas that of the existential unaccusative structure is the one in (27b). (27) a. [v Els joves [v do √cant-]] (en aquesta coral) b. v [v √cant- have (=be+ Pccr)] [PP hi [Pccr joves]]] Following Hale and Keyser (1993, 2002), the formation of unergatives can be argued to involve conflation of a nominal (or a simple root: e.g., √cant‘s[o]ng’)15, which occupies the complement position in (27a), with a null light verb (e.g. an agentive do): cf. sing – do song. The locative PP en aquesta coral ‘in this choir’, if present, would be an adjunct. The formation of the unaccusative argument structure in (27b) is quite different: a null possessive light verb have, which is conceived of as the result of conflating a null P with the more basic light verb be (see Kayne 1993; Rigau 1997, among others, for the proposal that have = be + P), subcategorises for a complement PP: the central coincidence relation P in (27b), which is crucial when dealing with possessive relations (see Hale 1986; Hale and Keyser 2002: chap. 7, among others), is conceived of as a birelational element that relates a possessor (hi ‘there’)16 with a possessee (joves ‘boys’). In (28) are some exam-

15 Conflation is understood as a local operation whereby the phonological matrix of a head is transmitted to its phonologically-defective sister head (cf. Hale and Keyser 2002 and AcedoMatellán 2010). Thus, for instance, in (27a) the phonological matrix of the root √cant conflates into its phonologically-defective (null) sister node v, giving rise to the verb cantar. 16 See Mateu and Rigau (2002) for the claim that the locative clitic hi ‘there’ acts as an impersonaliser in (27b). As predicted, Nominative case is then impossible in unaccusative existential constructions: e.g., *Hi ha(n) ells ‘there are theynom’ (see Rigau 1997, for more relevant discussion).

202

Jaume Mateu and Mar Massanell i Messalles

ples of the haver-hi construction (cf. Fr. Il y a; Sp. Hay; Cat. Hi ha; for some prominent syntactic analyses of this existential construction, see Hoekstra and Mulder 1990; Moro 1997; and Rigau 1997, among others). The main difference between the examples in (28) and the ones in (22c) and (24b), repeated below in (29), is that the latter involve an additional conflation of the root √cantwith the null light verb have (cf. 27b). (28) a. (En aquesta coral) hi han joves {que in this choir loc.cl have.3pl boys who canten/cantant} sing.3pl/singing

(Central Catalan)

b. (En aquesta coral) hi ha joves {que in this choir loc.cl has boys who canten/cantant} sing.3pl/singing

(NW Catalan)

(29) a. (En aquesta coral) hi canten joves. in this choir loc.cl sing.3pl boys

(Central Catalan)

b. (En aquesta coral) hi canta joves. in this choir loc.cl sings boys ‘There are boys singing in this choir.’ (28) a. (En aquesta coral) hi han joves {que in this choir loc.cl have.3pl boys who canten/cantant} sing.3pl/singing

(NW Catalan)

(Central Catalan)

b. (En aquesta coral) hi ha joves {que in this choir loc.cl has boys who canten/cantant} sing.3pl/singing

(NW Catalan)

(29) a. (En aquesta coral) hi canten joves. in this choir loc.cl sing.3pl boys

(Central Catalan)

b. (En aquesta coral) hi canta joves. in this choir loc.cl sings boys ‘There are boys singing in this choir.’

(NW Catalan)

A constructional approach to auxiliary selection

203

As shown by Mateu and Rigau (2002), the conflation of √cant- with have depicted in (27b) is similar to that of √cant- with go involved in the unaccusative structure of The boys danced into the kitchen, analysed in (30). Both (27b) and (30) are unaccusative argument structures: there is no argument occupying the specifier position of v, which corresponds to the external argument of unergatives or transitives (cf. 27a). (30) [v [v √dance go] [sc/pp the boys into the kitchen]]] As is well-known, the flexibility or “elasticity” of manner-of-motion verbs in Germanic languages exemplified in (30) has been argued to favour a constructional approach to argument structure (e.g., see Borer 2005 and Acedo-Matellán 2010, among others).17 Two other relevant examples (e.g., the unergative

17 It is often said that a Romance language like Italian also shows a similar polysemy (e.g., cf. xva–xvb) but at the same time it is clear that it lacks the regular Germanic polysemy, as exemplified in (ic): see Mateu (2002); Folli and Ramchand (2005); and Mateu and Rigau (2010), for different analyses and further discussion of why Romance lacks Germanic constructions like those ones in (xvic) through (xvin). (xv)

a.

Gianni ha {corso / ballato} per molte ore. Gianni has ran danced for many hours

b.

Gianni è corso via / Gianni è corso alla cucina. Gianni is run away / Gianni is run to.the kitchen ‘Gianni ran away. / Gianni ran to the kitchen.’

c.

*Gianni è ballato via / *Gianni è ballato alla cucina. Gianni is danced away / Gianni is danced to.the kitchen. ‘Gianni danced away. / Gianni danced to the kitchen’.

(xvi) a. b. c. d. e. f. g. h. i. j. k. l. m. n.

The The The The The The The The The The The The The The

boy boy boy boy boy boy boy boy boy boy boy boy boy boy

danced. danced a polka. danced into the kitchen. danced away. danced the puppet across the stage. danced the night away. danced his debts off. danced his feet sore. danced his way into the hearts of America. danced himself tired. danced his butt off. danced the hell out of that mambo. danced up a storm. outdanced her.

(Italian)

204

Jaume Mateu and Mar Massanell i Messalles

and causative/transitive structures) are analysed in (31). From a constructional perspective, there seems to be no obvious sense in which a verb (e.g. dance) is either unergative, unaccusative or causative. Rather, the only verbs are the abstract event-structural operators, which can be phonologically realised as full light verbs if no conflation operation applies (e.g., cf. do, make, cause, be, go, become, etc.) or as “lexical” verbs if they become conflated with a root, which can occupy a complement position, as in the unergative structure in (31a), or an adjunct one, as in the unaccusative structure in (30) or in the causative one in (31b). (31) a. [v The boy [v do √dance]]

unergative structure

b. [v The boy [v √dance cause] [sc/PartP his debts off]]] causative structure Concerning the structural semantics of light verbs, in (31a) v is read as do since it takes a DP as external argument and a root as complement. In both (30) and (31b) the light verb takes a Small Clause Result such as a telic PP (see Hoekstra 1988, 1998) and is read as go/become if there is no external argument or as cause if there is one (for further discussion on the structural meaning of syntactic argument structures, see Mateu 2002; Acedo-Matellán 2010; and Harley 2011, among others). With this background on a constructional approach to argument structure, let us now analyse the Italian counterparts of the Catalan existential constructions analysed above, which will also be argued to be unaccusative. The data in (32b,d), taken from Lonzi (1985) and also revisited by Levin and Rappaport Hovav (1995: 276–277; ex. 106–107), have been said to be counterexamples to Burzio’s (1986) claim that ergative (i.e., unaccusative) verbs are the only monadic verbs that admit ne-cliticisation of their argument. Following Lonzi (1985), Levin and Rappaport Hovav (1995: 275) point out that “a variety of verbs that take the auxiliary avere ‘have’ do permit ne-cliticization, but only when they are found in a simple tense; ne-cliticization is not possible when these verbs are found in a complex tense in which the auxiliary is expressed”. (32) a. *Di ragazze, of girls, fabbriche di factories of

ne hanno lavorato molte nelle part.cl have.3pl worked many in.the Shanghai. Shanghai.

(Italian)

b. Di ragazze, ne lavorano molte nelle fabbriche di Shanghai. of girls, part.cl work.3pl many in.the factories of Shanghai. ‘There are many girls working in the factories of Shanghai.’

A constructional approach to auxiliary selection

205

c. *Di ragazzi, ne hanno russato molti nel corridoio of boys, part.cl have.3pl snored many in the corridor del treno. of.the train d. Di ragazzi, ne russavano molti nel corridoio del treno. of boys, part.cl snored many in.the corridor of.the train ‘There were many boys snoring in the corridor of the train.’ Lonzi (1985), apud Levin and Rappaport Hovav (1995: 276–277; ex. 106–107) Levin and Rappaport Hovav (1995: 277) conclude that “phenomena said to involve ‘surface unaccusativity’ (…) are not unaccusative diagnostics strictly speaking, but rather to a large extent receive their explanation from discourse considerations” (see also Lonzi 1985 and Maling, Calabrese, and Sprouse 1994, for similar remarks). In particular, they point out that “unergative verbs are found in this construction under circumstances similar to those that sanction the appearance of English unergative verbs in locative inversion – that is, in contexts where the verb describes a characteristic activity or process of the entity it is predicated of” (276).18 This said, it is important to realise that Levin and Rappaport Hovav’s (1995) discourse-based observation does not account for the fact that avere-selection is not allowed in the examples in (32a,c). Such a restriction is not found in languages that lack auxiliary selection: for example, HAVE is possible in their Catalan counterparts: see (33), for example.19 (33) a. De dones, a les fàbriques, n’hi of women, in the factories, part.loc

(Central Catalan)

18 But see Culicover and Levine (2001) for a critical review of Levin and Rappaport Hovav’s (1995) discourse-based analysis of locative inversion. According to the former, the traditional unaccusative diagnostic provided by locative inversion must be regained once this unaccusative construction is separated from heavy NP inversion constructions with unergative verbs. 19 As pointed out above, the locative marker hi ‘there’ is obligatory in this existential construction: (xvii) a.

b.

*En {treballen / han treballat} moltes. part. work.3pl / have worked many *En {treballa / ha treballat} moltes. part. work.3sg / has worked many

(Central Catalan)

(NorthW. Catalan)

206

Jaume Mateu and Mar Massanell i Messalles

{treballen/han treballat} moltes. work.3pl/have.3pl worked many b. De dones, a les fàbriques, n’hi (NorthW. Catalan) of women, in the factories, part.loc {treballa / ha treballat} moltes. works / has worked many ‘There are many women working/who have worked in the factories.’ c. N’hi {treballen / han treballat} moltes. part.loc work.3pl / have worked many d. N’hi {treballa / ha treballat} moltes. part.loc work.3sg / has worked many

(Central Catalan) (NorthW. Catalan)

Perhaps one could claim that Levin and Rappaport Hovav’s (1995) remark on Italian only holds for imperfective tenses, since the latter can be regarded as idoneous for expressing habitual activities. However, the following triplet from Centineo (1996: 230–231, fn. 6) shows that this is not the case, since in the socalled passato remoto (lit. ‘remote past’) these alleged unergative verbs are also compatible with ne-cliticisation: see (34c). (34) a. Ce ne nuota tanta di gente in quella piscina. there part.cl swims much of people in that pool

(Italian)

b. ??Ce ne ha nuotato molta di gente in quella piscina. there part.cl has swum much of people in that pool c. Ce ne nuotò molta di gente in quella piscina. there part.cl swam much of people in that pool Centineo (1996: 230–231, fn. 6) In contrast, the ungrammaticality of (32a,c) and (34b) is actually predicted by Burzio’s (1986) correlation between ne-cliticisation in monadic verbs and unaccusativity. In this sense, notice Centineo’s (1996: 231, fn. 6) observation in (35): (35) “(…) it must also be added that some of the native speakers consulted about these data attempted to use essere as the auxiliary for (iv) ”. Indeed, avere is not be the expected auxiliary in (34b) if the Italian construction in (34) is unaccusative.20 Following our analysis of similar data from Cata20 See Torrego (1989), Rigau (1997), and Mateu and Rigau (2002), for different implementations of such a proposal.

A constructional approach to auxiliary selection

207

lan (see above), we assume that the constructions (rather than the verbs) in (32b,d), repeated below in (36), are unaccusative, whereby it is not surprising that they involve an obligatory SC-like PP: see the syntactic analysis in (37a), which involves a conflation process similar to the one analysed in (30), repeated in (38). (36) a. Di ragazze, ne lavorano molte nelle fabbriche di Shanghai. of girls, part.cl work.3pl many in.the factories of Shanghai b. Di ragazzi, ne russavano molti nel corridoio del treno. of boys, part.cl snored many in.the corridor of.the train (37) a. [v [v √lavor- be] [PP/SC molte ragazze [P’ nelle fabbriche di Shanghai]]] b. There are many girls in the factories of Shanghai who are working. (38) a. [v [v √dance- go] [PP/SC the boys [P’ into the kitchen]]] b. The boys went into the kitchen dancing. In (37a) a root designating an event (i.e., √lavor ‘work’) is conflated with a null light verb be, which subcategorises for a Small Clause whose inner predicate has a locative nature.21 Given the unaccusative structure in (37a), which lacks an external argument occupying the specifier position of the v(erbal) head, avere-selection is expected to be blocked. As noted, this result is compatible with (and actually predicted by) Burzio’s (1986) classical analysis.

21 In fact, things turn out to be more complex. For example, the data in (xviii), taken from Maling, Calabrese, and Sprouse (1994), do not involve any surface locative PP. However, Maling et al. (1994: 5) point out that (xviiib) is possible only on a very specific reading -namely, many people are calling in one specific place relevant to the speaker. A similar comment could be argued to be appropriate for (xviiia). Alternatively, temporal phrases like domani ‘tomorrow’ in (xviiia) or la domenica ‘on Sunday’ in (xviiib) could play an important role as well. The relevant conclusion seems to be that a spatiotemporal element is compulsory in order to license these existential constructions. (xviii) a.

Domani ne parleranno molti. (Italian) Tomorrow ne will.speak.3pl many

b.

Ne telefonano molti, di tifosi, la domenica! ne phone.3pl many of fans on Sunday

208

Jaume Mateu and Mar Massanell i Messalles

4 Concluding remarks Insertion of roots into syntactic argument structures is not as free as some radical constructionalists predict (e.g., OC venir ‘come’ cannot be associated to an unergative argument structure) but is much freer than proponents of projectionist approaches predict (e.g., so-called “core unergative verbs” like Cat. cantar ‘sing’ or work ‘treballar’ can enter into unaccusative argument structures; see Mateu and Rigau 2002). The relevant conclusion seems to be that there are no unaccusative verbs but rather unaccusative structures, which does not exclude that some roots tend to be lexically associated to some specific constructions (cf. Mateu 2002; Borer 2005; Ramchand 2008; and AcedoMatellán 2010, among others). In this paper, some diachronic and syncronic evidence has been claimed to favour the constructional approach to argument structure and, in particular, to auxiliary selection. Constructional approaches to auxiliary selection can also be claimed to be more compatible with the accounts that posit the intervention of non-lexical level factors. For example, they appear to be more adequate to account for the following contrasts from Quebec French (QF), drawn from Manente (2009): cf. (39). (39) a. Jean a arrivé / parti / entré à huit heures. Jean has arrived / left / entered at eight hours b. Jean est arrivé (= là) / parti (= absent) / entré (= dedans). Jean is arrived (there) / left (= absent) / entered (= inside) Interestingly, this use of BE in QF can be related to the stative resultative interpretation that McFadden and Alexiadou (2010) attribute to Earlier English I am come. Similarly, the contrast in (40) given by Manente (2009: 43) can be explained on the basis that BE is related to a perfect-of-result reading where the target state holds of the subject, as in Earlier English. (40) Maintenant qu’il {*a / est} arrivé / entré chez lui, il ne now that.he has / is arrived / entered at.his.house, he not voudra plus jamais ressortir want.fut any longer exit We conclude that (i) unaccusativity is not a property of verbs but rather of constructions (see also Moro 1997 or Rigau 1997, among others, as well as Gillmann, this volume) and (ii) auxiliary selection is not a matter of verbs but

A constructional approach to auxiliary selection

209

rather of constructions (see also McFadden and Alexiadou 2010 and Rosemeyer 2014: chapter 6, among others).

References Acedo-Matellán, Víctor. 2010 Argument structure and the syntax-morphology interface: A case study in Latin and other languages. Ph.D. dissertation, Departament de Lingüística General, Universitat de Barcelona. Downloadable at http://www.tdx.cat/handle/10803/ 21788. Aranovich, Raul. 2003 The semantics of auxiliary selection in Old Spanish. Studies in Language 27(1): 1–37. Batlle, Mar. 2002 L’expressió dels temps compostos en la veu mitjana i la passiva pronominal. El procés de substitució de l’auxiliar ésser per haver [The expression of compound tenses in the middle voice and the pronominal passive. The process of replacement of auxiliary ésser by haver]. Barcelona: Institut d’Estudis Catalans/ Publicacions de l’Abadia de Montserrat. Bentley, Delia and Thórhallur Eythórsson. 2003 Auxiliary selection and the semantics of unaccusativity. Lingua 114: 447–471. Burzio, Luigi. 1986 Italian Syntax. Dordrecht: Foris. Cennamo, Michela. 2002 La selezione degli ausiliari perfettivi in napoletano antico: Fenomeno sintattico o sintattico-semantico? Archivio Glottologico Italiano 87(2): 175– 222. Cennamo, Michela. 2008 The rise and development of analytic perfects in Italo-Romance. In: Thórhallur Eythórsson (ed.), Grammatical Change and Linguistic Theory: the Rosendal Papers, 115–142. Amsterdam, Philadelphia: John Benjamins. Centineo, Giulia. 1996 A lexical theory of auxiliary selection in Italian. Probus 8: 223–271. Culicover, Peter W. and Robert D. Levine. 2001 Stylistic inversion in English: a reconsideration. Natural Language and Linguistic Theory 19: 283–310. Folli, Raffaella and Gillian Ramchand. 2005 Prepositions and results in Italian and English: an analysis from event decomposition. In: Henk Verkuyl, Henriette de Swart and Angeliek van Hout (eds.), Perspectives on Aspect, 81–105. Dordrecht: Springer. Gillmann, Melitta. (this volume) Auxiliary selection in closely related languages: the case of German and Dutch. Hale, Kenneth L. 1986 Notes on world view and semantic categories: some Warlpiri examples. In: Peter Muysken and Henk van Riemsdijk (eds.), Features and Projections, 233–254. Dordrecht: Foris. Hale, Ken L. and Samuel J. Keyser. 1993 On argument structure and the lexical expression of syntactic relations. In: Ken L. Hale and Samuel J. Keyser (eds.), A View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 53–109. Cambridge, MA: MIT Press. Hale, Ken L. and Samuel J. Keyser. 2002 Prolegomenon to a Theory of Argument Structure. Cambridge, MA: MIT Press. Harley, Heidi. 2011 A minimalist approach to argument structure. In: Cedric Boeckx (ed.), The Handbook of Linguistic Minimalism, 426–447. Oxford: Oxford University Press. Hoesktra, Teun. 1988 Small clause results. Lingua 74: 101–139.

210

Jaume Mateu and Mar Massanell i Messalles

Hoekstra, Teun. 1999 Auxiliary selection in Dutch. Natural Language and Linguistic Theory 17: 67–84. Hoekstra, Teun and René Mulder. 1990 Unergatives as copular verbs: locational and existential predication. The Linguistic Review 7: 1–79. Kayne, Richard. 1993 Toward a modular theory of auxiliary selection. Studia Linguistica 47: 3–31. Keller, Frank and Antonella Sorace. 2003 Gradient auxiliary selection and impersonal passivization in German: an experimental investigation. Journal of Linguistics 39: 57– 108. Ledgeway, Adam. 2003 The distribution of the perfective auxiliary avere in Early Neapolitan: split intransitives conditioned by modal factors. Archivio Glottologico Italiano 88: 29–71. Levin, Beth and Malka Rappaport Hovav. 1995 Unaccusativity: at the Syntax-Lexical Semantics Interface. Cambridge, MA: MIT Press. Levin, Beth and Malka Rappaport Hovav. 2005 Argument Realization. Cambridge, New York: Cambridge University Press. Lonzi, Lidia. 1985 Pertinenza della struttura tema-rema per l’analisi sintattica. In: Harro Stammerjohann (ed.), Theme-Rheme in Italian, 99–120. Tübingen: Gunter Narr. Maling, Joan, Andrea Calabrese and Richard Sprouse. 1994 Domani ne parleranno molti: the independence of ne-cliticization and essere-selection in Italian. Unpublished manuscript, Brandeis University, Harvard University and Indiana University. Manente, Mara. 2009 Les verbes aller et venir et leur incompatibilité avec l’emploi résultatif. Écho des Études Romanes 5: 27–40. Massanell i Messalles, Mar. 2009 Beneficios de los corpus informatizados para la investigación diacrónica: el caso del CICA para la GCA y los auxiliares de perfecto. In: Laura Romero Aguilera and Carolina Julià Luna (eds.), Tendencias actuales en la investigación diacrónica de la lengua, 147–158. Barcelona: Publicacions i Edicions de la Universitat de Barcelona. Massanell i Messalles, Mar and Jaume Mateu. (in prep.) L’auxiliar dels temps de perfet [Auxiliary verbs in perfect tenses]. In: Josep Martines and Manuel Pérez-Saldanya (eds.), Gramàtica del Català Antic. Mateu, Jaume. 2002 Argument structure: relational construal at the syntax-semantics interface. Ph.D. dissertation, Departament de Filologia Catalana, Universitat Autònoma de Barcelona. Downloadable at http://seneca.uab.cat/clt/publicacions/tesis/index.html. Mateu, Jaume. 2005 La selecció de l’auxiliar en els verbs intransitius del català antic: alguns aspectes descriptius i explicatius [Auxiliary selection in Old Catalan intransitive verbs: some descriptive and explanatory issues]. Caplletra 38: 211–240. Mateu, Jaume. 2009 Gradience and auxiliary selection in Old Catalan and Old Spanish. In: Paola Crisma and Giuseppe Longobardi (eds.), Historical Syntax and Linguistic Theory, 176–193. Oxford: Oxford University Press. Mateu, Jaume and Gemma Rigau. 2002 A minimalist account of conflation processes: parametric variation at the lexicon-syntax interface. In: Artemis Alexiadou (ed.), Theoretical Approaches to Universals, 211–236. Amsterdam, Philadelphia: John Benjamins. Mateu, Jaume and Gemma Rigau. 2010 Verb-particle constructions in Romance: a lexicalsyntactic account. Probus 22: 241–269. Moro, Andrea. 1997 The Raising of Predicates. Cambridge: Cambridge University Press.

A constructional approach to auxiliary selection

211

McFadden, Thomas. 2007 Auxiliary selection. Language and Linguistics Compass 1(6): 674– 708. McFadden, Thomas and Artemis Alexiadou. 2010 Perfects, resultatives, and auxiliaries in Earlier English. Linguistic Inquiry 41(3): 389–425. Ramos, Joan Rafael. 2000 Ésser, estar i haver-hi en català antic. Estudi sintàctic i contrastiu [Ésser, estar, and haver-hi in Old Catalan: a contrastive and syntactic study]. Montcada i Reixac: Institut Interuniversitari de Filologia Valenciana/Publicacions de l’Abadia de Montserrat. Rigau, Gemma. 1997 Locative sentences and related constructions in Catalan: ésser / haver alternation. In: Amaya Mendikoetxea and Myriam Uribe-Etxebarria (eds.), Theoretical Issues at the Morphology-Syntax Interface, 395–421. Bilbao, Donostia, San Sebastián: Euskal Herriko Unibertsitatea/Gipuzkoa Foru Aldundia. Rosemeyer, Malte. 2014 Auxiliary Selection in Spanish. Gradience, Gradualness, and Conservation. Amsterdam, Philadelphia: John Benjamins. Shannon, Thomas. 1995 Toward a cognitive explanation of perfect auxiliary variation: some modal and aspectual effects in the history of Germanic. American Journal of Germanic Linguistics & Literatures 7(2): 129–163. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76: 859–890. Sorace, Antonella. 2004 Gradience at the lexicon-syntax interface: evidence from auxiliary selection and implications for unaccusativity. In: Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 243–268. Oxford: Oxford University Press. Torrego, Esther. 1989 Unergative-unaccusative alternations in Spanish. MIT Working Papers in Linguistics 10: 253–272.

Peter Öhl, Bergische Universität Wuppertal

Periphrasis as a precursor of analytic inflection: auxiliation in the (pre-)history of German 1 Introduction In the literature on auxiliaries and auxiliation (i.e. recategorisation of a lexical verb as an auxiliary), the uses of the terms periphrastic and analytic with respect to inflection are often not properly distinguished. In the terminology of systematic language typology, analytic inflection denotes a subtype of morphological marking where inflectional paradigms are not established by specific affixes but by means of morphologically autonomous words, i.e. auxiliaries or particles. Two central tense forms in English and German may serve as an example: in these languages, virtually every full verb can be “inflected” for future or perfect tense by means of specific auxiliaries (like will/shall and have, and werden and have/be). Thus, these forms are construed analytically in these languages. The term periphrastic indeed has a similar notion, meaning the expression of one function by more than one word. However, not every kind of periphrasis is suitable to form a complete paradigm in the sense of being valid for a whole class of elements like the class of lexical verbs. Very often, periphrastic constructions using lexical elements occur in cases where there is a gap in the grammatical paradigms of a language. Examples from different periods of different languages clearly demonstrate that conventionalised periphrases can indeed be considered the historical precursor of analytic forms of inflection. However, grammaticalisation must be completed if we want to speak of analytic inflection as a paradigmatic means of functional marking. Taking the German analytic perfect tense with haben (‘have’) as an example, one can state that it has developed from a periphrastic construction where HAVE originally was used as a full verb that was complemented by a predicatively used past participle. Similarly, the BE-perfect developed from the predicative construction with the copula sein. The change from conventionalised periphrases to a fully grammaticalised analytic inflectional pattern was not completed until the 11 th century. Due to the possibility of observing the development of the German analytic perfect tense in statu nascendi, it is a very suitable case example for an explanative account of auxiliation via lexical periphrasis. The paper is structured as follows. Since we found periphrastic forms and analytic inflection not properly distinguished in the literature, it is first necessary to define the terms as applied in this paper. This is done and discussed in the following section 2. In section 3, I discuss the relevant factors of syntactic

214

Peter Öhl

change; I also develop a model of auxiliation that integrates both usage based and acquisition based aspects. In section 4, I exemplify this model with the development of the German perfect tense, discussing the have-perfect first and continuing with some remarks on the auxiliation of the copula. The conclusion in section 5 is a case for an integrative model of language variation and change adequately considering and assessing both performance-based factors and the conditions related to language acquisition.

2 Analytic inflection vs. periphrasis The use of the term analytic in the context of morphological marking was first introduced in an early work on the typology of languages by Schlegel (1818: 16), and has, in modern structural typology of morphological systems, achieved the status of a scientific term, distinguishing inflectional paradigms where particles (PTCs) and auxiliaries (AUXs) are used to denote grammatical features, from those called synthetic, where this function is fulfilled by affixes (AFFs). The term periphrastic is very often used synonymously with analytic, neglecting the fact that it has a much wider range. Originally, periphrasis is a term from poetic and rhetoric arts, meaning the employment of an idiomatic expression, e.g. to avoid repetition. Periphrases very often replace a one-word expression with more than one word (cf. Bußmann 2008). In fact, its meaning is very close to that of paraphrasis, both terms originating in a prefixed form of the Greek word meaning ‘speak’: (1) a. gr. φράζειν b. gr. περί c. gr. παρά

‘speak, say’ ‘around’ ‘aside’

In the traditional use of the terms, the major difference is that periphrastic forms become common (or conventionalised), whereas a paraphrasis occurs more randomly (e.g. in order to explain the lexical meaning of a word). Thus, examples from lexicography and phraseology may well serve for illustration. (2) a. lexical form: policeman b. paraphrasis: man belonging to a department of government concerned with the keeping of public order c. periphrasis: guardian of the law

(cf. OED)

The use of the periphrasis in this sense is also quite common to express more abstract or grammatical meanings in the world’s languages, especially if the

Periphrasis as a precursor of analytic inflection

215

grammatical system of a language lacks an inflectional paradigm for a specific grammatical category. This can easily be illustrated with some examples from German in comparison to English and French. Standard German does not distinguish the verbal aspect systematically1 – unlike English, where the so-called progressive tense is realised by a fully grammaticalised form of analytic inflection. Instead, a construction with a temporal adverb can be applied, which is a periphrasis in the terms explained above. (3) Er ist gerade dabei, sein Fahrrad zu reparieren. he is just there-at his bicycle to repair-inf ‘He is just repairing his bicycle.’ There is also a more colloquial way of using a prepositional phrase in a predicative construction with the copula: (4) %Er ist [PP beim/am [[DP (sein) Fahrrad] reparieren ]. he is at.def his bicycle repair-inf ‘He is repairing his bicycle.’ Unlike in English, these periphrastic forms are contextually (e.g. sentences with perfective verbs sound odd even if they are not punctual) and grammatically (e.g. neither of these periphrases occurs in the passive) restricted: (5) ?Der Zug ist gerade dabei, am Bahnhof anzukommen. the train is just ‘there-at’ at.def station to-arrive.inf ‘The train is just arriving at the station.’ (6) a. *Das Buch ist dabei, gelesen zu werden. the book is ‘there-at’ read.pii to aux(pass) b. *Das Buchi ist [PP am [[DP xi] gelesen werden]] . the book is at.def read.pii aux(pass) ‘The book is being read.’ This means they are not paradigmatic, in the sense of inflecting a lexical class by means of a grammatical rule. Taking paradigmaticity as an indicator of

1 Although it does distinguish sentential aspect, which is part of the discussion below. 2 Note that I gloss both the past participle and the passive participle as PII (second participle); this is because they are homoformous anyway and very often polyfunctional or ambiguous.

216

Peter Öhl

grammaticalisation (cf. Lehmann 1995: 123), one may thus conclude that many of the periphrastic forms are not fully grammaticalised, whereas analytic inflectional forms are. If this is the case, one can further state the status of the progressive as analytic inflection in languages like French, where the source of grammaticalisation was an idiomatised adverbial paraphrasis with the prepositional phrase [PP en train] (lit. ‘in move’)3: (7) a. Il est en train de réparer son vélo. he is in move of repair-inf his bicycle ‘He is repairing his bicycle.’ b. Le train est en train d’arriver the train is in move of arrive-inf ‘The train is arriving.’ c. Le livre est en train d’être lu. the book is in move of be-inf read.pii ‘The book is being read.’ It can also be stated that in some German varieties (e.g. Allemanic or several varieties spoken in the Rhine area), the progressive periphrasis with the preposition am has been fully grammaticalised and can be regarded as analytic inflection. This is in addition evident from a change that has reanalysed the syntactic structure, making the predicative reading unavailable (loss of syntagmatic variability in the terms of Lehmann 1995: 123 ff.): in these dialects, am (originally a contracted form of the preposition an and the neuter definite determiner in the dative case) is no longer a preposition but a particle adjacent to the verb. It denotes – together with the infinitive – the progressive aspect (cf. Bhatt and Schmidt 1993). The following examples are abstracted from my own field research:4

3 However, to my knowledge, this conventionalisation took place before the lexicalisation of the word train as denoting a vehicle; originally, it just meant something like the noun ‘move’. 4 Note that these just reflect the dialect syntax, not the dialect phonology or lexicology which would differ substantially between the varieties where this grammatical form is used. Note also that the transcription of German dialect data is still not normalised in the literature. In several Alemannic subvarieties, the sentences in (6) would be put forward roughly like: (i)

a. b. c.

Der isch sei Rädle am richte. Dr Zuag isch grad am oakomme. Des Buach isch scho am glea werre.

Periphrasis as a precursor of analytic inflection

(8)

217

a. %Er ist sein Fahrrad am reparieren. he is his bicycle ptc repair-inf ‘He is repairing his bicycle.’ b. %Der Zug ist gerade am ankommen. the train is just ptc to-arrive.inf ‘The train is arriving.’ c. %Das Buch ist bereits am gelesen werden. the book is already ptc read.pii aux(pass) ‘The book is already being read.’

It is not difficult to find more examples, which can be explained in a similar way. While in English and French one can distinguish the close future tense by a specific grammaticalised use of the verb GO, German has conventionalised periphrastic forms that are not yet fully grammaticalised and are thus contextually restricted. Since the expression to be going to is fully grammaticalised in English (I do not go into the details of the semantics of the construction; for discussion cf. e.g. Wekker 1976), it is no problem to also use it together with the full verb variant of go. The same is true for French aller ‘go’. (9)

a. He’s going shopping. b. He’s going to5 go shopping.

(10) Il ne va pas aller à la maison. he neg go.3sg neg go to the home ‘He is not going to go home.’ In German, the combination of the verb go with an infinitival verb phrase is indeed also highly conventionalised and associated with an idiomatised connotation of the (close) future tense. However, since it is still a periphrasis employing a full verb, doubling of the verb go is at least stylistically marked, and many speakers judge it as unacceptable.

5 Hopper and Traugott (2003: 1) mention as a further indicator of full grammaticalisation that the collocation going+to can be phonologically contracted only in the case of analytic temporal inflection, not if to is used as a lexical preposition. (i) He’s going to/ %gonna go shopping. (ii) He’s going to/ *gonna college. This may be interpreted as a reanalysis of going+to as one functional element.

218

Peter Öhl

(11) a. Er geht einkaufen. he goes shop-inf b. *Er geht einkaufen gehen. he goes shop-inf go-inf (12) *Er geht nicht heimgehen. he goes neg home-go Other periphrastic forms expressing close future in German involve the idiomatic use of prepositional phrases like auf dem Weg ‘on the way’ or im Zuge ‘in the move’: (13) a. Er ist auf dem Weg [PP zum Einkaufen] / einzukaufen. he is on the way to-def shop-inf / to-shop-inf b. Er ist im Zuge, einen Artikel zu verschicken. he is in-def move a article to post Again, these periphrases are contextually restricted; e.g. im Zuge cannot be combined with a non-animated subject. (14) Das Buch ist auf dem Weg / *im Zuge zu erscheinen. the book is (on the way / in the move) to appear On the other hand, since neither those periphrases nor the idiomatised use of gehen are grammatical markers in the sense of explicitly denoting a grammatical feature, they are not complementary and can be combined in one sentence. (15) Er ist auf dem Weg / im Zuge, einkaufen zu gehen. He is (on the way / in the move) to-shop-inf to go In French, it is in fact also possible to combine en train and auxiliarised aller, even though both of them have to be regarded as grammatical markers. However, as shown above, en train has been grammaticalised not as denoting close future but progressivity. Thus, it does not compete with aller faire quelque chose. Both can combine in order to mark close future with the progressive aspect. (16) Il est en train d’aller achêter quelque chose. he is in move of go buy some thing ‘He is going to buy something.’

Periphrasis as a precursor of analytic inflection

219

A last example for periphrases that are strongly conventionalised but, from a systematic point of view, not yet cases of analytic inflection, are passive-like expressions in German. In all German varieties, verbs denoting reception are used in contexts of object-subject-diathesis, which is very similar to grammatical passivisation. This is often referred to as dative-passive or recipient-passive in the literature (cf. Wegener 1991). However, it can easily be shown that neither of the verbs used in this kind of periphrasis is fully grammaticalised. All of them are more or less contextually restricted: (17) a. Er bekam / kriegte / erhielt ein Buch geschenkt. he got / got / recieved a book give-as-present.pii ‘He was given a book as present.’ b. Er bekam / ?kriegte / *erhielt die Wahrheit gesagt. he got / got / recieved the truth tell.pii ‘He was told the truth.’ c. *Er bekam / kriegte / erhielt alles geglaubt. he got / got / recieved everything believe.pii ‘He was believed everything.’ Passivisation is not possible for any of these verbs of reception if the dependent verb lacks a direct object, even if this verb governs a dative object such as helfen ‘help’. This is at least true for Standard German. However, there are in fact German varieties where passivisation of helfen is possible with auxiliarised kriegen, e.g. in Luxemburgian6. (18) %Ich habe von denen geholfen gekriegt. I have from them helpV.pii got I was helped by them.’ This means that in these varieties the periphrasis get+PII seems to be fully grammaticalised as an analytic passive form where the indirect object can gain the function of the sentence subject. The former verb of reception has lost its argument structure, serving as an auxiliary for this specific form of passivisation. In all of the cases discussed here, it is quite evident that there is a strong correlation between periphrasis, i.e. a conventionalised way of expressing

6 Thanks to Carolin Döhmer for this piece of information.

220

Peter Öhl

meaning indirectly, and fully grammaticalised analytic forms. This is not only the reason why analytic tenses are often referred to as periphrastic in traditional accounts of syntactic change. In fact, it may also be the core of the gradual nature of auxiliation and similar processes of grammatical change often referred to in the literature (cf. Lehmann 1995: 22 ff., and many more). It is one of the purposes of this paper to turn this trivial correlation into a theoretically relevant observation. It is certainly not true that all periphrases lead to grammaticalisation. Neither shall I attempt to argue that all cases of analytic inflection must have a periphrasis as a precursor. Rather, I aim to answer one of the conditional questions of language change (the “general question of language change” in the terms of Coseriu 1974): under what circumstances can periphrasis become analytic inflection? I shall try to give an explanation in terms of an integrative model of language change, considering both formal and functional factors of grammar and speech. After a more general discussion of syntactic change in the following sections, I shall exemplify this model with data from the history of German, i.e. the rise of the so-called periphrastic perfect tense with have and be as auxiliaries.

3 Syntactic change For several decades, the scientific discourse on diverse phenomena of grammatical change has been coined by competing generative and functionalist approaches. Even though I do not deny my conviction of the explanatory potential of the generative model, this paper is not at all meant to advocate only the generative approach, nor to pull functional approaches to pieces. Rather, it is meant to argue for an integrative model which draws on appropriate answers provided by both approaches to the relevant questions. It is uncontroversial that functional approaches focus on the role the system of language plays for language use. As a consequence, it is mainly usagebased explanations that result from the respective observations, which then lead to assumptions as represented by statements like the following: “Of course, it is us using the language who change the language, by adapting it to our needs” (Nübling et al. 2006: 4; transl. PÖ).7 Even though this view is not at all beside the point (as seen above, speakers use their creativity to express grammatical notions), there is an important restriction often pointed out by

7 Original: Selbstverständlich sind wir, die wir Sprache verwenden, diejenigen, die die Sprache verändern, indem wir sie unseren Bedürfnissen anpassen.

Periphrasis as a precursor of analytic inflection

221

generative approaches, which has in fact become one of their leading arguments: the basic properties of the linguistic semiotic system are not determined by its communicative function and its basic rules are not open to manipulation by the speaker (Öhl 2006: 235 ff.; cf. Grewendorf 1999: 319 f.). Let us take the verbal aspect as an obvious example: the grammatical feature of the verbal aspect itself does not determine whether it may be linguistically represented by an analytic verb form, like in English (see above), by a grammaticalised adverbial, like in French (see above), or by affixes, like in Russian (cf. e.g. Leiss 2000: 216 ff.). Instead, it is the basic structural properties of grammar that provide us with the options we can choose to express aspect and other grammatical features. Moreover, speakers cannot freely choose between these options when producing their sentences. They have access only to the options of the grammar constraining their language.8 That is why speakers cannot invent grammatical rules or freely change the rule system of their languages. What they can do, however, is make creative use of the grammar at their disposal. (19) a. Er ist gerade dabei, sein Fahrrad zu reparieren. he is just there-at his bicycle to repair-inf b. *Er ist sein Fahrrad reparierend. he is his bicycle repairing ‘He is repairing his bicycle.’ The explanation put forward for these restrictions by generative grammarians is well known and thus just repeated here in short: the rules of grammar are not just produced by our common cognitive skills but result from the properties of our language module, i.e. the way it calculates structures. This determines not only the principles of language but also the parameters by which the grammars of natural languages systematically differ. These parameters provide options that are chosen during language acquisition on the basis of the linguistic input received from the parental generation and can hardly be changed after they have been fixed. Grammar acquisition from the generative point of view is illustrated in the following graph where UG (universal grammar) stands for the innate properties of the language faculty that are relevant for building grammatical structures; a definition of the term parameter is given below.

8 For the purposes of explaining the general aspects of first-language-grammar, second language acquisition and the controversies about the options of exceptional late grammar acquisition can be neglected.

222

Peter Öhl

(20) Grammar 1 (~ lang. competence of the parental generation)

Output 2 performance Grammar 2 (~ lang. competence of the children‘s generation)

performance Output 1

+ parameter setting Language Module (~ UG: Principles and Parameters)

(cf. Öhl 2006: 231; Cook and Newson 2007: 28 ff.; etc.) (21) Grammatical Parameters (GenGr) Grammatical Parameters are variables of the grammatical system which are set to structurally specific values on the basis of universal and innate principles and the data the child finds in the language s/he is exposed to during first language acquisition. This parameter setting is a prerequisite for the consistent inventory of rules for a specific/individual grammar. What is important to keep in mind here is that we as researchers do not have direct access to grammars. Instead, we examine the data produced both by the children’s and the parents’ generation. Looking at ambiguous data from periphrasis as described above, it seems obvious that the data must be examined quite carefully in order to decide whether a grammatical change has taken place between the two generations. And, what is not only important but factorial, the child’s analysis of the output produced by the parental generation may lead to the setting of parameter values that differ from those of the parent’s grammar, if the data is ambiguous – e.g. between lexical periphrasis and analytic inflection. The factors potentially triggering aberrant parameter setting have often been discussed in the literature (cf. Lightfoot 1991, 1999; Roberts 1993, 2007; Roberts and Roussou 2003; Gelderen 2005, 2009, etc.), and this paper is intended to add some aspects from language use. Before turning to that, I would like to briefly describe the parameters of syntax that are important for the following discussions. One of the most basic ones that distinguishes the syntax of languages like German from languages like English is the head position V° in the VP (cf. Cook and Newson 2007: 41 ff.; Roberts 2007: 92 ff.):

Periphrasis as a precursor of analytic inflection

223

(22) a. Paul has [VP eatenV° [DP an apple]] b. Paul hat [VP [DP einen Apfel ] gegessenV°] Paul has an apple eaten Another parameter concerns the potential positions of the finite verb which is, in generative grammar, mostly referred to in terms of verb movement. Potential verb positions are in the VP, in the IP (inflection phrase) dominating the VP, and in the CP (complementiser phrase) dominating both (details can be looked up e.g. in Cook and Newson 2007). In German main clauses (which contrast with most of the embedded ones owing precisely to this property), the finite verb moves from its base position to a higher position in the CP, producing a FIN-second-structure (23a). The normal position of a finite modal in languages like English is in the IP, following the subject (23b). Modals can also move to the CP in English (23c), however only in non-declaratives. While full verbs also move to the CP in FIN-second languages like German (23d), they can’t even leave the VP in English (23e). This is why, in some cases, do-support is necessary, e.g. if negation intervenes between V° and I°. (23)

a. [CP Leider [C’ kannv [IP Paul [VP keine Äpfel essen ] tv ]]] (FIN-second) b. Unfortunately, [IP Paul [I’ can [NegP not [VP eat apples ]]]] (FIN in I°) c. [CP Canv [IP Paul [I’ tv [VP eat apples ]]]] ? d. [CP Leider [C’ isstv [IP Paul [VP keine Äpfel tv ]]]] e. *Unfortunately, [IP Paul [I’ eatsv [NegP not [VP tv apples ]]]] (*V to I) f. Unfortunately, [IP Paul [I’ doesv [NegP not [VP eat apples ]]]] ( AUX in I)

A further parameter standing in relation to both of those just mentioned concerns the way in which functional categories are represented. I am using a notion adapted from a proposal by Roberts and Roussou (1999): if a functional feature (like tense) is realised by an affix attached to a lexical category, this means a lexical head has to move to the corresponding functional head position (specific functional heads that are located either in the domain of CP or of IP). If a functional feature is realised by a particle or an auxiliary, they do not have to move (even though they may move by a further operation). (24) Parametrisation of Functional Categories a. AFF (→ movement of lexical heads) b. ptc, AUX (→ no movement of lexical heads) (cf. Roberts and Roussou 1999)

224

Peter Öhl

Note that an auxiliary is nothing but a special case of an inflected functional element expressing an additional feature with an affix; this feature (e.g. agreement) may trigger movement. The difference can be illustrated with the synthetic preterite contrasted with the analytic perfect tense in German. In the former case, the inflected lexical verb moves to the IP where tense and agreement are located and may move on to the CP in main clauses. In the latter case, the lexical verb stays in its base position and the inflected AUX hatte is inserted in I°, representing preterite tense and agreement (alternatively: it is moved from T° to Agr°).9 (25) a. dass [IP sie [[NP den Studenten ] ti cmp she det student ‘… that she praised the student.’

VP]

lob-te IP] praise.pst.3sg

b. dass [IP sie [[NP den Studenten ] gelobt VP] hat-te IP] cmp she det student praise.pii aux.pst.3sg ‘… that she had praised the student.’ In Latin, the active perfect tense is represented by a finite verb form, more precisely an affix carrying the features of tense, aspect and agreement. This means the lexical verb would move to IP. The passive perfect tense, however, is an analytic form with the copula used as an AUX, thus constructed in a way similar to the analytic perfect in German. (26) a. lauda-vit, lauda-verat; lauda-tus sum b. (dass) ich gelob-t hab-e/ ha-t-te/ word-en bin The following example shows that tense and aspect can also be represented by particles, in languages lacking agreement like Tok Pisin. (27) wanpela man i bin skulim mi long Tok Pisim one man prog ant teach me in Tok Pisin ‘A man was teaching me in Tok Pisin.’ (Tok Pisin; Lightfoot 1991: 177)

9 What exact feature is represented by the AUX have is the topic of several discussions (cf. Grewendorf 1995; Musan 2002). Perfect tense is encoded by the AUX and the perfect participle compositionally. For the time being, I would just like to state that AUX represents anteriority, which I shall specify further below.

Periphrasis as a precursor of analytic inflection

225

Generative researchers of grammatical change such as Roberts (1993, 2007), Roberts and Roussou (2003: 194 ff.) or Gelderen (2004a) treat grammaticalisation of AUXs in terms of structural economy, as a by-product of structural simplification caused by eliminating syntactic movement in first language acquisition. Under this view, children seek the least expensive way of designing syntactic structures: a verb turns into an auxiliary simply because movement to IP is uneconomical. This has also been referred to as merge over move (Roberts and Roussou 2003) or the Late Merge Principle (see the discussion in Gelderen 2011: ch. 1.2.3). Since this view neglects a whole range of findings of grammaticalisation theory, such as the role of metaphor and metonymy as cognitive processes, controversies with functional accounts of grammaticalisation were unavoidable, where economy is treated in its relation to expressivity. This view is grounded on the assumption of a competition between the interests of speaker and hearer (“ease of production” vs. “ease of perception”, cf. Haspelmath 1998: 320; Hopper and Traugott 2003: 65 f.); it has in fact a long standing tradition, as reflected by the following quote from Martinet (1955): “The whole development of language is determined by the omnipresent contradiction of the communicative and expressive needs of human beings on the one hand, and, on the other hand, their tendency to restrict their mental and physical activities to a minimum” (Martinet [1955] 1981: 85; transl. PÖ).10 What these authors call expressivity is a factor of language change that must not be neglected in a generative approach, either. The way children interpret lexical elements may not only block structural simplification (cf. Öhl 2009a: 419 ff.), it may also cause grammatical change, when speakers make use of the grammatical options creatively and if children interpret the lexical material as grammatical markers, as in the case of periphrasis in the notion as defined above. Both structural simplicity and expressivity can be regarded as aspects of cognitive economy, assuming that explicit encoding of information is less costly with respect to both speech reception and production. And, given that language acquisition relies on parsing the output of speech production, this should be valid also for the development of a child’s grammar. This view is explicitly argued for in Öhl (2009b: 419 ff.) and formulated as two competing

10 Original: Die gesamte Sprachentwicklung wird bestimmt von dem stets vorhandenen Widerspruch zwischen den kommunikativen und den expressiven Bedürfnissen des Menschen einerseits und andererseits seiner Neigung, seine geistige und physische Aktivität auf ein Minimum zu beschränken.

226

Peter Öhl

cognitive strategies that are constitutive both for language use and language acquisition. (28) Minimal Effort in Computation (MEC) Use just as many operations as are necessary to design a structure converging with the features to encode. (29) Maximal Explicitness (MEX) Find the maximal amount of features converging with a consistent interpretation of a structural description. This means it is not just structural economy but also the informational potential of the input that is factorial for language acquisition and, following from that, for grammatical change. Another question the acquisition-based accounts must face concerns how such spontaneous and individual changes can spread over a speech community within a rather short period. Usage-based accounts seem to provide a much more intuitive explanation for how and why innovative expressions gradually become part of the grammatical system of a language, especially if grammatical rules are treated as usage-based generalisations over constructions (cf. Croft 2000); more traditional accounts (cf. Keller 1990) simply speak of conventionalisation of patterns of usage, which is, of course, an oversimplification. Usage-based changes are certainly one pillar of language development. However, the attempt at explaining grammatical change solely on the basis of speech production necessarily results in an overestimation of the speaker’s power to manipulate the rules of grammar. There are obvious formal criteria which massively constrain functionally motivated changes. Regarding the findings of generative approaches, grammatical change, i.e. changes in the basic rule system that is not accessible to the speaker, cannot simply be ascribed to creating and conventionalising ways of expression (cf. Öhl 2007, 2008). One of the first researchers to consider both generative and usage-based ideas of explaining grammatical change was David Lightfoot; the integrative approach used here owes very much to his discussion of the contrast between graduality and abruptness in language change (cf. Lightfoot 1979, 1991; 1999: 77 ff.). The quintessence of this discussion is: what changes gradually is not grammar itself but the way it is used in speech production; speech production, however, comes into play with the role of the “input” for language acquisition. Speakers make use of the options for manipulating the structure of clauses, which may become common and spread over a speech community. Given that sentences provide the triggers for parameter setting, this can result in the loss

Periphrasis as a precursor of analytic inflection

227

of robust input for parameter setting, i.e. input that makes the child fix a parameter’s value (cf. Lightfoot 1991: 63 ff.): “A sentence S expresses a parameter P if a grammar must fix P to a definite value in order to assign a well formed representation to S” (Lightfoot 1991: 19). Lightfoot (1999) further introduces the term cue for pieces of structure children parse in order to find parameter values for the acquisition of grammar. These cues are not only relevant for a cognitively economical way of acquiring grammar by avoiding the parsing of whole sentences before setting parameters that concern minor levels of syntactic structure, but they may also be misleading if they trigger a parameter setting that would be revised if a larger piece of structure had been parsed.11 To put it in the terms used in this paper, speakers producing sentences make creative use of the range of possibilities within a frame that is set by the inventory of lexical expressions and grammatical structures, constrained by the regular structure building operations (speech production; restricted options of enhancing expressivity; no manipulation of the basic rule system). Language learners interpret the input in order to acquire an inventory of lexical expressions and a system of regular structure-building operations. Modification of usage conventions in speech production changes the input diachronically and may manipulate the grammar of a language indirectly.12 So, what can be considered as gradual are variation, expansion and conventionalisation;13 regularisation, however, i.e. the real grammatical change, takes place abruptly during language acquisition. Since the patterns that become common in a speech community may change or even remove the triggers for parameter setting of a whole generation of children acquiring a grammar, it is not the grammatical change but the modified input that can spread. This is how a periphrasis can become a precursor of analytic inflection. An example that has been extensively discussed in the recent literature is the development of the French future, which not only involves auxiliation of the verb HAVE but also its further grammaticalisation as a suffix. Note that this change was possible only in the earlier stages of French when both the IP and

11 Note that this description of Lightfoot’s ideas is very undetailed and simplified; of course there is much more behind it, as explained in Lightfoot (1999). 12 Another kind of conventional change that is not discussed here but may play an interesting role by changing the input for language acquisition comes into the play with prescriptivism; cf. the discussion in Gelderen (2004b). 13 What is neglected here is the possibility of a sequence of micro-reanalyses, which, over a longer period of observation, would also give the impression of a gradual change of grammar; cf. Gelderen (2010).

228

Peter Öhl

the VP were, contrary to what is assumed to be the case in Modern French, head final. (30) Reanalysis of the future suffix in French I'

VP

I'

I°

VP

V°

. . . cantare . . .

t1



VP

.... habeoi

I'

I°

V° cantare



VP

.... habeo

I°

V° ti chanti-erai

(cf. Roberts 1993) The first stage of the process of change modelled here is the periphrastic use of have in a gerundive construction describing a deontic relation. Presumably, it was the connotation of futurity implied by deonticity that lead to the grammaticalisation as a future tense auxiliary which then represented a functional head like I°; note that such changes are often referred to as grammaticalisation of an implicature (cf. Rolf 1995). In the vulgar Latin varieties, this newly developed analytic form ousted the synthetic one that had been inherited from classical Latin (cf. Stotz 1998: 325 f.) and, after phonological reduction, became the source of further grammaticalisation, the development of a new future suffix. (31) Phonological reduction of Lat. habere cantare habeo > *[cantar aio] > chanterai (cf. Haspelmath 1998: 348 f.) Note that in our terms, the aspects of structural simplification, i.e. loss of movement, phonological reduction and affigation, are grounded on a cognitive strategy like MEC in (28) above, whereas the interpretational aspect, i.e. the fixation of the grammatical denotion as [FUT], is grounded on MEX (29). Since such processes of change are grounded on both principles of grammar and universal cognitive strategies determining speech production and language acquisition, it does not come as a surprise that they can occur not only in various languages, but even several times within one language where they affect the same functional paradigm. This is often referred to as cyclicity of grammaticalisation.14 So, the classical Latin future suffix developed from a periphrasis with 14 For the term cyclicity in general, cf. Abraham (2010) and Gelderen (2011); Gelderen also discusses numerous other linguistic cycles; for the future cycle in particular, cf. Gelderen (2011: ch. 7.4) and Abraham (2010: 264 f.).

Periphrasis as a precursor of analytic inflection

229

the subjunctive form (see below; I thank Martin Kümmel for this piece of information) of the copula BE in a way comparable to the grammaticalisation of habere. And, without wanting to go too far, one can state that French aller is a good candidate for the development of a new future auxiliary (see above example 10). (32) Development of future tense in Latin/French Proto Latin Class. Latin *kanta bhu-mos > canta-bimus sing be.subj-1pl sing-fut-1pl canta-re habe-mus > sing-inf hav-pres.1pl (GERUNDIVE > FUTURE)

French † chant-erons sing-fut-1pl (FUTURE) allons chant-er go-1pl sing-inf

(adapted from Eckardt 2006: 5) I shall now turn to an account of the rise and the development of the analytic perfect tense in German, which will also provide more evidence for the assumptions argued for so far.

4 Development of the German perfect tense 4.1 Have-perfect (cf. Öhl 2009a) The starting point for the development of the German analytic perfect tense with the AUX haben were predicative structures with PII in Old High German. They still exist in Modern German and are sometimes referred to as habenconfiguratives (cf. Businger 2011). There are also Modern English equivalents as follows: (33) a. He wants to have his car washed by no one else.

(passive-like HAVE-configurative) b. I have one apple (that is) (un)peeled. (depictive object predicative) c. We have everything ready and done. (object predicative in complex predication)

Beside the lexical verb haben, the relevant elements in German HAVE-configuratives are a direct object and a predicative element completing the construction

230

Peter Öhl

as the coda (i.e. closing element). This can be noted as a canonical schema, as suggested by Businger (2011: 30). (34) Canonical schema of HAVE-configuratives in German Subject – HAVE – NPacc – Coda (cf. Businger 2011: 30)15 The clause structure of these configuratives crucially differs from that of the analytic perfect tense since HAVE is a lexical verb generated in the single V° position, whereas HAVE as an AUX has its usual position in the IP and may even be generated in a functional head in the I-system, like T° (see fn.19 below). (35) a. dass [sie that they haben VP] have

[[NP die Getränke]] [AP (un)gekühlt] ([PP im Hause]) the drinks (un)chilled in-the house werden IP] will

b. dass [sie [[NP die Getränke] (*un)gekühlt VP] haben IP] In contrast, only transitive and ergative verbs were able to form a PII in early Old High German; the reason is that they were not yet used as part of an analytic tense form but only in configuratives with a direct object as the referential element. As long as there is such a restriction, one should assume predicative use even if a perfective interpretation is possible as an interpretatio moderna. (36) a. phígboum habeta sum gipflanzotan in sinemo uuingarten fig tree had someone planted in his vineyard 16 (Tat 102,2) b. in buah si iz duent [AP PRO zisamene gihaltan] zi habanne in book they it do together held to have (Otfr III 7, 54)

15 Note that, due to the German head final VP and IP, this would turn to Subject – NPACC – ‘Coda’ – HAVE in the basic clause structure that is found in subordinate clauses and with infinite forms of the AUX: (i) (ii)

dass wir alles fertig und erledigt haben Er will sein Auto von niemandem anders gewaschen haben.

16 Abbreviated references are clarified in the list of references under p. 245.

Periphrasis as a precursor of analytic inflection

231

The reanalysis of the analytic perfect from such configuratives has been accounted for both by generative and functionalist researchers. While Abraham (1992) suggested that HAVE became an AUX by means of a one-step-reanalysis of a small clause structure, Grønvik (1986) assumed that the use of HAVE as an AUX spread gradually from transitives to other verb classes by means of conventionalisation. Since both views in my opinion oversimplify the actual circumstances, I argued in (Öhl 2009a) that it was just the use of aspectually marked predicative constructions that increased steadily until the end of the 9th century. Only then did these structures become input for the learners’ reanalysis. In my terms, the development of such predicative constructions (i.e. reanalysis as complex predicates) and the further development of the auxiliary (i.e. recategorisation of V° as I°) can be explained on the grounds of a cognitive strategy like MEC in (28) above, whereas both the increase of use as an aspectual construction and the reinterpretation by the learner as a temporal form can be explained in terms of MEX in (29). Let us now turn to the analysis of the relevant data. HAVE-configuratives with PII occur in various early Germanic sources: (37) a. þin agen geleafa þe hæfþ gehældene your own belief you has healed (HomS 8,15: 24f) b. þa he ða hæfde þa wísan onfogne when he then had the leaders welcome.pii-acc.pl (Beda 344, 27) c. habde sie farfangene fiundo craftu had her cought away fiend(gen) might(dat) (Hel 3032) d. sie eigun mir ginomanan liabon druhtin minan they have me.dat taken.acc beloved.acc lord my.acc (Otfr V 7, 29) e. ir den christanun namun intfangan eigut you det christian name receive.pii have/own (Exh 9,5) f. pi daz er in worolto kiuuerkot hapeta prep dem he in world.dat shaped had (Musp 36)

232

Peter Öhl

Indicators of a predicative reading are the use of eigan ‘own’ as a suppletive verb form17 and nominal agreement at the PII. The latter was lost due to reduction in the course of the history of the language, which also supported the reanalysis of the PII from a nominal to a verbal form. I would like to briefly discuss Abraham’s (1992) proposal that the PII was reanalysed from the head of a small clause (presumably an AP) to a V° heading the VP of the whole sentence, which could be modelled as follows: (38) a. dass das Pferd die Fesseln bandagiert hat that the horse the fetlock bandaged-up has IP

VP

b.



V'

DP

I'

DP

SUBJ/– [das Pferd]

AP

V°

SUBJ/– [das Pferd]

VP

I°

DP

A°

hat

DP

V°

hat

OBJ [die Fesseln]

PII [bandagiert]

OBJ [die Fesseln]

PII [bandagiert]

There is some evidence that the PII as a secondary predicate did not head a small clause in the relevant constructions. The PII instead formed a complex verb together with haben, which is an option especially in OV-languages like German, where secondary predicates following the direct object are adjacent to the verb. I suggest that the formation of a complex predicate of two adjacent predicative parts can be considered a case of structural simplification according to MEC. First of all, note that haben-configuratives in Modern German18 do not behave at all like small clauses. A small clause (SC) is an autonomous domain of

17 eigan ‘own’ is used as a suppletive form for habēn in Pres.pl forms in OHG texts until Notker Teutonicus (~ 1000 AD; cf. Oubouzar 1975: 10 ff.). 18 I concede that this is not direct evidence about OHG grammar; however, a significant difference between OHG and NHG haben-configuratives cannot be grounded merely on theoretical assumptions.

Periphrasis as a precursor of analytic inflection

233

adverbial modification. In sentences with haben-configuratives, an adverbial immediately preceding the secondary predicate (i.e. the coda) modifies the verbal complex (VC; like rasiert haben and rasiert halten below). (39) a. Aus Unkenntnis wähnte sie [SC ihren Mann by unawareness considered she her husband geschäftehalber in Berlin]. for-business-reasons in Berlin

(SC)

b. Seit damals hati /hälti er stets den Kopf aus since then has/holds he ever the head for Hygienegründen [VK rasiert ti]. hygienic-reasons shaved

(VC)

Second, small clauses can be moved to the specifier position in front of the finite verb, which may be stylistically marked but not ungrammatical, unlike the fronting of the coda of the haben-configurative together with the complement: (40) a. ?[SC Ihren Mann geschäftehalber in Berlin] wähnte sie (nur) aus Unkenntnis. b. *[Den Kopf] [aus Hygienegründen] [VK rasiert ti ] hati /hälti er seit damals stets.

(SC) (VC)

Third, a small clause predicate cannot be fronted to the specifier position together with the main clause predicate. A complex like rasiert haben or rasiert halten, however, can: (41) a. *In Berlin gewähnt hat sie aus Unkenntnis ihren Mann geschäftehalber. b. [VK Rasiert haben/halten] könnte man den Kopf (z. B.) aus Hygienegründen.

(SC) (VC)

Since complexes like Recht haben und rasiert haben are just as coherent as verborgen halten and rasiert halten, they cannot be coordinated with phrases that otherwise could be complements of the lexical verb haben: (42) a. *Er hielt es verborgen und [ihm eine Rede]. coherent

234

Peter Öhl

b. *Er hält den Kopf rasiert und [einen lustigen Hut in der Hand]. coherent (43) a. *Er hat Recht und [ein loses Mundwerk]. coherent b. *Er hat den Kopf rasiert und [einen lustigen Hut in der Hand]. coherent Complex verbs like verborgen halten are already attested in the Old High German sources. (44) hialt uns (…) dar giborgan kept us there concealed (Otfr IV 55, 42) I assume that in a similar way, the full verb haben was not immediately grammaticalised as an AUX, but as a functional verb in aspectually marked complex predicates with an internal argument position. (45) a. dass ich ein Beispiel [V° parat [V° habe]] that I an example ready have b. Er hat es damals [V° verborgen [V° gehalten]] he has it back-then concealed kept Since the theta grid of the verb haben and other transitive verbs is parallel, they are able to unify their argument structures. Thus, the complex formation was originally only an option with transitive verbs (Öhl 2009a: 286 ff.). (46) a. was er in der Welt [V° geschaffen [V° hatte]] what he in world.dat shaped had b. dass ihr den christlichen Namen [V° empfangen [V° habt]] that you the christian name received have Structures with haben+PII of intransitive verbs, i.e. with PII that cannot be used as object predicatives, did not occur in OHG texts before Notker Teutonicus (∼ 1000 AD).

Periphrasis as a precursor of analytic inflection

235

(47) a. tar habet si imo geantwurtet sinero frago then has she him answered his question.dat (Notk I: 284, 26) b. habe ich keweinot so filo have I cried that much (Notk II: 15,30) c. so habet er gelogen thus has he lied (Notk I: 544,29) In these texts we find the situation comparable to Modern German, where there are examples that are ambiguous between the predicative and the analytic inflectional reading, and others that are unambiguous cases of the analytic perfect. (48) a. uuanda du gemíchellichot hábest dinen námen (ambiguous) when you greaten-pii have-2sg your name ‘when you are having your name greatened’ ‘when you have greatened your name’ (Notk Ps. III, 997) b. tar habet si imo geantwurtet sinero frago (unambiguous) then has she him answered his-gen question-gen ‘then she has given him an answer to his question’ (Notk I, 284, 26) Let us have a look at the integrity19 of haben in different constellations with a PII in order to illustrate the potential of the modified input for grammatical change. For easier modelling, I will use more examples from Modern German: (49) a. Das Zebra hat vier Hufe, in der (pii as postposed attribute) the zebra has four hooves in the Regel gewetzt(e). rule whetted-(agr) ‘The zebra has four hooves that are, as a rule, whetted.’

19 For the loss of integrity as a parameter of grammaticalisation, cf. Lehmann (1995: 123 ff.).

236

Peter Öhl

b. [PROi gewetzt] hat seine Hufei das whetted has its hooves the Zebra, zebra

(pii as predicative attribute)

[PRO beschlagen] liegen sie in der Regel nur beim shod lie they in the rule only with.def Hauspferd vor. house-horse ahead ‘Whereas the zebra has hooves that are whetted, they are, as a rule, shod only in the case of the domestic horse.’ c. dass ein Zebra immer [seine Hufe (pii in a verbal complex) [V gewetzt hält/hat]] ‘that a zebra always keeps/has its hooves whetted’ d. dass das Zebra seine [VP Hufe gewetzt] hat ‘that the zebra has whetted its hooves’

(analytic perfect)

In fact, there may be some chance for haben+PII to be reanalysed as an inflectional form from a construction with the possessive reading (49a). However, the more alternative constellations with haben+PII there are, the higher also the frequency in the input for language acquisition. Moreover, with the existence of input with desemanticised haben where perfectivity (or anteriority20) is implicit (49b+c), there are even potential triggers for resetting the relevant parameter according to MEX, turning the former full verb into an IP-element (cues in the sense of Lightfoot 1999). Here it should be emphasised that it is not the construction with have but the perfective/anterior semantics of the PII itself that makes an aspectual/temporal reading of the sentences like (49b+c) possible. Thus, the reason for the

20 Note that use of the PII does not necessarily denote perfectivity; that is why, in German, the analytic tense form haben+PII, even though it is called the perfect tense, often just denotes anteriority, e.g. with punctual resultatives (cf. Öhl 2014; detailed discussion can be found also in Musan 2002). (i)

Er hat den Ball gerade ins Tor geschossen. he has the ball just into-def goal shot ‘He just shot the ball into the goal.’

The non-perfective semantics are also the reason why have+PII is not used in the translation to English, where the use of this construction diachronically developed in a different way.

237

Periphrasis as a precursor of analytic inflection

aspectual/temporal contrast of sentences like the Old High German ones in (37d and e) above and sentences like the following is grounded on the use of the PII. (50) a. was er in der Welt schuf what he in world.dat shaped b. dass ihr den christlichen Namen empfingt that you the Christian name received

(preterite) (preterite)

The difference between explicit inflection and implicit aspectual/temporal reading is illustrated here, again with Modern German counterparts: (51) a. was er in der Welt geschaffen/ parat/ zur Verfügung hatte b. dass ihr den christlichen Namen empfangen/ zur Verfügung habt

(implicit perfectivity) (implicit anteriority)

Further reanalysis of the structure (the PII then heading the VP) and recategorisation of haben created the new paradigm of analytic tense, where the AUX presumably just represents anteriority, as a head in the I-system,21 whereas perfectivity is an additional feature that may be provided by the PII, depending on the semantics of the verb (see fn.18). (52) dass [IP ihr [VP den christlichen Namen empfangenV° ] habtI° ] that you the Christian name received have(aux) When haben became an AUX and no longer selected a direct object, it could also be used with intransitive verbs, which was not possible in the precursing periphrases with haben as a full verb. Thus, the use of the construction haben+PII signalling aspectual or temporal markedness (i.e. perfectivity/anteriority) gradually expanding in the course

21 Note that this is also kind of simplifying, given that the AUX also occurs in the infinitive: (i)

Er soll den Aufsatz gestern geschrieben haben. he shall the paper yesterday write.pii aux ‘He is said to have written the paper yesterday.’

There are several ways of explaining this, e.g. by a split-IP-model with an infinitival AUX in T°, whereas finite forms are always in Agr°.

238

Peter Öhl

of the 9th and 10th centuries finally provided the input for auxiliarising haben, which is a case of abrupt grammatical change. As said above (section 3), changes like this are grounded on universal principles of grammar and cognition. That is why they can occur in all languages sharing the conditional prerequisites, such as the existence of a verb like have. Desemanticisation of possessive HAVE is in fact a change converging in several languages, e.g. in Latin, where secondary predication together with habere already existed in the classical period. (53) a. Necdum omnia (…) edita facinora habent not-yet all(acc.pl) detect(pii.acc.pl) crime(gen.pl) have(3pl) ‘The did not yet have all of the crimes detected.’ (Livius XXXIX, 16, 3; cf. Salvi 1987: 229) b. Hannibal quia fessum militem proeliis Hannibal because exhaust(pii.acc) army by combats operibusque habebat, … labour-koor had ‘Since Hannibal had an army exhausted by combats and labour, …’ (cf. Thielmann 1885: 376) Structures like these provided the input for the rise of the analytic perfect tense in later periods of Romance. However, it is certainly not adequate to interpret these data as early occurrences of a so-called “periphrastic perfect”, as suggested e.g. by Thielmann (1885). They are simply periphrases using the lexical verb habere – like the haben-configuratives addressed above. This is also made evident by their occurrence together with the synthetic perfect form of habere (for a more detailed discussion cf. Öhl 2009a: 273 ff.). (54) … quam semper cognitam habui what.acc.fem always think.pii.acc.fem have.perf.1.sg ‘(things) that I have had (?as) thought.’ (anonymous; cf. Grandgent 1962: 55)

4.2 Remarks on the auxiliation of the copula Since the former copular verb BE is used as a perfect auxiliary with certain verbs not only in German but also in several other Germanic and Romance languages, some concluding remarks on (the rise of) auxiliary choice seem to be necessary.

Periphrasis as a precursor of analytic inflection

239

The grammaticalisation of BE22 as an AUX was a process similar to that of have. Since there is no obvious interdependence of the two processes, the auxiliation of both of them may be regarded as a case of convergence. The development of the copula was treated in a functionalist framework by Dik (1987), whose explanation is similar to my account of the development of the input for parameter resetting, however, without being explicit about the question of how a conventionalisedform becomes regularised as part of the grammar: “Innovative aspectual forms (were) reinterpreted as temporal or diathetic later on” (Dik 1987: 80). (55) Caesar victus est. Caesar beaten is ‘Caesar has been beaten.’ (vgl. Dik 1987: 69) Note that the analytic form esse+PII was used in Latin only for the passive of the perfect tense and of the so-called deponentia (i.e. verbs that are inflected like passives even though they have an active meaning). Several of those were just ergative verbs, however, other ergative verbs could be inflected synthetically for the perfect active (see below), just like the transitive and unergative verbs. In versions of Bible verses in older Germanic languages where there was no perfect tense, Latin sentences with the perfect of ergatives and deponentia were translated using a predicative construction with BE+PII. This is illustrated below with synopses of Bible verses, each with the version from Luther’s Bible for comparison. (56) a. qui venerant ex omni castello Galilaeae

(Vulg Lk 5,17)

b. die komen waren aus allen Merckten in Galiläa ‘who had come from all towns in Galilee’

(Luth)

c. þaiei wesun gaqumanai us allama haimo who were come.pii.nom.pl from all homes Galeilaias Galilee.gen ‘who were people having arrived from all homes of Galilee’

(Wulf)

22 I do not even attempt to explain the development of the copula from a former verbum substantivum which may be comprehensible in a quite intuitive way but is not at all historically reconstructable.

240

Peter Öhl

(57) a. defuncti sunt enim qui querebant animam pueri

(Vulg Mt 2,20)

b. Sie sind gestorben, die dem Kinde nach dem leben stunden. ‘They have died, those who sought the boy’s life.’ c. arstorbane sint thie thar suohtun thes die.pii.nom.pl are rel there sought dem.gen knehtes sela knave.gen soul ‘They are dead, those who sought the boy’s life.’

(Luth)

(Tat 11,1)

The crucial difference between ergative verbs and other intransitive verbs is that their PII can be used as a predicative, just like that of transitive verbs. It not only denotes a predication over the direct object but also over the subject of a sentence. Thus, like the transitive verbs, ergative verbs could produce a PII for predicative use long before it was used in order to make an analytic tense form. In Old High German, the copula occurs not only with adjectives but also with the present and the past participle of various verbs. (58) a. thaz er sculdig ist widar got that he guilty is against god (be guilty as VC?)

(Exh 41 f.)

b. Gotes geist ist sprehhendi God’s spirit is speaking (is ≈ exists?)

(Is 4.2.5)

c. dhasz christ iu ist langhe quhoman that christ you is long come.pii

(Is 26.14)

While – unlike in English – there was no grammaticalisation of the periphrasis BE+PI (58b) in OHG (on the periphrasis in Old English cf. e.g. Nickel 1966), BE+PII was grammaticalised as an analytic tense form for ergative verbs.23 Like in other languages, the asymmetry of auxiliary selection also persisted 24 in German – with some language specific variation having as it developed

23 I concede that this statement is also somewhat oversimplifying; however, this is not the right place to repeat the large quantity of discussion on auxiliary selection. I refer to the discussion in Öhl (2009b: 295 ff.) and some more representative work like Haider and RindlerSchjerve (1987) and Grewendorf (1989) and, of course, the papers in this volume. 24 On persistence as a characteristics of grammaticalisation processes, cf. Hopper and Traugott (2003: 94 ff.).

Periphrasis as a precursor of analytic inflection

241

in all of the languages (see the other articles in this volume). I give just two telling examples: (59) a. Ich bin in der Schule geblieben. b. Je suis resté à l’école. (60) a. Ich bin zur Schule gerannt. b. J’ai couru a l’école. As shown by these sentences, auxiliary selection is parallel in German and French with the verb STAY, but there is a difference with the verb RUN. This may have been caused by a change of semantic conceptualisation of the verb RUN in one of these languages (cf. Öhl 2009: 300). As is well known, in languages like Modern English there is a generalised AUX used for the analytic past tense forms. This is due to a diachronic change ousting BE as a perfect tense auxiliary (cf. Denison 1993). (61) a. We have stayed at school. b. We have run to school. In formal terms, this means that have was grammaticalised a further time, such that the selection of specific verbs was lost and its formal properties were reduced to the expression of the temporal feature. Again I refer also to the discussion in other papers in this volume.

5 Conclusions Due to alternating performance- and acquisition-based changes, the grammaticalisation of the perfect tense in German cannot be explained by purely formal or functional methods. Functionally motivated changes are certainly one pillar of language change. However, there are obvious formal criteria which constrain the options of variation. Changes in the basic rule system of a language that is not accessible to the speaker cannot simply be ascribed to speech variation (cf. Öhl 2007, 2008). On the other hand, usage-based changes in language systems are often neglected in accounts based only on language acquisition. In processes of grammaticalisation, change of what has been called the core grammar in the generative framework is often initialised by functional variation at what has been called the fringe, i.e. the areas of a grammatical system that are accessible to manipulation in linguistic performance. One ex-

242

Peter Öhl

ample is the rise of periphrastic forms using lexical material creatively but within constraints given by the present grammar. Only if taken as input for the acquisition of grammatical rules can these forms be regularised as part of a changed grammatical system. It is characteristic of such a kind of change that the options of using such newly developed analytic forms significantly differ from those for the original periphrasis, e.g. by the generalisation over unergative verbs. My long-term objective that I intend to reach by broadening the database of examination to other areas of change (as I did when accounting for change in complementiser systems in Öhl 2009b) is an integrative model of language variation and change that adequately considers and assesses both performance-based factors and the conditions related to language acquisition.

References Abraham, Werner. 1992 Event structure accounting for the emerging periphrastic tenses and the passive voice in German. In: Garry W. Davis and Gregory K. Iverson (eds.), Explanation in Historical Linguistics, 1–15. Amsterdam, Philadelphia: John Benjamins. Abraham, Werner. 2010 Methodische Überlegungen zu Grammatikalisierung, zyklischem Wandel und dem Wechsel von Analytik zu Synthetik – und zyklisch weiter zu Analytik (?). In: Dagmar Bittner and Livio Gaeta (eds.), Kodierungstechniken im Wandel. Das Zusammenspiel von Analytik und Synthese im Gegenwartsdeutschen, 249–274. (Linguistik – Impulse und Tendenzen 34.) Berlin, New York: De Gruyter. Bhatt, Christa and Claudia M. Schmidt. 1993 Die am + Infinitiv-Konstruktion im Kölnischen und im umgangssprachlichen Standarddeutschen als Aspekt-Phrasen. In: Werner Abraham and Josef Bayer (eds.), Dialektsyntax, 71–98. (Linguistische Berichte, Sonderheft 5.) Opladen: Westdeutscher Verlag. Businger, Martin. 2011 als Vollverb. Eine dekompositionale Studie. Berlin, New York: De Gruyter. Bußmann, Hadumod. 32002 Lexikon der Sprachwissenschaft. Stuttgart: Kröner. Cook, Vivian and Mark Newson. 2007 Chomsky's Universal Grammar: an Introduction. Oxford: Blackwell. Coseriu, Eugenio. 1974 Synchronie, Diachronie und Geschichte. Das Problem des Sprachwandels. München: Fink. Croft, William. 2000 Explaining Language Change: an Evolutionary Approach. Harlow: Longman. Denison, David. 1993 English Historical Syntax: Verbal Constructions. London: Longman. Eckardt, Regine. 2006 Meaning Change in Grammaticalization: an Enquiry into Semantic Reanalysis. Oxford: Oxford University Press. Gelderen, Elly van. 2004a Grammaticalization as Economy. (Linguistik Aktuell/Linguistics Today 71.) Amsterdam, Philadelphia: John Benjamins. Gelderen, Elly van. 2004b Economy, innovation, and prescriptivism: from spec to head and head to head. The Journal of Comparative Germanic Linguistics 7(1): 59–98.

Periphrasis as a precursor of analytic inflection

243

Gelderen, Elly van. 2005 Principles and parameters in change. In: Leonie Cornips and Karen Corrigan (eds.), Syntax and Variation: Reconciling the Biological and Social, 179–198. Amsterdam, Philadelphia: John Benjamins. Gelderen, Elly van. 2009 Grammaticalization from a Biolinguistic Perspective. In: Rudie Botha and Chris Knight (eds.), The Prehistory of Language, Volume I, 225–243. Oxford: Oxford University Press. Gelderen, Elly van. 2010 Features in reanalysis and grammaticalization. In: Elizabeth Traugott and Graeme Trousdale (eds), Gradience, Gradualness, and Grammaticalization, 129–147. Amsterdam, Philadelphia: John Benjamins. Gelderen, Elly van. 2011 The Linguistic Cycle. Language Change and the Language Faculty. Oxford: Oxford University Press. Grandgent, Charles H. [1934]1962 An Introduction to Vulgar Latin. New York: Hafner. Grewendorf, Günther. 1989 Ergativity in German. Dordrecht: Foris. Grewendorf, Günther. 1995 Präsens und Perfekt im Deutschen. Zeitschrift für Sprachwissenschaft 14(1): 72–90. Grewendorf, Günther. 1999 Das funktionalistische Paradox. Zum Problem funktionaler Erklärungen in der Linguistik. In: Herbert E. Wiegand (ed.), Sprache und Sprachen in den Wissenschaften. Geschichte und Gegenwart, 313–336. Berlin, New York: De Gruyter. Grønvik, Ottar. 1986 Über den Ursprung der aktiven Perfekt und Plusquamperfektkonstruktion im Deutschen. Oslo: Solum. Haider, Hubert and Rositta Rindler-Schjerve. 1987 The parameter of auxiliary selection. Italian-German contrasts. Linguistics 25: 1029–1055. Haspelmath, Martin. 1998 Does grammaticalization need reanalysis? Studies in Language 22: 315–352. Hopper, Paul J. and Elizabeth C. Traugott. 22003 [1993] Grammaticalization. Cambridge: Cambridge University Press. Keller, Rudi. 1990, 32003 Sprachwandel: von der unsichtbaren Hand in der Sprache. Tübingen: Francke. Lehmann, Christian. 1995 [1982] Thoughts on Grammaticalization. München: Lincom Europa. Leiss, Elisabeth. 2000 Artikel und Aspekt. Die grammatischen Muster von Definitheit. Berlin, New York: De Gruyter. Lightfoot, David. 1979 Principles of Diachronic Syntax. Cambridge: Cambridge University Press. Lightfoot, David. 1991 How to Set Parameters. Arguments from Language Change. Cambridge, Mass.: MIT Press. Lightfoot, David. 1999 The Development of Language: Acquisition, Change, and Evolution. Malden, Mass.: Blackwell. Martinet, André. [1955] 1981 Sprachökonomie und Lautwandel. Eine Abhandlung über die diachronische Phonologie. Translated by Claudia Fuchs. Stuttgart: Klett-Cotta. Musan, Renate. 2002 The German Perfect: Its Semantic Composition and Its Interactions with Temporal Adverbials. Dordrecht: Kluwer. Nickel, Gerhard. 1966 Die Expanded Form im Altenglischen: Vorkommen, Funktion und Herkunft der Umschreibung beon/wesan + Partizip Präsens. Neumünster: Wachholtz. Nübling, Damaris, Antje Dammel, Janet Duke and Renata Szczepaniak. 2006, 3 2010 Historische Sprachwissenschaft des Deutschen. Eine Einführung in die Prinzipien des Sprachwandels. Tübingen: Narr Franke Attempo Verlag.

244

Peter Öhl

Öhl, Peter. 2006 Über Sinn und Nutzen einer Generativen Grammatiktheorie. In: Ružena Kozmová (ed.), Sprache und Sprachen im Mitteleuropäischen Raum. Vorträge der Internationalen Linguistik-Tage Trnava 2005, 229–243. Trnava: Univerzita sv. Cyrila a Metoda. Öhl, Peter. 2007 Rezension: Lyle Campbell, Historical Linguistics – an Introduction. Linguistische Berichte 209: 99–103. Öhl, Peter. 2008 Rezension: Damaris Nübling et al., Historische Sprachwissenschaft des Deutschen. Eine Einführung in die Prinzipien des Sprachwandels. Sprache & Sprachen 37: 53–57. Öhl, Peter. 2009a Die Entstehung des periphrastischen Perfekts mit haben und sein im Deutschen – eine längst beantwortete Frage? Formale und funktionale Erklärungsansätze für die Auxiliarisierung. Zeitschrift für Sprachwissenschaft 28: 265– 306. Öhl, Peter. 2009b Sprachwandel und kognitive Ökonomie: Zur Grammatikalisierung und Substitution von Satzkonnektoren. Linguistische Berichte 220: 393–438. Öhl, Peter. 2014 Predicate classes: a study in compositional semantics. In: Klaus Robering (ed.), Events, Arguments, and Aspects. Topics in the Semantics of Verbs, 235–264. (Studies in Language Companion Series 152.) Amsterdam, Philadelphia: John Benjamins. Oubouzar, Erika. 1975 Über die Ausbildung der zusammengesetzten Verbformen im deutschen Verbalsystem. Beiträge zur Geschichte der deutschen Sprache und Literatur 95: 5–96. Roberts, Ian. 1993 A formal account of grammaticalization in the history of Romance futures. Folia Linguistica Historica 13: 219–258. Roberts, Ian. 1997 Directionality and word order change in the history of English. In: Ans van Kemenade and Nigel Vincent (eds), Parameters of Morphosyntactic Change, 397– 426. Cambridge: Cambridge University Press. Roberts, Ian. 1999 Verb movement and markedness. In: Michel DeGraff (ed.), Language Creation and Language Change, 287–328. Cambridge, Mass.: MIT Press. Roberts, Ian. 2007 Diachronic Syntax. Oxford: Oxford University Press. Roberts, Ian and Anna Roussou. 2003 Syntactic Change. A Minimalist Approach to Grammaticalization. Cambridge: Cambridge University Press. Rolf, Eckard. 1995 Zur Grammatikalisierung konversationeller Implikaturen. In: Frank Liedtke (ed.), Implikaturen: Grammatische und pragmatische Analysen, 87–102. Tübingen: Niemeyer. Salvi, Giampaolo. 1987 Syntactic restructuring in the evolution of Romance auxiliaries. In: Martin Harris and Paolo Ramat (eds.), Historical Development of Auxiliaries, 225–236 Berlin, New York: De Gruyter. Schlegel, August Wilhelm von. 1818 Observations sur la langue et la littérature provençales. Paris: Librairie grecque-latine-allemande. Stotz, Peter. 1998 Handbuch zur lateinischen Sprache des Mittelalters, Volume 4: Formenlehre, Syntax und Stilistik. München: Beck. Thielmann, Philipp. 1885 Habere mit dem Partizip Perfekt Passiv. Archiv für lateinische Lexikographie und Grammatik II: 372–432, 509–549. Wegener, Heide. 1991 Der Dativ – ein struktureller Kasus? In: Gisbert Fanselow and Sascha W. Felix (eds.), Strukturen und Merkmale syntaktischer Kategorien, 70–103. (Studien zur deutschen Grammatik 39.) Tübingen: Narr. Wekker, Herman Christiaan. 1976 The Expression of Future Time in Contemporary British English: an Investigation into the Syntax and Semantics of Five Verbal Constructions Expressing Futurity. Amsterdam: North-Holland.

Periphrasis as a precursor of analytic inflection

245

Primary sources (listed according to the abbreviations used above) Beda

=

Exh

=

Hel

=

HomS

=

Is

=

Liv

=

Luth

=

Musp

=

Notk

=

Otfr

=

Tat

=

Vulg

=

Wulf

=

Miller, Thomas (ed.). 1890; repr. 1959 The Old English Version of Bede’s Ecclesiastical History of the English People / Alfred the Great. Edited with a translation and introduction by Thomas Miller. London: Trübner. Exhortatio ad plebem christianam. In: Elias von Steinmeyer. 1963 Die kleineren althochdeutschen Sprachdenkmäler, 49–54. Berlin; Zürich: Weidmann. Behagel, Otto. 91984 Heliand und Genesis. Edited by Burkhart Taeger. Tübingen: Niemeyer. Thorpe, Benjamin. 1844, 1846 Ælfrices Bocgild. The Homilies of the Anglo-Saxon Church. The First Part, Containing The Sermones Catholici, or Homilies of Ælfric. In the original Anglo-Saxon, with an English version. 2 vols. London: Richard & John E. Taylor. Eggers, Hans. 1964 Der althochdeutsche Isidor. Nach der Pariser Handschrift und den Monseer Fragmenten neu herausgegeben. Tübingen: Niemeyer. Feix, Josef (ed.). 31991 Ab urbe condita. Titus Livius, Römische Geschichte (lateinisch-deutsch). München: Artemis-Verlag. Volz, Hans (ed.). 21973 Die Ganze Heilige Schrifft Deudsch/ Martin Luther. Darmstadt: Wissenschaftliche Buchgesellschaft. Muspilli. In Wilhelm Braune and Ernst A. Ebbinhaus (eds.). 17 1994 Althochdeutsches Lesebuch, 86–89. Tübingen: Niemeyer. Tax, Petrus W. (ed.). 21986–90 Die Werke Notkers des Deutschen. Tübingen: Niemeyer. Piper, Paul (ed. ). 1878 Otfrieds Evangelienbuch. 1. Theil: Einleitung und Text. Paderborn: Schöningh. Sievers, Eduard (ed.). 21961 Tatian. Lateinisch und Altdeutsch mit ausführlichem Glossar. Darmstadt: Wissenschaftliche Buchgesellschaft. Colunga, Alberto and Laurentio Turrado (eds.). 1959 Biblia Vulgata. Salamanca. Hans C. von Gabelentz and Julius Loebe. 1843 Ulfilas. Veteris et novi testamentis versionis Gothicae cum glossario et grammatica linguae Gothicae. Leipzig: Brockhaus.

Section 3: Mechanisms of Gradual Change: BE > HAVE and HAVE > BE

Rolf Kailuweit, Albert-Ludwigs-Universität Freiburg

BE or HAVE in Contemporary Standard French – residua of semantic motivation 1 Introduction Split-intransitivity – i.e. the distinction between unaccusative and unergative verbs1 – has been an intensively discussed topic in general linguistics during the last decades. As far as Romance languages are concerned, the most obvious marking of split-intransitivity is the use of two different auxiliaries in compound tenses. In contemporary French for example, most intransitive verbs take HAVE in compound tenses, but some verbs take BE. (1) a. Le roi a rit the king have-prs.3sg laugh-pcpt ‘The king has laughed.’ b. Le train est arrivé the train be-prs.3sg arrive-pcpt ‘The train has arrived.’ Nevertheless, Romance intransitive verbs do not split naturally into two auxiliary selection classes. As already pointed out by Rosen (1984), there is considerable intralinguistic and interlinguistic variation. On the one hand, there are for instance several intransitive verbs that allow for BE and HAVE in French: (2) a. Ce texte est paru le 5 août this text be-prs.3sg appear-pcpt the 5 august ‘This text came out on August 5th.’ (FRANTEXT, JOCARD)

1 The term split-intransitivity was coined by Merlan (1985) and spread by Van Valin (1990). In comparison with the terms inaccusativity/inergativity introduced by Perlmutter (1978) in the Relational Grammar framework, it is more neutral from a typological point of view. Due to the fact that inaccusativity/inergativity are well established terms in the literature, I will continue to use them in this paper.

250

Rolf Kailuweit

b. Il y a quelques mois, un livre a paru Ago some months a book have-prs.3sg appear-pcpt ‘Some months ago, a book came out.’ (FRANTEXT, FEBVRE) On the other hand, several intransitive verbs allowing for BE in one language are exclusively used with HAVE in another. There are, for example, obvious mismatches between French and German: (3) a. Christ est ressuscité Christ be-prs.3sg revive-pcpt (FRANTEXT: DÉON) b. Christ a ressuscité Christ have-prs.3sg revive-pcpt (FRANTEXT: BONNEFOY) c. Christus ist / *hat auferstanden Christ be/have-prs.3sg revive-pcpt ‘Christ is resurrected.’ (4) a. Elle a couru tout droit à la police She have-prs.3sg run-pcpt directly to the police (FRANTEXT: GARY) b. Sie ist direkt zur Polizei gelaufen She be-prs.3sg directly to the police run-pcpt ‘She ran directly to the police.’ On the basis of this kind of data, Rosen (1984) denied that the unaccusativeunergative distinction could be motivated semantically. In addition, from a diachronic point of view many Romance languages and varieties show a tendency to substitute BE-selection with HAVE-selection: in Spanish, Portuguese and Catalan,2 BE-selection is no longer an option in contemporary use (Vincent 1982; Posner 1996). In French the number of BE-selecting verbs has diminished considerably since the 18th century (Fournier 2002). This seems to be clear counter-evidence to a semantic motivation of split-intransitivity. However, several authors have tried to reformulate semantic criteria for the overall distinction of two classes of intransitive verbs. Dowty (1991),

2 With the exception of a few northern dialects.

BE or HAVE in Contemporary Standard French – residua of semantic motivation

251

Schwarze (1996, 1998) and Aranovich (2003) pointed out that the subject of unaccusative verbs is never a prototypical agent. Centineo ([1986] 1996), Van Valin (1990) and Levin and Rappaport Hovav (1995) highlighted the differences at the level of Aktionsart (activities vs. other Aktionsart classes). As far as the variation of auxiliary choice in Romance languages is concerned, Labelle (1992) claimed that BE-selection is a sufficient but unnecessary condition for inaccusativity in French, while Centineo ([1986] 1996), Van Valin (1990) and Bentley (2006: 44) declared that HAVE-selection is only limited to the Aktionsart class of activities in Italian. In recent years, Sorace (Sorace 2000; Legendre and Sorace 2003)3 has worked out an integrative approach to auxiliary selection that claims to explain interlinguistic and intralinguistic variation as well as the direction of diachronic change. In the present paper I shall deal with the motivation of BE-selection in Contemporary Standard French. In section 2 an overview of the number of BEselecting verbs in Contemporary Standard French is given in comparison to Italian and Old Spanish. Section 3 will account for the fact that in earlier stages of French the BE-construction and the HAVE-construction had different functions: it will be shown that the BE-construction was used as a resultative, the HAVE-construction as an anterior. This distinction has been lost in Contemporary Standard French while the number of BE-selecting verbs has diminished considerably. Section 4 will revisit Sorace’s Auxiliary Selection Hierarchy (ASH). Following Kailuweit (2011), I shall prove that the different features that determine the affinity for the unaccusative or unergative pole have to be weighed to describe the seven semantic classes of the hierarchy and their order in a coherent way. Finally, in section 5 the spread of Contemporary Standard French’s BE-residua over the ASH-classes will be analysed. I shall prove that ASH is a useful tool to describe the distribution of BE-residua, although ASH’s semantic explanation does not apply for a certain number of outliers. It will be shown that change of location is a necessary but insufficient condition for BEselection in Contemporary Standard French since a change of location component can be detected in the semantics of all BE-selecting verbs.

2 BE in contemporary standard French The construction BE + participle has several functions in Contemporary Standard French. BE is the only auxiliary used in French passive constructions which

3 For over 10 years, Sorace has tested her approach in synchrony and diachrony, taking several languages and varieties into consideration. See Sorace (this volume) for further references.

252

Rolf Kailuweit

can be formed with transitive verbs only.4 All reflexive constructions take BE in the compound tenses. A small number of intransitive verbs constructs compound tenses with BE. Burzio claims that Italian passives and reflexives behave syntactically as BE-intransitives (unaccusatives): “All and only the verbs that can assign a θrole to the subject can assign (accusative) case to an object” (Burzio 1986: 178). Nonetheless, in French there are extraction dissimilarities between passives and reflexives (Abeillé and Godard 2002): (5) a. *C‘est parti soudain à Rome que Jean est It be-prs.3sg leave-pcpt suddenly to Rome that John be-prs.3sg b. C‘est It be-prs.3sg a have-prs.3sg

complètement détruite que sa maison completely destroy-pcpt that his house par les bombes by the bombs

Reflexives always take BE in contemporary standard French, independent of the semantic class of the predicate or its transitive or intransitive status. Hence there is no variation at all that can be explained on semantic grounds. Following Legendre and Sorace (2003), I shall only account for intransitives in this paper. The similarities and dissimilarities of different BE-constructions will not be dealt with. From a synchronic point of view, French and Italian are the only national Romance languages that use BE for the formation of compound tenses of a subgroup of intransitive verbs. BE-selecting intransitives are also found in Occitan, Corsican, Sardinian, Rhaeto-Romanic and in some Catalan and Romanian dialects. It has been suggested in the literature that the differences concerning the spread of BE-selecting verbs in contemporary standard French and Italian are not significant. Following Salvi (1988), Schwarze (1996) only counts 92 BE-selecting verbs in Italian, indicating that this number may be slightly greater if one also considers derivatives. However, it is easy to prove that the number of BE-selecting verbs in Italian is much higher. According to the CD-ROM edition of the Dizionario Italiano Sabatini Colletti (DISC), 1,175 out of 3,293 intransitive verbs allow for BE in compound tenses. Of course, this number has

4 Some verbs that take an indirect or oblique allow for passive constructions with an expletive il (‘it’) as a dummy subject. This construction is mainly used in formal speech: i. …ces noms de famille donnèrent naissance à une foule de dérivés dont il sera parlé ci-après (FRANTEXT) (‘these family names gave birth to masses of derivatives which will be talked about below’).

BE or HAVE in Contemporary Standard French – residua of semantic motivation

253

to be revisited. Several verbs are antiquated or restricted to dialectal or highly formal use. Nonetheless, Italian native speakers whom I confronted with this list marked more than 500 current Italian intransitives as BE-selecting (Kailuweit 2011). It is obvious that there are considerably fewer BE-selecting verbs in contemporary standard French. Arrivé (2006) lists about 100 verbs that allow for the selection of BE. The largest part is marked as also allowing for the selection of HAVE. Grevisse and Goosse (2008: 1032)5 refer to this list, indicating that many of these verbs are doubtful candidates for BE-selection, since their BEconstructions seem to consist of a copula and an adjectivised participle. In a rather laconic way they comment that these verbs take HAVE if they are “really” conjugated (Grevisse and Goosse 2008: 1033). Sankoff and Thibault (1977: 85–94) proposed a number of criteria to distinguish between Canadian French BE + participle as an auxiliary for compound tenses – anteriors in Bybee, Perkins, and Pagliuca’s (1994) terminology – and as an adjectival construction with a copula. Adverbials referring to the manner in which an event is realised are evidence against the copula + adjectivised participle interpretation. In addition, adverbials expressing the motivation of a subject referent for an action favour an eventive interpretation. Other criteria refer to the fact that the copula construction necessarily denotes a resultant state that persists at reference time. If this is clearly not the appropriate interpretation, BE functions as an anterior: (6) J’avais ça même quand on est arrivé I-have-pst.ipfv.1sg that same when one be-prs.3sg arrive-ptcp.m.sg ici à Pointe Saint-Charles here at Pointe Saint-Charles ‘The same thing happened to me when we arrived here at Pointe SaintCharles.’ (Sankoff and Thibault 1977: 88) Quand must be translated as ‘in the instant in which’. This is due to the difference in the temporal and aspectual morphology in the subordinate clause, and the main verb. The situation expressed by the BE + participle construction is conceptualised as bounded. A resultative interpretation would have required

5 In 1936, Maurice Grevisse published the first edition of the French reference grammar Le bon usage. After his death in 1980, his colleague André Goose updated the work and published the 12 th and 13 th editions. The 14 th edition (2007) was completely revised. I shall refer to the 14 th reprinted edition from 2008.

254

Rolf Kailuweit

an imperfective BE (était arrivé) (also see the discussion of this example in Rosemeyer 2014). Note that there is a metonymic link between anteriors and resultatives (Rosemeyer 2012, 2014): a resultant state has been brought about by a previous event and a previous event may lead to a resultant state that persists at reference time. While the HAVE-construction in Contemporary Standard French always focuses on the previous event, the BE-construction is ambiguous. The degree of adjectivisation of the participle on the one hand and on the other the possibility of using the BE-construction when a resultative reading is ruled out by the context has to be determined for each verb. In Kailuweit (2011), I discussed the lists of Arrivé (2006) and Grevisse and Goosse (2008) in detail. The results are the following: the selection of BE is obligatory for about 20 verbs in contemporary standard French. About 20 verbs allow for variation between the BE-construction and the HAVE-construction in an anterior reading. The BE-construction of about 60 verbs marked as varying in Arrivé (2006) seems to be a copula + adjectivised participle construction. As far as previous stages of Spanish are concerned, Benzing (1931) compiled a list of 38 verbs that allowed for the BE-construction in Old Spanish. For most verbs there was variation until the 16th or 17th century. After that period, only HAVE is used as an auxiliary in compound tenses. Later studies have added only four verbs to this list (Elvira González 2001). In his PhD dissertation, Rosemeyer (2014) shows that Old Spanish BE-constructions should be interpreted as resultatives. However, due to the metonymic relation of anteriors and resultatives, the BE-construction underwent a functional change and was reanalysed as an anterior in Early Modern Spanish. The BE-anterior was completely substituted by the HAVE-construction during the 17th century and also disappeared from the resultative construction, in which a different copula (estar) prevailed. In contemporary Spanish, BE-constructions are restricted to passives of transitive verbs. Therefore, Early Modern Spanish seems to show a quite similar picture to Contemporary Standard French as far as the distribution of HAVE and BE as auxiliaries in compound tenses are concerned. In the next section, I shall raise the question of whether there is evidence for a process of reanalysis in earlier stages of French that parallels Rosemeyer’s findings for Spanish.

BE or HAVE in Contemporary Standard French – residua of semantic motivation

255

3 Functional split in earlier stages of French For a subgroup of the BE-selecting verbs, Grevisse and Goosse (2008: 1033) indicate a rule that may explain the variation of BE and HAVE-constructions: “Certains verbes intransitifs […] se conjuguent avec avoir quand ils expriment l’action – et avec être quand ils expriment l’état résultant de l’action […] Dans beaucoup de cas, cette règle est plus théorique que pratique. [‘Certain intransitive verbs […] are conjugated with avoir (‘to have’) when they express action – and with être (‘to be’) when they express the state resulting from the action […] In many cases, this rule is more theoretical than practical.’] (Grevisse and Goosse 2008: 1033, translation, RK). When speaking of individual verbs, they generally highlight that the use of one auxiliary is antiquated, dialectal or vulgar. The reference dictionary of Contemporary Standard French, Le Grand Robert de la langue française (2007), also comments on the following verbs as candidates for a resultative-anterior opposition: (7) accoucher (‘give birth’), accourir (‘run together’), descendre (‘descend, decrease’), éclore (‘emerge’), monter (‘ascend, increase’), partir (‘leave’), passer (‘pass’), rester (‘stay’), ressusciter (‘revive’), résulter (‘result’), sortir (‘go out’), tomber (‘fall’) and the derivatives with re- (‘again’) (8) redescendre (‘descend again, decrease again’), remonter (‘ascend again, increase again’), repartir (‘leave again’, ‘go back’) repasser (‘pass again’), ressortir (‘go out again’), retomber (‘fall back’, ‘fall down again’) Grevisse and Goosse (2008: 1032) illustrate their “theoretical rule” with the following example in which the adverbial indicates a bound situation: (9) Il a reparti ce matin à six heures He have-prs.3sg go back-ptcp.m.sg this morning at six o‘clock ‘He has gone back this morning at six o‘clock.’ Le Grand Robert (2007) states apodictically that repartir take BE in the sense of ‘to go back’. The use of HAVE with repartir in the sense of ‘to reply’ is marked as antiquated and formal. In the FRANTEXT corpus an almost identical example is found with partir (‘leave’) that proves clearly that this verb can be use as an anterior in a bound situation:

256

Rolf Kailuweit

(10) Lui et Bucky étaient partis le matin six heures He and Bucky be-ipfv.3sg leave-ptcp.m.pl the morning six o‘clock ‘He and Bucky had left this morning at six o’clock.’ (FRANTEXT: CAMUS) Le Grand Robert (2007) considers the use of partir (‘leave’) with HAVE in contemporary French vulgar. According to Gevisse and Goosse (2008: 1033) and Le Grand Robert (2007), rester (‘stay’) is another example that illustrates a rule that was in force until the 19th century, but is obsolete today. Voltaire still used HAVE with rester to denote a previous state that did not persist at reference time: (11) J’ai resté huit jours à la maison I-have-prs.1sg stay-ptcp.m.sg eight days at the house ‘I stayed eight days at home.’ (ROBERT: VOLTAIRE) For most of the varying verbs listed in (7) and (8) the use of HAVE is banished from the contemporary standard. However, disparaître (‘disappear’) and accoucher (‘give birth’) generally take HAVE in all contexts, while ressusciter (‘revive’) and résulter (‘result’) show a variation of the two auxiliaries that seems to be free and does not confirm the anterior-resultative opposition (Kailuweit 2011). To sum up, the anterior-resultative opposition can be considered obsolete in Contemporary Standard French.6 For a small group of verbs, BE has become the only acceptable auxiliary in every context, while other verbs show variation that is no longer driven by the anterior-resultative opposition. From a diachronic point of view, it is obvious that several verbs underwent a change as far as the use of the BE-construction is concerned. However, the development of auxiliary use in French has not yet been studied in a systematic way. The most detailed reliable studies date back to late 19th and early 20th century.7 Foerster (1908), when working with the findings of Hofmann (1890), declared: “Im Altfrz. war das vorherrschende Hilfsverb bei intransitiven Verben être, heute ist es sein Rivale avoir” [In Old French the predominant auxiliary for intransitive verbs was be, nowadays it is its rival have] (Foerster 1908: 103, translation, RK).

6 However, the opposition seems to be still in force in substandard varieties (Blanche-Benveniste 1997) 7 Later studies, such as Gamillscheg (1957) or Togeby (1974) do not go into details.

BE or HAVE in Contemporary Standard French – residua of semantic motivation

257

A look at his data shows that the hypothesis is true for verbs of change of state and change of location. Purely stative verbs vary: durer (‘last’), reposer (‘repose’), séjourner (‘linger’) and vivre (‘live’) predominantly take HAVE, but sometimes BE, while gésir (‘lie’) only appears with HAVE in Foerster’s corpus. Verbs that do not have a state predicate in their semantic structure, such as pleurer (‘cry’), songer (‘dream’) or pécher (‘sin’), always take HAVE. As far as Classical French is concerned, Oudin’s grammar (1640: 214–232) lists 95 verbs with BE and 15 showing BE-HAVE variation. Some of them are change of location verbs, but most are anticausative variants of change of state verbs. Fournier (2002: 255–262) gives evidence for 33 varying change of state and change of location verbs. In addition, she gives an example of how the manner-of-motion verb courir (‘run’) takes BE in a telic construction. By adding an adverbial that denotes a goal to an atelic manner-of-motion verb, the construction becomes telic.8 (12) Monsieur le Chevalier était de retour. Je suis Monsieur le Chavalier be-ipfv.3sg back. I be-prs.1sg courrue ici run-ptcp.f.sg here ‘Monsieur le Chavalier was back. I have run over here.’ (SÉVIGNÉ, Fournier 2002: 256) Van Valin (1990) refers to Italian correre to illustrate that BE-selection in Italian requires a state predicate in the semantic structure of the construction. The diachronic data confirm that this rule also holds for French. Rosemeyer (2012: 139) points out with regard to Old Spanish, that “resultatives can only appear with predicates expressing complex situations that involve a transition to a resultant state”. This is obviously the case for change of location and change of state verbs as well as for manner-of-motion verbs allowing for template augmentation. Nonetheless, there are also purely stative verbs that appear in BE-constructions. Rosemeyer (2014) shows that stative verbs in Old Spanish receive a change of state interpretation when they appear in a BEconstruction. As far as Contemporary Standard French is concerned, the most obvious example of a BE-selecting verb that seems to have no change component in its meaning is rester (‘stay’). However, verbs denoting the continuation of a pre-existing state show a complex event structure. According to the formal-

8 This is an instance of template augmentation in the sense of Levin and Rappaport Hovav (2005).

258

Rolf Kailuweit

ism of Role and Reference Grammar, the semantic structure of rester (‘stay’) is quite similar to the semantic structure of partir (‘leave’). (13) a. partir (‘leave’): BECOME (NOT be.at’[x, y]) b. rester (‘stay’): NOT BECOME (NOT be.at’[x, y]) Verbs of continuation of a pre-existing state denote that at a point of reference a possible change of state or location has not occurred. According to Sokol (1999), they are bounded to the left and open to the right and can be described as telic in a border sense. Hence, rester (‘stay’) is not a counterexample to the hypothesis that the BE-construction had a resultative meaning in earlier stages of French. As we have seen, rester (‘stay’) as well as several verbs of change of location have generalized the use of BE in contexts in which no resultative interpretation is available. Other verbs, especially verbs of change of state, have generalised HAVE. A small group allows for variation that is no longer driven by the resultative-anterior opposition. Since the number of BE-selecting verbs has considerably decreased in Contemporary Standard French and the functional opposition between the BE-construction and HAVE-construction has disappeared, I consider the remaining BE-selecting verbs as BE-residua. According to Sankoff and Thibault (1977), the existence of BE-residua is only due to the prescriptive norm of Contemporary Standard French that tries to block the typologically-driven spread of HAVE as the only auxiliary in compound tenses: Nous croyons que l’utilisation de l’auxiliaire avoir […] correspond à une tendance vers la régularisation des conjugaisons. La petite classe de verbes pour lesquels les grammaires prescrivent l’usage exclusif de être ne nous semble ni assez homogène, ni suffisamment distincte des autres verbes pour expliquer une résistance à cette régularisation. [We believe that the use of the auxiliary avoir (‘have’) […] corresponds to a tendency toward the regularization of conjugations. The small class of verbs for which the grammarians prescribe the exclusive use of être (‘be’) seems to us neither sufficiently homogeneous nor distinct from other verbs to explain a resistance to this regularization.] (Sankoff and Thibault 1977: 106, translation, RK).

However, the question arises whether BE-residua form an arbitrary group or whether there is a systematic explanation of BE-selection in contemporary standard French that is not based on the resultative-anterior opposition. Legendre and Sorace (2003) apply the Auxiliary Selection Hierarchy (ASH) to contemporary standard French, although they only examine a subgroup of documented BE-residua (Kailuweit 2011). The next section will revisit the ASH approach from a theoretical point of view. In the last section of this paper I shall

BE or HAVE in Contemporary Standard French – residua of semantic motivation

259

discuss to which extent a revisited ASH is helpful to explain the distribution of BE-residua in Contemporary Standard French.

4 Auxiliary Selection Hierarchy revisited According to Legendre, Miyata, and Smolensky (1991), split intransitivity is both syntactically encoded and semantically determined. As far as auxiliary selection at the syntactical level is concerned, there is no gradience. A verb appears either in a HAVE-construction or in a BE-construction. Gradient could be the statistical distribution of the two-constructions of varying verbs or at a semantic level the distance of a specific verb or verb class to an unaccusative or unergative prototype. Legendre and Sorace (2003)9 use the Auxiliary Selection Hierarchy (Sorace 2000) to account for the BE-selection asymmetries between Italian and French. The hierarchy distinguishes seven classes of monadic (one-place) intransitive verbs that show decreasing affinity for BE-selection and increasing affinity for HAVE-selection. The class of change of location is located at the top of the hierarchy showing the highest affinity for BE-selection. Change of location is followed by change of state. In the middle of the hierarchy we find verbs of continuation of a pre-existing state and of existence of state. The classes of uncontrolled processes and of motional controlled processes are located in the lower part of the hierarchy. At the bottom, the class of non-motional controlled processes shows the lowest affinity for BE-selection and the highest to HAVEselection. The hierarchy allows us to determine the cut-off point for BE-selection in a specific language. If BE-selection is excluded for a specific class, classes located lower in the hierarchy will not allow for BE-selection. Table 1 illustrates the different cut-off points for French and Italian. According to Legendre and Sorace (2003), the cut-off point for French is between the classes of change of state and continuation of a pre-existing state; for Italian it is between the classes of existence of state and uncontrolled processes. Note that the notation “A*” which appears in the Italian classes of uncontrolled processes and motional controlled processes indicates that there is still variation (occasional BEselection) in theses classes. As far as French is concerned, Legendre and Sorace (2003) admit that two verbs of continuation of a pre-existing state – rester (‘stay’) and demeurer (‘stay’) – select BE. Nonetheless, they do not indicate the existence of these

9 See Legendre (2007) for a slightly modified approach.

260

Rolf Kailuweit

Tab. 1: Auxiliary selection in French and Italian (adapted from Legendre and Sorace 2003: 227). Auxiliary selected

Verb classes

French

Italian

E

E

E E*

E E

E* A

E E*

Change of state a. Change of condition: morire/mourir, etc. b. Appearance: apparire/apparaître, etc. c. Indefinite change in a particular direction: salire/monter, scendere/descendre appassire/faner, peggiorare/empirer, etc.

A

E*

Continuation of pre-existing state: durare/durer, etc.

A A

E E*

Existence of state: a. essere/être b. esistere/exister, bastare/suffire à

A A A

A* A A*

Uncontrolled processes a. Emission: risuonare/résonner, etc. b. Bodily functions: sudare/suer, etc. c. Involuntary actions: tremare, trembler, etc.

A

A*

Motional controlled processes: nuotare/nager, etc.

A

A

Non-motional controlled processes: lavorare/travailler, etc.

Change of location: arrivare/arriver, venire/venire, etc.

verbs in Table 1. Hence, the cut-off point cannot be considered a strict logical concept. Rather, it refers to a subjective evaluation of the data that depends on two aspects that are not made explicit. On the one hand, a class is declared E or A because of the statistical relation of BE-selecting and HAVE-selecting verbs. On the other hand, the statistical relation of BE-constructions and HAVEconstructions of the varying verbs in a specific class seems to be a decisive criterion. For an E*, varying verbs should prefer BE over HAVE, for an A* it should be the other way round. Another problem of the ASH consists in an independent semantic motivation of the classes and their order. In the original form (Sorace 2000), the hierarchy seems to be the result of a semasiological approach that starts from the meaning of prototypically BE-selecting and HAVE-selecting verbs. Legendre and Sorace (2003) present a semantic decomposition of the seven classes based on features that have been discussed as being relevant for split-intransitivity: +/–telic [TE], +/–motion [MO], +/–directed change [DIR], +/–protagonist control (or agentivity) [CON], +/–state [ST]. However, the five binary features are

261

BE or HAVE in Contemporary Standard French – residua of semantic motivation

Tab. 2: Featural composition of monadic intransitive verbs in French and Italian (adapted from Legendre and Sorace 2003: 227). Aux Aux Semantic/aspectual features → Fr

Ital

emergent verb classes ↓

E E E

E E E

Change of location: arriver/arrivare aller/andare venir/venire

TE

MO

DIR

CON

ST

+ + +

+ + +

+ + +

+/− +/− +/−

− − −

+

−

+

−

−

+

−

+

−

−

− −

+/− + − +

+/− −

− −

E

E

E

E

E A

E E

Change of state a) change of condition mourir/morire b) appearance: apparaître/apparire c) indefinite change in a particular direction: monter/salire , descendre/scendere faner/appassire, empirer/peggiorare

A

E

Continuation of a pre-existing state: durer/durare

−

−

−

−

+

A A

E E

Existence of state: être/essere exister/esistere, suffire/bastare

− −

− −

− −

− −

+ +

A A A

A A A

Uncontrolled processes: a) bodily functions: suer/sudare b) involuntary actions: trembler/tremare c) emission: résonner/risuonare

− − −

− − −

− − −

− − −

− − −

A

A

Controlled processes (motional): nager/nuotare

−

+

−

+

−

A

A

Controlled processes (non-motional): travailler/ lavorare

−

−

−

+

−

not sufficient to describe all classes and subclasses and their order in a coherent way. Table 2 shows Legendre and Sorace’s feature-value pairings for the classes and subclasses in Table 1. Verbs of continuation of a pre-existing state and existence of state cannot be distinguished, since the same values are assigned to the five features in both classes. The subclass faner/appassire (‘wilt’) and empirer/peggiorare (‘worsen’) as a non-controlled variant of indefinite changes in a particular direction show only one +/–value [DIR]. Since the classes of continuation of a pre-existing state and of existence of state receive also one +/–value for the feature [ST], faner/appassire (‘wilt’) and empirer/peggiorare

262

Rolf Kailuweit

Tab. 3: Revisited ASH (adapted from Kailuweit 2011: 417). TE = telic

MO = Movement

DIR = Directed change

CON = ST = control static

1

1

4

−2

1

+

+

+

+/−

−

4–6

+ +

− −

+ +

− −

− −

5 5

−

+/−

+

+/−

−

2–5

Continuation of a pre-existing state:

+

−

−

−

+

2

Existence of state

−

−

−

−

+

1

Uncontrolled processes a) bodily functions b) involuntary actions c) emission

− − −

− − −

− − −

− − −

− − −

0 0 0

Controlled processes (motional)

−

+

−

+

−

−1

Controlled processes (non-motional) −

−

−

+

−

−2

Change of location Change of state a) change of condition b) appearance c) indefinite change in a particular direction:

(‘worsen’) show the same degree of semantic unaccusativity as the stative classes. In addition, since motion is a feature of unaccusativity and control a feature of unergativity, the two +/–values for controlled motional processes cancel each other out. Therefore, the controlled motional processes do not differ in semantic unergativity from uncontrolled processes. Nonetheless, as for proto-agent and proto-patient features in Dowty (1991), it is quite obvious that the five features do not have the same significance. Telicity has been considered the central feature for unaccusativity (Levin and Rappaport Hovav 1995), control or agentivity for unergativity (Schwarze 1998). However, it would be too simple to reduce ASH to a continuum between a telic and an agentive pole. Giancarli (this volume) rightly states that these features do not apply in the same way to all classes and subclasses of the hierarchy. In Kailuweit (2011), I proposed a modified version of the ASH. The basic idea consists in weighing the features. I consider directed change a very strong feature of semantic unaccusativity and control a strong feature of semantic unergativity. Telicity, movement and stativity are weak features of unaccusativity. If the very strong unaccusativity feature is given for class or subclass, the

BE or HAVE in Contemporary Standard French – residua of semantic motivation

263

feature-value-pairing gets 4 points. If the strong unergativity-feature is given the paring gets -2 points. For a given weak unaccusativity-feature a pairing gets 1 point. The sum of the points for the five feature-value-pairings indicates the degree of semantic unaccusativity of a verbclass. Table 3 shows that the order of the seven classes can be justified by this approach. Note that the higher degree of unaccusativity of continuation of a pre-existing state in comparison with existence of state stems from the fact that the continuation class is telic in a broader sense (Sokol 1999). The first two classes – change of location and change of state – overlap as far as the degrees of unaccusativity are concerned. If verbs of change of location are controlled, e.g. sortir (‘go out’) with a human subject, the degree of unaccusativity is 4; if they are uncontrolled, e.g. arriver (‘arrive’), they reach a 6, which is the maximum degree of unaccusativity. The unaccusativity degree of change of state verbs is 5 with the exception of the subclass of indefinite change in a particular direction. (14) Je suis peut-être descendu encore vachement plus bas I be-prs.1sg perhaps descend-pcpt still bloody more deep ‘Perhaps, I still have gone bloody deeper down.’ (FRANTEXT: DEGAUDENZI) In (14), the subject-argument controls her or his own directed movement as an unbounded process. Hence the feature-value cluster is [TE–], [MO+], [DIR+], [CON+], [ST–] corresponding to a degree of unaccusativity of 3. For monter and descendre [CON+] implies [MO+]. With the reading ‘increase’ and ‘decrease’ the two verbs exclude a controlling subject. Therefore, 3 is the minimum degree of unaccusativity of these verbs. Even this subclass of change of state verbs thus differs in unaccusativity from the next lower class of continuation of a preexisting state. Telic verbs of directed change reach a considerably higher degree of unaccusativity than stative verbs in the middle of the hierarchy. This is in line with Mateu’s (2009) model which allows for five different classes on the basis of three features [T], [r] and [R]. Roughly speaking, [T] corresponds to the existence of a transition, the “coincidence relation” [r] marks the completion of an action (comparable to the notion of telicity), and [R] marks agentivity of the verb’s subject referent (Mateu 2009: 189–190). With the help of these features, Mateu (2009) distinguishes telic and atelic change of location or state ([T+, r±], continuation of a pre-existing state/existence of state [T–, r–], nonvolitional verbs of internal causation [R–] and volitional verbs of internal causation [R+]. If motion is not a decisive factor to distinguish the degree of unaccusativity of the first two classes of the ASH, why is it the case that most verbs that

264

Rolf Kailuweit

only allow for the BE-construction are verbs of change of location in Standard Contemporary French? We will come back to this question in the last section of this paper. Note that motion is used in Legrendre and Sorace (2003) to distinguish two kinds of controlled processes. While non-motional processes do not allow for BE-constructions in Romance languages which feature BE-HAVE-alternation,10 motional processes can be constructed with BE under certain conditions. We have already seen that like Italian correre (‘run’), courrir (‘run’) forms compound tenses with BE in older stages of French if the construction is telic. Hence, it seems to be telicity and not motion that licences the use of BE. Nonetheless, motion is a prerequisite for this type of construction and can therefore be considered a factor that decreases unergativity and enables BEselection. However, the class of controlled motional processes does not fit perfectly into a hierarchy of decreasing semantic unaccusativity. In Italian, several verbs of uncontrolled processes show free variation (Sorace 2000: 877–878). Generally, for these verbs HAVE-constructions are more frequent than BE-constructions. In the CODIS corpus for instance, only 5 of 17 occurrences of squillare (‘ring’) in compound tenses take BE. The variation does not seem to be semantically motivated: (15) a. alle sette e un quarto, è squillato at seven and a quarter be-prs.3sg ring-pcpt il telefono the telephone ‘At a quarter past seven, the telephone rang.’ b. In quel momento aveva squillato il telefono in that moment have-ipfv.3sg ring-pcpt the telephone ‘In that moment the telephone rang.’

(CODIS)

(CODIS)

In contrast, variation of Italian correre is semantically motivated and categorical. However, nuotare (‘swim’) does not allow for BE even in telic constructions (Sorace 2000: 875). German verbs of controlled motion categorically take BE (Keller and Sorace 2003). Hence, at least for German the order of the classes has to be modified (see Gillmann this volume). From a diachronic point of view, the ASH predicts that the spread of HAVE occurs from the bottom to the top. The loss of BE-selection in Spanish as well as the development towards Contemporary Standard French seem to confirm

10 According to Tuttle (1986), BE is the only auxiliary of several central Italian dialects.

BE or HAVE in Contemporary Standard French – residua of semantic motivation

265

the hypothesis that the higher a class in the hierarchy the longer it withstands the substitution process. Aranovich (2003) shows on the basis of Benzing’s (1931) data that the two highest classes – “change of location” and “change of state” – still allowed for BE-selection in the 17th century. Rosemeyer (2014) proves in a conscientious corpus study that from 1425, the chance of survival of BE-selection is higher for change of location verbs than for change of state verbs. Interestingly, this was not the case before 1425. In Old Spanish the classes of change of location and change of state show an identical affinity for BEselection. Motion is not a significant factor in split-intransitivity until Early Modern Spanish. Then the situation started to change. According to Rosemeyer (2014), the conservation of BE-selection with verbs of change of location is a frequency effect. In the first half of the 17th century, ir (‘go’) and venir (‘come’) are the most frequent verbs in his corpus, volver (‘return’) appears in position 8 and partir (‘leave’) in position 12. Since high frequency leads to entrenchment, highly frequent verbs resist the ongoing morphological change longer. In addition, the data suggest putting forward the hypothesis that speakers started to associate the BE-construction with a change of location, thus applying a new rule to BE-selection: when confronted with the choice between BE + past participle and HAVE + past participle, use BE + past participle if you wish to express a past change of location event (Rosemeyer 2014). Note that this rule also applies to less frequent verbs of change of location. Hence, the interplay of semantic parameters of unaccusativity is not universal. Their presence and weight depend on language-specific developments, as Rosemeyer (2014) shows for Spanish and Gillmann (this volume) for Dutch and German. Against the background of these insights, the last section of this paper will cast some light on the distribution of BE-selecting verbs in Contemporary Standard French. Is motion a decisive factor in the conservation of BE-selection in French and does frequency plays a role in establishing an analogous rule to the rule Rosemeyer proposes for Early Modern Spanish? We will see that French data confirm Rosemeyer’s findings for Spanish. In addition, the existence of several BE-selecting outliers which have not been accounted for in the literature can be explained by taking into consideration the change of location rule.

266

Rolf Kailuweit

5 BE-residua in Contemporary Standard French Based on the FRANTEXT corpus (1951–2000) and standard-setting grammars and dictionaries (Arrivé 2006; Grevisse and Goosse 2008; Le Grand Robert 2007), Kailuweit (2011) compiled an exhaustive list of French BE-residua (see Table 4). As in Table 1, A and E indicate the use of HAVE or BE as an auxiliary in compound tenses. If a small “a” in combination with a capital E is used, this

Tab. 4: French BE-residua in ASH (adapted from Kailuweit 2011: 410). change of location

change of state

Change of condition

accourir ‘run together’ (E/a), aller ‘go’ (E), arriver ‘arrive’ (E), convenir de ‘agree to’ (E/A), (dis)convenir de ‘deny’ (E/A), échapper qc à q ‘slip out’ (E/A), échoir à ‘hatch’ (E/a), entrer ‘enter’ (E), intervenir ‘intervene’ (E), (re)partir ‘leave’ (E), parvenir à ‘reach somewhere’(E), (re)passer ‘pass’ (E/a), provenir de ‘accrue’ (E), (re)sortir ‘go out’ (E), survenir ‘turn up’ (E), rentrer ‘come back’ (E), retourner ‘come back’ (E), (re)venir ‘come’ (E) accoucher de ‘give birth’ (A/e), décéder ‘die’ (E/a), (re)devenir ‘become’ (E), échoir ‘decay’ (E), expirer ‘decay’ (E), mourir ‘die’ (E), naître ‘be born’ (E), ressusciter ‘revive’ (E/A), trépasser ‘fade away’ (A/e)

Happening

advenir ‘happen’ (E)

Appearance

apparaître ‘appear’ (E/a), disparaître ‘disappear’ (A/e), éclore ‘emerge’ (E/a), paraître ‘appear’ (E/A), résulter de ‘result from’(E)

Indefinite change in a particular direction

(re)descendre ‘descend’ (E), (re)monter ‘arise’ (E/a), (re)tomber ‘fall’(E)

Continuation of pre-existing state

demeurer ‘stay’ (E/a), rester ‘stay’ (E)

Existence of state Uncontrolled processes Controlled processes (motional) Controlled processes (non-motional)

emission

émaner ‘emanate’ (E/A)

sauter ‘jump’ (A/e)

BE or HAVE in Contemporary Standard French – residua of semantic motivation

267

means that the BE-construction is clearly more frequent than the HAVE-construction. For (A/e) it is the other way around. More or less equal use of the two constructions is coded by (E/A). Not all of the verbs in Table 4 are monadic in the sense of Legendre and Sorace (2003), but at least in French the inclusion of dyadic BE-selecting verbs does not impede the applicability of the ASH. Note that the French prototypical BE-selecting verb aller is not monadic. It requires the presence of an obligatory locative argument: (16) Il vous a parlé de votre pays […] il y He you have-prs.3sg talk-pcpt of your country he there est allé souvent be-prs.3sg go-pcpt often ‘He has talked to you about your country […] he has often gone there.’ (FRANTEXT: BASTIDE) Dyadic (two-place) verbs are printed in bold. These verbs denote a change of location not literally, but metaphorically. Most of them are derivatives of change of location verbs. Parvenir (‘reach somewhere, come through’) is a derivative of venir (‘come’) and denotes a metaphorical arrival. The two derivatives of venir (‘come’) convernir de (‘agree to’) and disconvenir de, generally used in the expression ne pas disconvenir de qch (‘not to deny sth’), are more problematic. Although the eventuality of “agreeing” could be described as a controlled non-motional process, the change of location component of “coming together” is still detectable. Only this component explains why these verbs allow for BE-selection. Échapper qch à q takes BE in the sense of ‘slip out with, mention unintentionally’. It is a genuine metaphorical change of location verb, as échoir is in the sense of ‘fall to’. Note that choir (‘fall’), antiquated or very formal in Contemporary Standard French, takes HAVE (Le Grand Robert 2007). There are two outliers that are not dealt with in Legendre and Sorace (2003). We have seen that in the 18th century, the verb courir as an instance of the class of controlled motional processes still took BE in telic contexts. According to Grevisse and Goosse (2008: 1036), BE-selection with courir is not acceptable in Contemporary Standard French, but can still be heard from time to time in spoken discourse. Nonetheless, there is another verb of the class that is occasionally used with BE in telic contexts: sauter (‘jump’) (Grevisse and Goosse 2008: 1037).11 Hence, sauter (‘jump’) has to be included in the list of BE-selecting verbs in Contemporary Standard French:

11 Legendre and Sorace (2003) mention sauter (‘jump’) as a HAVE-selecting verb.

268

Rolf Kailuweit

(17) George est sauté au bas du cabriolet George be-prs.3sg jump-pcpt on the backseat of the cabriolet ‘George has jumped on the backseat of the convertible.’ (FRANTEXT, MAURIAC) The other outlier appears in the class of uncontrolled processes. According to Grevisse and Goosse (2008: 1037) émaner (‘emanate’) shows free variation between BE-selection and HAVE-selection. (18) Elle étincelait comme si la lumière fût émanée She shine-ipfv.3sg as if the light be-prs.3s emanate-pcpt d’elle from her ‘She was shining as if the light emanated from her.’ (ROBERT, GAUTIER) Are we dealing with arbitrary exceptions or is there a general organising principle? My claim is that such an organising principle exists as a necessary, but insufficient condition for BE-selection. In the narrower or wider sense, all BEselecting verbs in Contemporary Standard French can be considered verbs of change of location. Verbs of manner of motion turned into change of location verbs with a goal argument-adjunct added by template augmentation. This explains the occurrence of sauter (‘jump’) with BE in telic contexts, but it seems arbitrary that only this verb is accepted in the BE-construction in Contemporary Standard French, while other verbs of the same class are not. As far as the class of uncontrolled processes is concerned, it is curious that only the verb that gives name to the subclass of emission, émaner (‘emanate’), allows for BEselection. Nonetheless, emission could be reinterpreted as a change of location, in which case émaner (‘emanate’) would be a special case of sortir (‘go out’) and can be used as a hyperonym in the same contexts: (19) … les électrons expulsés par l’action de la lumière the electrons expel-pcpt by the action of the light – photoélectrons – sortent de la matière photolectrons go out-ipfv.3pl from the materia ‘The electrons emanate from the materia as photoelectrons expelled by the action of the light.’ (FRANTEXT: BROGLIE) The BE-selecting verbs in the class of continuation of a pre-existing state – rester (‘stay’) and demeurer (‘remain’) – denote the negation of a change of

BE or HAVE in Contemporary Standard French – residua of semantic motivation

269

location in their literal, locative meaning. Moreover, the different types of BEselecting verbs of change of state in Contemporary Standard French are metaphorically related to change of location. BE-selecting verbs of indefinite change in a particular direction – descendre (‘descend, decrease’), monter (‘ascend, increase’) and tomber (‘fall’) – denote a directed movement in their literal meaning. Verbs of appearance as well as verbs denoting birth and death can be reinterpreted as verbs of coming and going from a localist viewpoint. In addition, several BE-selecting verbs in Contemporary Standard French are derivatives of prototypical verbs of change of location, mostly of venir (‘come’). Note that BE-selection was not restricted to change of location in earlier stages of French. Oudin ([1640] 1972) compiled a large number of anticausative BE-selecting or varying verbs of change of state which do not suggest a localist reinterpretation: (20) Anticausatives in Oudin ([1640] 1972: 214–226)12 blondir ‘become fairer’, durcir ‘harden’, embellir ‘grow more attractive’, empirer ‘worsen’, engoudir ‘nump’, enlaider ‘become ugly’, espassir ‘thicken’, faner ‘wilt’, guérir ‘heal’, hausser ‘raise’, jaunir ‘yellow’, moisir ‘rot, stagnate’, mollir ‘soften’, noircir ‘blacken’, rougir ‘go red’, veillier ‘grow old’, etc. If change of location was not a prerequisite for BE-selection in Classical French, it seems reasonable to suppose that Contemporary Standard French developed as Early Modern Spanish did more than three centuries before.13 In line with Rosemeyer (2014), I put forward the hypothesis that the creation of a new rule for BE-selection was a frequency effect that occurred at the same time as a general typologically-driven trend to HAVE-selection. It is beyond the scope of this paper to prove this process in detail. I do not provide statistical evidence for the use of verbs of change of location in the 19th century. Nonetheless, contemporary frequency dictionaries suggest that Rosemeyer’s observation for Early Modern Spanish also holds for Contemporary Standard French. According to Julliand, Brodin, and Davidovitch (1970), venir (‘come’) is the

12 Oudin (1972: 215) defines causative/anti-causative pairs in the following way: “ceux qui sont actifs transitifs, & neutres tout ensemble” [those [verbs] that are active transitive and neutral at the same time]. 13 The change may explain the “surprising” result of Heidinger’s study (this volume) that not only [TEL+] and [MO+], but also [CON+] increase the probability of a BE-construction with monter (‘ascend, increase’). If it is true that [MO–] correlates with [CON–], then the observed impact of [CON+] may be a side effect of the impact of [MO+].

270

Rolf Kailuweit

fourth most frequent verb after the transitive verbs avoir (‘have’),14 dire (‘say’) and savoir (‘know’). Aller (‘go’) is the tenth, rester the 14th, passer the 16th and arriver the 21st most frequent. Revenir (‘come again’) is the 25th, partir (‘leave’) the 35th, entrer (‘enter’) the 36th and sortir (‘go out’) the 40th most frequent. The only HAVE-selecting intransitive verb in this field is parler (‘speak’) in 12th place. The data in Quasthoff, Fiedler, and Hallsteinsdóttir (2013) confirm these findings, although the dictionary is not easy to work with, since it is a full form dictionary which does not indicate lemma frequencies of verbs. However, a look at the infinitives shows that aller (‘go’), passer (‘pass’), partir (‘leave’), sortir (‘go out’) and venir (‘come’) are among the most frequent 1,000 words, partir being the most frequent form followed by aller and passer. HAVE-selecting parler (‘speak’) appears in 740th place after the five verbs of change of location. Therefore, BE-selecting verbs of change of location are among the most frequent verbs in Contemporary Standard French. It seems reasonable to assume that the BE-constructions of these verbs are entrenched in a subject-external, objective way, guiding conventionalised language use.15 However, not only BE-selection has become the only acceptable way of expressing compound tenses in Standard Contemporary French for these particular verbs. It is also associated with change of location as a general rule that also applies to some former BE-selecting verbs that have a change of location component in their semantics, but do not feature among the most frequent verbs. It has to be pointed out that this is of high theoretical importance. A usage-based approach that takes frequency into account does not necessarily exclude the assumption of general rules that apply for high frequent and low frequent forms at the same time. Frequency can play a role in cases of system instability. It does not only lead to conservation, but also helps to reorganise the system. As far as French is concerned, the instability is caused by the grammatical change that causes a BE-construction to develop from a resultative to a fully grammaticalised anterior construction. As we have seen in section 2 of this paper, this process that occurred in Early Modern Spanish (Rosemeyer 2014) is a rather recent development in French: the “theoretical” distinction between resultative BE-constructions and anterior HAVE-constructions for “some verbs” which show variation (Grevisse and Goose 2008: 1033) turns out to be the last laconic

14 The auxiliary use of avoir (‘have’) is not counted. 15 My own approach to entrenchment does not focus on the speaker’s mind, but on conventionalized interchange in form of language games that is based on intertextuality (see Kailuweit, 2013 and Rosemeyer, this volume).

BE or HAVE in Contemporary Standard French – residua of semantic motivation

271

reminiscence of prescriptive grammar applied to an antiquated stage of Standard French. The new rule, “select BE with a verb of change of location”, has not yet made its way into the reference works. As a necessary but insufficient condition – the rule “select BE only if there is a change of location component in the semantics of the verb” is reality from a descriptive point of view. However, apart from the prototypical (and highly frequent) verbs of change of location, which all select BE exclusively, the question remains unanswered whether there are replenishing rules that could explain why certain former BE-selecting verbs of lower ASH-classes have become BE-residua and other verbs of the same semantic classes have not.

6 Conclusion and outlook In the present paper I have shown that the distribution of BE-selecting verbs in Contemporary Standard French is not arbitrary. In earlier stages of French the opposition between BE-selection and HAVE-selection depended on resultativity. BE was used in a resultative construction while HAVE was used to express (pure) anteriority. Note that there is contiguity of resultative and anterior constructions. All resultatives imply an anterior event that brought about a change, but only some anterior events lead to a resultant state that holds at reference time. Between the 18th and the early 20th century the system changed. For one thing, BE-constructions were reinterpreted as anteriors that did not necessarily imply a result state holding at reference time. For another, HAVE-constructions were generalised with verbs that have a resultative meaning, such as change of state verbs. All verbs allowing for BE in anterior constructions in Contempory Standard French formerly used BE in resultative constructions. Therefore, they can be considered BE-residua. The BE-construction is not “productive”. In Contemporary Standard French, there are no recent loan verbs or verbs resulting from word formation processes that allow for BE in compound tenses. Sorace’s ASH in the feature-based version of Legendre and Sorace (2003) helps to compare the distribution of BE-residua in Contemporary Standard French to the distribution of BE-selection in other Romance and Germanic languages that encode split-intransitivity by means of auxiliary selection. However, the features have to be considered to allow for a coherent distinction of ASH’s seven semantic classes. An examination of the features (Kailuweit 2011) confirms the insight of Mateu (2009) that the semantic feature “motion” does not universally increase the affinity of a verb for unaccusativity. Rosemeyer (2014) has shown that the establishment of a new rule in Early Modern Span-

272

Rolf Kailuweit

ish – “select BE with a verb of change of location” – is a frequency effect. The same new rule applies to Contemporary Standard French. Evidence was provided to back the claim that frequency also plays a decisive role in Contemporary Standard French in establishing “change of location” as the organising principle of BE-selection. All BE-residua in Contemporary Standard French can be interpreted as change of location verbs in a nearer or broader sense. The rule applies not only to the most frequent verbs, but also to less frequent verbs that have a change of location component in their semantic structure. However, the rule is a necessary but insufficient condition for BE-selection. It explains why there are some outliers in lower ASH classes that allow for BEselection, but the question remains unanswered why other verbs of these classes with a change of location component in their semantic structure do not allow for BE-selection. Some questions remain open for further research. From a diachronic point of view it is puzzling why a considerable number of verbs taking BE in resultative contexts did not allow for HAVE when no resultative interpretation was available. In line with Rosemeyer’s findings for Old Spanish (Rosemeyer 2012, 2014), I put forward the hypothesis that the spread of the HAVE-construction in French was a long-lasting process. With regard to verbs that were not affine to the original contexts of HAVE (controlled, non-resultative processes), anteriority was probably encoded with the past tense, especially with the imperfective. From a synchronic point of view, the question is raised as to how resultativity can be expressed when the resultative use of BE no longer exists. Unlike Spanish, French has not developed a new resultative construction with a specific auxiliary. First of all, the contiguity relation of resultativity and anteriority comes into play. Nowadays, the HAVE-construction may refer to a former event without excluding that the result of this event still holds at reference time. As we have seen, Arrivé (2006) and Grevisse and Goosse (2008) list about 50 verbs of Contemporary Standard French that allow for a BE-construction which is clearly not an anterior. Is this construction (still) a resultative construction and if so, why is this construction limited to a relatively small number of verbs? In my opinion, we should distinguish the copula + adjectivised participle construction of Contemporary Standard French from the former resultative construction. The resultative construction refers to a result state which is contiguous to a former event that led to a change. The copula + adjective construction denotes a state without referring to its coming about. A verb like croître (‘grow’) allowed for a resultative construction in Classical French: (21) Le Marquis est un peu crû, mais ce n’est the marquees be-prs.3sg a little grow-pcpt but this not be-prs.3sg

BE or HAVE in Contemporary Standard French – residua of semantic motivation

273

pas assez pour se récrier ; sa taille ne sera point neg enough for refl shout for joy his size not be-prs.3sg neg comme celle de son père as that of his father ‘The maquees has grown in stature, but not as much as to shout for joy. He will never have the stature of his father.’ (Sevigné, Fournier 2002: 257) However, in Contemporary Standard French, the participle of croître (‘grow’) can no longer be combined with a BE-construction. (22) Il *est / a crû He has grown (23) Il est (plus) grand He is tall(er) The participle of croître (‘grow’) has not developed into an adjective, unlike the participles of other verbs of change of state that appear in the lists of Arrivé (2006) and Grevisse and Goosse (2008). In Contemporary Standard French, the result state of “being grown” can be referred to with the HAVE-construction or with an adjective such as grand (‘tall’). Further research will have to focus on French change of state verbs and their development in the 18th and 19th centuries. This will help us to understand the reorganisation of split-intransitivity that led to the contemporary system.

References Abeillé, Anne and Danielle Godard. 2002 The syntactic structure of French Auxiliaries. Language 78(3): 404–452. Aranovich, Raúl. 2003 The semantics of auxiliary selection in Old Spanish. Studies in Language 27(1): 1–37. Arrivé, Michel. 2006 Bescherelle – La conjugaison pour tous. Édition entièrement revue sous la responsabilité scientifique de Michel Arrivé. Paris: Hatier. Bentley, Delia. 2006 Split-Intransitivity in Italian. Berlin, New York: De Gruyter. Benzing, Joseph. 1931 Zur Geschichte von ser als Hilfszeitwort im Spanischen. Zeitschrift für romanische Philologie 51: 385–460. Blanche-Benveniste, Claire. 1997 La notion de variation syntaxique dans la langue parlée. Langue française 115(1): 19–29. Burzio, Luigi. 1986 Italian Syntax. Dordrecht: Reidel.

274

Rolf Kailuweit

Bybee, Joan L., Revere Perkins and William Pagliuca. 1994 The Evolution of Grammar. Tense, Aspect and Modality in the Languages of the World. Chicago: University of Chicago Press. Centineo, Giulia. [1986]1996 A lexical theory of auxiliary selection in Italian. Probus 8: 223 –271. (Originally in Davis [California] Working Papers in Linguistics 1 1986). Dowty, David. 1991 Thematic proto-roles and argument selection. Language 67(3): 547–619. Foerster, August. 1908 Avoir und Être als Hilfsverba beim intransitiven Zeitwort in ihrer Entwicklung vom Alt- zum Neufranzösischen. Darmstadt: Otto. Fournier, Nathalie. 2002 Grammaire du français classique. Paris: Belin. Gamillscheg, Ernst. 1957 Historische französische Syntax. Tübingen: Niemeyer. Grevisse, Maurice and André Goosse. 142008 Le bon usage: grammaire française. Bruxelles: De Boeck/Duculot. Hofmann, Fritz. 1890 Avoir und estre in den umschreibenden Zeiten des altfranzösischen intransitiven Zeitworts. Berlin: Mayer & Müller. Juilland, Alphonse, Dorothy Brodin and Catherine Davidovitch. 1970 Frequency Dictionary of French Words. The Hague, Paris: Mouton. Kailuweit, Rolf. 2005 Linking: Syntax und Semantik französischer und italienischer Gefühlsverben. Tübingen: Niemeyer. Kailuweit, Rolf. 2011 Le choix de l’auxiliaire: Être ou avoir en français standard contemporain. In: Renata Enghels, Machteld Meulleman and Clara Vanderschueren (eds.), Peregrinatio in Romania: Artículos en homenaje a Eugeen Roegiest con motivo de su 65 cumpleaños, 397–420. Gent: Academia Press. Kailuweit, Rolf. 2013 Radical Role and Reference Grammar (RRRG) A sketch for remodelling the Syntax-Semantics-Interface. In: Elke Diedrichsen and Brian Nolan (eds.), Linking Constructions into Functional Linguistics – The Role of Constructions in Grammars, 103– 141. Amsterdam, Philadelphia: John Benjamins. Keller, Frank and Antonella Sorace. 2003 Gradient auxiliary selection and impersonal passivization in German: an experimental investigation. Journal of Linguistics 39: 57– 108. Labelle, Marie. 1992 Change of state and valency. Linguistics 28: 375–414. Le Grand Robert de la langue française. CD-ROM. 2007 Paris: Le Robert/Sejer. Legendre, Géraldine, Yoshiro Miyata and Paul Smolensky. 1991 Unifying syntactic and semantic approaches to unaccusativity: a connectionist approach. In: Proceedings of the 17 th Annual Meeting of the Berkeley Linguistic Society. General Session & Parasession on the Grammar of Event Structure, 156–167. Berkeley: Berkeley Linguistic Society. Legendre, Géraldine and Antonella Sorace. 2003 Auxiliaires et intransitivité en français et dans les langues romanes. In: Danièle Godard (ed.), Les langues romanes. Problèmes de la phrase simple, 185–234. Paris: Editions du CNRS. Legendre, Géraldine. 2007 Optimizing auxiliary selection in Romance. In: Raúl Aranovich (ed.), Split Auxiliary Systems. A Cross-linguistic Perspective, 145–180. Amsterdam, Philadelphia: John Benjamins. Levin, Beth and Malka Rappaport Hovav. 1995 Unaccusativity: at the Syntax-Lexical Semantics Interface. Cambridge, MA: MIT Press. Mateu, Jaume. 2009 Gradience and auxiliary selection in Old Catalan and Old Spanish. In: Paola Crisma and Giuseppe Longobardi (eds.), Historical Syntax and Linguistic Theory, 176–193. Oxford: Oxford University Press.

BE or HAVE in Contemporary Standard French – residua of semantic motivation

275

Merlan, Francesca. 1985 Split intransitivity: functional oppositions in intransitive inflection. In: Johanna Nichols and Anthony Woodbury (eds.), Grammar Inside and Outside the Clause: Some Approaches to Theory from the Field, 324–362. Cambridge: Cambridge University Press. Oudin, Antoine. 1972 [2/1640] Grammaire françoise rapportée au langage du temps. Genève: Slatkine. Perlmutter, David M. 1978 Impersonal passives and the unaccusative hypothesis. Proceedings of the Berkeley Linguistics Society 4: 157–189. Posner, Rebecca. 1996 The Romance Languages. Cambridge, UK: Cambridge Universtiy Press. Quasthoff, Uwe, Sabine Fiedler and Erla Hallsteinsdóttir (eds.). 2013 Frequency Dictionary French. Leipzig: Leiziger Universtätsverlag. Roegiest, Eugeen. 2001 La grammaire relationelle. In: Günter Holtus, Michael Metzeltin und Christian Schmidt (eds.), Lexikon der Romanistischen Linguistik, Bd. I/1, 393–407. Tübingen: Niemeyer. Rosemeyer, Malte. 2012 How to measure replacement: Auxiliary selection in Old Spanish bibles. Folia Linguistica Historica 33(1): 135–174. Rosemeyer, Malte. 2014 Auxiliary Selection in Spanish. Gradience, Gradualness, and Conservation. Amsterdam, Philadelphia: John Benjamins. Rosen, Carol. 1984 The interface between semantic roles and initial grammatical relations. In: David Perlmutter and Carol Rosen (eds.), Studies in Relational Grammar 2, 38–77. Chicago, IL: University of Chicago Press. Sabatini, Francesco and Vittorio Colletti. 21999 Dizionario Italiano. Edizione in CD-Rom. Firenze: Giunti. Salvi, Giampaolo. 1982 La frase semplice. In: Lorenzo Renzi (ed.), Grande grammatica italiana di consultazione Vol.1, 29–113. Bologna: Il Mulino. Sankoff, Gillian and Pierette Thibault. 1977 L’alternance entre les auxiliaires avoir et être en français parlé à Montréal. Langue française 34: 81–108. Schwarze, Christoph. 1996 The syntax of Romance auxiliaries. In: Miriam Butt and Tracy Holloway (eds.), LFG-Workshop. Proceedings of the First LFG Conference, 418–433. Grenoble: Rank Xerox Research Centre. Schwarze, Christoph. 1998 A lexical-functional analysis of Romance auxiliaries. Theoretical Linguistics 24: 83–105. Sokol, Monika. 1999 Das Zusammenspiel der Verbalkategorien und der französischen Futura. Tübingen: Niemeyer. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76: 859–890. Togeby, Knud. 1974 Précis Historique de Grammaire Française. Copenhague: Akademisk Forlag. Tuttle, Edward F. 1986 The spread of ESSE as universal auxiliary in Central Italo-Romance. Medievo Romanzo 11: 229–287. Van Valin, Robert D. 1990 Semantic parameters of split intransitivity. Language 66: 221– 260. Vincent, Nigel. 1982 The development of the auxiliaries HABERE and ESSE in Romance. In: Nigel Vincent and Martin Harris (eds.), Studies in the Romance Verb, 71–97. London: Biddles Ltd., Guildford and Kings Lynn Croom Helm Ltd.

Steffen Heidinger, Universität Graz

The auxiliary selection of French monter ‘move upward’ from the 16 th to the 20 th century 1 Introduction

1

The overall topic that this study is embedded in is the diachronic development of compound tense auxiliary selection of intransitive verbs in French.2 The pivotal points of this development are (i) that in present-day French the vast majority of intransitive verbs select avoir ‘have’ (henceforth A) and only relatively few verbs select être ‘be’ (E) and (ii) that the situation was different at earlier stages of the language: many more verbs selected E in Old and Middle French. The situation in present-day French is thus the result of a decrease in verbs selecting E and an increase in verbs selecting A (cf. section 2 of this paper). In order to find out more about this development I will investigate in detail the development of the intransitive use of the verb monter ‘move upward’ (which is one of the verbs that still selects both E and A in present-day French). As is well known, in languages with more than one auxiliary (i.e. with split auxiliary systems) the distribution of the auxiliaries can be determined (or at least influenced) by both semantic and syntactic factors (cf. McFadden 2007 for an overview). Syntactic distinctions that play a role in languages with a split auxiliary system are the distinction between transitive, reflexive and intransitive verbs, and within the class of the intransitive verbs the distinction between unergative and unaccusative verbs. French transitive verbs differ from reflexive verbs (or verbs in reflexive constructions) in that transitive verbs only select A, while reflexive verbs select E. (1) transitive: A Jean a cassé le vase. Jean A.3.sg.prs break.pst.ptcp the vase ‘John broke the vase.’

1 I wish to thank the participants of the workshop “Auxiliary selection: gradience and gradualness” (Freiburg, June 2012), Malte Rosemeyer, Rolf Kailuweit and an anonymous reviewer for their helpful comments on earlier versions of this paper. 2 For the sake of brevity, compound past tense auxiliary selection will be referred to by the term auxiliary selection in the remainder of this paper.

278

Steffen Heidinger

(2) reflexive: E a. Jean s’ est lavé. J. refl E.3.sg.prs wash.pst.ptcp ‘Jean washed (himself).’ b. La branche s’ est cassée. the branch refl E.3.sg.prs break.pst.ptcp ‘The branch broke.’ In German, on the other hand, transitive and reflexive verbs select the same auxiliary, namely haben ‘have’. What French and German have in common is that intransitive verbs are the syntactic class that shows the highest degree of variation in auxiliary selection. It does not come as a surprise that intransitive verbs are also the class for which the auxiliary selection has been studied most extensively in such languages (e.g. French, Italian, Dutch, German). Given that different auxiliaries are selected by the verbs of one syntactic class, the distinction between transitive, reflexive and intransitive verbs cannot shed light on the auxiliary selection of intransitive verbs. Within syntactic approaches the split auxiliary selection of intransitive verbs has been linked to the distinction between unergative and unaccusative verbs (Perlmutter 1978, 1989; Burzio 1986; Loporcaro 2007).3 But even in syntactic approaches, verb semantics must be taken into consideration since the distribution of intransitive verbs in the unaccusative and the unergative class is presumably semantically determined (cf. Perlmutter 1978; Levin and Rappaport Hovav 1995). Semantic features that have been mentioned as relevant for split intransitivity and/or the auxiliary selection of intransitive verbs are amongst others telicity, the semantic role of the subject, and the semantic verb class (which in turn often incorporates distinctions based on telicity and the properties of the subject; cf. Levin and Rappaport Hovav 1995; Sorace 2000; Legendre and Sorace 2003; Cennamo and Sorace 2007; Kailuweit 2011). In the present paper, the syntactic distinction between unaccusative and unergative verbs does not play a role. Given the diachronic orientation of the present investigation, the majority of data comes from stages of the language for which no intuitions of native speakers are available. As a consequence, the diagnostics which have been used in the literature to distinguish between unaccusative and unergative verbs cannot be applied in a reasonable way (e.g. absolute constructions or en-clitization; for French, cf. Legendre and Sorace

3 Conversely, auxiliary selection has been used as a diagnostic for the distinction between the two syntactic subclasses of intransitive verbs.

The auxiliary selection of French monter ‘move upward’

279

2003; Labelle 1992). Instead I will look at some of the semantic features that have been assumed to determine split intransitivity (syntactic or not). More precisely, I will consider the features [±agent], [±telic], and [±motion] (cf. section 3.1). The goals of the paper are (i) to describe the overall development of the auxiliary selection of French monter from the 16 th to the 20 th century and (ii) to verify the relevance of the three above-mentioned semantic features in this development. As concerns the relevance of the semantic distinctions, the first issue is whether changes occur with respect to these features in the auxiliary selection of monter from the 16 th to the 20 th century: – Does the frequency with which the auxiliaries occur in a given context (e.g. E in telic vs. atelic contexts) change? – Does the frequency with which monter selects the two auxiliaries in a given context (e.g. telic contexts) change over time? The second issue concerns the predictive power of these semantic features for the auxiliary selection of monter and whether their predictive power changes. This issue is directly linked to the second of the above questions. For example, is telicity a good predictor for monter in the selection of E vs. A? The third and last issue is whether the consideration of the semantic features mentioned above sheds light on the overall quantitative development of the auxiliary selection of monter.

2 Main characteristics of French auxiliary selection 2.1 Present-day hexagon French The variety of French which is spoken nowadays in mainland France has a binary auxiliary system, with E and A as the two compound tense auxiliaries (cf. McFadden 2007 for a recent overview of auxiliary systems and especially Loporcaro 2007 on triple auxiliation). The distribution of E and A is such that transitive verbs select A (cf. 1), reflexive verbs select E (cf. 2), and intransitive verbs select E, A or both (cf. Legendre and Sorace 2003; Manente 2008; Giancarli 2011; Kailuweit 2011).4, 5 4 In several varieties of French spoken outside mainland France, especially those in Northern America, A has been generalized even more and has become the only auxiliary (cf. Manente 2008: 42–47). 5 The auxiliary selection of French monter is further discussed in Manente (2013) – a paper which came to the author’s attention only after completion of this paper and which is thus not taken into account here.

280

Steffen Heidinger

For transitive and reflexive verbs the distribution is thus rather straightforward. Concerning reflexive verbs it should be added that E is selected independently of the specific type of reflexive construction: in both (2a.), which is a true reflexive construction, and (2b.), which is a reflexive anticausative, E is selected as the auxiliary, not A. Intransitive verbs provide the most interesting syntactic context for the study of French auxiliary selection. As mentioned above, three subsets of French intransitive verbs can be distinguished based on their auxiliary selection. The largest subset is the one with verbs that only select A: this set includes most verbs of change of state, activity verbs, motion and manner-ofmotion verbs, examples of which are (intransitive) casser ‘break’, travailler ‘work’, courir ‘run’. (3) Jean a/ *est travaillé toute la. journée. Jean A/ E.3.sg.prs work.pst.ptcp all the day ‘Jean worked the whole day long.’ The subset of French intransitive verbs that only select E is restricted to about ten to twenty verbs, one of them being aller ‘go’. (4) Jean est/ *a allé à Paris. Jean E/ A.3.sg.prs go.pst.ptcp to Paris ‘Jean went to Paris.’ The size of the set of verbs that only select E varies slightly in the literature; in (5) the list proposed in Grevisse and Goose (2008) is given. (5) Set of French intransitive verbs selecting exclusively E: aller ‘go’, arriver ‘arrive’, décéder ‘die’, devenir ‘become’, entrer ‘enter’, mourir ‘die’, naître ‘be born’, partir ‘leave’, rester ‘stay’, retourner ‘return’, sortir ‘go out’, tomber ‘fall’, venir ‘come’ (Grevisse and Goose 2008: 1033) The third subset contains verbs which select both E and A; examples are descendre ‘go down’, monter ‘go up’, passer ‘pass’. The size of this last subset varies considerably depending on whether the authors include only eventive or also resultative uses. For example, Grevisse and Goose (2008: 1033) provide a rather long list of verbs including many verbs of change of state. But with these verbs the construction être + past participle can only have a resultative meaning and the verbs should not be on a list of verbs that select E as the compound tense auxiliary (cf. also Heidinger 2010: 188–189 and Kailuweit 2011: 408).

The auxiliary selection of French monter ‘move upward’

281

(6) Some French intransitive verbs selecting both E and A: accroître ‘grow/increase’, augmenter ‘increase’, baisser ‘go down’, changer ‘change’, commencer ‘begin’, diminuer ‘decrease’, éclater ‘burst’, embellir ‘embellish’, empirer ‘worsen’, enlaidir ‘turn ugly’, finir ‘end’, grandir ‘grow in size’, grossir ‘grow/gain weight’ (Grevisse and Goose 2008: 1033) A verb that undoubtedly can have both auxiliaries in eventive intransitive uses is monter, as in (7). (7) a. le fleuve a monté très vite … the river A.3.sg.prs rise.pst.ptcp very fast ‘the river rose very quickly’ (Rambaud 1997, Frantext) b. le bruit de la mer est monté jusqu’ à moi the noise of the sea E.3.sg.prs rise.pst.ptcp up to me ‘the noise of the sea rose up to me’ (Soupault 1983, Frantext) Given the situation in present-day hexagon French (as described above and repeated in (8) below), the question arises as to which principles govern the auxiliary selection of French intransitive verbs. Recently, Kailuweit proposed an account of French auxiliary selection for which he applies a gradual notion of unaccusativity (based on the semantic features inherent telicity, motion, directedness, controllability, and stativity) (cf. Kailuweit 2011: 415). Depending on the specification of these semantic features, intransitive verbs have a higher or lower degree of unaccusativity. Crucially, the probability that a verb selects E positively correlates with the degree of unaccusativity (Kailuweit 2011: 416– 417). (8) Distribution of E and A with intransitive verbs in present-day hexagon French: (i) only A (= largest set) casser ‘break’, travailler ‘work’, courir ‘run’, etc. (ii) only E (restricted to about 20 verbs) aller ‘go’, arriver ‘arrive’, venir ‘come’, etc. (iii) A and E (restricted to about 30 verbs) descendre ‘move downward’, monter ‘move upward’, passer ‘pass’, etc.

282

Steffen Heidinger

2.2 Diachronic development: decrease in verbs that select être During the history of French, the auxiliary selection of intransitive verbs has changed considerably. Mackenzie (2006: 129–144) describes this process from the perspective of the auxiliary E and distinguishes three stages: in the first stage, the sequence E + past participle was not used to express the perfect, but was a resultative copula describing the resultant state after an event. Because of the semantics of the construction, only the past participles of telic verbs could appear in this construction. In the second stage, the construction was interpreted as a perfect tense and E was grammaticalized as a perfect auxiliary. In the early period of being a perfect auxiliary E was only selected by telic verbs (which is a heritage of the aspectual restriction of the resultative copula). In the third stage, the auxiliary selection was conventionalized independently of the aspectual properties of verbs. This dissociation of the auxiliary E and the resultant state constraint led to an increase in verbs that select A, and to a decrease in verbs that select E. Although Mackenzie’s (2006) description leaves some open questions – especially with respect to its empirical foundation – it nevertheless seems to be correct that the overall development is a decrease in verbs that select E and an increase in verbs that select A. Further evidence for the wider use of E at earlier stages can be found in the traditional Romance literature. Benzing (1931: 451) notes that E was the dominant auxiliary of intransitive verbs in Old French. Clédat (1903: 38–39) states that verbs of change of state frequently selected E as a perfect auxiliary in the Middle Ages (“l’ancienne langue”), but only select A in present-day French. Förster (1908: 103) describes Middle French (1300–1500) as an intermediate period between Old French (dominance of E) and Modern French (dominance of A). Förster (1908: 69–100) further lists several verbs that selected E and A at older stages of French, but only select A in present-day French: approcher ‘approach’, avancer ‘move’, disparaître ‘disappear’, lever ‘rise’, tourner ‘turn’, finir ‘end’, périr ‘perish’, crever ‘burst’, rompre ‘break’, diminuer ‘decrease’, and multiplier ‘multiply’. Similarly, Oudin (1640) provides a list of about 100 verbs of change of state selecting E in 17th-century French and Gougenheim (1973: 119) notes that the possibility of choosing between E and A existed for more verbs in 16th-century French than in present-day French (cf. also Fournier 1998). Despite these hints, which all point in a similar direction, it must be noted that there is still no comprehensive corpus-based investigation of the development of French auxiliary selection. One problem that such a study faces is that for the sequence E + past participle the distinction between a passive use (resultative or eventive) on the one hand and an eventive intransitive use on

The auxiliary selection of French monter ‘move upward’

283

the other is not trivial (cf. Giancarli 2011: 19–28; also McFadden and Alexiadou 2010: 399–403 on the be-periphrasis and different types of perfects in Earlier English). Consider for example the data in (9) and (10) cited by Förster (1908) to illustrate the use of E as a perfect auxiliary with verbs of change of state at earlier stages of French. It is by no means obvious that these examples show eventive intransitive uses and not resultative constructions. Therefore, statements on the respective situation by contemporary grammarians such as Oudin (1640) seem more reliable. (9)

rompre ‘break’ a. […] par ou il est rompu p p it E.3.sg.prs break.pst.ptcp ‘where it broke’ (14 th century, Miracles de Nostre Dame, Förster 1908: 93) b. Mes biens mondains sont rompuz et my goods earthly E.3.pl.prs break.pst.ptcp and cassez break.pst.ptcp ‘my earthly goods shattered and broke’ (16 th century, Œuvres complètes de Gringore, Förster 1908: 93)

(10) tourner ‘turn’ a. Més ma vie est tournée en desespoir but my life E.3.sg.prs turn.pst.ptcp into despair ‘but my life turned into despair’ (14 th century, Froissart, Förster 1908: 83) b. Nostre ris est tourné en pleurs our laughter E.3.sg.prs turn.pst.ptcp into tears ‘our laughter turned into tears’ (16 th century, Le Mystère du vieil testament, Förster 1908: 84) Despite these methodological issues we can retain two facts from the literature presented above: first, in present-day French only a very small number of intransitive verbs select E. Second, the situation was different at earlier stages of the language, at which time E was selected by many more verbs. The assumed diachronic development was thus that the number of verbs that selected E decreased and the number of verbs that selected A increased.

284

Steffen Heidinger

3 Auxiliary selection of French monter ‘move upward’ from the 16 th to the 20 th century 3.1 Relevant features of monter After the short overview on the development of auxiliary selection in French given in section 2 we will now focus on the development of one single verb, namely monter ‘move upward’. As the examples in (11) to (13) show, the verb can be used in different syntactic constructions such as transitive, intransitive or reflexive. (11) intransitive […] je suis monté au premier étage […] I E.1.sg.prs go.up.pst.ptcp at. the first floor ‘I went up to the first floor.’ (Sartre 1983; Frantext) (12) transitive 6 J’ ai monté une seconde fois l’ escalier […] I A.1.sg.prs go.up.pst.ptcp a second time the stairway ‘I went up the stairway a second time.’ (Forest 1999; Frantext) (13) reflexive […] leurs économies se montaient à cent quatre-vingt-huit their savings refl go.up.3.pl.pst p one.hundred.and.eighty francs […] francs ‘their savings grew to one hundred and eighty francs’ (Martin du Gard 1922; TLFi)

6 Note that in French avoir is selected even if the sentence is only syntactically, but not semantically transitive (cf. Hopper and Thompson 1980 on semantic transitivity). In German, for example, the equivalent of (12) is formed with the auxiliary sein ‘be’ despite the fact that the sentence is syntactically transitive: (i) Ich bin ein zweites Mal die Stiegen hinaufgestiegen I E.1.sg.prs a second time the steps go.up.pst.ptcp ‘I went up the stairway a second time.’

The auxiliary selection of French monter ‘move upward’

285

The main reason why this verb is interesting for a diachronic study of auxiliary selection is that the verb can have various meanings in its intransitive use and these meanings cut across distinctions that are relevant for auxiliary selection: [±agent], [±telic], and [±motion]. Crucially, these semantic distinctions do not necessarily correlate with syntactic differences (such as the transitive and intransitive use), but can also be found in the intransitive use itself. For example, one could argue that monter has different meanings in (11) and (7a.) despite the fact that in both cases the verb is used intransitively. Some of the meanings that can be attributed to monter in its intransitive use are ‘go up’, ‘go up to x’, ‘ascend (to heaven)’, ‘enter’, ‘get in (a coach)’, ‘get on (a horse)’, ‘increase’, ‘come up’ (this list is based on the corpus data presented in section 3.3). Of course, such variation in meaning raises questions about the structure of the lexical entry of the verb or the root. Depending on the theoretical framework and the theory about the relation between the lexical semantics and sentence semantics, one could deduce most of the meanings in the examples from an underspecified lexical entry, which is then augmented by contextual information. Or one might postulate a rather rich lexical entry with several subentries, such that the different meanings correspond to different subentries. Obviously all of the meanings have in common that they are related to a concrete or abstract upward movement of an entity. The fact that the verb monter has several meanings in its intransitive use does not yet make it especially interesting for the study of auxiliary selection. The crucial point is that these meanings cut across the above-mentioned distinctions that have proven relevant for auxiliary selection (cf. also Manente (2008: 89–91) on the semantic properties of monter). The first of these distinctions concerns the semantic role of the subject and here I distinguish between agent subjects and patient subjects.7 The second distinction has to do with the type of event denoted by the verb: does it or does it not describe a motion

7 The use of the terms agent and patient in this paper needs some clarification. Haegeman (1994: 49) defines agent, patient and theme as follows: an agent is “the one who intentionally initiates the action expressed by the predicate”, a patient is “the person or thing undergoing the action expressed by the predicate”, and a theme is “the person or thing moved by the action expressed by the predicate”. I will depart from these definitions in that I will not distinguish between two types of undergoers depending on whether the predicate describes a motion event, and therefore lump together patient and theme under the notion patient. Note further that I will classify as agents the intentionally acting subjects of monter as a verb of change of location, such as il in (14a.), and not only the subjects of monter as a verb of manner of motion. Finally, the classification of a subject as an agent or a patient relies on animacy only insofar as inanimate subjects can never be [+agent]; animate subjects, however, can be either an agent or a patient depending on their role in the event denoted by monter.

286

Steffen Heidinger

event? The third and final distinction concerns telicity where I distinguish between telic vs. atelic contexts.8 The examples in (14) show that monter can appear with agents and patients as subjects, that it may denote a motion event and a non-motion event, and finally that it can appear in telic and atelic contexts. Note however that the binary features [±agent], [±telic], [±motion] do not always combine like in these two examples; in (15) monter has a patient subject, describes a non-motion event, but appears in a telic context. (14) a. agent-subject / motion / telic context […] il a monté dans la voiture […] he A.3.sg.prs go.up.pst.ptcp p the coach ‘He got into the coach.’ (Rétif de la Bretonne 1784; Frantext) b. patient-subject / non-motion / atelic context […] mais les prix ont monté […] but the prices A.3.pl prs go.up.pst.ptcp ‘But the prices increased.’ (Mendès-France 1984; Frantext) (15) patient-subject / non-motion / telic context la rente a monté à 90. the income A.3.sg.prs go.up.pst.ptcp p 90 ‘The income went up to 90 [Francs?].’ (Sand 1849; Frantext) Two more relevant properties of monter are that it selects both être and avoir and that it has a rather high text frequency and therefore sufficient data can be retrieved to trace the diachronic development in text corpora.

3.2 Data source The analysis of the development of monter is based on data extracted from the text corpus Frantext. Frantext is a large diachronic compilation of French texts.

8 The term telic context refers to cases where a bound event is described irrespective of whether the boundedness is triggered by the verb itself, by other elements of the sentence (e.g. locative or temporal adjuncts) or by contextual information.

The auxiliary selection of French monter ‘move upward’

287

Tab. 1: Corpus searches in Frantext. Queries

Hits checked

Time period

(monté|montée|montés|montées); 1500–1800; 1554 Hits

Hits 1–100 Hits 300–400 Hits 700–800 Hits 1400–1500

1500–1560 1623–1641 1715–1731 1783–1794

(&cavoir|&cêtre) (monté|montée|montés|montées); 1801–1850; 604 Hits

Hits 1–200; 300–400; 500–600

1801–1850

&cavoir (monté|montée|montés|montées); 1851–2000; 677 Hits [all checked]; &cêtre (monté|montée|montés|montées); 1851–2000; 2252 Hits [checked: years 1880–1884, 1930–1934, 1980–1983]

1880–1884 1930–1934 1980–1983

One asset of the corpus is its size: it contains about 250 million words (as of May 2012); a second important aspect is the diachronic coverage from the 12 th century to present-day French (although the size of the corpus increases towards present-day French). The fact that the corpus is not syntactically annotated does not pose a problem in the present context given that we focus on one verb only (for which lexically specified queries are possible). Table 1 shows how the data was extracted from Frantext. It shows the queries, the number of resulting hits in the corpus, the number of hits that were actually looked at and the resulting diachronic coverage. For example, the first query included the past participle forms of monter from 1500 to 1800; this query yields 1554 hits in the corpus. The next column indicates the hits that were looked at, and the last column shows the periods that are covered by the hits that were looked at. The total number of examples that were analysed amounts to 802 (this does not include all the irrelevant examples that had to be discarded manually, e.g. adjectival uses of the past participle of monter or uses of montée as a noun). For the period from 1851 to 2000 two different searches were conducted; one for the auxiliary E and one for A. For A the whole period was considered while for E only three spot checks were looked at. This difference is motivated by the fact that monter selects E much more frequently than A. In order to retrieve a reliable amount of data for the analysis of A+monter, a longer period of time had to be examined than for E. At some points in the presentation of the results, data for the auxiliary A is marked with an asterisk; this indicates that these numbers refer to this longer period of time.

288

Steffen Heidinger

3.3 Results 3.3.1 Quantitative development of E and A with monter After the preliminary remarks in sections 3.1 and 3.2 we can now turn to the diachronic development of the auxiliary selection of intransitive monter. I begin my presentation of the results with the overall quantitative development, i.e. how often monter selects E and A respectively as a compound tense auxiliary from the 16 th to the 20 th century. As shown in Figure 1 and Table 2, E dominates over the whole period under scrutiny. The percentage of E ranges between 77 % and 92 %. Although there is no continuous development, the overall trend is a slight increase in E (and a decrease in A) towards present-day French. This is an interesting result because the development of monter runs against the assumed general development of French auxiliary selection presented in section 2 (i.e. decrease in E and increase in A). Tab. 2: Auxiliary selection of monter (%).

1500–1560 1623–1641 1715–1731 1783–1794 1800–1850 1880–1884 1930–1934 1980–1983

E

A

85.11 77.08 80.00 89.74 77.31 85.48 91.80 91.78

14.89 22.92 20.00 10.26 22.69 14.52 8.20 8.22

Fig. 1: Auxiliary selection of monter (%).

The auxiliary selection of French monter ‘move upward’

289

3.3.2 Qualitative aspects 3.3.2.1 General remarks In section 3.1 three distinctions that are presumably relevant for auxiliary selection have been introduced: motion vs. non-motion, agentive- vs. non-agentive subjects, telic vs. atelic contexts. Further it has been shown that intransitive monter is compatible with both parts of all three dichotomies: it can be used as a motion and a non-motion verb, it can have agentive and non-agentive subjects and it can appear both in telic and atelic contexts. Concerning the qualitative aspects of the development I analyzed the impact of these distinctions on the auxiliary selection of monter (cf. section 1 for the exact research questions). In the analysis, each of the three distinctions was looked at from two perspectives: a context-centred perspective and an auxiliary-centred perspective. In the case of the context-centred perspective, it was investigated how often the two auxiliaries are selected in a given context (cf. 16). For example, all cases where monter is used as a motion verb were examined and it was verified how often the verb selects E in this context and how often A. In the case of the auxiliary-centred perspective I considered how often a given auxiliary appears in one of two possible contexts (cf. 17). For example, I looked at all the occurrences of E+monter and then verified how often the auxiliary appears in contexts where monter is a motion verb as opposed to the contexts where monter is a non-motion verb. Independently of the two perspectives, I also considered the diachronic development of the auxiliary selection of monter. (16) context-centred perspective

(17) auxiliary-centred perspective

3.3.2.2 Semantic role of the subject: agent vs. patient The discussion of the qualitative developments starts at the beginning of the sentence, namely with the subject. It has already been shown above that mont-

290

Steffen Heidinger

er can appear with agent-subjects or with patient-subjects; further examples are given in (18) and (19). (18) S = agent Pour lui, ravie, elleAG était montée à Paris. for him delighted she E.3.sg.pst go.up.pst.ptcp p Paris ‘For him, delighted, she had gone to Paris.’ (Navarre 1981; Frantext; modified) (19) S = patient monté à la tête. […] l’ alcoolPAT m’ est the alcohol me E.3.sg.prs go.up.pst.ptcp p the head ‘The alcohol went to my head.’ (Sartre 1983; Frantext; modified) From the context-centred perspective, the question is how often monter selects E and how often A if the subject of the sentence is an agent and whether the Tab. 3: Auxiliary selection of monter with an agent-subject (%).

1500–1560 1623–1641 1715–1731 1783–1794 1800–1850 1880–1884 1930–1934 1980–1983

E

A

86.05 74.19 82.50 90.91 80.91 90.70 98.00 100.00

14.63 25.00 17.39 12.12 18.14 6.00 5.26 0.00

Fig. 2: Auxiliary selection of monter with an agent-subject.

The auxiliary selection of French monter ‘move upward’

291

Tab. 4: Percentage of agent-subjects for E+monter and A+monter.

1500–1560 1623–1641 1715–1731 1783–1794 1800–1850 1880–1884/1851–1900* 1930–1934/1901–1950* 1980–1983/1951–2000*

E

A*

92.50 62.16 82.50 85.71 88.56 73.58 87.50 86.57

85.71 72.73 70.00 75.00 71.19 46.38 51.28 9.38

Fig. 3: Percentage of agent-subjects for E+monter and A+monter.

proportion changes diachronically.9 As shown by Figure 2, if the subject is an agent, E dominates over A at all times. The diachronic trend from the 16 th century to the present is that the percentage of E increases and the percentage of A decreases. There are also relevant changes with respect to the properties of the subject if one looks at the data from the auxiliary-centred perspective. Auxiliary-centred means to verify for E and A how often they appear with an agent-subject and how often with a patient-subject. Concerning E+monter, Figure 3 shows that the percentage of the use with agent-subjects remains relatively stable at

9 It would be desirable also to verify the auxiliary selection of monter if the subject is a patient. But since these cases are rather infrequent (for some spot checks n < 5) the comparison between E and A in this respect does not allow for a reliable analysis. The same holds for the use of monter in non-motion events and atelic contexts, where the choice of monter between E and A has also been studied only for motion events and telic contexts.

292

Steffen Heidinger

a high level over the whole period. Concerning A+monter, however, we see that the use with agent-subjects decreases dramatically towards present-day French, which in turn means an increase in patient-subjects.

3.3.2.3 Motion vs. non-motion The second distinction considered is about whether monter describes a motion event or not. Examples showing that monter can describe, but is not limited to, motion events are given in (20) and (21). (20) motion quand elle est montée en voiture when she E.3.sg.prs go.up.pst.ptcp p coach ‘when she got into the coach’ (Balzac 1847; Frantext) (21) non-motion (change of state) […] mais les salaires ont monté […] but the salaries A.3.pl.prs go.up.pst.ptcp ‘But the salaries increased’ (Sabatier 1980; Frantext) Beginning again with the context-centred perspective, Figure 4 shows how often monter selects E and A if it is used as a motion verb and how the percentage of E and A changes diachronically. Again, E dominates over A at all times and there is the same diachronic trend as in the case of the agent-subjects, namely that the percentage of E increases and that the percentage of A decreases towards present-day French.

Tab. 5: Auxiliary selection of monter as a motion verb (%).

1500–1560 1623–1641 1715–1731 1783–1794 1800–1850 1880–1884 1930–1934 1980–1983

E

A

86.05 75.00 81.40 88.89 78.49 84.75 91.53 97.10

13.95 25.00 18.60 11.11 21.51 15.25 8.47 2.90

The auxiliary selection of French monter ‘move upward’

293

Fig. 4: Auxiliary selection of monter as a motion verb.

Tab. 6: Percentage of monter as a motion verb for E+monter and A+monter.

1500–1560 1623–1641 1715–1731 1783–1794 1800–1850 1880–1884/1851–1900* 1930–1934/1901–1950* 1980–1983/1951–2000*

E

A*

92.50 81.08 87.50 91.43 98.01 94.34 96.43 100.00

85.71 90.91 80.00 100.00 91.53 81.16 84.62 37.50

Fig. 5: Percentage of monter as a motion verb for E+monter and A+monter.

From an auxiliary-centred perspective, there are also relevant changes. Auxiliary-centred means to verify how often E and A appear with monter as a motion

294

Steffen Heidinger

verb and how often with monter as a non-motion verb. Concerning E+monter, Figure 5 shows that the percentage of the use with monter as a motion verb is always at a very high level, but still increases towards present-day French. For the combination of A+monter, the use with monter as a motion verb decreases towards present-day French, which in turn means an increase of the uses with monter as a verb of change of state.

3.3.2.4 Telic vs. atelic event The last distinction to be considered is the one between telic and atelic contexts. Examples of the use of monter in a telic and an atelic context are given in (22) and (23). Note that these examples are not the typical examples of telic and atelic contexts in my data: a typical telic event is one with monter as a motion verb where the motion is delimited by a locative adjunct. In (22) however, the motion event is not delimited by a locative adjunct but by the fact that the verb means ‘come in’ and the event is thus delimited in the sense that it is over once the subject has entered the respective location (which in this case is only contextually given). The typical atelic event shows monter as a verb of change of state with the meaning ‘increase’ and without any hint that the increase is bound. In the atelic context in (23), however, monter describes a motion event, but the motion is not delimited; neither overtly by means of a locative adjunct nor contextually. Crucially, what (23) is about is not the achievement of a certain locus. Rather, it focuses on how the subjects carried out the event denoted by monter. (22) telic context il est monté tout à l’heure pour réclamer ses loyers; he E.3.sg.prs go.up.pst.ptcp just p claim his rent ‘He just came in to claim the rent.’ (Muger 1848; Frantext) (23) atelic context […] haletants tous deux parce qu’ils avaient monté breathless all two because they A.3.pl pst go.up.pst.ptcp trop vite. too fast ‘both of them breathless, because they had come up too fast’ (Duhamel 1941; Frantext)

The auxiliary selection of French monter ‘move upward’

295

Tab. 7: Auxiliary selection of monter in telic contexts (%).

1500–1560 1623–1641 1715–1731 1783–1794 1800–1850 1880–1884 1930–1934 1980–1983

E

A

85.37 75.00 82.61 87.88 81.86 94.00 94.74 100.00

14.63 25.00 17.39 12.12 18.14 6.00 5.26 0.00

Fig. 6: Auxiliary selection of monter in telic contexts.

Tab. 8: Percentage of monter as a telic verb for E+monter and A+monter.

1500–1560 1623–1641 1715–1731 1783–1794 1800–1850 1880–1884/1851–1900* 1930–1934/1901–1950* 1980–1983/1951–2000*

E

A*

87.50 89.19 95.00 82.86 96.52 88.68 96.43 94.03

85.71 100.00 80.00 100.00 72.88 52.17 46.15 37.50

As concerns the context-centred perspective, Figure 6 shows how often monter selects E and A if it is used in a telic context. E clearly dominates over A at all

296

Steffen Heidinger

Fig. 7: Percentage of monter as a telic verb for E+monter and A+monter.

times and the dominance of E increases towards present-day French (conversely the percentage of A decreases). Turning to the auxiliary-centred perspective, the data also show relevant diachronic changes (cf. Table 8 and Figure 7). Concerning the combination of monter and E, Figure 7 shows that telic contexts clearly predominate and the percentage of these remains relatively stable at a high level over the whole period. Concerning monter and A, telic contexts strongly decrease towards present-day French (which in turn means that the atelic contexts increase).

3.3.3 Discussion In section 1, the main goal of this study was formulated as follows: to describe the overall development of the auxiliary selection of French monter from the 16 th to the 20 th century and to verify the relevance of the three above-mentioned semantic features in this development. Concerning the first issue, E clearly dominates over A during the whole period of investigation and there is a slight increase in the frequency of E towards present-day French. Regarding the second issue, the relevance of the semantic features [±agent], [±telic] and [±motion] is reflected in our data in several ways. First, we have seen that the frequency with which monter selects the two auxiliaries in certain contexts changes over time. The three contexts we looked at were [+agent], [+telic], and [+motion]. In all three contexts, the frequency of E clearly increases at the expense of A, to a maximum of 97.10 % in the case of [+motion] and 100 % in the case of [+agent] and [+telic]. Second, as a consequence of the increase in E in the above-mentioned semantic contexts, the predictive power of the features [+agent], [+telic], and [+motion] for the use of

The auxiliary selection of French monter ‘move upward’

297

E also increases during the period under investigation.10 Third, with regard to the development of the auxiliary A, the binary features [±agent], [±telic], and [±motion] have proven relevant as they show that the contexts in which the auxiliary appears have changed considerably from the 16 th to the 20 th century. While in the beginning of the period under investigation A appeared predominantly in the same contexts as E – i.e. with agent subjects, in motion events, and in telic contexts – the situation is completely different in present-day French: A appears in the majority of cases with patient subjects, in atelic contexts, and describing non-motion events. Finally, the consideration of the semantic features may also shed light on the overall quantitative development of the auxiliary selection of monter, i.e. the increase in the frequency of selection of E. Crucially, the contexts in which the frequency of E increases at the expense of A ([+agent], [+telic], and [+motion]) are those contexts in which the verb monter is found more often (as opposed to ([‒agent], [‒telic], and [‒motion]).11 We may thus conclude that the overall increase in E is driven by its increase in the contexts in which monter most frequently occurs. The case of monter thus shows that conserving frequency effects (where items with a high usage frequency resist a linguistic change longer than items with a lower usage frequency; cf. Bybee 2006) may be observed within the different usage contexts of a single lemma, and not only when comparing different lemmas with different usage frequencies.12

10 With respect to their predictive power for the selection of E (as opposed to A), a logistic regression model yields the following order among the factors: [+telic] > [+agent] > [+motion] (calculated based on the data for the spot checks 1500–1560, 1623–1641, 1715–1731, 1783–1794, 1800–1850, 1880–1884, 1930–1934, and 1980–1983; rank discrimination indexes are C = 0.648 and Somers’ Dxy = 0.296). Although all three factors favour the use of E, only [+telic] and [+agent] show statistically significant effects (< 0.05). (I am grateful to Malte Rosemeyer for his help with the statistics.) The lack of a statistically significant effect for [+motion] may be explained as follows: the class of [+motion] not only includes uses of monter as a change-oflocation verb, but also as a manner-of-motion verb with the meaning ‘to ride (a horse)’; in this latter use of monter, which is especially frequent in the older data, there is a preference for the selection of A. An annotation of the data using [±change-of-location] instead of [±motion] would most probably yield a significant effect of [+change-of-location] for the selection of E. 11 The distribution of monter in the relevant semantic contexts is as follows: [+agent] vs. [‒agent] = 80.94 % vs. 19.06 %, [+telic] vs. [‒telic] = 89.22 % vs. 10.78 %, [+motion] vs. [‒motion] = 93.75 % vs. 6.25 %; these values are calculated based on the data for the spot checks 1500–1560, 1623–1641, 1715–1731, 1783–1794, 1800–1850, 1880–1884, 1930–1934, and 1980–1983. 12 Rosemeyer’s (2014: section 7.3, under review) work on the development of Spanish auxiliary selection has revealed a conserving effect related to semantic verb classes with different frequency of use: the more frequent class of change-of-location verbs preserved the declining auxiliary ser ‘be’ longer than the less frequent class of change-of-state verbs.

298

Steffen Heidinger

4 Conclusions In this paper I have presented a corpus-based description of the development of the auxiliary selection of monter from the 16 th to the 20 th century. The quantitative development is that the frequency of E increases at the expense of A. Given that the overall development of E and A with French intransitive verbs is that A spreads at the cost of E (cf. section 2, and the literature cited there), the attested development of monter seems at first glance rather exceptional. However, once certain qualitative aspects are considered, i.e. the semantic contexts in which the two auxiliaries are chosen, the development of monter can also be interpreted in the context of the overall development. Crucially, the contexts in which the role of E is strengthened are typical E-contexts in Sorace’s (2000) auxiliary selection hierarchy in that monter is used as a verb of change of location.13 Although the development of monter runs against the overall development in French, the contexts where E resists the spread of A are the ones where we would most expect it. The development of monter thus shows the consolidation of a decreasing form in a restricted domain. The decreasing form is E as a compound tense auxiliary in French. The restricted domain is that the sentence has an agent-subject and the verb is used as a motion verb and in a telic context. The consolidation is shown by the fact that towards present-day French, E+monter becomes more frequent in this restricted domain, while the use of A+monter decreases. The mirror image of the consolidation of E is that A decreases in domains that it has already conquered (at least partially). The decrease of A in these contexts thus shows that developments in this domain of grammar are not necessarily unidirectional.

13 Concerning the development of Spanish auxiliary selection, Rosemeyer (2014: 186–187, under review) shows convincingly that one important factor in the spread of Sp. haber ‘have’ at the expense of ser ‘be’ is the similarity to the original usage context of haber; ser survives the longest in contexts which are most distinct from the original usage contexts of haber. In the case of Fr. monter this principle seems to apply only partially. As we have seen, agentive subjects are a factor that favours the use of E over A despite the fact that agentive subjects form a similarity to the original usage context of A (as an auxiliary for transitive verbs). The fact that in the case of monter agentive subjects are not a factor favouring A is in line with Sorace’s (2000: 864) observation that the preference of verbs of change of location for the selection of E holds independent of the semantic role of the subject.

The auxiliary selection of French monter ‘move upward’

299

References Benzing, Josef. 1931 Zur Geschichte von ser als Hilfszeitwort im Spanischen. Zeitschrift für Romanische Philologie LI: 385–460. Burzio, Luigi. 1986 Italian Syntax. Dordrecht: Reidel. Bybee, Joan L. 2006 From usage to grammar: the mind’s response to repetition. Language 82: 711–733. Cennamo, Michela and Antonella Sorace. 2007 Auxiliary selection and split intransitivity in Paduan: variation and lexical-aspectual constraints. In: Raúl Aranovich (ed.), Split Auxiliary Systems. A Cross-linguistic Perspective, 65–99. Amsterdam, Philadelphia: John Benjamins. Clédat, Léon. 1903 Le participe passé, le passé composé et les deux auxiliaires. Revue de Philologie Française XVII: 19–62. Förster, August. 1908 Avoir und être als Hilfsverba beim intransitiven Zeitwort in ihrer Entwicklung vom Alt- zum Neufranzösischen. Darmstadt: Otto. Fournier, Nathalie. 1998 Grammaire du français classique. Paris: Belin. Frantext = ATILF (CNRS, Université Nancy 2) Base textuelle de Frantext. Nancy. www.frantext.fr. Giancarli, Pierre-Don. 2011 Les auxiliaires ÊTRE et AVOIR: Étude comparée corse (esse/avè), français de France (être/avoir), acadien (être/aouèr) et anglais (be/have). Rennes: Presses Univ. de Rennes. Gougenheim, Georges. 1973 Grammaire de la langue française du seizième siècle. Paris: Picard. Grevisse, Maurice and André Goosse. 142008 Le bon usage: grammaire française. Bruxelles: De Boek & Larcier. Haegeman, Liliane. 21994 Introduction to Government & Binding Theory. Oxford: Blackwell. Heidinger, Steffen. 2010 French Anticausatives: a Diachronic Perspective. Berlin, New York: De Gruyter. Hopper, Paul and Sandra A. Thompson. 1980 Transitivity in grammar and discourse. Language (56): 251–299. Kailuweit, Rolf. 2011 Le choix de l’auxiliaire: Être ou avoir en français standard contemporain. In: Renata Enghels, Machteld Meulleman and Clara Vanderschueren (eds.), Peregrinatio in Romania: Artículos en homenaje a Eugeen Roegiest con motivo de su 65 cumpleaños, 397–420. Gent: Academia Press. Labelle, Marie. 1992 Change of state and valency. Journal of Linguistics 28(2): 375–414. Legendre, Geraldine and Antonella Sorace. 2003 Auxiliaires et intransitivité en français et dans les langues romanes. In: Danièle Godard (ed.), Les langues romanes; problèmes de la phrase simple, 185–234. Paris: Editions du CNRS. Levin, Beth and Malka Rappaport Hovav. 1995 Unaccusativity: at the Syntax-Semantics Interface. Cambridge, MA: MIT Press. Loporcaro, Michele. 2007 On triple auxiliation in Romance. Linguistics 45(1): 173–222. Mackenzie, Ian E. 2006 Unaccusative Verbs in Romance Languages. Basingstoke: Palgrave Macmillan. Manente, Mara. 2008 L’Aspect, les auxiliaires ‘être’ et ‘avoir’ et l’hypothèse inaccusative dans une perspective comparative français/italien. Ph.D. thesis, Université Paris 8 / Università Ca’ Foscari Venezia.

300

Steffen Heidinger

Manente, Mara. 2013 Variabilité aspectuelle et influence analogique dans le choix de l’auxiliaire avec les verbes intransitifs de changement d’état monter et descendre. Journal of French Language Studies 23(3): 423–434. McFadden, Thomas. 2007 Auxiliary selection. Language and Linguistics Compass 1(6): 674– 708. McFadden, Thomas and Artemis Alexiadou. 2010 Perfects, resultatives, and auxiliaries in earlier English. Linguistic Inquiry 41(3): 389–425. Oudin, Antoine. 1640 Grammaire françoise, rapportée au langage du temps. Paris: A. de Sommaville. www.archive.org, retrieved 13. 4. 2012. Perlmutter, David M. 1978 Impersonal passives and the unaccusative hypothesis. Proceedings of the Berkeley Linguistics Society 4: 157–189. Perlmutter, David M. 1989 Multiattachment and the Unaccusative Hypothesis: The perfect auxiliary in Italian. Probus 1: 63–119. Rosemeyer, Malte. 2014 Auxiliary Selection in Spanish. Gradience, Gradualness, and Conservation. Amsterdam, Philadelphia: John Benjamins. Rosemeyer, Malte. Under review Tracing the conserving effect of frequency. In: Heike Behrens and Stefan Pfänder (eds.), Again on Frequency. Effects in Language. Berlin, New York: De Gruyter. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76: 859–890. Sorace, Antonella. 2004 Gradience at the lexicon-syntax interface: evidence from auxiliary selection. In: Artemis Alexiadou, Elena Anagnostopoulu and Martin Everaert (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 243–268. Oxford: Oxford University Press. TLFi = ATILF, CNRS and Université de Lorraine (eds.) Le Trésor de la Langue Française informatisé; www.atilf.fr.

Malte Rosemeyer, Albert-Ludwigs-Universität Freiburg

Entrenchment and discourse traditions in Spanish auxiliary selection 1 Introduction Usage-based linguistics (UBL) assumes that frequency of use has a crucial influence on the development of linguistic elements. Among other frequency effects, Bybee (2006) argues for a conserving effect of frequency. Thus, a linguistic element with a high token frequency is subject to entrenchment: the repeated activation of the mental representation of that element leads to a strengthening of its mental representation. In language change, entrenchment is said to counteract processes of language change. In situations where a linguistic element is in the process of being replaced by a second, more productive linguistic element, high-frequency instantiations of the replaced element will resist the change longer than low-frequency instantiations. As a cognitivist approach to language, UBL is mostly concerned with the cognitive foundations causing the frequency effects observed in language use and language change. However, when applying the notion of entrenchment to the study of language change, a problem with this cognitivist perspective immediately arises: the context-dependency of entrenchment. It is unclear whether entrenchment has to be modelled as a function of the global token frequency of an instantiation of a construction, or as the token frequency of that instantiation in specific linguistic contexts. From a methodological perspective, diachronic studies of entrenchment thus always face the problem of which token frequencies to count. This problem could potentially cause huge biases in these studies. In a recent paper, Kabatek (2013) discusses the question of how to conduct quantitative studies in diachronic corpora and puts special emphasis on the problem of discourse traditions (Schlieben-Lange 1983; Koch 1997). A language as represented in historical texts is not a system but a conglomerate of several systems, each corresponding to a historically determined way of text organisation, i.e. a discourse tradition. Diachronic studies of language change tend to neglect this influence of text organisation on the use of language in these texts and consequently assume an ideal, generalised language that does not exist. Kabatek notes that diachronic studies of entrenchment commit the same error, since they typically consider the product of a language (i.e. the texts) to be the language itself (Kabatek 2013: 20). If the rules of language use are specifically bound to discourse traditions, entrenchment should also be dependent on discourse traditions.

302

Malte Rosemeyer

In this paper, I develop a new perspective on the question of frequency effects in language change. I propose that entrenchment not only varies according to discourse traditions, but also that the very notion of entrenchment is inseparable from discourse traditions. I suggest that entrenchment plays a pivotal role in the creation of discourse traditions and that in turn, discourse traditions enable entrenchment. In order to demonstrate the relevance of these assumptions, I conduct a quantitative analysis of the development of Spanish auxiliary selection in three different discourse traditions: historiographical texts, administrative documents and private letters. The findings from the study support earlier studies which claim that the disappearing ser + PtcP construction was conserved with verbs that have a high token frequency – in particular, verbs that express change-of-location predicates. In addition, they show that the strength of this conserving effect varies according to the discourse tradition. As a result, the analysis demonstrates that it is indeed possible to study entrenchment in historical texts if close attention is paid to the question of discourse traditions. This paper is structured as follows. In section 2, I elaborate on the notion of the context-dependency of entrenchment processes and consequently, conservation in language change. In section 3, I give a brief overview of recent findings on the development of auxiliary selection in Spanish. After an introduction to the data used in this study (section 4), I conduct a quantitative analysis of auxiliary selection in these data (section 5). The final section 6 discusses the relevance of these findings for studies of frequency effects and discourse traditions.

2 Discourse traditions and entrenchment The term discourse tradition was coined in Romance linguistics and usually refers to historical conventionalised patterns of text production (SchliebenLange 1983; Koch 1997; Oesterreicher 1997; Jacob and Kabatek 2001; Wilhelm 2001: 467; Aschenberg 2003: 5; Kabatek 2005a, 2005b; Schrott and Völker 2005; Kabatek 2013). As an example, consider the following quote from Jacob and Kabatek (2001: viii): [Discourse traditions] are historical-normative patterns that have been established in a society and which speakers follow in the production of discourse. The existence of these categories means that each discourse, and consequently, each historical text, is not only part of a specific language (of several languages) but is rather situated in an intertextual succession constituted from a series of repeating elements. These elements refer to both the level of the “context” (situational, medial or institutional constellations) and the level

Entrenchment and discourse traditions in Spanish auxiliary selection

303

of forms on the text surface (e.g. specific text passages, formulaic use of constructions and language) (Jacob and Kabatek 2001: viii, translation MR).

Studies on discourse traditions take a genuinely pragmatic approach to the description of language: they claim that the use of a linguistic element is to a great extent determined by its usage context. One and the same linguistic element can have a different function in different discourse traditions. This consideration has important implications, as differences in discourse traditions can govern linguistic phenomena as different as polysemy (Blank 2005), the function of grammatical elements such as verbal periphrases (Jacob 1994, 2001), or the use of politeness markers (Held 2005). The discourse tradition approach to the description of language can be related to the increasing recognition in historical linguistics that language change is intricately related to discourse contexts. Many historical linguists claim that the discourse context plays an instrumental role in linguistic innovation processes. For instance, recent studies on grammaticalisation processes emphasise the importance of the discourse context for the reanalysis of a linguistic element and consequently, the emergence of a new grammatical meaning (Diewald 2002; Heine 2002; Diewald 2006). If a grammatical element is used in a novel discourse context, its grammatical meaning changes to fit the discourse context. This idea basically corresponds to the notion that the reanalysis of a grammatical element in a specific context is caused by a pragmatic implicature (Dahl 1985: 10–11). Grammaticalisation can then be described as the conventionalisation of this conversational implicature (Traugott 1989; Traugott and König 1991; Hopper and Traugott 2003: 81–93). As suggested by this parenthesis on the importance of discourse traditions for the description of mechanisms of language change such as grammaticalisation, the notion of discourse traditions has a direct impact on the concept of how language change works. Kabatek (2013: 15–16) argues that if we accept the idea of discourse traditions, we must also take seriously the fact that there are distinctions between (a) a language system and a text and (b) language systems that serve as the basis for the generation of texts and completed texts. Kabatek’s first point emphasises that a language is a much less homogeneous entity than a text, given that a language can be characterised as a conglomerate of several systems characterised by factors such as diatopic, diastratic and diaphasic variation. However, historical linguists often neglect this problem and study a text as if it were representative of “the language” used in that time period. The second point addressed by Kabatek is the core assumption of the study of discourse traditions: the inherent intertextuality of language use. Thus, all levels of the use of a language system – grammatical, phonetic and semantic rules – are intrinsically historically determined. In Kabatek’s words,

304

Malte Rosemeyer

“every text, from the salutation to the novel, evokes anterior texts, even if it does not repeat elements belonging to these texts” (Kabatek 2013: 16, translation MR). It follows that it is impossible to argue for an abstract development of a language distinct from the development within discourse traditions: discourse traditions or text genres do not merely modify language use like a sociolinguistic variable, for instance, but directly determine language use. There are interesting parallels between the perspective on language use drawn by researchers on discourse traditions and researchers working in the paradigm of usage-based linguistics. In the following, I give a brief introduction to UBL and in particular, its model of the conserving effect of frequency in order to demonstrate these similarities. UBL claims that the speakers’ experience with language has a direct influence on their language use (Bybee and Hopper 2001; Bybee 2006; Diessel 2007, 2011; Pfänder et al. 2013). The frequency with which speakers perceive a linguistic element alters the form of the linguistic element, causing frequency effects such as chunking (Newell 1990: 185–193; Ellis 1996; Bybee 2010: 33–56). When modelling chunking, many UBL approaches make use of exemplar theory (Medin and Schaffer 1978; Kuhl 1994; Lacerda 1995; Johnson 1997; Pierrehumbert 2001, 2002). In essence, exemplar theory is a theory of categorisation. It proposes that when a person perceives a novel stimulus, s/he will categorise this stimulus according to its similarity to other stimuli that s/he has previously encountered. This similarity relationship determines the way these exemplars are stored in the mind: in the storage process, similar exemplars are stored proximally to each other, whereas dissimilar exemplars are stored less proximally to each other. From a neurological point of view, this locationist metaphor can be understood as specifying the strength of activation of the links between the stored exemplars. As a result, exemplar theory proposes that categories be represented as clusters of specific tokens perceived by a person that in one way or another resemble each other. The categories formed by exemplars are thus gradient. Similarity between exemplars as modelled in UBL can also be a function of context. In particular, a repeated co-occurrence of two linguistic elements corresponds to a similarity between these elements in terms of situational contexts (trivially, these elements are perceived at two points in time that are close to one another). The repeated simultaneous activation of two linguistic elements will therefore lead to a strengthening of the link between the exemplars. This process results in chunking: Due to the repeated simultaneous activation of two linguistic elements, the speaker will start conceptualising these two elements as forming a chunk, i.e. a new category. Chunking results in entrenchment (Langacker 1987: 59), i.e. an improvement of both accessibility and

Entrenchment and discourse traditions in Spanish auxiliary selection

305

production of the chunked element. Important evidence for the psychological reality of chunking can be found in studies of L1 acquisition. These studies suggest that chunking may be a first step in the creation of grammatical rules, given that chunking leads to the creation of a complex hierarchical structure between the chunked elements and the resulting chunk (Tomasello 1992; Lieven, Pine, and Baldwin 1997; Goldberg 2006; Behrens 2009). The notions of chunking and entrenchment are crucial for the explanation of the conserving effect of frequency (Bybee 2006). When subject to chunking, a morphologically complex form will typically decrease in analysability. In Bybee’s words, “the more a sequence of morphemes or words is used together, the stronger the sequence will become as a unit and the less associated it will be to its component parts” (Bybee 2010: 48). As an example, consider Poplack’s (2001) study of subjunctive mood in Canadian French. Although in Canadian French the subjunctive is currently being replaced with the indicative, the raw numbers of use would appear to indicate that the subjunctive is still very frequent. Within the envelope of variation isolated by Poplack, the subjunctive is used in 77 per cent of the cases (Poplack 2001: 411). Poplack, however, demonstrates that this numerical dominance of the use of subjunctive is misleading, since the productivity of the subjunctive is restricted to a very small but frequent group of both matrix verbs and embedded verbs (Poplack 2001: 412). Although in theory any verb can be used in subjunctive mood, in Poplack’s data only four verbs occur regularly with the subjunctive: être (‘be’), aller (‘go’), avoir (‘have’) and faire (‘do’). It is evident that these verbs have a very high overall token frequency. Consequently, Poplack claims that although the token frequency of the standard subjunctive variant is elevated, virtually all its uses are concentrated among a handful of highly favoring matrix verbs collocated with a small cohort of frequent and irregular embedded verbs. Outside of these few contexts, in which its use has become ritualised, selection of the subjunctive is very rare (Poplack 2001: 414).

Due to the high token frequency of specific instantiations of the complex subordinating construction (for instance, falloir + que + subjunctive), these syntagms are subject to an entrenchment process. This entrenchment process leads to a loss of internal analysability. In other words, the high-frequency instantiations of the construction start forming a category on their own. This means that over time, their productivity gets less bound to the productivity of the construction of which they are instantiations. As a result, the high-frequency syntagms are affected less by the replacement process operating on the Canadian French subjunctive. This is the conserving effect of frequency. Sections 3 and 5 of this paper will demonstrate a scenario very similar to the one suggested by Poplack for the development of Spanish auxiliary selection.

306

Malte Rosemeyer

When comparing the descriptions of the discourse-traditional and the usage-based approach to language change given in this section, striking similarities between the two approaches arise. In particular, both approaches assume that the experience of language users with language has a direct influence on the way linguistic elements are employed. This overlap is evident in the importance of the notions of conventionalisation and intertextuality for the two research approaches. First, both approaches assume that the co-occurrence of linguistic elements is conventionalised in language use over time. The crucial factor for this conventionalisation process is repeated exposure. Indeed, the emergence of discourse traditions can be conceptualised as a type of entrenchment on a higher level of linguistic complexity: the repeated use of a specific constellation of linguistic elements in a specific discourse situation will lead to the strengthening of the connections between the mental representations of these linguistic elements. Second, the UBL and discourse tradition approach share a focus on intertextuality as developed in post-structuralism (Barthes 1967; Foucault 1988 [1969], 2002 [1969]). These studies put into perspective the role of the author in text production. If any text “is made of multiple writings, drawn from many cultures and entering into mutual relations of dialogue, parody, contestation” (Barthes 1967), the author plays a less innovative role than commonly assumed. Rather, texts are the product of preceding texts. As already noted in Kabatek’s (2013) discussion of the discourse tradition approach, intertextuality is a core assumption of the discourse tradition approach. Intertextuality is also crucial for UBL. The exemplar model of language production assumes that in the production of a linguistic element, the language user necessarily evokes earlier language events involving this linguistic element. Entrenchment only differs from this “regular” model of language production in that it cancels out the abstraction processes involved in linguistic productivity and thus leads to the explicit copying of the entrenched linguistic element. This means that entrenchment can be conceptualised as a process by which the use of a linguistic element comes to increasingly depend on intertextuality. This discussion of the notions of conventionalisation and intertextuality has demonstrated broad similarities between the discourse tradition approach to language and UBL. I have suggested that discourse traditions can be understood as a kind of “macro-entrenchment”: the repeated co-occurrence of a constellation of linguistic elements leads to the conventionalisation of this constellation and consequently, a discourse tradition. In turn, the UBL notion of entrenchment is intrinsically intertextual in nature and can be understood as a “micro-discourse tradition”: just as the use of a linguistic element within a

Entrenchment and discourse traditions in Spanish auxiliary selection

307

certain discourse tradition depends on the previous use of that element in that discourse tradition, the use of an entrenched linguistic element depends on the repeated previous use of that linguistic element. The identification of the similarities in the approaches described in this section offers a potential solution to a problem that is highly relevant to historical linguistics, i.e. the problem of the “accumulative episteme” (Jacob 2001) of medieval writing. Medieval authors typically take their material from several sources, often copying entire text passages. With regard to Spanish, this problem can be illustrated clearly for the discourse tradition of historiographical texts. The historiographical texts attributed to Alfonse X are composed of passages taken from works by classical Latin authors such as Lucan, Ovid and Pliny the Elder; religious texts like the bible; texts written by Eusebius of Cesarea; as well as French and Arabian literature and epic poems in Romance (Fernández-Ordóñez 2004: 390). This method of text production is indeed a defining property of medieval literacy (Burke 2004: 97). In the case of Spanish historiographical texts, however, it continues well into the 17 th century: Many of the later historiographical texts both explicitly adopt the model of the Alfonsine chronicles and copy from them (Von Hoegen 2000: 30). Indeed, this indicates that these Spanish texts belong to the same discourse tradition. It also constitutes a problem because many of the tokens from these texts are copies from earlier tokens. For this reason, Jacob (2001: 156) argues that these tokens are not representative of the language of the writer and consequently, that quantitative analyses of medieval texts are impossible to conduct. From the perspective developed in this section, however, the phenomenon of explicit copying is not qualitatively different from other mechanisms that lead to the constitution of a discourse tradition. In other words, within the Spanish discourse tradition of historiographical texts, these texts refer to previous texts both implicitly (e.g. by topic and narrative style) and explicitly (explicit copies). Effectively, copied tokens mirror the principle of intertextuality. Likewise, an interaction between entrenchment and copying can be assumed. During the process of copying, old-fashioned linguistic forms are sometimes modernised. Evidence of this phenomenon is found in Rodríguez Molina (2006b). He demonstrates that when copying tokens belonging to the ser + PtcP construction in medieval texts, the copiers often replace these tokens with the more modern haber + PtcP (cf. section 3). The discussion of entrenchment in this chapter suggests that these replacement mechanisms might not be aleatory. Rather, it can be assumed that entrenchment has an important effect on whether or not a linguistic element is modernised when copied: more entrenched elements may be less susceptible to modernisation than less en-

308

Malte Rosemeyer

trenched elements. This is due to (a) the strengthened mental representation of these entrenched elements and (b) the fact that arguably, entrenched linguistic elements are more likely to be perceived as typical of the discourse tradition. The theoretical perspective developed in this section thus suggests that there is no principled distinction between the mechanism of entrenchment in spoken and written discourse traditions. Just as speakers use entrenched linguistic elements because of repeated exposure to these elements in spoken language, writers use entrenched linguistic elements due to repeated exposure to these elements in specific discourse traditions in written language. Intertextuality follows similar rules in written language as in spoken language. If there is a difference regarding entrenchment in written and spoken language, it is quantitative in nature. In particular, this is due to differences in the way discourse traditions in written and spoken language are constituted. Spoken discourse traditions are mainly determined by factors regarding the situational context. For instance, due to the use of stereotypical and recurrent patterns, it is licit to characterise the dialogues between customer and clerk at grocery store counters as a discourse tradition. This discourse tradition is basically determined by the situational context: The speakers do not situate their speech in the discourse tradition in order to explicitly refer to earlier language usage events. Rather, entrenchment mainly takes place on a subconscious level. In contrast, references to earlier usage events in written discourse traditions are caused by more deliberate processes: Writers sometimes might reflect on the fact that within their discourse tradition, a certain linguistic element is entrenched, and often deliberately use this element in order to follow the stylistic prescripts of that discourse tradition. Due to the fact that written texts persist longer in time than spoken texts, this also means that archaisms are more likely to occur in written language than in spoken language. As a result, written discourse traditions are typically expected to be more conservative than spoken discourse traditions. Crucially, this assumption also leads to a sub-division within written texts. Thus, some written texts share more characteristics with spoken texts than others. Following Koch and Oesterreicher’s terminology, some texts are characterised by communicative proximity, whereas other texts are characterised by communicative distance (Koch and Oesterreicher 1985, 1990, 2001). As illustrated by the grocery store example, it is this parameter which governs the conservativism of some discourse traditions in relation to other discourse traditions. Summarising these considerations, I establish two hypotheses: 1. The cognitive mechanism of entrenchment plays a pivotal role in the emergence of discourse traditions both in the written and the spoken modality.

Entrenchment and discourse traditions in Spanish auxiliary selection

2.

309

By repeatedly using elements that are typical of a certain discourse situation, a writer or speaker signals that her text belongs to a specific discourse tradition. Conversely, entrenchment can be conceptualised as a process by which a linguistic element increasingly comes to depend on intertextuality. This means that discourse traditions have an enabling function for entrenchment: Entrenchment operates within specific discourse traditions. Discourse traditions characterised by communicative distance display a more conservative language use than discourse traditions characterised by communicative proximity.

The next section of this paper will illustrate the relevance of these hypotheses for the example of Spanish auxiliary selection.

3 Auxiliary selection in Spanish In contrast to Modern Spanish, Old Spanish possessed two auxiliaries for the formation of the verbal periphrases commonly termed compound tenses in intransitive sentences: ser (‘be’) and haber (‘have’). The two construction types are illustrated in (1) and (2). In contrast, haber is used invariably in transitive sentences (3).1 (1) perdonamus a don Marcus e a los sobreditos cavalleros pardon.prs.1pl to Don Marcus and to the abovementioned knights el daño que nos es venido por la mala the damage that to.us be.prs.3sg come.ptcp.m.sg because.of the bad voz que nos pusieron voice that to.us put.pst.pfv.3sg ‘We forgive Marcus and the above-mentioned knights the damage caused by their attacks on our reputation.’ (Carta mediante la cual don Marco de Artieda y otros muchos piden perdon al Monasterio de San Salvador y este se lo concede, March 1262, apud CODEA)

1 Sources marked in capital letters correspond to the following corpora: CODEA = Corpus de Documentos Españoles Anteriores a 1700 (Grupo de Investigación de Textos para la Historia del Español 2012), BIBLE = Biblia Medieval (Enrique-Arias 2008), LETTERS = Cartas de Particulares en Indias de Siglo XVI (Fernández Alcaide 2009), HIST = Corpus de la selección de auxiliares en textos historiográficos (Rosemeyer 2014). The HIST corpus mainly consists of tokens extracted from the Corpus Diacrónico del Español (Real Academia Española 2012). For a more detailed description of the corpora, cf. section 4.

310

Malte Rosemeyer

(2) porque Sancho Pérez avía venido e because Sancho Pérez have.pst.ipfv.3sg come.ptcp.m.sg and entrado en la iglesia enter.ptcp.m.sg in the church ‘Because Sancho Pérez had come and entered the church’ (Carta mediante la cual Joán Gómez encomienda a Pascual Martínez la iglesia de Alhama y le entrega las llaves de dicha iglesia, September 13, 1417, apud CODEA) (3) todas las mejorías, e aprovechamientos e edificamientos all the improvements and improvements and constructions que en las dichas casas […] oviéredes that in the say.ptcp.f.pl houses have.pst.ipfv.sbj.2pl fecho, e labrado e mejorado make.ptcp.m.sg and work.ptcp.m.sg and improve.ptcp.m.sg ‘All the improvements that you have made in the said houses’ (Carta de venta de unas casas de Catalina Rodríguez a Juan Sánchez y Elvira Sánchez, August 29, 1447, apud CODEA) The variation between ser and haber with intransitive verbs is to a great degree determined by verb semantics (Benzing 1931; Yllera 1980). Ser + PtcP is typically used with intransitive verbs that express a change of state or a change of location (4–5), whereas haber + PtcP is typically used with intransitive verbs that express states or activities (6–7). This means that in Old Spanish texts, examples such as (2) are rather atypical. (4) cuerpo de ombre | que maten o que sea body of man that kill.prs.sbj.3pl or that be.prs.sbj.3sg por si muerto because.of refl die.ptcp.m.sg ‘Corpse of a man who has been killed or who has died on his own’ (Bible E8/6, 13 th c., apud BIBLE) (5) Et eillos aun, no heran llegados, qua|ndo and they yet not be.pst.ipfv.3pl arrive.ptcp.m.pl when vino al rey mandado que absalon come.pst.pfv.3sg to.the king message that Absalon matara a todos los hijos kill.pst.ipfv.sbj.3sg to all the sons ‘And they had not yet arrived, when the King received a message which said that Absalon had killed all of the sons.’ (Bible E8/6, 13 th c., apud BIBLE)

Entrenchment and discourse traditions in Spanish auxiliary selection

311

(6) assaz auedes estado en esta montanna enough have.prs.2pl stay.ptcp.m.sg in this mountain ‘You have stayed long enough on this mountain.’ (Bible E8/6, 13 th c., apud BIBLE) (7) quando el ouo fablado folgo el when he have.pst.pfv.3sg speak.ptcp.m.sg rejoice.pst.pfv.3sg the espiritu deillo spirit of.him ‘When he had spoken, he was happy.’ (Bible E8/6, 13 th c., apud BIBLE) This distinction in the distribution of the auxiliaries in older stages of Spanish has been explained as (a) a universal contrast between unaccusative and unergative verbs (Elvira González 2001), (b) an interaction between argument structure and verbal aspect (Aranovich 2003; Mateu 2009) and (c) a functional contrast between the resultative ser + PtcP construction and the perfective haber + PtcP construction (Rodríguez Molina 2006a; Rosemeyer 2012, 2014; Mateu and Massanell this volume). Since this paper is concerned with the development of Spanish auxiliary selection, I will not discuss this issue here but refer the reader to the studies mentioned above. After the first half of the 15 th century, writers start replacing the ser + PtcP construction with the haber + PtcP construction in active intransitive sentences. In texts dating back to the end of the 17 th century, ser + PtcP has mostly disappeared from these usage contexts. Especially since the beginning of the 16 th century there are many examples of the use of haber + PtcP in intransitive contexts expressing a change of state or change of location: (8) pues no me e muerto do muchas since not refl have.prs.1sg die.ptcp.m.sg give.prs.1sg many graçias a mi dios thanks to my God ‘I thank God because I have not died.’ (De Alonso de Herrojo a su mujer, Teresa González, vecina de Reina (Badajoz), March 10, 1583, apud LETTERS) (9) auia nueue meses que se auia have.pst.ipfv.3sg nine months that refl have.pst.ipfv.3sg ydo go.ptcp.m.sg ‘Nine months ago he had gone away.’ (De Celedon Favalis a su padre, Simón Favalis, en Madrid, March 20, 1587, apud LETTERS)

312

Malte Rosemeyer

(10) Vos, señor, habéis crecido mucho you master have.prs.2pl grow.ptcp.m.sg much ‘You, master, have grown much.’ (Historia de la vida y hechos del Emperador Carlos V, 1604–1611, apud HIST) (11) Andrea Gasparo Gorço, que por el Rey Católico Andrea Gasparo Gorço that because.of the king catholic había vuelto a Fez have.pst.ipfv.3sg return.ptcp.m.sg to Fez ‘Andrea Gasparo Gorço, who had returned to Fez because of the Catholic King’ (Historia de Felipe II, rey de España, 1619, apud HIST) In Aranovich (2003) and Rosemeyer (2014) this replacement process is assumed to result from actualisation (Andersen 2001; De Smet 2012). Following usagebased approaches to language change such as Bybee (2003) and Barðdal (2008), it can be suggested that the grammaticalisation of the haber + PtcP construction caused an increase in the productivity of the construction and particularly a gradual expansion of its type frequency. As a result, the writers start using verbs of change of state and change of location in the haber + PtcP construction. The actualisation process is determined by the similarity of these usage contexts to the original usage context of haber + PtcP, i.e. transitive sentences. In Rosemeyer (2014), I model this process as illustrated in Figure 1. Thus, in contexts classified as syntactically and semantically intransitive, predicates that do not contain a transition in their event template [–DIR] and are used in atelic configurations [–TE] are first to be affected by the actualisation process, followed by atelic predicates of change of state and change of location [+DIR,–TE] and lastly telic predicates of change of state and change of location [+DIR,+TE].2 Syntactically transitive, semantically transitive

Transitive sentences

→

Mixed forms

→

Locative adverbials in object position Sentences with dative arguments “Absolute” transitives Reflexives and reciprocals

Syntactically intransitive, semantically intransitive

[−DIR] [−TE]

[+DIR] [−TE]

[+DIR] [+TE]

Fig. 1: The actualisation of haber + PtcP (Rosemeyer 2014: 265).

2 Figure 1 also illustrates the predominance of the criterion of transitivity. For instance, compound tense of reflexive verbs, which can be characterised as syntactically transitive, are typi-

Entrenchment and discourse traditions in Spanish auxiliary selection

313

In addition to this actualisation process, frequency effects have had an important influence on the development of the replacement of ser + PtcP with haber + PtcP. In Rosemeyer (2013, 2014), evidence is given for the existence of conserving effects in the development of Spanish auxiliary selection. In line with the notion of entrenchment introduced in section 2, these studies demonstrate that in a corpus of historiographical texts, the use of ser + PtcP persists longer with high-frequency verbs than low-frequency verbs, since ser + PtcP syntagms formed with these verbs have a high token frequency and are therefore susceptible to entrenchment. In Rosemeyer (2014), it is also shown that there are systematic asymmetries in the frequency of use of the verbs investigated in the study (cf. section 4 for a list). In particular, the semantic parameter of a change-of-location meaning is positively correlated with the usage frequency of verbs. In other words, many of the high-frequency verbs express a change-of-location situation. Examples are verbs such as ir (‘go’), venir (‘come’), pasar (‘pass’), partir (‘go away’) and volver (‘return’). The majority of late ser + PtcP tokens involve participles formed from these verbs. While the semantic parameter of change of location does not significantly influence Old Spanish auxiliary selection, this parameter starts determining auxiliary selection in Early Modern Spanish, favouring ser + PtcP over haber + PtcP. Synchronic evidence for this rule in the 16 th century is found in Rosemeyer (2013). This observation suggests a similar situation in Modern French, where the selection of être (‘be’) instead of avoir (‘have’) is often motivated by the parameter of change-of-location semantics (Kailuweit 2011). Interestingly, Heidinger’s (this volume) case study of the French verb monter (‘go up’) suggests a parallel between the developments in Spanish and French. Between the 16 th and 20 th century, être-selection becomes more probable, relatively speaking, depending on whether or not monter denotes a change of location. It seems reasonable to interpret this development as the result of a conserving effect: During the slow process of replacement of être with avoir, être-selection persists longer with change-of-location verbs due to their high token frequency. This development leads to the creation of a synchronic grammatical rule: “when expressing a change-of-location event, use être over avoir”. Similar observations apply for German (Gillmann this volume). In Rosemeyer (2014) I restricted my analysis to a sole discourse tradition, i.e. historiographical texts. This decision was taken (a) in order to avoid skewing of the data through the simultaneous analysis of several discourse tradi-

cally formed with haber already in Old Spanish. This observation illustrates the need to take into account both syntactic and semantic factors in the analysis of the development of auxiliary selection systems, as suggested by Loporcaro (2011, this volume).

314

Malte Rosemeyer

tions and (b) because that discourse tradition is most stable across the investigated time period and offers enough data to license a quantitative analysis. In accordance with the discourse tradition approach to language history introduced in section 2, one can assume that both the use of auxiliary selection and its development differ from discourse tradition to discourse tradition. In particular, it is to be expected that the conserving effect demonstrated in Rosemeyer (2014) is dependent on discourse tradition. Stolova (2009) provides preliminary evidence for this assumption. Stolova analyses 372 ser + PtcP tokens from texts after 1700 in the Corpus del Español (Davies 2002–). These tokens are manifestations of a fossilised use of ser + PtcP. They are formed from the verbs bajar (‘go down’), correr (‘run’), entrar (‘enter’), ir (‘go’), llegar (‘come’), partir (‘go away’), pasar (‘pass’), salir (‘leave’), subir (‘go up’), tornar (‘return’) and volver (‘return’). All of these verbs typically express a changeof-location event, which can be considered evidence for the conserving effect demonstrated in Rosemeyer (2014). Stolova also demonstrates how the use of these fossilised items perpetuates discourse traditions (Hypothesis 1). For instance, in the examples (12) and (13) taken from Stolova’s data a ser + PtcP token from the 16 th century (12) is copied into a 19 th century poem (13). (12) Como el ciervo huiste, habiéndome herido. / like the deer flee.pst.pfv.2sg have.prog.refl wound.ptcp.m.sg Salí tras ti clamando y eras leave.pst.pfv.1sg after you shout.prog and be.pst.ipfv.2sg ido go.ptcp.m.sg ‘Like a deer you fled after wounding me / I followed you shouting and you were gone’ (Cántico spiritual, 1542–1591, apud Stolova 2009: 395) (13) La vida transpusiste! …/ Hermanito querido; the life change.pst. pfv.2sg little.brother love.ptcp.m.sg Salí tras ti clamando … y eras leave.pst.pfv.1sg after you shout.prog and be.pst.ipfv.2sg ido! go.ptcp.m.sg ‘You turned my life upside down! / beloved little brother; I followed you shouting … and you were gone’ (En la muerte de un hermano niño, 1811–1863, apud Stolova 2009: 395) Furthermore, Stolova suggests that the use of these fossilised ser + PtcP syntagms differs according to discourse tradition. In the legal texts in her corpus,

Entrenchment and discourse traditions in Spanish auxiliary selection

315

ser + PtcP typically appears in the fossilised form era llegado el caso (‘the case had arisen’) (14). In narrative texts, ser + PtcP is typically used to indicate time, as in era llegado septiembre (‘September had come’) (15). While the latter two uses of ser + PtcP are metaphorical, ser + PtcP typically expresses the literal event of a change of location in historiographical texts, like in (16). (14) y no era llegado el caso de and not be.pst.ipfv.3sg arrive.ptcp.m.sg the case of examinarse si ésta incurrió realmente en las examine.inf.refl if she commit.pst.pfv.3sg really in the contravenciones violations ‘And it had not been examined whether she had indeed committed these violations of the law.’ (Revista de derecho y jurisprudencia, 20 January 1927, apud Stolova 2009: 396) (15) había pasado el verano y era have.pst.ipfv.3sg pass.ptcp.m.sg the summer and be.pst.ipfv.3sg llegado septiembre arrive.ptcp.m.sg September ‘Summer had passed and September had come.’ (La gaviota, 1849, apud Stolova 2009: 396) (16) El capitán le respondió que había poco the captain to.him answer.pst.pfv.3sg that have.pst.ipfv.3sg little tiempo que era llegado á la isla time that be.pst.ipfv.3sg arrive.ptcp.m.sg to the island ‘The captain answered him that he had only recently come to the island.’ (Viajes al Maluco de Fray García de Loaísa y Álvaro de Saavedra, 1837, apud Stolova 2009: 397) In summary, Stolova’s results suggest both similarities and differences between entrenchment and conservation mechanisms between different discourse traditions. On the one hand, the distribution of verbs in the ser + PtcP construction appears to be relatively homogeneous across discourse traditions: ser + PtcP is typically used with verbs expressing a change of location. On the other hand, the function of the conserved construction appears to differ according to discourse tradition. An explanation for this difference may reside in whether the construction expresses a metaphorical or literal meaning.

316

Malte Rosemeyer

4 Data The analyses conducted in this study are based on 4,680 haber + PtcP and ser + PtcP tokens from Spanish texts dated between 1260 and 1688. According to the criteria developed in Rosemeyer (2014), only tokens that fall into the envelope of variation between the two constructions were included in the corpus, leading to the exclusion of, for example, syntactically transitive tokens. Likewise, the study is based on the 43 verbs selected in Rosemeyer (2014) for the investigation of the development of Spanish auxiliary selection. These verbs vary systematically according to the semantic features “change of state” [DIR], “telicity” [TE], “change of location” [MO], control by the subject referent [CON], as well as overall usage frequency of the verb. Table 1 summarises these verbs and their typical semantic values regarding the features [DIR] and [TE] which have special relevance for the description. The corpus is divided into three smaller subcorpora, each representing a specific discourse tradition. The first and biggest subcorpus – HIST – is taken from the data collected in Rosemeyer (2014). HIST represents the auxiliary selection in 44 Spanish historiographical texts from 1270 to 1698 and consists of 4,158 tokens. These texts mainly stem from the Corpus Diácronico del Español (Real Academia Española 2012). The selection of the texts was guided by the

Tab. 1: Verbs investigated in this study. Predicate class Description

[+DIR,+TE]

[+DIR,–TE]

[–DIR,–TE]

Verbs and their typical translation

acaecer (‘happen’), ahogar (‘drown’), alzar (‘rise’), aparecer (‘appear’), avenir (‘happen’), ayuntar (‘gather’), cabalgar (‘mount’), caer (‘fall down’), Telic change of state cuntir (‘happen’), espantar (‘become frightened’), or change of finar (‘end’, ‘die’), hundir (‘sink’), maravillar location (‘become astonished’), morir (‘die’), nacer (‘be born’), pasar (‘pass’), perecer (‘perish’), quebrar (‘break’), sentar (‘sit down’), tornar (‘return’), venir (‘come’), volver (‘return’) andar (‘walk’), caminar (‘walk’), correr (‘run’), Atelic change of corromper (‘corrode’), crecer (‘grow’), descender (‘go state or change of down’), escapar (‘escape’), exir (‘leave’), huir (‘flee’), location ir (‘go’), menguar (‘diminish’), partir (‘leave’), subir (‘go up’, ‘rise’) arrepentir (‘repent’), durar (‘last’), fincar (‘stay’), State or persistence holgar (‘stay’), morar (‘stay’, ‘live’), quedar (‘stay’), of pre-existing state sobrar (‘remain’), yacer (‘lie’)

Entrenchment and discourse traditions in Spanish auxiliary selection

317

criteria established in Fernández-Ordóñez (2006), thus guaranteeing a relative authenticity of the manuscripts used in the editions of the texts. The second subcorpus – CODEA – represents auxiliary selection in Spanish documents from the royal chancellery, ecclesiastical institutions and municipal administrations, as well as less formal letters between dignitaries from these institutions. It consists of 100 tokens taken from the Corpus de Documentos Españoles Anteriores a 1700 (Grupo de Investigación de Textos para la Historia del Español 2012). These are all of the tokens formed from the verbs in Tab. 1 found in these texts. The small number of occurrences is a result of (a) the small overall size of the corpus and (b) the fact that the use of haber + PtcP and ser + PtcP is less frequent in that corpus. However, these tokens can be regarded as more representative because the texts in the CODEA corpus are based on original manuscripts. Thus, from a philological point of view these data are more reliable than the data collected in HIST. Likewise, the third subcorpus – LETTERS – offers more reliable data than the HIST corpus. LETTERS shows auxiliary selection in private letters from Spanish immigrants in the American colonies between 1500 and 1600 and consists of 422 tokens. The corpus is based on the edition of these letters from original manuscripts by Fernández Alcaide (2009). As pointed out in Fernández Alcaide’s study, the corpus provides important data, since private letters constitute a discourse tradition characterised by a high communicative proximity (in terms of Koch and Oesterreicher 1985, 1990, 2001). Thus, [the corpus is characterised] by a form of discourse organisation that is essentially based on the fact that the utterances are produced with less planning than in official or more formal texts, i.e. a text characterised by communicative distance; because the linguistic interchange takes place between family members or at least persons who know each other; because of the communicative necessities that cause these letters […]; because of the emotional intensity that the letters sometime acquire … (Fernández Alcaide 2009: 313– 314, translation MR).

The discourse traditions contained in the corpora can be distinguished according to two parameters, i.e. communicative proximity vs. communicative distance and narrative vs. documental text function. These parameters are modelled as scalar here, as illustrated in Figure 2:

LETTERS communicative proximity narrative text function

HIST

CODEA communicative distance documental text function

Fig. 2: Categorisation of the subcorpora according to communicative proximity and text function.

318

Malte Rosemeyer

Regarding the parameter of communicative proximity vs. communicative distance addressed in the quote from Fernández Alcaide (2009), the data from the LETTERS corpus are characterised by communicative proximity, while the data from the CODEA corpus are characterised by communicative distance and the HIST corpus occupies an intermediate position. Likewise, the texts from the LETTERS corpus typically fulfil a narrative function, whereas the texts from the CODEA corpus usually document decisions. Texts from the HIST corpus again occupy an intermediate position, since the discourse tradition of historiographical texts typically oscillates between the need for an objective documentary account of events and narrative expressivity. The analysis of the data in the next section will demonstrate the relevance of specifying the discourse traditions represented by the three subcorpora according to these criteria.

5 The context-dependency of conserving effects in Spanish auxiliary selection In this section, I demonstrate the relevance of the hypotheses (1) and (2) repeated below. 1.

2.

The cognitive mechanism of entrenchment plays a pivotal role in the emergence of discourse traditions both in the written and the spoken modality. By repeatedly using elements that are typical of a certain discourse situation, a writer or speaker signals that her text belongs to a specific discourse tradition. Conversely, entrenchment can be conceptualised as a process by which a linguistic element increasingly comes to depend on intertextuality. This means that discourse traditions have an enabling function for entrenchment: Entrenchment operates within specific discourse traditions. Discourse traditions characterised by communicative distance display a more conservative language use than discourse traditions characterised by communicative proximity.

I first analyse the differences in the usage frequency of haber + PtcP and ser + PtcP in the three corpora and then go on to analyse the differences in the function of haber + PtcP and ser + PtcP. The analysis demonstrates a correlation between the frequency with which change-of-location events are described in a discourse tradition, and the entrenchment of ser-selection with the corresponding verb class. This finding explains why ser-selection seems to be no less frequent in the LETTERS corpus than in the other corpora.

Entrenchment and discourse traditions in Spanish auxiliary selection

319

5.1 Usage frequency of haber + PtcP and ser + PtcP Figure 3 illustrates the approximate joint absolute usage frequency of haber/ ser + PtcP-construction per million words in the three corpora. The frequency measure was normalised using the formula: freg =

n of tokens n of words in the corpus

× 1,000,000

Within the envelope of variation selected for this study, the use of haber + PtcP and ser + PtcP constructions is most frequent in historiographical texts (HIST), followed by administrative documents (CODEA) and private letters (LETTERS). Arguably, this variation results from the differences in text function between the three discourse traditions. The verbs investigated in this study (cf. Table 1)

Fig. 3: Joint usage frequency of haber + PtcP and ser + PtcP in HIST, CODEA and LETTERS3.

3 The application of the formula is straightforward for the LETTERS and CODEA corpora because no randomisation procedure was applied. However, due to the great token number in the HIST corpus, frequent haber + PtcP and ser + PtcP types were randomised in that corpus. As a result, an approximation method had to be introduced in order to calculate the overall number of haber + PtcP and ser + PtcP tokens in that corpus. This approximation method has the following formula: n of reviewed tokens n of tokens = overall n of tokens encountered using a search string / n of randomly selected tokens

320

Malte Rosemeyer

are typical for situations described in historiographical texts. These texts often elaborately describe successions of events. In contrast, administrative texts and private letters usually refer to events in isolation and tend to describe the emotional or legal consequences of these events. Theoretically, these differences in the joint usage frequency of the two constructions should not lead to differences in the overall distribution of haber + PtcP and ser + PtcP between the three discourse traditions. As elaborated in section 3, the most important predictors of auxiliary selection in Spanish are the semantic parameters “change of state” [DIR] and “telicity” [TE]. Tables 2–

Tab. 2: Development of auxiliary selection over time according to [DIR] and [TE] in HIST. 1200–1299 H

S

1300–1399

%S H

S

%S

1400–1499 H

S

1500–1599

%S H

S

1600–1699

%S H

S

%S

[+DIR,+TE] 26 405 94.0

41 397 90.6 144 551 79.3 468 189 28.8 426 75 15.0

[+DIR,–TE] 15

75 83.3

33 128 79.5

92 150 62.0 187

[–DIR,–TE] 56

16 22.2

43

98

TOTAL

28 39.4

19 16.2

46 19.7 251 15

88

3

3.3

93

5.6

0

0.0

97 496 83.6 117 553 82.5 334 720 68.3 743 238 24.3 770 90 10.5

Tab. 3: Development of auxiliary selection over time according to [DIR] and [TE] in CODEA. 1200–1299

1300–1399

1400–1499

H

S

%S

H

S

%S

H

[+DIR,+TE] 0

4

100.0 0

5

[+DIR,–TE] 0

1

100.0 (NA)

[–DIR,–TE] 0

1

100.0 (NA)

(NA)

TOTAL

6

100.0 0

0

5

1500–1599

1600–1699

S

%S H

S

%S H

S

%S

100.0 1

5

83.3 19

15

44.1 10

1

9.1

1

0

0.0 18

3

14.3

2

0

0.0

5

0

0.0

9

0

0.0

71.4 42

18

30.0 21

1

4.8

100.0 2

5

Tab. 4: Development of auxiliary selection over time according to [DIR] and [TE] in LETTERS. 1200–1299

1300–1399

1400–1499

H

H

H

S

%S

S

%S

S

1500–1599

%S H

[+DIR,+TE] (NA)

(NA)

(NA)

[+DIR,–TE] (NA)

(NA)

(NA)

61

[–DIR,–TE] (NA)

(NA)

(NA)

33

TOTAL

(NA)

(NA)

(NA)

S

1600–1699

%S H

205 113 35.5 (NA) 10 14.1 (NA) 0

0.0 (NA)

299 123 29.2 (NA)

S

%S

Entrenchment and discourse traditions in Spanish auxiliary selection

321

Tab. 5: Auxiliary selection in the 16th-century section of HIST, CODEA and LETTERS. Corpus

H

S

%S

HIST CODEA LETTERS

743 42 299

238 18 123

24.3 30.0 29.2

4 illustrate the influence of these semantic parameters on auxiliary selection over time in the three subcorpora. In these tables, “H” refers to haber + PtcP tokens, whereas “S” refers to ser + PtcP tokens. First, the data summarised in Tables 2–4 demonstrate that only in the 16 th century are there enough tokens to license a quantitative comparison of the use of auxiliary selection in the three discourse traditions. Second, the distribution of haber + PtcP and ser + PtcP according to [DIR] and [TE] is similar in all of the corpora. Thus, ser-selection is most frequent with telic predicates of change of state or change of location [+DIR,+TE], followed by atelic predicates of change of state or change of location [+DIR,–TE] and lastly predicates of state or continuation of a pre-existing state which are always atelic [–DIR,–TE]. However, there also seem to be slight quantitative differences. In particular, it seems that in the 16 th century, ser-selection is relatively more frequent for the predicate class [+DIR,+TE] in CODEA (44.1 per cent) and LETTERS (35.5 per cent) than in HIST (18.8 per cent). In contrast, ser-selection in the predicate classes [+DIR,–TE] and [–DIR,–TE] is relatively more frequent in HIST than in CODEA and LETTERS. This suggests that in CODEA and LETTERS, the grammatical productivity of the ser + PtcP construction is restricted to a more specific predicate class than in HIST. The analyses in the next section will demonstrate that this finding is actually an effect of differences in the frequency with which change-of-location predicates are expressed in the three discourse traditions. Third, a comparison of auxiliary selection in the 16th-century section of the three corpora (marked by a grey shade in Tables 2–4) seems to suggest that ser-selection is relatively less frequent in historiographical texts than in the other two discourse traditions. For sake of clarity, the overall distribution of the two constructions in the 16th-century section of the three corpora is summarised in Table 5. Although there indeed seem to be differences in the distribution of auxiliary selection in the three corpora, these differences do not reach statistical significance (χ² = 4.216 (2), p = 0.121). Consequently, the data from the corpus do not warrant the conclusion that there is a difference in the overall distribution of haber + PtcP and ser + PtcP in the three discourse traditions.

322

Malte Rosemeyer

This result is surprising because it contradicts Hypothesis 2. The discourse tradition represented by the LETTERS corpus, i.e. private letters, is characterised by a high degree of communicative proximity. Thus, it would be expected that the historical process by which ser + PtcP came to be replaced with haber + PtcP is most advanced in this discourse tradition. Although in line with this hypothesis, the grammatical productivity of ser + PtcP is more restricted in the LETTERS corpus, the overall frequency of ser-selection does not appear to reflect this difference. The analysis of the functional differences between the two constructions in the three discourse traditions will, however, offer an explanation for this finding that allows Hypothesis 2 to be maintained. In summary, in this section two observations regarding the 16th-century data were made that have no straightforward explanation. First, the raw data in Tables 2–4 suggests that ser-selection with telic change-of-state or changeof-location predicates [+DIR,+TE] is relatively more frequent in CODEA and LETTERS than in HIST. Second, it was shown that in terms of the overall usage frequency of haber + PtcP and ser + PtcP, the distribution of the two constructions does not appear to differ. This finding contradicts the hypothesis that discourse traditions characterised by communicative proximity display a more modern language use than discourse traditions characterised by communicative distance.

5.2 Functional differences in auxiliary selection In this section, I demonstrate how the assumption of conserving effects in language change can provide an explanation for these two findings. In addition, the discussion illustrates the context-dependency of entrenchment postulated in section 2. A more detailed analysis of the distribution of auxiliary selection according to predicate class suggests that the semantic parameter “change of location” [MO] can explain the contradictory results from the last section. Predicates of change of location are a subclass of predicates of change of state (cf. Table 1). Consequently, variation in auxiliary selection according to change-of-location predicates can explain the apparent variation in auxiliary selection according to change-of-state predicates. In Table 6, I restrict the data to 16th-century haber + PtcP and ser + PtcP tokens that express a change of state [+DIR,–MO] or change of location [+DIR,+MO] and illustrate the percentage with which a token from this predicate class in the three corpora expresses a change of location [+DIR,+MO]. Compared to other situations of change of state, historiographical texts describe fewer events that can be characterised as change-of-location events

Entrenchment and discourse traditions in Spanish auxiliary selection

323

Tab. 6: Usage frequency of change-of-location predicates [+DIR, +MO] relative to the usage frequency of change-of-state predicates [+DIR,–MO] in HIST, CODEA and LETTERS in the 16 th century.

HIST CODEA LETTERS

[+DIR,+MO]

[+DIR,–MO]

% [+DIR,+MO]

503 37 313

387 18 76

56.5 67.3 80.5

Tab. 7: Auxiliary selection in the 16 th century according to subcorpus and the predicate classes [+DIR,+MO], [+DIR,+MO] and [–DIR,–MO]. HIST

[+DIR,+MO] [+DIR,–MO] [–DIR,–MO] TOTAL

CODEA

LETTERS

H

S

%S

H

S

%S

H

S

%S

378 277 88 743

125 110 3 238

24.9 28.4 3.3 24.3

27 10 5 42

10 8 0 18

27.0 44.4 0.0 30.0

192 74 33 299

121 2 0 123

38.7 2.6 0.0 29.2

than legal texts and private letters. While the difference between HIST and CODEA does not reach statistical significance (χ² = 2.028 (1), p = 0.155), the difference between HIST and LETTERS does (χ² = 4.264 (1), p = 0.039*). I summarise this observation in (17). (17)

Observation 1: In comparison to historiographical texts in the 16 th century, private letters from the 16 th century describe significantly more events of change of location than other events of change of state.

In the light of observation 1, it is worth investigating whether the differences between the three corpora in terms of the relative usage frequency of ser + PtcP with predicates marked as [+DIR,+TE] are epiphenomenal and actually derive from the differences in the usage frequency of predicates marked as [+DIR,+MO]. Table 7 illustrates 16th-century auxiliary selection in the three corpora according to whether the predicate refers to a change of location [+DIR,+MO], a change of state [+DIR,–MO] or a state viz. continuation of a preexisting state [–DIR,–MO]. In the 16th-century section of the HIST and CODEA corpora, the difference between change-of-state and change-of-location predicates does not have a significant influence on auxiliary selection. Neither in the HIST corpus nor in the CODEA corpus does the difference between the conditions [+DIR,+MO] and

324

Malte Rosemeyer

[+DIR,–MO] reach statistical significance (HIST: χ² = 1.259 (1), p = 0.262, CODEA: χ² = 0.971 (1), p = 0.324). In contrast, the difference between these two types of predicates has a crucial influence on auxiliary selection in the LETTERS corpus. First, only in two of 76 cases in the LETTERS corpus in which the predicate refers to a change of state is ser used as the auxiliary (2.6 per cent). Second, in 121 of 313 cases in which the predicate expresses a change of location in the LETTERS corpus, ser is used as the auxiliary (38.7 per cent). This effect reaches a very high statistical significance (χ² = 35.061 (1), p < 0.001***). I summarise this observation in (18): (18)

Observation 2: When writing about events of change of location, the authors of the texts in the LETTERS corpus use ser + PtcP significantly more often than when writing about other events of change of state. No similar statistically significant effect can be observed in the 16th-century sections of the HIST corpus and the CODEA corpus.

A last observation drawn from Table 7 refers to the differences between the three discourse traditions with respect to the usage frequency of ser + PtcP with predicates expressing a mere change of state [+DIR,–MO], as well as a state or a continuation of a pre-existing state [–DIR,–MO]. With these predicates, ser-selection is least frequent in the private letters. Only the difference in auxiliary selection in the domain of predicates marked as [+DIR,–MO] between the HIST corpus and the LETTERS corpus reaches statistical significance (χ² = 22.547 (1), p < 0.001***). I summarise this third observation in (19). (19)

Observation 3: When describing situations of change of state [+DIR,–MO], ser-selection is significantly less frequent in LETTERS than in the 16th-century section of HIST.

The three observations in (17–19) explain the contradictory findings from the last section, namely the facts that ser-selection is as frequent in the LETTERS corpus as in the other corpora and that ser-selection is more frequent in predicates marked as [+DIR,+TE] in the LETTERS corpus than in the HIST corpus. Observation 3 suggests that the parameter [DIR], i.e. the fact that the situation refers to a change of state, is not relevant for auxiliary selection in the LETTERS corpus. This means that the high frequency of ser-selection in the predicate classes [+DIR,+TE] and [+DIR,–TE] in the LETTERS corpus results from the fact that the use of ser-selection is very frequent with change-of-location predicates in the LETTERS corpus (observation 2) and that in the LETTERS corpus, more than 80 per cent of the predicates marked as [+DIR] are change-of-location predicates (observation 1). This result can be explaned in terms of the interplay between entrenchment and discourse traditions as postulated in Hypothesis 1. Crucially, the observations 1–3 suggest a positive correlation between the frequency with which

Entrenchment and discourse traditions in Spanish auxiliary selection

325

a predicate of change of location is used in a discourse tradition and the frequency of ser-selection with change-of-location predicates in that discourse tradition. In the data investigated in this study, the conserving effect caused by entrenchment is strongest in the discourse tradition in which change-oflocation predicates are most frequent, i.e. private letters. Although in Rosemeyer (2013, 2014) I demonstrated a similar trend for historiographical texts, from a synchronic perspective this trend is not strong enough to result in a significant difference in auxiliary selection between change-of-state and change-of-location predicates. The detailed analysis of the data in this section has also demonstrated that Hypothesis 2 can be maintained when describing the development of auxiliary selection in the three discourse traditions. In the LETTERS corpus, the grammatical productivity of ser + PtcP has apparently receded to the degree that it is exclusively used with a row of entrenched syntagms formed from change-oflocation verbs. In contrast, the type frequency of ser + PtcP is higher in the HIST and CODEA corpora. This suggests that in comparison to the discourse traditions represented by these latter corpora, the replacement of ser with haber has progressed faster in discourse traditions characterised by communicative proximity – in this case, private letters. The high absolute frequency of ser-selection in the LETTERS corpus results from the fact that in that discourse tradition, change-of-location predicates are very frequent.

6 Summary and discussion In this paper, I have analysed the differences in auxiliary selection between three discourse traditions in Spanish – historiographical texts, administrative documents and private letters. An analysis of these differences in the 16th-century section of the corresponding corpora suggests a positive correlation between the frequency with which a predicate of change of location is used in a discourse tradition and the frequency of ser-selection with change-of-location predicates in that discourse tradition. On the one hand, these results confirm the analysis proposed in Rosemeyer (2014): The conserving effect of frequency specifically affects change-of-location predicates, since verbs expressing this meaning are frequent in all three investigated discourse traditions. On the other hand, the results also suggest that entrenchment and the ensuing conserving effect in language change differ quantitatively across the three examined discourse traditions. The analysis has revealed a difference in the distribution of ser + PtcP and haber + PtcP in 16th-century historiographical

326

Malte Rosemeyer

texts and private letters. While the semantic parameter “change of state” is an important predictor of auxiliary selection in historiographical texts, the authors of the private letters appear to regard the semantic parameter “change of location” as the crucial parameter of auxiliary selection. Extremely few ser + PtcP tokens in the LETTERS corpus do not display a change-of-location semantics. I have argued that this finding needs to be explained by assuming contextdependent entrenchment: Given that change-of-location events are more frequent in private letters and that due to the communicative proximity of these texts the change has progressed further, change of location has become the most important predictor of auxiliary selection. This assumption also has explanatory power for Stolova’s (2009) finding that there are slight differences in the function of fossilised ser + PtcP tokens across different discourse traditions. As the discussion of the data from the LETTERS corpus has shown, entrenchment or fossilisation can lead to the reanalysis of the function of the entrenched syntagms. This reanalysis, however, can be claimed to depend on the involved discourse function. Consequently, the study at hand proposes that it is necessary to reconcile the cognitivist approach taken in usage-based linguistics (UBL) and the pragmatic approach taken in studies of discourse traditions. This reconciliation is advantageous for either theoretical perspective. Regarding the usage-based approach to language change, my proposal leads to a more precise definition of how entrenchment works in language change. In order to refine the notion of the frequency effect of entrenchment in language change, it is crucial to take into account the fact that entrenchment operates within specific experiential domains. The notion of discourse traditions offers a possible model for these experiential domains. Thus, accounting for discourse traditions in the study of frequency effects allows us to explain differences in usage frequency between different types of texts which remain obscure if we assume that entrenchment effects operate globally. Regarding the discourse tradition approach to language change, my proposal offers a way of getting a clearer idea of how discourse traditions emerge. In order to study their emergence, it is not only necessary to study “text-internal factors” such as recurrent topics, but also phenomena on the “text surface”. Thus, writers recur to linguistic elements entrenched within a certain discourse tradition in order to situate their texts within that discourse tradition. As a result, the notion of entrenchment can be exploited in order to study how discourse traditions emerge and are consolidated over time. In addition, these assumptions are highly relevant for the methodology of the study of historical texts, since they suggest that in addition to the concept of discourse traditions, the concept of entrenchment is crucial for the assessment of the representativity of tokens from historical texts.

Entrenchment and discourse traditions in Spanish auxiliary selection

327

References Andersen, Henning (ed.). 2001 Actualization: Linguistic Change in Progress. Amsterdam, Philadelphia: John Benjamins. Aranovich, Raúl. 2003 The semantics of auxiliary selection in Old Spanish. Studies in Language 27(1): 1–37. Aschenberg, Heidi. 2003 Diskurstraditionen – Orientierungen und Fragestellungen. In: Heidi Aschenberg and Raymund Wilhelm (eds.), Romanische Sprachgeschichte und Diskurstraditionen, 1–18. Tübingen: Narr. Barðdal, Jóhanna. 2008 Productivity: Evidence from Case and Argument Structure in Icelandic. Amsterdam, Philadelphia: John Benjamins. Barthes, Roland. 1967 The death of the author. Aspen 5–6. Available online at http:// www.ubu.com/aspen/aspen5and6/threeEssays.html-barthes, retrieved 16. 1. 2012. Behrens, Heike. 2009 Usage-based and emergentist approaches to language acquisition. Linguistics 47(2): 383–411. Benzing, Joseph. 1931 Zur Geschichte von ser als Hilfszeitwort bei den intransitiven Verben im Spanischen. Zeitschrift für romanische Philologie 51: 385–460. Blank, Andreas. 2005 Polysemie: Sprachbewußtsein, semantische Relationen, Diskurstraditionen. In: Daniel Jacob, Thomas Krefeld and Wulf Oesterreicher (eds.), Sprache − Bewusstsein − Stil. Theoretische und historische Perspektiven, 31–46. Tübingen: Narr. Burke, James. 2004 Medieval Spanish prose. In: David T. Gies (ed.), The Cambridge History of Spanish Literature, 95–104. Cambridge: Cambridge University Press. Bybee, Joan L. 2003 Mechanisms of change in grammaticization: the role of frequency. In: Richard Janda and Brian Joseph (eds.), The Handbook of Historical Linguistics, 624–647. Oxford: Blackwell. Bybee, Joan L. 2006 From usage to grammar: the mind’s response to repetition. Language 82(4): 711–733. Bybee, Joan L. 2010 Language, Usage, and Cognition. Cambridge, New York: Cambridge University Press. Bybee, Joan L. and Paul J. Hopper (eds.). 2001 Frequency and the Emergence of Linguistic Structure. Amsterdam, Philadelphia: John Benjamins. Dahl, Östen. 1985 Tense and Aspect Systems. Oxford: Blackwell. Davies, Mark. 2002– Corpus del español (100 million words, 1200s-1900s). Available online at http://www.corpusdelespanol.org, retrieved 23. 3. 2013. De Smet, Hendrik. 2012 The course of actualization. Language 88: 601–633. Diessel, Holger. 2007 Frequency effects in language acquisition, language use, and diachronic change. New Ideas in Psychology 25: 108–127. Diessel, Holger. 2011 Review article of ‘Language, usage and cognition’ by Joan Bybee. Language 87: 830–844. Diewald, Gabriele. 2002 A model for relevant types of contexts in grammaticalization. In: Ilse Wischer and Gabriele Diewald (eds.), New Reflections on Grammaticalization, 103– 120. Amsterdam, Philadelphia: John Benjamins. Diewald, Gabriele. 2006 Context types in grammaticalization as constructions. In: Doris Schönefeld (ed.), Constructions, Special Volume 1 (9). Ellis, Nick. 1996 Sequencing in SLA: phonological memory, chunking and points of order. Studies in Second Language Acquisition 18: 91–126.

328

Malte Rosemeyer

Elvira González, Javier. 2001 Intransitividad escindida en español: El uso auxiliar de ser en español medieval. EdLing 15: 201–245. Enrique-Arias, Andrés (dir.). 2008 Biblia Medieval. Available online at http:// www.bibliamedieval.es, retrieved 26. 5. 2011. Fernández Alcaide, Marta. 2009 Cartas de particulares en Indias del siglo XVI. Edición y estudio discursivo. Madrid, Frankfurt am Main: Iberoamericana, Vervuert. Fernández-Ordóñez, Inés. 2004 Alfonso X en la historia del español. In: Rafael Cano Aguilar (ed.), Historia de la lengua española, 381–422. Barcelona: Ariel. Fernández-Ordóñez, Inés. 2006 La Historiografía medieval como fuente de datos lingüísticos. Tradiciones consolidadas y rupturas necesarias. In: José Jesus Bustos Tovar and José Luis Girón Alchonchel (eds.), Actas del VI Congreso Internacional de Historia de la Lengua Española 11, 1779–1807. Madrid: Arco Libros. Foucault, Michel. 1988 [1969] Was ist ein Autor? In: Michel Foucault (ed.), Schriften zur Literatur, 7–31. Frankfurt am Main: Fischer. Foucault, Michel. 2002 [1969] Archeology of Knowledge. London and New York: Routledge. Goldberg, Adele E. 2006 Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press. Grupo de Investigación de Textos para la Historia del Español. 2012 Corpus de Documentos Españoles Anteriores a 1700, Available online at http://demos.bitext.com/codea/, retrieved 10. 12. 2012. Heine, Bernd. 2002 On the role of context in grammaticalization. In: Ilse Wischer and Gabriele Diewald (eds.), New Reflections on Grammaticalization, 83–101. Amsterdam, Philadelphia: John Benjamins. Held, Gudrun. 2005 Der Einfluss von Höflichkeit auf die mittelalterliche Briefkunst – exemplarische Überlegungen zur Entwicklung von Textstruktur und Syntax vom dictamen zur freien Briefpraxis. In: Angela Schrott and Harald Völker (eds.), Historische Pragmatik und historische Varietätenlinguistik in den romanischen Sprachen, 45–62. Göttingen: Universitätsverlag Göttingen. Hopper, Paul J. and Elizabeth Closs Traugott. 2003 Grammaticalization. 2. Ed. Cambridge: Cambridge University Press. Jacob, Daniel. 1994 Die Auxiliarisierung von habere. Habilitationsschrift, Albert-LudwigsUniversität Freiburg. Jacob, Daniel. 2001 Representatividad lingüística o autonomía pragmática del texto antiguo? El ejemplo del pasado compuesto. In: Daniel Jacob and Johannes Kabatek (eds.), Lengua medieval y tradiciones discursivas en la Península Ibérica, 153–176. Frankfurt am Main: Vervuert. Jacob, Daniel and Johannes Kabatek (eds.). 2001 Lengua medieval y tradiciones discursivas en la Península Ibérica: Descripción gramatical, pragmática historica, metodología. Frankfurt am Main: Vervuert. Johnson, Keith. 1997 Speech perception without speaker normalization. In: Keith Johnson and John W. Mullennix (eds.), Talker Variability in Speech Processing, 145–165. San Diego, Cal.: Academic Press. Kabatek, Johannes. 2005a Las tradiciones discursivas del español medieval: historia de textos e historia de la lengua. Iberoromania 62: 28–44. Kabatek, Johannes. 2005b Tradiciones discursivas y cambio lingüístico. Lexis 29(2): 151– 177. Kabatek, Johannes. 2013 ¿Es posible una lingüística histórica basada en un corpus representativo? Iberoromania 77(1): 8–28.

Entrenchment and discourse traditions in Spanish auxiliary selection

329

Kailuweit, Rolf. 2011 Le choix de l’auxiliaire: Être ou avoir en français standard contemporain. In: Renata Enghels, Machteld Meulleman and Clara Vanderschueren (eds.), Peregrinatio in Romania: Artículos en homenaje a Eugeen Roegiest con motivo de su 65 cumpleaños, 397–420. Gent: Academia Press. Koch, Peter. 1997 Diskurstraditionen: zu ihrem sprachtheoretischen Status und ihrer Dynamik. In: Barbara Frank, Thomas Haye and Doris Tophinke (eds.), Gattungen mittelalterlicher Schriftlichkeit, 43–79. Tübingen: Narr. Koch, Peter and Wulf Oesterreicher. 1985 Sprache der Nähe – Sprache der Distanz. Mündlichkeit und Schriftlichkeit im Spannungsfeld von Sprachtheorie und Sprachgebrauch. Romanistisches Jahrbuch 36: 15–43. Koch, Peter and Wulf Oesterreicher. 1990 Gesprochene Sprache in der Romania: Französisch, Italienisch, Spanisch. Tübingen: Niemeyer. Koch, Peter and Wulf Oesterreicher. 2001 Gesprochene Sprache und geschriebene Sprache. Langage parlé et langage écrit. In: Günter Holtus, Michael Metzelin and Christian Schmitt (eds.), Lexikon der Romanistischen Linguistik, I,2, 584–627. Tübingen: Niemeyer. Kuhl, Patricia K. 1994 Learning and representation in speech and language. Current Opinion in Neurobiology 4: 812–822. Lacerda, Francisco. 1995 The perceptual-magnet effect: An emergent consequence of exemplar-based phonetic memory. In: Kjell Elenius and Peter Branderup (eds.), Proceedings of the XIIIth International Congress of Phonetic Sciences, II, 140–147. Stockholm: KTH, Stockholm University. Langacker, Ronald W. 1987 Foundations of Cognitive Grammar. Theoretical Prerequisites. Stanford: Stanford University Press. Lieven, Elena V. M., Julian M. Pine and Gillian Baldwin. 1997 Lexically-based learning and early grammatical development. Journal of Child Language 24: 187–219. Loporcaro, Michele. 2011 A euroversal in a global perspective: auxiliation and alignment. In: Peter Siemund (ed.), Linguistic Universals and Language Variation, 55–91. Berlin, New York: De Gruyter. Loporcaro. (this volume) Perfective auxiliation with reflexives in Medieval Romance: syntactic vs. semantic gradients. Mateu, Jaume. 2009 Gradience and auxiliary selection in Old Catalan and Old Spanish. In: Paola Crisma and Giuseppe Longobardi (eds.), Historical Syntax and Linguistic Theory, 176–193. Oxford: Oxford University Press. Mateu, Jaume and Mar Massanell. (this volume) A constructional approach to auxiliary selection: Evidence from existential constructions. Medin, Douglas L. and Marguerite M. Schaffer. 1978 Context theory of classification learning. Psychological Review 85(3): 207–238. Newell, Allen. 1990 Unified Theories of Cognition. Cambridge: MIT Press. Oesterreicher, Wulf. 1997 Zur Fundierung von Diskurstraditionen. In: Barbara Frank, Thomas Haye and Doris Tophinke (eds.), Gattungen mittelalterlicher Schriftlichkeit, 19– 42. Tübingen: Narr. Pfänder, Stefan, Heike Behrens, Peter Auer, Daniel Jacob, Rolf Kailuweit, Lars Konieczny, Bernd Kortmann, Christian Mair and Gerhard Strube. 2013 Erfahrung zählt. Frequenzeffekte in der Sprache – ein Werkstattbericht. Zeitschrift für Literaturwissenschaft und Linguistik 43(169): 7–32. Pierrehumbert, Janet B. 2001 Exemplar dynamics: Word frequency, lenition and contrast. In: Joan L. Bybee and Paul J. Hopper (eds.), Frequency and the Emergence of Linguistic Structure, 123–136. Amsterdam, Philadelphia: John Benjamins.

330

Malte Rosemeyer

Pierrehumbert, Janet B. 2002 Word-specific phonetics. In: Carlos Gussenhoven and Natasha Warner (eds.), Laboratory Phonology 7, 101–139. Berlin, New York: De Gruyter. Poplack, Shana. 2001 Variability, frequency, and productivity in the irrealis domain of French. In: Joan L. Bybee and Paul J. Hopper (eds.), Frequency and the Emergence of Linguistic Structure, 405–428. Amsterdam, Philadelphia: John Benjamins. Real Academia Española. 2012 Banco de datos (CORDE) [en línea]. Corpus diacrónico del español. Available online at http://www.rae.es, retrieved 20. 4. 2012. Rodríguez Molina, Javier. 2006a Ser + participio en español antiguo: perífrasis resultativa, no tiempo compuesto. In: José Jesus Bustos Tovar and José Luis Girón Alchonchel (eds.), Actas del VI Congreso Internacional de Historia de la Lengua Española, II, 1059–1072. Madrid: Arco Libros. Rodríguez Molina, Javier. 2006b Tradición manuscrita y gramática histórica: los tiempos compuestos en los textos medievales. In: Lola Pons Rodríguez (ed.), Historia de la lengua y crítica textual, 19–67. Madrid, Frankfurt: Iberoamericana Vervuert Verlag. Rosemeyer, Malte. 2012 How to measure replacement: Auxiliary selection in Old Spanish bibles. Folia Linguistica Historica 33(1): 135–174. Rosemeyer, Malte. 2013 Tornar and volver: The interplay of frequency and semantics in compound tense auxiliary selection in Medieval and Classical Spanish. In: Elly Van Gelderen, Jóhanna Barðdal and Michela Cennamo (eds.), Argument Structure in Flux. The Naples-Capri Papers, 435–458. Amsterdam, Philadelphia: John Benjamins. Rosemeyer, Malte. 2014 Auxiliary Selection in Spanish. Gradience, Gradualness, and Conservation. Amsterdam, Philadelphia: John Benjamins. Schlieben-Lange, Brigitte. 1983 Traditionen des Sprechens: Elemente einer pragmatischen Sprachgeschichtsschreibung. Stuttgart: Kohlhammer. Schrott, Angela and Harald Völker. 2005 Historische Pragmatik und historische Varietätenlinguistik. Traditionen, Methoden und Modelle in der Romanistik. In: Angela Schrott and Harald Völker (eds.), Historische Pragmatik und historische Varietätenlinguistik in den romanischen Sprachen, 1–24. Göttingen: Universitätsverlag Göttingen. Stolova, Natalya I. 2009 Los corpus diacrónicos al servicio del estudio de los arcaísmos gramaticales. In: Andrés Enrique-Arias (ed.), Diacronía de las lenguas iberorrománicas: nuevas aportaciones desde a lingüística de corpus, 385–400. Madrid, Frankfurt am Main: Iberoamericana, Vervuert. Tomasello, Michael. 1992 First Verbs: A Case Study of Early Grammatical Development. Cambridge: Cambridge University Press. Traugott, Elizabeth C. 1989 On the rise of epistemic meanings in English: An example of subjectification in semantic change. Language 65: 31–55. Traugott, Elizabeth C. and Ekkehard König. 1991 The semantics-pragmatics of grammaticalization revisited. In: Elizabeth C. Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, 1, 189–218. Amsterdam, Philadelphia: John Benjamins. Von Hoegen, Saskia. 2000 Entwicklung der spanischen Historiographie im ausgehenden Mittelalter. Am Beispiel der Crónicas de los Reyes de Castilla Don Pedro I, Don Enrique II, Don Juan I y Don Enrique III von Pero López de Ayala, der Generaciones y semblanzas von Fernán Pérez de Guzmán und der Crónica de los Reyes Católicos von Fernando del Pulgar. Frankfurt am Main: Peter Lang. Wilhelm, Raymund. 2001 Diskurstraditionen. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language Typology and Language

Entrenchment and discourse traditions in Spanish auxiliary selection

331

Universals / La typologie des langues et les universaux linguistiques / Sprachtypologie und sprachliche Universalien. An International Handbook / Manuel international / Ein internationales Handbuch, I, 467–477. Berlin, New York: De Gruyter. Yllera, Alicia. 1980 Sintaxis histórica del verbo español: Las perífrasis medievales. Zaragoza: Universidad de Zaragoza.

Melitta Gillmann, Universität Hamburg

Auxiliary selection in closely related languages: 1 the case of German and Dutch 1 Introduction Since Perlmutter (1978) and Burzio (1986), the selection of the perfect auxiliaries have or be2 with intransitives has been explained universally by unaccusativity. It has been argued that in languages like German, Italian and Dutch be occurs with unaccusatives, whereas have is used with unergatives. However, variation within and across languages poses a challenge for the so-called Unaccusativity Hypothesis, which is based on a strictly bipartite system. This problem has been overcome by Sorace’s Auxiliary Selection Hierarchy (ASH) (Sorace 2000, and much subsequent work). Sorace proposes that auxiliary selection is a gradient phenomenon. Depending on their semantic class, verbs are more or less categorical in the choice of have or be. Verbs of non-motional process and change of location constitute the “core classes”, i.e. core unergatives and core unaccusatives which are cross-linguistically consistent in auxiliary selection and categorically require either have or be. The remaining verb classes take an intermediate position on the ASH as they allow variation to different degrees. Within these intermediate classes, languages differ regarding where they draw the dividing line between have and be selection. This explains why auxiliary choice is not identical cross-linguistically and languages exhibit different degrees of variation. This paper also approaches the issue of cross-linguistic variation and the gradient organisation of auxiliary selection: by comparing the closely related languages German and Dutch, certain particularities of both languages will be revealed and will be traced back to stepwise, i.e. gradual, change. Therefore, auxiliary selection in West Germanic will be situated in the context of gradual grammaticalisation processes (section 2). Afterwards, section 3 compares contemporary German and Dutch with respect to auxiliary selection. Although both languages generally apply similar principles in the choice of have or be,

1 I would like to thank Renata Szczepaniak, Gunther De Vogelaer, Malte Rosemeyer and Peter Öhl for their constructive comments on an earlier version of this article. 2 Capitalised have and be indicate the corresponding auxiliaries across languages, while small letters in italics denote auxiliaries in individual languages (e.g. Germ haben vs Du hebben).

334

Melitta Gillmann

there are considerable differences which, interestingly, follow the gradient stages put forward by the ASH. Overall, it turns out that in Dutch auxiliary selection is characterised by telicity to a higher extent than in German. Drawing from a self-conducted corpus study, it will be demonstrated that these differences, in particular, concern manner-of-motion verbs which almost consistently select sein in contemporary German (section 4). Within this semantic group a feature [+ locomotion] has evolved that triggers the productive spread of sein ‒ at the expense of telicity. In the last section (section 5), recent cases of context extension in Dutch are examined. While transitive sentences pose a heavy constraint on sein selection in German, certain two-placed (and even transitive) verbs have currently adopted zijn in Dutch. It will be argued that these differences are facilitated by case syncretism which has affected Dutch to a greater extent than German and has made analogical extension to sentences with direct objects possible.

2 Auxiliary selection and grammaticalisation in West Germanic Sorace (e.g. 2000) argues that (cross-linguistic) variation with auxiliary selection is characterised by gradience. In Traugott and Trousdale (2010), the relation between synchronic gradience and diachronic gradualness typically resulting from grammaticalisation processes is discussed. The following section ties in with that issue; first, it is concerned with how gradient variation can generally be traced back to gradual change (2.1). Afterwards, the gradient organisation of auxiliary selection in West Germanic is situated into the context of perfect grammaticalisation and gradual change (2.2).

2.1 Gradience and gradual context extension The following is concerned with the question of how synchronic gradience generally emerges in the course of grammaticalisation. In order to illustrate the relationship between gradual change and gradient variation, some of the processes involved in grammaticalisation are introduced. Here, grammaticalisation is conceived of as a process in which a lexical item generalises in meaning and becomes a grammatical marker within a certain construction (e.g. Bybee 2010). As a result, a new grammatical construction arises. Crucial to this notion of grammaticalisation is that the emerging construction spreads to new contexts (Hopper and Traugott 2003; Himmel-

Auxiliary selection in closely related languages: the case of German and Dutch

335

mann 2004). This context extension, however, does not occur abruptly, but proceeds in small steps, i.e. gradually. In the course of the gradual spread, certain semantic and morphosyntactic features characterising the historical source are dropped, which is often referred to as “desemantisation” and “decategorisation” (e.g. Hopper 1991). The gradualness of the change may be traced back to the fact that the extension to new contexts is often motivated by “analogisation” based on similarity (Traugott and Trousdale 2010: 37; also see De Smet 2012). This means the emerging construction first spreads to contexts which highly resemble the historical source. Consequently, a reanalysed construction does not only lose aspects of its source through extension, but also retains particular features, socalled “features of persistence” (Hopper 1991: 22). This tension between extension and persistence causes synchronic gradience. Gradience in syntactic and morphosyntactic distribution often reflects the gradual development of a construction in a current language system (Traugott and Trousdale 2010). From a synchronic point of view, gradience is best captured by a prototype approach representing the cognitive organisation of categories (e.g. Langacker 1987; Lakoff 1987). There is a central representative of a category characterised by a bundle of features. More recent members of that category will group around this prototype with a gradiently increased distance. Section 3 argues that the ASH, as put forward by Sorace (e.g. 2000), can be conceived of as a prototype model. The most central representatives of be + [PP]3 correspond to the historical source.

2.2 Gradience in auxiliary selection resulting from historical change While former approaches mainly addressed the issue of auxiliary selection in Germanic from a synchronic point of view, recent work increasingly focuses on diachronic questions (see McFadden and Alexiadou e.g. 2010 and Ļęcki 2010 for English; Larsson 2009 for Scandinavian; Coussé 2014 for Dutch). Here, these insights are situated into the context of grammaticalisation which diachronically paves the way for the synchronic principles of auxiliary choice. According to Bybee and Dahl (1989) and Bybee, Perkins, and Pagliuca (1994: 51–105), the development of the present perfect or general past category universally follows a three-step pathway of grammaticalisation (fig. 1).

3 PP = P(ast) P(articiple)

336 Resultative He is gone.

Melitta Gillmann

Present Perfect Phil has already done his homework.

General Past Germ Caesar hat den Rubicon überschritten. ʻCaesar crossed the Rubicon.ʼ

Fig. 1: Universal grammaticalisation path of the past tense (according to Bybee, Perkins and Pagliuca 1994, simplified).4

The pathway presented in Figure 1 emanates from a resultative construction highlighting the resultant state of a previous event. This function can be demonstrated by frozen uses of the English be + [PP] construction, e.g. He is gone. In the example, the focus lies on speech time and the construction may be classified as present tense. McFadden and Alexiadou (e.g. 2010) argue that the construction never shifted from a resultative to eventive (or experiential) reading in English language history (e.g. *I am just arrived.). According to this, be + [PP] has never grammaticalised as perfect.5 The resultative reading is achieved compositionally by combining the copula be and the past participle which serves as predicative adjective (McFadden and Alexiadou 2010: 393). Thus, each constituent of the construction bears its lexical meaning and shows a high degree of autonomy (Lehmann 1995). It is important to note that the resultative reading can only be yielded with telic predicates bearing an inherent endpoint. Atelic verbs are not available in the resultative construction as they do not contain a target state in their lexical meaning.6 Furthermore, be + [PP] is restricted to intransitive contexts. This limited domain correlates with the weak context extension of the resultative construction and reflects its low degree of grammaticalisation. In further development, the construction broadens its functional domain gradually. The focus shifts from the resultant state to the previous event. Consequently, in this stage the construction also expands to atelic verbs. With respect to verb class, it is no longer liable to limiting constraints. The spread to atelic verbs indicates that the construction has been reanalysed as present perfect. However, this present perfect is always characterised by a certain amount of current relevance7 (Dentler 1998: 136–140). In line with this current rele-

4 Note that despite the discrete stages depicted in Figure 1, here grammaticalisation is conceived of as a gradual process and its sequential phases are regarded as overlapping steps involving continuous dimension. 5 A corpus study in Old High German and Old Saxon vernaculars shows that in these languages even the first occurences of be + [PP] are (weakly) grammaticalized (for more detail see Gillmann in prep.). 6 For the notion of target state see Kratzer (2000). 7 For discussion of the notion of “current relevance”, see Dahl and Hedin (2000), Zeman (2010: 204–207) and Gillmann (in prep.: chap. 5.3).

Auxiliary selection in closely related languages: the case of German and Dutch

337

vance, the construction is often combined with adverbials referring to recent past situations such as just or already. Today, English have + [PP] has reached this stage of grammaticalisation (She has just arrived).8 According to McFadden and Alexiadou (2010), the differences of grammaticalisation achieved by the constructions with have and be account for the fact that the English perfect only exhibits the auxiliary have. In German and Dutch, on the other hand, the grammaticalisation has advanced further. Here, both have and be + [PP] have lost their obligatory current relevance and may refer to general past situations (e.g. Du Vroeger hebben we regelmatig gefietst ʻIn the past we cycled regularlyʼ). Owing to this, a combination with adverbials of definite past time is possible (e.g. Du gisteren/Germ gestern ʻyesterdayʼ or Du vroeger/Germ früher ʻearlierʼ). As both have and be + [PP] have evolved into general past markers in Continental West Germanic (e.g. Dammel, Nowak, and Schmuck 2010: 346–350), they are expected to have gradually lost features associated with their historical sources by spreading to new contexts. The grammaticalisation of these analytic past markers, however, is special because two originally unrelated constructions diachronically converged in function and constricted themselves in their extension. The constructions emerged from two opposite resultative source domains: have + [PP] from telic transitive sentences9 and be + [PP] from telic intransitives. Crucially, have + [PP] extended faster than be + [PP] in the history of all West Germanic languages and gained terrain over more contexts earlier (see Kuroda 1999 and Gillmann in prep. for German; Coussé 2014 for Dutch; Ļęcki 2010 and McFadden and Alexiadou e.g. 2010 for English). Therefore, be + [PP] has only marginally expanded and today occurs less frequently. This paper focuses on the extension of be + [PP] in German and Dutch from a synchronic point of view. It is concerned with how the closely related languages differ and how they might be explained in terms of gradual change, i.e. extension and persistence.

3 Telicity – a more reliable predictor for auxiliary selection in Dutch than in German As argued in section 2.1, synchronic gradience may best be captured by a prototype model. Sorace᾿s (e.g. 2000) ASH can be conceived of as a prototype model. 8 Note that this construction diachronically also arises from a resultative construction, which can be exemplified by sentences like She has her mouth closed. These sentences additionally contain the notion of possession due to the predicate have (for a critical view on the notion of possession in the resultative source construction, see Kuroda 1999). 9 See footnote 5.

338

Melitta Gillmann

The telic change-of-location and change-of-state verbs constitute the most central representatives, i.e. the prototype, for be + [PP].10 From a historical perspective, this synchronic prototype corresponds to the contexts of the source construction characterised by the features [+telicity] and [−transitivity]. The “intermediate” verbs grouped around these prototypical verb classes show a lower degree of telicity and, consequently, are less categorical in auxiliary choice. Telicity has been identified to be the crucial determinant for auxiliary selection in Dutch (Zaenen 1988, 1993; Dowty 1991; van Hout e.g. 2004). By comparing Dutch to German, it becomes apparent that telicity plays a less important role for auxiliary selection in German. These differences may be illustrated with an application of the ASH (Figure 2). ASH (Sorace e.g. 2000)

Dutch

German

Change of location Change of state Continuation of state Existence of state Uncontrolled process Motional process Non-motional process

be be have/be have/be have have/be have

be be /have have/be have/be have be (!) have

Core Unergatives

Core Unaccusatives

Fig. 2: Auxiliary Selection Hierarchy in Dutch and German.

As predicted by the ASH, there is no variation within the group of core verbs: verbs of change of location invariably select be (1a, b), while non-motional process verbs always take have (2a, b). (1) a. Das Mädchen ist verschwunden. The girl be-3sg disappeared ʻThe girl has disappeared.ʼ b. Kerstfeest is gekomen. Christmas be-3sg arrived ʻChristmas has arrived.ʼ (2) a. Ich habe zu viel gearbeitet. I have-1sg too much worked ʻI used to work too much.ʼ

→ change of location

Germ

→ change of location

→ non-motional process

Du

Germ

10 The transitive prototype of have + [PP] is not captured by the ASH, as the hierarchy is limited to intransitive verbs; see Shannon (1990, 1995) and Coussé (2014).

339

Auxiliary selection in closely related languages: the case of German and Dutch

b. Wij hebben nog nooit zo gelachen. We have-1pl yet never so laughed ʻWeʼve never laughed like that before.ʼ

→ non-motional process

Du

Nevertheless, the comparison reveals that German differs from Dutch regarding change-of-state verbs: these verbs select be more consistently in Dutch than in German. For instance, the verbs toenemen/zunehmen ʻincreaseʼ and afnemen/ abnehmen ʻdecreaseʼ take haben in German, but zijn in Dutch. Historically, these verbs were only used transitively, in the sense of ʻto gain sth.ʼ or ʻto take sth. awayʼ and accordingly selected have (Kern 1912: 70–71). Only from 14th century were they used intransitively. In Dutch, this change in argument structure was followed by an auxiliary shift from hebben to zijn. Obviously, a diachronic leveling according to semantic verb class took place here. German, in contrast, has kept haben as the only auxiliary despite the new usage. The exceptional behaviour of these change-of-state verbs in German might be accounted for by their status as “degree achievements” (e.g. Dowty 1979: 88–90; Hay, Kennedy, and Levin 1999; Levin and Rappaport-Hovav 2005: 95–96) which do not necessarily denote a bounded event and hence are not prototypically telic.11 But this explanation does not apply to a second semantic group of change-of-state verbs that differ with respect to the selected auxiliary in German and Dutch: telic verbs denoting beginning or ending occur with zijn in Dutch, but (almost)12 invariably take haben in German (3a, b). It becomes apparent that telicity is a more reliable predictor for BE selection in Dutch than in German. (3) a. De film is geëndigd/begonnen. The movie be-3sg ended/begun ʻThe movie has ended/started.ʼ b. Der Film hat geendet/angefangen. The movie have-3sg ended/begun ʻThe movie has ended/started.ʼ

Du

Germ

Certain degree achievements like dry or rot, however, which are ambiguous between a telic and atelic reading, vary in terms of auxiliary selection in both

11 Nevertheless, according to Levin and Rapparport-Hovav (1995) degree achievements are unaccusative in either reading and consequently would be expected to select sein in German. 12 In some Low German dialects anfangen ʻto beginʼ selects sein. Apparently, in Northern German varieties auxiliary selection is more sensitive to verbal aspect than in Southern German; correspondingly, see the verbs of maintenance of position below.

340

Melitta Gillmann

languages. Interestingly, here the auxiliary switch conveys aspectual features (Szczepaniak 22011: 139): with these degree achievements, be is preferred when the result is emphasised (4b/5b),13 whereas have highlights the ongoing event (4a/5a): (4) a. Die Wäsche hat schon 2 Stunden getrocknet. The laundry have-3sg already two hours dried ʻThe laundry has been drying for two hours already.ʼ b. Die Wäsche ist getrocknet, du kannst sie abhängen. The laundry be-3sg dried you can-2sg her take-down ʻThe laundry is dry, you can take it off the clothesline.ʼ (5) a. Ik heb gisteren lekker uitgeslapen. I have-1.sg yesterday lovely slept-in ῾Yesterday, I had a good night᾿s rest.᾿ b. Vandaag ben ik uitgeslapen. Today be-1.sg I slept-in ῾Today, I am well rested.᾿

Germ

Germ

Du

Du

Within the group of continuation of state and existence of state, have is selected as a rule. Basically the only exceptions in both languages are the highly frequent copula verbs bleiben/blijven ʻto remainʼ and sein/zijn ʻto beʼ. Apparently, there has also been a weak spread to atelic verbs in Dutch. This spread proceeds along the gradient stages of the ASH as atelic verb classes which are relatively close to the telic core verbs are the first to be affected. Within these intermediate verb classes the adaption of be is driven by token frequency: highly frequent verbs are the first to adopt the auxiliary. Within the class of continuation-of-state and state verbs, the standard languages do not differ considerably from each other. Interestingly, non-standard varieties give further evidence for the above-mentioned observation that telicity determines auxiliary selection in Dutch to a greater extent than in German. In certain non-standard varieties of Dutch, especially in Flemish varieties such as those spoken in Belgium, both copulas allow hebben beside zijn (6a, b).14

13 For a similar effect in France, see Kailuweit (2011) and Heidinger (this volume). 14 According to de Rooij (1988: 26–28), the auxiliary alternation in these varieties agrees with sentential aspect. Whenever the duration of the state is highlighted, hebben is preferred. In a corpus study within the Spoken Dutch Corpus (also see section 4.1), no significant correlation between duration and hebben selection could be identified. However, in the varieties which exhibit auxiliary alternation with the copula zijn, hebben is preferred whenever zijn serves as

Auxiliary selection in closely related languages: the case of German and Dutch

341

In Southern German, in contrast, states even show a stronger tendency towards sein than in the standard language. Here, maintenance of position verbs such as stehen ʻto standʼ and kleben ʻto stickʼ take sein despite their atelic meaning (7a, b). These forms are highly accepted in Southern German, which is reflected by the fact that they even occur in regional newspapers.15 (6) a. Ik heb twee jaar in Frankrijk geweest. I have-1sg two year in France been ʻI have been living in France for two years.ʼ b. Ik ben op de top van de berg geweest. I be-1sg on the top of the mountain been ʻI have been to the top of the mountain.ʼ (7) a. Ein Kaugummi war auf dem Teppich geklebt. A chewing-gum be-past-3sg on the carpet stuck ʻA wad of chewing gum stuck to the carpet.ʼ b. In der Zeitung ist gestanden, dass […] In the newspaper be-3sg stood that ʻThe newspaper reported that […]ʼ

Fl

Fl

Germ

Germ

So far, the general conclusion may be drawn that auxiliary selection is determined by telicity in Dutch to a greater extent than in German. As argued above, telicity diachronically originates in the resultative source construction. Though zijn + [PP] has reached the stage of general past reading (see Figure 1), it is still strongly associated with telicity. Interestingly, the differences between German and Dutch are gradient and follow the steps provided by the ASH. The observation that telicity plays a more important role for auxiliary selection in Dutch than in German is particularly confirmed by manner-of-motion verbs. Therefore, in the next section, these verbs are examined in more detail.

auxiliary of the progressive construction zijn + [aan het Vinf] (for more details, see Gillmann in prep.). 15 In accordance with this, Werth et al. (2011) show in a recent ERP study that, within the maintenance-of-position group, Southern German speakers exhibit no significant differences in their acceptability judgments between the auxiliaries haben and sein, whereas Northern German speakers significantly disapprove of the usage of sein. 16 Interestingly, geweest hebben prevailed in historical Dutch. Only from the 16th century on has the copula verb increasingly occurred with the auxiliary zijn; see Coussé 2014.

342

Melitta Gillmann

4 Auxiliary selection with manner-of-motion verbs: a corpus study 4.1 Telicity conditioning auxiliary selection in Dutch Manner-of-motion verbs are special in that they are ambiguous between telic and atelic aspect. Combined with an element delimiting the event, i.e. a directional phrase, they are telic; without they are atelic. It is well known that with Dutch motional process verbs the choice of hebben or zijn consistently follows sentential aspect. Zijn is selected whenever the action is telicised (8a). In all other cases hebben occurs (8b). (8) a. Stijn is naar huis gelopen. [telic] Stijn be-3sg to home walked ʻStijn walked home.ʼ

Du

b. Stijn heeft urenlang gelopen. [atelic] Stijn have-3sg for-hours walked ʻStijn was walking for hours.ʼ

Du

This strong alignment of auxiliary selection to telicity within the group of manner-of-motion verbs is unique to Dutch. Sorace mentions that “Dutch is the most systematic language in this respect: all manner-of-motion verbs switch from hebben to zijn when they are embedded in a predicate that has been telicised by a directional phrase” (Sorace 2000: 875). In order to analyse the auxiliary behaviour of verbs with variable auxiliary selection, I conducted a corpus investigation in the Spoken Dutch Corpus. The corpus study shows that Sorace᾿s statement holds for spoken varieties of Dutch. Table 1 provides the results yielded for the motional process verbs gelopen ʻwalkedʼ, gevlogen ʻflownʼ, gereden ʻdrivenʼ and gezwommen ʻswumʼ.

Tab. 1: Auxiliary selection with intransitive manner-of-motion verbs in the Spoken Dutch Corpus. Type

zijn in % (absolute value)

hebben in % (absolute value)

gelopen ʻrunʼ gereden ʻdrivenʼ gevlogen ʻflownʼ gezwommen ʻswumʼ

59.5 % 68.9 % 68.1 % 10.0 %

40.5 % 31.1 % 31.9 % 90 %

(116) ( 51) ( 32) ( 2)

(79) (23) (15) (18)

Auxiliary selection in closely related languages: the case of German and Dutch

343

The corpus study shows that auxiliary selection within the group of manner-of-motion verbs goes along consistently with aspectual features. Atelic sentences clearly prefer hebben (see fig. 3).

Fig. 3: Auxiliary selection with atelic manner-of-motion verbs.17

Zijn is significantly used in telic sentences containing for example a prepositional phrase which constitutes the terminus of the event. However, Figure 4 reveals that hebben is not excluded in sentences denoting a motion limited by a directional phrase.

Fig. 4: Auxiliary selection with motion verbs with a delimiting element.18

17 Some instances of the verb lopen ʻto runʼ deviate from this pattern. This concerns idioms such as Het is anders gelopen ʻIt came differentʼ which exclusively offer the auxiliary zijn, although these sentences denote atelic events. 18 Note, that due to the so-called “counterfactual effect” (McFadden and Alexidaou 2010) verbs which usually appear with zijn may switch to hebben in counterfactual mood in spoken varieties of Dutch. For more details, see Shannon (1995).

344

Melitta Gillmann

Hebben rarely occurs in contexts with a directional phrase. In these sentences the boundaries of the event are defocused. Thus, sentences such as (9) are indeed atelic despite the directional phrase. (9) nadat ze uren in de regen hadden gelopen naar auto of trein. Du After they hours in the rain had-3.pl walked to car or train ῾After they had been walking in the rain for hours in order to reach their cars or the train᾿ The investigation of manner-of-motion verbs confirms that zijn + [PP] is highly sensitive to telicity. As argued above, telicity is a feature retained from the resultative source. Apparently, in Dutch, context extension is not very advanced in terms of verbal aspect, and zijn + [PP] is strongly restricted to telic contexts. According to Paul (1905) and Sapp (2011), also in historical stages of German manner-of-motion verbs used to shift the auxiliary when the motional process was telicised. However, the following section shows that this does not apply to Contemporary German.

4.2 [+Locomotion] triggering sein selection in German Similar to Dutch, auxiliary selection of manner-of-motion verbs was determined by sentential aspect in historical stages of German (see Paul 1905 for

Tab. 2: Auxiliary selection with manner-of-motion verbs in German newspaper texts. Type

sein in % (absolute value)

haben in % (absolute value)

gesprungen ʻjumpedʼ gelaufen ʻrunʼ geflogen ʻflownʼ gefahren ʻdrivenʼ geschwommen ʻswumʼ geritten ʻridden (on horseback)ʼ getanzt ʻdancedʼ

99.5 % (1313) 99.0 % (602)17 91.7 % (1358) 90.3 % (1450) 89.1 % (391) 74.3 % (249) 4.9 % (26)

0.5 % (7) 1.0 % (6) 8.3 % (123) 9.7 % (156) 10.9 % (48) 25.7 % (84) 95.1 % (505)

19 Tab. 2 presents the results of a pre-study. In the final investigation similiar tendencies are yielded (see Gillmann in prep.). Note that the absolute data provided for the verbs gelaufen, gefahren and geflogen represent a sample. For practical reasons, only a limited subset of these verbs could be examined. In fact, the corpus contained at least 2,000 instances of each of the three verbs.

Auxiliary selection in closely related languages: the case of German and Dutch

345

Middle High German; Sapp 2011 for Early New High German). Contemporary German, however, deviates from that pattern. As has been noted before, German manner-of-motion verbs show a strong propensity to the auxiliary sein (e.g. Keller and Sorace 2003: 701). This observation is confirmed by a corpus investigation within the corpora of the IDS Mannheim (DeReKo) which contain newspaper texts from different regional areas of Germany (for more detail, see Gillmann 2011, in prep.). The following items have been examined: At first glance, it is striking that apart from tanzen ʻto danceʼ – which can be rather classified as a controlled process verb – all the motional process verbs behave homogeneously and significantly tend towards sein. This tendency becomes even more apparent when sentences without an accusative are considered. Here, sein occurs in almost 100 % of cases. Only schwimmen ʻto swimʼ and reiten ʻto ride on horsebackʼ provide the auxiliary haben in 4 % and 3.1 % in intransitive use (see Figure 5).

Fig. 5: Auxiliary selection with intransitive manner-of-motion verbs in German newspaper texts.

The results yielded in the corpus study suggest that the most frequently occurring motion verbs show the strongest tendency towards sein. While geschwommen and geritten which are limited to around 440 and 333 occurrences in the corpus allow haben in 3–4 % of cases, haben never occurs and even appears to be ungrammatical with the frequently used 20 verbs gelaufen and gefahren. This does not seem to be a coincidence. According to Bybee (2010) the repetition of a sequential unit leads to the chunking of its elements. Consequently, the construction loses transparency and is stored as a whole. It appears that

20 See footnote 19.

346

Melitta Gillmann

the auxiliary sein which was originally restricted to telic manner-of-motion verbs through repetition was generalised also in atelic use.21 From these results the conclusion may be drawn that motional process verbs have lexicalised sein. As a consequence, auxiliary choice with manner-of-motion verbs is not sensitive to telicity anymore (Gillmann 2011, in prep.). Once again, it becomes apparent that verbal aspect is not a reliable predictor for haben and sein selection in German. Instead, the investigation shows that the only contexts obligatorily requiring haben are prototypical transitive sentences (Hopper and Thompson 1980). These sentences contain an individuated object in combination with an additional directional phrase which delimits the event and provides a terminus for the change of location (10a). Thus, the object is characterised as an incremental theme (Dowty 1991: 567–571). Furthermore, sentences with a definite or individuated object strongly tend towards haben selection (10b). Although this object does not “measure out the event” (Tenny 1995: 51), it can be considered a patient (or theme) since it is the argument of which the change of location is predicated. In contrast, in combination with an incorporated, non-referential object sein is strongly preferred (10c). These sentences are intransitive. The same applies to sentences containing a path object: in these contexts, sein is used invariably (10d).22 Only geschwommen and geritten allow both haben and sein with path objects. (10) a. Er hat/ *ist die Kinder in die Schule gefahren. He have-3sg/ *be-3sg the children to the school driven →prototypical transitive ʻHe drove the children to school.ʼ

Germ

b. Sie hat/ ?ist den Mercedes gefahren. She have-3sg/ ?be-3sg the Mercedes driven →individuated object ʻShe drove the Mercedes.ʼ

Germ

c. Sie ist/ ?hat Mercedes gefahren. She be-3sg/ ?have-3sg Mercedes driven →incorporated object ʻShe drove Mercedes.ʼ

Germ

21 Similarly, Rosemeyer (2014) explains the resistance of motion verbs to lose ser-selection in Old Spanish by frequency. 22 Interestingly, path objects trigger selection of avoir in French: Jʼai monté une seconde fois lʼescalier. ʻI went up the stairs a second time.ʼ (see Heidinger this volume).

Auxiliary selection in closely related languages: the case of German and Dutch

d. Er ist/ ??hat einen Umweg gefahren. →path-object He be-3sg/ ??have-3sg a detour driven ʻHe took a detour by car.ʼ

347 Germ

According to Shannon (e.g. 1990, 1995) and Arnett (2007), highly transitive sentences may be considered the prototypical domain of haben + [PP]. Also from a historical point of view, haben + [PP] originates in transitive sentences (Dieninghof 1904: 15; Grønvik 1986: 30), whereas sein + [PP] arises from intransitive predicates. Apparently, the contexts of the historical source construction have remained most consistent in the selection of haben up to today. This can be accounted for in terms of persistence (Hopper 1991). Intransitive manner-of-motion verbs, however, strongly prefer the auxiliar sein. Randall et al. (2004) and Randall (2007) point out that the verb tanzen requires sein even in atelic use, whenever a motional displacement is denoted. She explains this behaviour with the semantic feature [+locomotion] triggering sein selection in German. Obviously, this explanation can be applied to motional verbs in general: it appears that sein + [PP] is strongly associated with the feature of [+locomotion]. There are certain pieces of evidence in favour of a semantic feature of [+locomotion] which productively triggers sein selection in German. For instance, particular experiencer verbs denote [+locomotion] when they are combined with a directional phrase (11a). Although these verbs generally select haben (11b), when referring to a motional process they require sein (11c). (11) a. Er keucht nach Hause. He wheeze-3sg to home ʻShort of breath, he is walking home.ʼ

Germ

b. Nach dem Aufstieg hat/ *ist er ganz schön. After the climb have-3sg/ *be-3g he pretty-much gekeucht wheezed ʻAfter the climb, he was basically gasping for air.ʼ

Germ

c. Er *hat/ ist nach Hause gekeucht. He *have-3sg/ be-3sg to home wheezed ʻShort of breath, he walked home.ʼ

Germ

The most important evidence in favour of the semantic feature of [+ locomotion], however, comes from the productive spread of sein to new motional process verbs. Verbs referring to non-prototypical motional processes which usual-

348

Melitta Gillmann

ly take haben in standard language increasingly occur with the auxiliary sein in Colloquial German (12a, b, c): (12) a. Ich bin eingeparkt. I be-1sg parked ʻI have parked (the car).ʼ

Germ

b. Bist du schon mal gesurft? Be-2sg you already once surfed ʻHave you surfed before?ʼ

Germ

c. Früher bin ich immer gependelt. In-the-past be-1sg I always commuted ʻIn the past, I used to commute (to work).ʼ

Germ

All in all, it becomes apparent that German manner-of-motion verbs are special. Unlike their Italian and Dutch counterparts, they have generalised sein due to high token frequency of particular manner-of-motion verbs. This is in line with the observation that the impact of telicity as a determinant of auxiliary selection has decreased in German. Resulting from the generalisation of the auxiliary within the context of motional process, a semantic feature [+locomotion] has emerged which is strongly associated with sein + [PP]. [+Locomotion] propels the extension of sein to new motion verbs at the expense of telicity. The productive spread to motion verbs such as surfen ῾to surf ᾿ and pendeln ῾to commute᾿ in (11) is based on similarity to manner-of-motion verbs such as gelaufen, gefahren and geflogen which have generalised sein in all contexts ‒ regardless of the aspect of the predicate. Owing to this similarity, the spread to manner-of-motion verbs only constitutes a small step and can be described as gradual. From a universal perspective, the auxiliary behaviour of manner-of-motion verbs in German is striking. Since these verbs are characterised by [+agentivity] and [−telicity], they would be expected to generalise haben. It becomes clear that the extension of sein + [PP] has advanced further in German than in Dutch. In the course of this extension, the role of telicity as a productive determinant of auxiliary selection has decreased. However, transitivity imposes a heavy constraint on the extension of sein + [PP] in German.

5 Recent cases of context extension in Dutch: zijn selection with two-placed verbs Section 4 showed that transitive sentences pose a heavy constraint on the selection of sein in German. Dutch, in contrast, is less restrictive concerning argu-

Auxiliary selection in closely related languages: the case of German and Dutch

349

Tab. 3: Two-placed verbs selecting zijn in Dutch.21 Verb Class

Verbs

(1) Experiencer [−control] (2) Beginning and Ending (3) Change of Location

vergeten ʻto forgetʼ, verliezen ʻto loseʼ, verleren ʻto unlearnʼ beginnen ʻto beginʼ, ophouden ʻto ceaseʼ passeren ʻto passʼ, volgen ʻto followʼ

ment structure. What sets Dutch apart from other languages which exhibit alternation of have and be is that certain groups of two-placed and even transitive verbs diachronically switched from hebben to zijn (for discussion, see Lieber and Baayen 1997; Hoekstra 1999). From a historical perspective, zijn penetrating into transitive sentences is a fairly young development basically starting in the 18 th/19 th centuries (Kern 1912). As Table 3 illustrates, this process concerns certain formally and/or semantically determined verb classes. First, there is a group of experiencer verbs formally marked by the prefix ver-: vergeten ʻto forgetʼ, verliezen ʻto loseʼ and verleren ʻto unlearnʼ. These verbs share a subject that displays a low degree of control. Regardless of the obligatory direct object, they show variable auxiliary selection in Contemporary Dutch. According to van Es and van Caspel (1974: 41–42), the choice of the auxiliary zijn emphasises that the event is not volitionally intended by the subject argument and, consequently, detracts from its control.24 Similarly, Hoekstra explains the zijn-selection of these verbs in terms of two internal arguments (Hoekstra 1999: 75–76). Second, verbs of beginning and ending have to be mentioned. As demonstrated above, these verbs regularly take zijn in intransitive use, which is to be expected due to their telic meaning. However, they show a strong tendency towards zijn selection even in combination with an additional verbal or nominal complement (Hitler is de oorlog begonnen. ʻHitler has initiated the war.ʼ). This behaviour is surprising since these sentences may be classified as prototypical transitive denoting an “effective carrying over of an activity from an A[gent] to a patient” (Hopper and Thompson 1980: 279). It appears that prototypical transitive sentences pose a less categorical constraint for be selection in Dutch than in German (see chapter 4.1). The following focuses on the last group, i.e. two-placed motion verbs exhibiting the auxiliary zijn. The motion verbs provided in (13) show a strong

23 For the sake of brevity, the following focuses on motion verbs as presented in (3). 24 This explanation challenges van Hout᾿s (2004: 60) claim that agentivity does not figure in Dutch auxiliary selection at all.

350

Melitta Gillmann

propensity towards zijn. Interestingly, they exhibit the same auxiliary alternation as the manner-of-motion verbs examined in 4.1; hebben may occur when the duration of the event is highlighted, see (13d) (Lieber and Baayen 1997: 807). (13) a. Ik ben zijn voetsporen gevolgd. I be-1sg his footprints followed ‘I followed his footprints.᾿

Du

b. De groep is de stad genaderd. The group be-3sg the town neared ῾The group approached the city.᾿

Du

c. Hij is me gisteren tegengekomen. He be-3sg me yesterday came-towards ῾Yesterday, he came towards me.᾿

Du

d. De weg heeft de rivier gevolgd.23 The road have-3sg the river followed ῾The road followed the river.᾿ (from Lieber and Baayen 1997: 807)

Du

It is striking that many studies on Dutch auxiliary selection refer to sentences such as (13a–d) as “transitive” (e.g. Honselaar 1987: 61–63; Lieber and Baayen 1997: 809–810; Hoekstra 1999: 75–76; Haeseryn et al. 1997: 79–80). Apparently, here the notion of transitivity is based on syntactic features, i.e. the presence or absence of a direct object. Especially in the sentences (13b, c), the object is not affected by the action, but rather serves as a goal encoding the terminus of the event. These sentences do not score high in terms of semantic transitivity. In German, this low degree of transitivity is reflected in argument structure containing a dative-marked object, i.e. an indirect object. That is to say, the German sentences in (14) are not even syntactically transitive. gefolgt. (14) a. Ich bin seinen Fußspuren I be-1sg his footprints-dat followed ῾I followed his footprints.᾿ genähert. Die Gruppe hat sich der Stadt The group have-3sg itself the town-dat neared ῾The group approached the city.᾿

Germ

Germ

25 Note that, volgen ῾to follow᾿ favours hebben over zijn, e.g. Du Ik heb de lezing gevolgd. ῾I participated in the lecture᾿, when it denotes a non-motional process.

Auxiliary selection in closely related languages: the case of German and Dutch

Er ist mir gestern entgegengekommen He be-3sg me-dat yesterday walked-towards ῾Yesterday, he walked towards me.᾿

351 Germ

The same applied to historical Dutch, which explains how the selection of zijn originally came about. While earlier Middle Dutch exhibited a stable case system which morphologically distinguished four cases, “[b]y the end of the fifteenth century Dutch case endings had almost completely eroded” (Van der Wal 1993). Up to today, there has been no morphological distinction between direct and indirect object. In consequence, the verbs in (13) are conceived as transitive in Contemporary Dutch. This is reflected by the fact that volgen ῾to follow᾿ allows passivation (15a) which is ungrammatical in German (15b): (15) a. Ik werd gevolgd door de Amerikaanse geheime dienst. Du I became-1sg followed by the American secret service ῾I was hunted by the CIA.᾿ b. *Ich wurde gefolgt vom amerikanischen I became-1sg followed by-the American Geheimdienst. secret service ῾I was hunted by the CIA.᾿

Germ

It becomes clear that the occurrence of zijn in sentences such as (13) does not result from a diachronic extension to syntactically transitive sentences. The transitive verbs select zijn because they had been intransitive originally containing an indirect object. After the case marking system had been given up, the formal distinction between indirect and direct objects was abandoned. In consequence, the verbs are today conceived as transitive. Nevertheless, they did not drop the auxiliary zijn. This means that a diachronic change in argument structure gave rise to these two-placed verbs selecting zijn.26 Sentences such as (13) provided models for analogisation of two-placed verbs with an originally accusative marked object. For instance, the semantically related verb passeren ῾to pass sth.᾿ has diachronically adopted zijn. This is surprising since the verb exhibits a relatively high degree in transitivity in

26 Though, note that in earlier Middle Dutch the verb volgen selected hebben exclusively. Only from the 14th century on did it switch to zijn, which according to Kern can be traced back to a change in the aspectual semantics of the verb (Kern 1912: 189–192; also see Paul 1905: 190).

352

Melitta Gillmann

German and only selects haben. In Dutch, passeren exhibits a clear preference for zijn today (16a).27 (16) a. zodra je de fictieve grens Sommerville-Cambridge Du as-soon-as you the fictitious border Sommerville-Cambridge was gepasseerd was-3sg passed ῾as soon as you had crossed the fictitious border Sommerville-Cambridge᾿ b. ’k heb daar gepasseerd op de autostrade en ’k I have-1sg there passed on the highway and I had er nog niet op gedacht. had-1sg there not yet of thought ῾I was driving by on the highway without thinking of it.᾿

Du

c. en als laatste daar hebben we toch even naast and as last there have-1pl we just alongside gepasseerd zeer mooie plaats in Brugge passed very beautiful square in Brugge ῾and finally we walked past a very beautiful square in Brugge᾿

Du

In the Spoken Dutch Corpus, only eight of 55 instances exhibit the auxiliary hebben (16b, c). Interestingly, the selection of hebben is not motivated by the presence or absence of a direct object. Similar to intransitive manner-of-motion verbs, the choice of the auxiliary rather depends on the aspect of the sentence (also see Kern 1912: 158, 189–190, 213; Honselaar 1987: 62). Hebben appears when the durational course of the action is emphasised and the boundaries are defocused. This means that hebben is preferred when an object denoting an extensional path occurs or when the object delimiting the motional event is omitted (16b), in which case the action is atelic. Zijn, in contrast, appears when the path is conceived of as transitional (16a). It becomes clear that, similar to intransitive manner-of-motion verbs, auxiliary variation with passeren is motivated by telicity rather than its degree of transitivity. According to Kern (1912: 213), is gepasseerd occurred alongside heeft gepasseerd already in the 16th century. In contrast to the verbs examined above (13), this change in auxiliary selection constitutes a case of context extension. The fact that the auxiliary switch is motivated by telicity suggests that passeren

27 Note that the German equivalent passieren selects haben due to its relatively high degree of transitivity.

Auxiliary selection in closely related languages: the case of German and Dutch

353

adopted zijn through analogisation to intransitive motion verbs as presented in section 4.1 However, it is striking that zijn spread to passeren despite the presence of a direct object. It is very likely that the adoption of zijn was possible only because a structural model existed provided by verbs such as volgen ῾to follow᾿ or naderen ῾to approach᾿. After the loss of the case marking system by the end of the 15th century, these verbs with an originally dative-marked object have increasingly been conceived as transitive. Consequently, they provided a structural model for the analogical extension of zijn to verbs with a direct object. Thus, the existence of these semantically related two-placed verbs selecting zijn created the conditions for the spread of zijn to the transitive passeren. This means that the loss of morphological case marking facilitated the extension of zijn into sentences containing a direct object. After motion verbs such as passeren had adopted zijn, constraints concerning argument structure had, apparently, been loosened. From the 18th to the 19th century, zijn spread to other two-placed and even transitive predicates, see the experiencer verbs (e.g. vergeten ῾to forget᾿) and verbs of beginning and ending (e.g. beginnen ῾to begin᾿) in Table 5. Presumably, the fact that Dutch also exhibits zijn with two-placed (and transitive) verbs can be traced back to case syncretism which neutralised the distinction between direct and indirect objects. This explains why transitivity poses a higher constraint for context extension of be + [PP] in German; as the German case system has remained stable at four cases throughout the German language history (see for example Dammel and Gillmann 2014), sentences such as (14) have always been formally distinct from transitive sentences. Consequently, a model for analogical extension was missing, which prevented the spread of sein to transitive sentences. All in all, the diachronic spread of zijn within the group of two-placed verbs can be described as gradual. Zijn did not spread to transitive sentences abruptly, but gradually affected predicates with an increasing degree of transitivity. Interestingly, in Contemporary Dutch, two-placed motion verbs exhibit the same auxiliary alternation as the manner-of-motion verbs examined in 4.1 Sentential aspect is the decisive determinant for the choice of hebben or zijn. Apparently, this auxiliary behaviour is motivated by analogy to intransitive manner-of-motion verbs.

6 Summary − language-specific extension Overall, it becomes apparent that be + [PP] diachronically expanded to new contexts in German and Dutch, respectively. This spread, which often occurred

354

Melitta Gillmann

through analogisation, proceeded in small steps and can be described as gradual. In the course of the context extension, the closely related languages have both reduced and preserved features associated with the resultative source construction. However, they differ regarding which features have been subject to change. While in German the role of telicity has been reduced, in Dutch zijn is currently spreading to transitive sentences. This diverging course of the extension is reflected by the distribution of the auxiliaries in the contemporary languages. In German, auxiliary selection often neglects verbal aspect: there are telic intransitives that select haben (e.g. zunehmen ʻincreaseʼ, abnehmen ʻdecreaseʼ or the verbs of beginning and ending) as well as certain atelics that take sein. Concerning the latter, manner-of-motion verbs are of particular interest as they categorically take sein in German regardless of the sentential aspect. With reference to Randall (2007), it has been argued that here a new feature [+locomotion] has evolved which productively determines the selection of sein within the group of manner-of-motion verbs. However, prototypical transitive sentences which constitute the central domain of haben + [PP] disallow sein even within the group of manner of motion. Dutch auxiliary selection, in contrast, is highly sensitive to telicity. As opposed to German, a diachronic levelling may be observed which brings auxiliary choice in line with sentential aspect. Here, only a few highly frequent states (e.g. is geweest ʻhas beenʼ) deviate from the aspectual pattern. A corpus investigation shows that manner-of-motion verbs productively shift to zijn whenever the action is telicised. This, interestingly, also applies to two-placed motion verbs which are characterised by auxiliary variation, depending on whether the action is highlighted in duration or the transition is emphasised. As opposed to German, Dutch allows zijn with certain two-placed and (even transitive) verbs. It has been argued that the diachronic spread to twoplaced verbs has been facilitated by case syncretism which affected Dutch to a much higher extent than German. Owing to the loss of case morphology, the formal distinction between direct and indirect object had disappeared by the end of the 15 th century. This gave rise to models for the analogical extension of zijn to syntactically and even semantically transitive predicates. Thus, the fact that Dutch had undergone case syncretism to a much higher extent than German created the conditions for the spread of zijn to transitive predicates. Though the extension to new contexts proceeded in small steps and turned out to be gradual, different features of the historical source have been subject to change in German and Dutch. It becomes clear that, although grammaticalisation follows universal pathways, the language system and previously existing patterns have an influence on the course of the extension of an emerging

Auxiliary selection in closely related languages: the case of German and Dutch

355

construction since they may serve as models. As a result, slightly varying conditions may lead to different distributions of a construction in closely related languages.

References Arnett, Carlee. 2007 Transitivity parameters and auxiliary selection by L2 students of German. In: Raúl Aranovich (ed.), Split Auxiliary Systems. A Cross-Linguistic Perspective, 25–46. Amsterdam, Philadelphia: John Benjamins. Burzio, Luigi. 1986 Italian Syntax. A Government-Binding Approach. Dordrecht: Reidel. Bybee, Joan and Östen Dahl. 1989 The creation of tense and aspect systems in the languages of the world. Studies in Language 13(1): 51–103. Bybee, Joan Revere Perkins and William Pagliuca. 1994 The Evolution of Grammar. Tense, Aspect, and Modality in the Languages of the World. Chicago: University of Chicago Press. Bybee, Joan. 2010 Language, Usage and Cognition. Cambridge: Cambridge University Press. Coussé, Evie. 2014 Lexical expansion in the have and be perfect in Dutch. A constructional prototype account. In: Diachronica 31(2): 159–191. Dahl, Östen and Eva Hedin. 2000 Current relevance and event reference. In: Dahl, Östen (ed.), Tense and aspect in the languages of Europe, 385–401. Berlin, New York: De Gruyter. Dammel, Antje, Jessica Nowak Jessica and Mirjam Schmuck. 2010 Strong verb paradigm leveling in four Germanic languages. Journal of Germanic Linguistics 22(4): 337–359. Dammel, Antje and Melitta Gillmann. 2014 Relevanzgesteuerter Umbau der Substantivflexion im Deutschen. Spiegelt Diachronie Typologie? Beiträge zur Geschichte der Deutschen Sprache und Literatur (PBB) 136(2): 173–229. Dentler, Sigrid. 1998 Gab es den Präteritumschwund? In: John Ole Askedal and Wilhelm Zickfeldt (eds.), Historische germanische und deutsche Syntax. Akten des internationalen Symposiums anlässlich des 100. Geburtstages von Ingerid Dal, Oslo, 27.9.–1.10. 1995, 133–147. (Osloer Beiträge zur Germanistik.) Frankfurt am Main, New York: Peter Lang. DeReKo = http://www.ids-mannheim.de/cosmas2/. Dieninghoff, Joseph. 1904 Die Umschreibungen aktiver Vergangenheit mit dem Participium Praeteriti im Althochdeutschen. Bonn: Georgi. Dowty, David R. 1979 Word Meaning and Montague Grammar. The Semantics of Verbs and Times in Generative Semantics and in Montague’s PTQ. (Synthese Language Library.) Dordrecht: Reidel. Dowty, David R. 1991 Thematic proto-roles and argument selection. Language 67(3): 547– 619. Es, Gustaaf van and Paulus P. van Caspel. 1974 De verbale groepvorming in de zin I [Verbal clusters in the sentence I]. Groningen: Rijksuniversiteit te Groningen. Gillmann, Melitta. 2011 Die Grammatikalisierung des sein-Perfekts. Beiträge zur Geschichte der deutschen Sprache und Literatur (PBB) 133(2): 203–234.

356

Melitta Gillmann

Gillmann, Melitta. (in prep.) Die Perfektkonstruktionen haben + V-PP und sein + V-PP aus gebrauchsbasierter Perspektive. Eine Korpusuntersuchung im Althochdeutschen, Altsächsischen und Neuhochdeutschen. Ph.D. thesis, University of Hamburg. Grønvik, Ottar. 1986 Über den Ursprung und die Entwicklung der aktiven Perfekt- und Plusquamperfektkonstruktionen des Hochdeutschen und ihre Eigenart innerhalb des germanischen Sprachraumes. Oslo: Solum Forlag. Haeseryn, Walter. 1997 Algemene nederlandse spraakkunst [A Comprehensive Grammar of Dutch]. Groningen: Nijhoff. Haspelmath, Martin. 1998 Does grammaticalisation need reanalysis? Studies in Language 22: 315–351. Hay, Jennifer, Christoper Kennedy and Beth Levin. 1999 Scalar structure underlies telicity in ‘degree achievements’. In: Tanya Mathews and Devon Strolovitch (eds.), Proceedings of Semantics and Linguistic Theory (SALT) 9, 127–144. Ithaca. Heidinger, Steffen. (this volume) The auxiliary selection of French monter ‘move upward’ from the 16 th to the 20 th century. Himmelmann, Nikolaus. 2004 Lexicalization and grammaticization. Opposite or orthogonal? In: Walter Bisang, Nikolaus Himmelmann and Björn Wiemer (eds.), What Makes Grammaticalization? A Look from its Fringes and its Components, 21–42. (Trends in linguistics. Studies and monographs.) Berlin, New York: De Gruyter. Hoekstra, Teun. 1999 Auxiliary selection in Dutch. Natural Language and Linguistic Theory 17: 67–84. Honselaar, Wim. 1987 Zijn v. Hebben in het samengesteld perfectum [Be vs. have in the compound perfect]. De nieuwe taalgids 80(1): 55–68. Hopper, Paul J. and Sandra Thompson. 1990 Transitivity in grammar and discourse. Language 56(2): 251–299. Hopper, Paul J. 1991 On some principles of grammaticization. In: Elizabeth Closs Traugott and Bernd Heine (eds.), Approaches to Grammaticalization, 17–36. Amsterdam, Philadelphia: John Benjamins. Hopper, Paul J. and Elisabeth Closs Traugott. 2003 Grammaticalization. (Cambridge Textbooks in Linguistics.) Cambridge, New York: Cambridge University Press. Hout, Angeliek van. 2004 Unaccusativity as telicity checking. In: Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 60–83. Oxford: Oxford University Press. Kailuweit, Rolf. 2011 Le choix de lʻauxiliaire. Être ou avoir en français standard contemporain. In: Renata Enghels, Machteld Meulleman and Clara Vanderschueren (eds.), Peregrinatio in Romania. Articulos en homenaje a Eugeen Roegiest con motivo de su 65 cumpleanos, 397–420. Gent: Academia Press. Kern, Johan Hendrik. 1912 De met het participium praeteriti omschreven werkwoordsvormen in’t Nederlands [The periphrastic forms with past participle in Dutch]. Amsterdam: Müller. Kratzer, Angelika 2000. Building statives. In: Lisa J. Conathan, Jeff Good, Darya Kavitskaya, Alyssa B. Wulf and Alan C. L. Wulf (eds.), Proceedings of the Twenty-Sixth Annual Meeting of the Berkeley Linguistics Society. General Session and Parasession on Aspect (2000). Berkley: Berkeley Linguistics Society, 385–399. Kuroda, Susumu. 1999 Die historische Entwicklung der Perfektkonstruktionen im Deutschen. (Beiträge zur germanistischen Sprachwissenschaft.) Hamburg: Buske. Lakoff, George. 1987 Women, Fire, and Dangerous Things. What Categories Reveal about the Mind. Chicago: University of Chicago Press.

Auxiliary selection in closely related languages: the case of German and Dutch

357

Langacker, Ronald W. 1987 Foundations of Cognitive Grammar. Stanford, Calif: Stanford University Press. Larsson, Ida. 2009 Participles in time. The Development of the Perfect Tense in Swedish. Ph.D. thesis, University of Göteborg. Łęcki, Andrzej M. 2010 Grammaticalisation Paths of ‘Have’ in English. (Studies in English Medieval Language and Literature.) Frankfurt am Main, Oxford: Peter Lang. Lehmann, Christian. 1995 Thoughts on Grammaticalization (LINCOM Studies in Theoretical Linguistics.) München: LINCOM Europa. Levin, Beth and Malka Rappaport Hovav. 1995 Unaccusativity. At the Syntax-Lexical Semantics Interface. Cambridge, Mass.: MIT Press. Levin, Beth and Malka Rappaport Hovav. 2005 Argument Realization. Cambridge: Cambridge University Press. Lieber, Rochelle and Harald Baayen. 1997 A semantic principle of auxiliary selection in Dutch. Natural Language and Linguistic Theory 15: 789–845. McFadden, Thomas and Artemis Alexiadou. 2010 Perfects, resultatives and auxiliaries in Early English. Linguistic Inquiry 41(3): 389–425. Öhl, Peter. 2009 Die Entstehung des periphrastischen Perfekts mit haben und sein im Deutschen – eine längst beantwortete Frage? Zeitschrift für Sprachwissenschaft 28: 265–306. Paul, Hermann. 1905 Die Umschreibung des Perfektums im Deutschen mit haben und sein. München: Verlag der K. Akademie. Randall, Janet, Angeliek van Hout, Harald Baayen and Jürgen Weissenborn. 2004 Acquiring unaccusatives. A cross-linguistic look. In: Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.), The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 332–353. Oxford: Oxford University Press. Randall, Janet. 2007 Parameterized auxiliary selection. A fine-grained interaction of features and linking rules. In: Raúl Aranovich (ed.), Split Auxiliary Systems. A CrossLinguistic Perspective, 207–235. Amsterdam, Philadelphia: John Benjamins. Rooij, Jaap de. 1988 Van hebben naar zijn. Het gebruik van hebben en zijn in de voltooide tijden (actief) van zijn, gaan, vergeten en verliezen in standaardtaal, ouder Nederlands en dialect [The shift from hebben to zijn. The selection of hebben and zijn in the present perfect (ative) of the verbs zijn, gaan, vergeten and verliezen in standard language, historical Dutch and dialects]. Amsterdam: P. J. Meertens-Instituut. Rosemeyer, Malte. 2014 Auxiliary Selection in Spanish. Gradience, Gradualness, and Conservation. Amsterdam, Philadelphia: John Benjamins. Rosemeyer, Malte. (this volume) Entrenchment and discourse traditions in Spanish auxiliary selection. Sapp, Christopher. 2011 Auxiliary selection in the Early New High German perfect tenses. Groninger Arbeiten zur Germanistischen Linguistik 53(2): 29–43. Shannon, Thomas F. 1990 The unaccusative hypothesis and the history of the perfect auxiliary in Germanic and Romance. In: Andersen Henning and Konrad Koerner (eds.), Historical Linguistics 1987, 461–499. Amsterdam, Philadelphia: John Benjamins. Shannon, Thomas F. 1995 Toward a cognitive explanation of perfect auxiliary variation. American Journal of Germanic Linguistics & Literatures 7(2): 129–163. Smet, Henrik De. 2012 The course of actualization. Language 88: 601–633. Sorace, Antonella. 2000 Gradients in auxiliary selection with intransitive verbs. Language 76(4): 859–890.

358

Melitta Gillmann

Sorace, Antonella and Frank Keller. 2003 Gradient auxiliary selection and impersonal passivization in German. Journal of Linguistics 39(1): 57–108. Spoken Dutch Corpus = . Szczepaniak, Renata. 2011 Grammatikalisierung im Deutschen. Eine Einführung. Tübingen: Narr. Tenny, Carol. 1995 How motion verbs are special. Pragmatics & Cognition 3(1): 31–73. Traugott, Elizabeth Closs and Graeme Trousdale (eds.). 2010a Gradience, Gradualness and Grammaticalization. Amsterdam, Philadelphia: John Benjamins. Traugott, Elizabeth Closs and Graeme Trousdale. 2010b Gradience, gradualness and grammaticalization. How do they intersect? In: Elizabeth Closs Traugott and Graeme Trousdale (eds.), Gradience, Gradualness and Grammaticalization, 19–44. Amsterdam, Philadelphia: John Benjamins. Wal, Marijke van der. 1993 Teaching Latin and observing the Dutch vernacular. Exercitium Puerorum (1485). In: Jan Noordegraaf (ed.), Five Hundred Years of Foreign Language Teaching in the Netherlands, 1–9. Amsterdam: Stichting Neerlandistiek u. a. Werth, Alexander, Ina Bornkessel-Schlesewsky, Simon Kasper, Markus Philipp and Jürgen Erich Schmidt. 2011 Regionale Effekte in der neuronalen Verarbeitung von Auxiliaren im Deutschen. Unpublished manuscript. Zaenen, Annie. 1988 Unaccusative verbs in Dutch and the syntax-semantics interface. Technical Report SSL 123: 317–335. Zaenen, Annie. 2011 Unaccusativity in Dutch. Integrating syntax and lexical semantics. In: James Pustejovsky (ed.), Semantics and the Lexicon, 129–161. Dordrecht, London: Springer. Zeman, Sonja. 2010 Tempus und “Mündlichkeit” im Mittelhochdeutschen. Zur Interdependenz grammatischer Perspektivensetzung und “Historischer Mündlichkeit” im mittelhochdeutschen Tempussystem. Berlin, New York: De Gruyter.

General index Abruptness 226–227, 238, 335, 353 Absolute construction 3, 52–54, 278 Accomplishment (verb of) see Verb class Achievement (verb of) see Verb class Activity (verb of) see Verb class Actualisation 12, 16, 312–313 Adverb see Adverbial Adverbial 5–6, 11–12, 36, 56, 93, 137, 140– 141, 148–149, 151, 153, 157–158, 165, 174, 179, 215–216, 221, 232–233, 253, 255, 257, 268, 312, 337, 350 – Agent-oriented 174, 253 – Frequency 151, 157–158, 165 – Goal 5–6, 11, 257, 268, 312, 350 – Manner 140, 149, 253 – Time 56, 93, 137, 141, 148, 179, 215–216, 255, 337 Affix 133, 214, 221, 223–224 Agent (semantic role) see Semantic role Agentivity 5, 11, 14, 24, 26–31, 34, 37–38, 43–44, 54, 67–68, 71, 82–85, 105, 112– 113, 161, 168, 171–172, 176, 184, 198, 259–264, 266–268, 272, 281, 338, 345, 348–349 Agreement 49, 50–51, 55–56, 61, 62, 115, 117, 180, 193–194, 200, 224, 232 Aktionsart 5, 124–125, 251 Allomorphy 7, 10, 12, 16 Analogisation see Actualisation Anteriority 13, 90, 94, 124, 136–138, 224, 236–237, 271–272 Anticausative 103, 114, 257, 269 Aorist 90–93, 133, 135 Aoristic drift 92 Appearance (verb of) see Verb class Apposition 89 Auxiliary Selection Hierarchy 7–16, 24–25, 43, 55, 66–70, 73, 79–87, 90, 92, 117, 146–147, 161, 168, 170–171, 178, 184– 185, 192, 251, 258–267, 271, 298, 333– 335, 337–341 Beginning and ending (verb of) see Verb class

Case syncretism 15, 334, 353–354 Causative 37, 103, 114, 168, 175, 204, 257, 269 Change of location (verb of) see Verb class Change of state (verb of) see Verb class Charlemagne Sprachbund 2 Chunking 304–305, 345 Cleft 89 Clitic climbing 195 Clitic pronoun 106, 108, 110–111, 114, 118 Coercion see Template Augmentation Combinatorial Primitives 36 Communicative distance 308–309, 317– 318, 322, 326 Communicative proximity 308–309, 317– 318, 322, 326 Completion see Telicity Conflation 201–204, 207 Constructionalist approach 11–13, 24, 36, 190, 204, 208 Control see Agentivity Conventionalisation 216, 231, 303, 306 Copula 2, 7–8, 73, 137, 179–180, 189, 191– 192, 213–215, 224, 229, 238–241, 253– 254, 272, 282, 336, 340–341 Core grammar 241 Creativity 220, 225, 227 Cyclicity 228 Dative 9, 216, 219, 312, 350, 353 Deadjectival verb 175 Decategorisation 335 Decomposition 12, 14, 260 Definiteness 3, 13, 53, 186–195, 200, 346 Degree achievement (verb of) see Verb class Dementia 31 Deponens 239 Desemantisation 335 Detransitivisation 114 Diasystem 303 Directed movement (verb of) see Verb class Discourse tradition 14–16, 302–304, 306– 309, 313–314, 317–318, 325–326

360

General index

Economy 73, 225–226 Electroencephalography 33–34 Ellipsis 11, 87–89 Emission (verb of) see Verb class Entrenchment 265, 270, 301–302, 305– 306, 308–309, 325–326 EQUI deletion 52 Ergative see Unaccusative Hypothesis Event structure 10, 28, 204 Event time 136 Event-related brain potentials (ERPs) 32–35 Exemplar Theory 304 Existence (verb of) see Verb class Existential construction 12–13, 16, 185, 189–194, 197–205, 207 Experiencer (semantic role) see Semantic role Experiencer (verb of) see Semantic role Experiential perfect 12, 92, 124–128, 130, 132, 134–140, 142, 147, 151–153, 155, 157–158, 177, 336 Expressivity see Creativity Extended Now Theory 136–137 External argument 79, 81, 106–108, 114, 174–175, 200, 203–204, 207 Eye-tracking 31–32 First-phase Syntax 36, 169–170 Fluid intransitivity 25 Focus 34, 336, 344, 352 Frequency effect 14, 165, 269, 272, 326 Future 133, 135, 217–218, 227–229 Goal (semantic role) see Semantic role Government and Binding Theory 2, 79 Grammatical parameter 221–224 – Finite verb position parameter 223 – Head position parameter 222–223 Grammaticalisation 12–13, 123, 135, 213– 220, 225, 227–229, 235, 239–241, 303, 312, 334–337 Idiom 214, 216–218, 343 Implicature 228, 303 Inchoative 175 Inflection 213–220, 227, 236–237 Integrity (grammaticalisation parameter) 235

Intermediacy 198 Internal argument 81, 106–108, 175, 200, 234 Intersective Gradience 37–38 Intertextuality 270, 303, 306–307, 309, 318 Irrealis effect 8–9, 12, 66–67, 123–128, 138, 142, 343 Language acquisition 29–30, 37, 184, 221– 222, 225–228, 236, 241–242, 305 Language attrition 30 Language module 221–222 Late Merge Principle 225 Lexical Functional Grammar 5 Light verb 201–204, 207 Linking rule 24, 54 Locative 97, 199–201, 267, 269, 294, 312 Locative inversion 205 Locomotion see Change of location Manner adverbial 12, 149, 253, 257 Manner of motion (verb of) see Verb class Markedness 45, 49, 147, 217, 233, 237 Memory (neurolinguistics) 33–34 Merge Over Move Principle 225 Metaphor 14, 225, 267, 315 Metonymy 225, 254 Middle voice 98, 100, 103, 116, 133–134, 154 Modality – Deontic 228 – Epistemic 101, see Irrealis effect Mood 66, 180, 343 – Optative 135 – Subjunctive 139, 305, Negation

52, 131, 223

Paraphrasis 214, 216 Participial clause 52–56, 146–147, 153, 176 Participle 13, 52, 146, 153, 159, 191, 194 – Active participle 148, 152, 162–163, 169, 172, 175, 177 – Passive participle 179, 215, 230, 232– 236 – Past participle 35, 55–56, 61, 115–117, 193, 213, 215, 230, 232–236, 240, 282, 336

General index

– Perfect participle 133, 224 – Resultant state participle 150, 151, 153– 154, 158–160, 177, 179 – Stative participle 149 – Target state participle 153 Partitive cliticisation 3, 26, 200, 204, 206, 278 Passato remoto 206 Passive 30, 79, 103, 106–108, 110–114, 116, 133, 145, 176, 215, 219, 229, 239, 250–252, 282 – Adjectival passive 139 – Dative passive 219 – Impersonal passive 173–174 – Mediopassive see Middle voice – Passive perfect 134, 224 – Recipient passive 219 – Stative passive 128,139–140, 150–153, 158–159, 178 Patient (semantic role) see Semantic role Perfect – Passive perfect see Passive – Pluperfect 66, 68, 70, 90, 133, 138, 148, 180 – Present perfect 90, 125, 141, 179, 335– 336 – Synthetic perfect 133–135, 141, 224, 237–238 Perfectivity 13, 236–237 Periphrasis 13, 73, 123, 126, 128, 214–220, 222 Persistence 240, 335, 337, 347 Politeness 303 Polysemy 203, 303 Position (verb of) see Verb class Possessive 179–180, 201, 236 Post-structuralism 306 Postverbal subject 3, 30, 199–200 Prefixation 27, 29, 34–36, 132, 214, 349 Prepositional phrase 27, 215, 216 Process (verb of) see Verb class Progressive 215–219, 341 Projectionist approach 24–25, 36, 208 Proto-Role approach 5–6, 11 Prototypicality 5, 12, 80–82, 346 Pseudocleft 12, 149–150 Quasi-reflexive

47, 58, 60, 64

361

Re- (French prefix) 255 Recategorisation 214, 231 Reception (verb of) see Verb class Reduplication 132–133 Reference time 136–137 Reflexive 9, 11, 44–49, 58–74, 80, 86–87, 94–96, 101–118, 252, 277–280, 284, 312 Relational Grammar 2, 48–49, 79, 249 Relative clause 89–90 Replacement 11, 13–14, 16, 46–48, 131– 132, 141, 192, 194, 305, 307, 312–313, 325 Representativity 326 Restructuring 195 Resultant state 5, 150–153, 154, 158–160, 177–179, 253–254, 271, 282, 336 Resultative 7–8, 12–15, 56, 73, 124, 126, 130–136, 137–142, 146–153, 156, 158– 160, 165–166, 169–180, 189, 208, 236, 251, 253–258, 272, 280, 282–283, 311, 336–337, 344, 354 Retroherent unaccusative 60 Role and Reference Grammar 50–51, 62, 258 Rule 15, 24, 44–46, 54, 90, 215–216, 221– 222, 226–227, 241–242, 255–257, 265, 269, 271–272, 301, 303, 305 Salience 33–34 Scalarity 11, 80, 87–94, 117 Semantic role 5, 45, 62, 96, 106, 109–110, 118, 170, 179, 278, 285, 289–292, 298 Sentential argument 186 Sign language 26 Small clause 204, 207, 231–233 Source (semantic role) see Semantic role Split Intransitivity Hierarchy 24–25, 81–82 Standard Average European 2 State (verb of) see Verb class Stativity 46, 149, 180, 262–263, 281 Subsective Gradience 37–38 Suffixation 36, 106, 154, 175, 227–228 Synthetic preterite see Perfect Target state 150–153 Telicity 11, 14, 16, 24, 26–29, 31, 34, 36, 44, 46, 52–58, 82–86, 92, 98–99, 176,

362

General index

184, 198, 262–264, 278–279, 281, 286, 320, 334, 337–344, 346, 348, 352, 354 Template augmentation 7–8, 35–36, 137, 257, 268 Thematic role see Semantic role Theme (semantic role) see Semantic role Theta-role see Semantic role Token frequency 301–302, 305, 313, 348 Topic 105–106 Topicalisation 104, 118 Transitivity 53, 284, 312–313, 338, 348, 350, 353 Two-placed verb 334, 348–353 Type frequency 312, 325 Unaccusative see Unaccusative Hypothesis Unaccusative Hypothesis 2–3, 8, 10, 12, 23–24, 28–31, 38, 48–51, 73, 79, 117, 175, 333 Unaccusativity diagnostics 4, 24, 29, 80, 161, 174, 176, 205 Underspecification 10, 28, 32, 81, 85, 285 Unergative see Unaccusative Hypothesis Universal Grammar 221–222, 238 Usage-Based 226, 241, 270, 301, 304, 326 Valency 73, 86, 104–106, 108, 110, 118 Variation 2, 4–10, 15, 24–27, 30, 36, 58, 61, 65–72, 85–87, 94, 97, 101–106, 136, 143, 146–147, 156–157, 160–161, 164– 166, 168–169, 173, 175–180, 184, 227, 241–242, 249, 251–252, 256–259, 264, 268, 278, 285, 303, 310, 322, 333–334, 338, 352, 354 Verb movement 223 Verb class – Accomplishment 5–6 – Achievement 5 – Activity 5, 8, 24, 35, 82, 84, 184, 205, 280

– Agent 5–6, 24, 50, 96–100, 107–108, 112, 114, 116, 251, 262, 279, 285–286, 290, 296–298, – Appearance 13, 16, 183–184, 190, 192, 269 – Beginning and ending 339, 349, 354 – Change of location 7, 9, 14–16, 27, 34, 57, 71–73, 162, 164, 166–168, 171–173, 175–176, 251, 257–272, 310–316, 321– 326, 333–334, 338, 344- 349 – Change of state 7, 9, 27, 34–35, 68, 71– 72, 85, 97, 100, 112, 162, 164, 166–168, 170–176, 183, 257–266, 269, 271, 273, 282–283, 292, 294, 310–311, 316, 320– 325 – Degree achievement 339–340 – Directed movement 263, 269 – Emission 25, 28, 43, 73, 268 – Existence 13, 16, 101, 183–184, 192, 259, 263, 340 – Experiencer 97, 100, 347, 349, 353 – Goal 96–100, 106–118 – Manner of motion 7–8, 183, 198, 203, 257, 268, 285, 297, 334, 342–348, 350, 352–354 – Patient 5–6, 11, 24, 46, 79, 96–100, 116, 262, 285–286, 289–292, 297, 346, 349 – Position 28, 341 – Process 26, 28, 38, 97–101, 205, 259– 261, 263–265, 268, 272, 333, 338–339, 342, 345–348, – Reception 219 – Source 11, 96–100, 103, 106–117 – State 5, 7–8, 10, 25, 28, 45, 85, 99–100, 162, 173, 183, 190–191, 257, 259, 262– 263, 268–269, 310, 340, 354 – Theme 97, 186, 188–189, 191, 193–194, 285, 346 – Weather 97 Weather (verb of) see Verb class

Language index Acadian French 1, 11, 72, 92–94 Acehnese 50–51 Albanian 2 Basque

74, 84–87, 92, 94, 102, 107–108, 117, 125, 141, 145, 169, 180, 183–184, 198– 199, 203–204, 206–207, 251–253, 257, 259–261, 264, 278, 333, 348

26 Japanese

26

Canadian French 253, 305 Catalan 1, 9, 13, 48, 183–209, 250, 252 Chinese 26 Corsican 1, 11, 84–118, 252 Croatian 26 Czech 2

Latin 57, 224, 228–229, 238–239, 307 Leccese 61 Luxemburgian 219

Danish 1, 11, 145–146, 148, 153, 160, 162, 165, 169, 173, 179 Dutch 1, 5, 8–9, 15–16, 26–27, 85, 92, 123– 125, 129–130, 140–141, 145–146, 174, 265, 278, 333–355

Neapolitan 8–9, 11, 26, 63, 65–69, 71, 73, 123, 196 Norwegian 1, 12, 145–147, 153–154, 157– 158, 164–165, 168–169, 178

English 1, 7–9, 12, 25, 30, 90, 92, 123–128, 132, 134, 138–142, 145, 156, 170, 189, 205, 208, 213, 215, 217, 221, 222–223, 229, 236, 240–241, 283, 335–337 Flemish 340 Florentine 61, 63, 64 French 1, 3, 9, 11, 14, 16, 26, 45, 47, 61, 72, 83–84, 86, 90–94, 98, 102, 107–108, 115, 117, 123, 125, 130, 141, 145, 215– 218, 221, 227–229, 241, 249–273, 277– 298, 307, 313, 346 German 1, 8–9, 12–13, 15–16, 26–29, 34– 36, 85, 92, 124–125, 128–130, 138–142, 145–146, 150, 154, 157, 159, 169, 179, 199, 213–242, 250, 264–265, 278, 284, 313, 333–355 Greek 1–2, 12, 123, 132–136 Icelandic 1, 12, 145–148, 153–160, 162, 164–165, 168–170, 174, 177–179 Irish 26 Italian 1, 3, 5–6, 9, 25–27, 30–32, 36, 38, 43, 45, 47, 49, 51–55, 57–58, 60–61, 72,

Macedonian

Occitan

2

252

Paduan 26 Portuguese 9, 250 Procidano 180 Quebec French Rhaeto-Romanic Romanesco 61 Russian 221

208 252

Sardinian 61, 252 Sicilian 8, 11, 63, 67, 69–72, 73 Spanish 1, 5, 8–9, 11, 14–16, 26, 30, 36, 46–48, 58–60, 63–65, 67, 69–71, 73, 82, 123–124, 130–131, 183, 250, 254, 257, 264–265, 269–270, 272, 297, 299, 301–326, 346 Swedish 1, 8–9, 12, 123, 124, 128–129, 145–147, 149–180 Tok Pisin 224 Turkish 26