Quantitative Approaches to Grammar and Grammatical Change: Perspectives from Germanic 9783110401929, 9783110401752

Citation preview

Sam Featherston and Yannick Versley (Eds.) Quantitative Approaches to Grammar and Grammatical Change

Trends in Linguistics Studies and Monographs

Editor Volker Gast Editorial Board Walter Bisang Jan Terje Faarlund Hans Henrich Hock Natalia Levshina Heiko Narrog Matthias Schlesewsky Amir Zeldes Niina Ning Zhang Editors responsible for this volume Volker Gast and Amir Zeldes

Volume 290

Quantitative Approaches to Grammar and Grammatical Change Perspectives from Germanic Edited by Sam Featherston and Yannick Versley

ISBN 978-3-11-040175-2 e-ISBN (PDF) 978-3-11-040192-9 e-ISBN (EPUB) 978-3-11-040212-4 ISSN 1861-4302 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. © 2016 Walter de Gruyter GmbH, Berlin/Boston Typesetting: Compuscript Ltd., Shannon, Ireland Printing: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Acknowledgements

vi

Sam Featherston and Yannick Versley Introduction 1 Markus Bader Complex center embedding in German – The effect of sentence position Gisbert Fanselow, Jana Häussler and Thomas Weskott Constituent order in German multiple questions: Normal order and (apparent) anti-superiority effects

33

Jutta Hartmann, Andreas Konietzko and Martin Salzmann On the Limits of Non-Parallelism in ATB Movement: Experimental Evidence for Strict Syntactic Identity 51 Stefan Hofstetter Measure Phrase Constructions in English, German, and French: The (Non-)Occurrence of Antonyms and Effects of Evaluativity Tom Ruette and Dirk Speelman Interpreting aggregated distances. The case of Old High German texts

85

113

Augustin Speyer Relative Object Order in High and Low German

143

Achim Stein and Carola Trips Modeling language contact with diachronic crosslinguistic data Anna Volodina and Helmut Weiß Diachronic Development of Null Subjects in German

187

Susanne Winkler, Janina Radó and Marian Gutscher What Determines ‘Freezing’ Effects in was-für Split Constructions? Index

233

165

207

9

Acknowledgments We should like to thank here all of our colleagues from the research center SFB 833 Bedeutungskonstitution who have worked together with us in the production of this collection, in particular Beate Starke, who is both the brain and the heart of the centre, and Sophie von Wietersheim, Vivian Schreier, Alicia Bitzer, Sarah Becker and Hannah Gerbrich, who supported the editing process. We should also like to recognize the contribution of Wolfgang Konwitschny, who good-humouredly prevents disaster at the type-setting stage.

Sam Featherston and Yannick Versley

Introduction

There is a strong tradition in Linguistics of seeking to formulate theories on a solid basis of empirical evidence, even if this has not been quite so apparent recently. Both Chomsky’s (1965) insistence on using introspection as the primary source of evidence accessible to any native speaker and the subsequent preference for more objective sources of evidence that go beyond a single speaker can and should be understood as a quest for replicable, accessible and reliable evidence for hypotheses. The quest for reliable empirical evidence has since gone on to conquer many areas of linguistics without any sign of slowing, a fact that is pleasing to those among us who take the validation of linguistic theories seriously. This volume is a collection of work done by linguists who take this view and wish their work to stand on firm foundations. One probable factor in the increasing popularity of supporting linguistic claims with quantitative evidence is the fact that the contribution of data towards theory building can vary according to the preferences of individual linguists. We may perhaps distinguish the three approaches to the relationship between quantitative data and grammatical theory in current linguistics at the sentence level: the first is ‘data-driven’ linguistics, which we might characterize as being an approach which looks first at the data and attempts to describe and account for what it finds. It has the advantage of being scientific and objective in the sense that it delivers impartial generalizations, but critics might object that it forgoes the chance to obtain broader insights by trying to find merely descriptive patterns in the data. A contrasting approach is to lay greatest weight on the development of theory. This has been most common in theoretical syntax and semantics over the last couple of decades. In this work, linguists have tended to base their claims on a very few examples with little or no effort to find external confirmation. Practioners have often characterized this work as ‘theoretical’ (or perhaps ‘formal’), but we can confidently call it ‘data-light’. Critics of this approach might reproach it for being agenda-driven and seeking confirmation for positions held, rather than even-handedly seeking the truth. A more recent position could be termed ‘data-respecting’ linguistics. Colleagues taking this approach typically analyze sentence structures in theoretical models and seek to develop them, but they consider it essential to test their hypotheses empirically once they have made them. These linguists are theoretically informed and may well develop their hypotheses on the basis of theory, but they readily accept that a hypothesis is defeated if the data does not support it.

2 

 Sam Featherston and Yannick Versley

They may also analyze their data with an awareness of the additional nontheoretical factors which may affect it. This approach thus marries the previously more nearly exclusive theoretical and empirical approaches. So linguists interested in sentence structure and interpretation no longer need to choose between being theoretical or being empirical. The fact that more relationships between data and linguistics are possible means that there is room in the big tent of empirically working linguists for a wide range of different positions. The fact that data can be used to support theoretical analyses rather than replacing them has made the collection of data more attractive. This situation also allows different fields to relate to each other. A couple of decades ago people looking at sentence form on the basis of quantitative data would have tended to produce an account in terms of processing and report it at CUNY or at AMLAP, while people taking a more theoretical perspective would have phrased their conclusions in terms of grammatical theory and take it to NELS. These two groups would have had little to say to one another. But in the light of what we feel tempted to call the “empirical revolution in the study of sentence grammar”, this gulf has been narrowed. Even people whose primary concern is grammatical theory take note of processing effects and attribute certain effects to them. Correspondingly, workers focusing on the surface evidence can relate more to the constructs of the theoreticians, because the two layers of explanation have been brought into contact. And these workers too must account for the data gathered by the theoreticians. The papers in this volume are a snapshot of some of the work currently being done in this field. They vary upon this cline of data-orientation from papers such as Hofstetter and Fanselow, Häussler & Weskott at the one end to Volodina & Weiß and Ruette & Speelman at the other end. Hofstetter is an example of the penetration of data into places where the quantitative approach had not previously dared to tread, namely in the semantics of measure phrases. This is some way from the work of Ruette & Speelman, who mark the other end of the spectrum, since their article very clearly starts from the available data and attempts to do something new with it, in this case with a novel analysis type – Individual Differences Scaling. While these two papers are fairly distant from one another, the other papers fill up more or less all the available space between them, thus yielding a coherent grouping of evidence-based linguistic work. There are two other parameters of difference which might be noted among the papers included here. The first is that of data type. Six of the nine report what we might term ‘elicited data’, which is mostly that from experimental procedures. Another six of the nine make use of ‘observational data’, that is, data from linguistic use which occurred spontaneously in a naturalistic context. The overlap between the groups highlights an increasing trend for linguists to correlate more



Introduction 

 3

than one single data source. A finding is all the more worthy of attention if the effect of interest can be identified in multiple data types. The example of experimental results being replicated – or resembled – in corpus data yields particularly convincing confirmatory evidence because these two data types exemplify our major approaches to what our grammar might look like. If we assume that the object of study of a sentence grammar is something like a mental state or a competence, then experimental elicitation in real time and the observation of spontaneous production represent the two chief ways that we may learn about its nature. But there is a further approach to the study of language, one which is coming increasingly into vogue: the application of analyses from synchronic fields such as syntax and semantic processing to diachronic data. It is plausibly argued (e.g. Peter Koch p.c.) that language variation, whether synchronic or diachronic, is as valid an indicator and measure of the boundaries of the range of variation of language and the factors that determine possible variation as is experimental data. Any language form which exists or has existed must be a natural language, and it can therefore offer us a perspective on the grammar. In particular it can be very instructive to compare varieties of a single language and identify precisely where in the system the differences have developed. Historical stages of a language can be of particular interest in the investigation of the available space of variation because they allow us to control for the standardization which contemporary language varieties are affected by. For example, looking at the language of Chaucer or Shakespeare can provide us with insights into the modern language by giving us an idea what variation can exist within a grammar which has not been constrained by the requirements of mass communication, of school systems, and by the standardization in the Enlightenment period. Looked at like this our modern language can be seen as an atypical example of the way that languages work. In their natural state and without the enormous pressures of our highly developed large-scale cultures, languages may be more flexible that they appear to us. This is particularly true since some genres of older texts seem to give a degree of access to the spoken language of the period, because the distinction in styles between written and spoken idioms has not yet been fully established and/or the expectation that written texts will be polished and structurally transparent is not yet applied. Several of the papers in this volume take the historical approach, which seems currently to be enjoying something of a revival among younger researchers. The findings are interesting and persuasive. Synchronic syntacticians have long been uncomfortably aware that their models often do not readily allow them to capture the observable facts of grammatical change, whilst simultaneously knowing

4 

 Sam Featherston and Yannick Versley

that syntactic change has indeed occurred. The collection of quantities of data on just how grammatical changes have proceeded should be an important step towards producing improved synchronic syntax models which can accommodate processes of change. A final issue that we should like to mention is the question of the adherence of the papers and their authors to particular grammatical frameworks. It is true that there is a difference of emphasis between the papers reporting experimental studies and the others, in that the questions raised in the experimental studies are more explicitly phrased in terms of grammatical theory. But only one paper – Fanselow, Häussler & Weskott – makes use of explanatory mechanisms which might be regarded as uniquely theoretically motivated within the Generative Program. Hofstetter offers an encoding of his analysis in formal semantics, but in both cases, the reader knows what they mean, whether they share the author’s assumptions or not, and the linguistic point is made anyway. The other papers are all fairly theory-neutral; one might say that they belong to a post-generative consensus, by which we mean that they build upon work done within the generative tradition, without being bound by the goals or methods of the generative tradition in the narrow sense, where all structures are ‘generated’ by the iterative application of rewrite rules, such as VP → V, NP. The post-generative approach regards phenomena as inherently likely to concern the interaction between multiple components of grammar (understood broadly), including syntax, and seeks to distinguish between them. This is clearly so in the papers by Bader, Hartmann, Konietzko & Salzmann, and Winkler, Radó & Gutscher, all of which concern phenomena straddling the boundaries of syntax, information structure, and processing. Such attention to phenomena straddling modular boundaries is entirely typical of work within the currently very active field of empirical grammar research, where claims in the literature are subjected to empirical scrutiny, and the implications of the findings for explanatory models are analyzed. The priority in this approach is to start off by getting the facts right as a basis for explanation. And when a finding has to be accounted for, the authors of this group of papers weigh up whether a grammatical rule is required or whether it cannot be accounted for by the semantics or other grammatical module, or by an independently motivated processing effect. This is very close to the explanatory approach taken in the non-experimental papers. Most of the work here uses fairly newly available resources to establish facts about language varieties, using the technique of language comparison to sharpen the picture by providing a contrast. This group does not usually explicitly call things ‘processing effects’, but the model of language description they employ is not so far from the post-generative model, and I would guess that most explanatory factors advanced here would be accepted by both groups.



Introduction 

 5

All the papers thus share an interest in looking at and accounting for the primary data. All of them are interested in basing their accounts on firm foundations and making sense of the findings in terms that belong to the common ground within the linguistic community. They are thus contributing to the same project of grammatical explanation, where the data forms the constraining factor, not some factor such as elegance from within an explanatory framework. There are significant commonalities between the sort of explanations that you need to account for gradations of perceived well-formedness or processing ease, on the one hand, and historical change and dialectal (micro-)variation, on the other. It is thus not surprising that these previously fairly separate fields of linguistics are growing together fast, and that the data-driven aspect of their work is pushing it. When you look more closely at the primary data and respect the discipline of accounting for the data, you realize that the other person’s data and explanatory background is not so different. The paper by Bader is a classic of the experimental syntax type as he discusses both corpus frequencies and speeded grammaticality judgements to try to solve a puzzle in sentence grammar and processing. The mysterious ‘missing VP effect’ is the finding that certain ungrammatical structures with treble-embedded clauses are found to be more acceptable than grammatical ones. While this phenomenon has been noted, the empirical evidence relating to it is contradictory. Bader identifies the factor which caused the conflicting findings, which turns out to be the location of the embedding in the larger sentence structure. He uses this information to put forward a new account of the trigger for the phenomenon: the Discrimination Hypothesis. This paper is located at the boundary of syntax and processing and makes use of insights from both fields. Fanselow, Häussler & Weskott perform a similar disentanglement of factors in their work on the anti-superiority effect. The issue they address is the circumstances under which we find a reversal of the usual superiority effect – the preference for it to be the subject wh-item which is raised to sentence-initial position in a multi-wh-question. They note that multiple wh-questions based on psych verbs, where the Experiencer is not also the subject, seem to show a pattern which is not consistent which the normal formulation of the Superiority constraint in terms of non-crossing movement. On the basis of their experimental data, they argue that alternative bases for default ordering must be assumed, specifically on the basis of Case for multiple wh-questions. The third paper too has experimental data as its evidential base. Hartmann, Konietzko & Salzmann look at Across-the-Board movement to determine what degree of identity the extracted element(s) must have. This paper provides empirical evidence for a strict syntactic identity condition for Across-the-Board movement in German. In their data, Case mismatches are not tolerated and, contrary

6 

 Sam Featherston and Yannick Versley

to previous claims in the literature, the use of syncretic forms does not improve the mismatches. Examples such as Käse mag ich nicht und ist auch nicht gut für mich ‘Cheese, I don’t like and isn’t good for me’, where the topicalized element is subject of one clause and object of another, but the two forms are syncretic, are judged to be unacceptable. Their results therefore support accounts that predict the gaps to have to have identical morphological and syntactic features. Antonymous adjectives in measure phrase constructions are the subject of the paper by Hofstetter. It has routinely been asserted that we can put positive or neutral adjectives into measure phrase constructions such as three feet high, ten centimetres thick, but we cannot use antonymous adjectives – *three feet low, *ten centimetres thin. The contribution seeks to verify this assumption, testing for the generality of the findings by looking at three languages, and presenting the structures in three different syntactic conditions. The results show that the standard claims in the literature have only a very restricted validity and certainly cannot count as generalizations. Hofstetter analyzes the contribution which evaluativity makes to this result and develops a novel classification of gradable predicates on this basis. The data basis of the paper is low-tech but the results are very significant. The next paper – by Ruette & Speelman – is the most data-oriented in the collection and illustrates a data analysis method with an example from Old High German nominal inflection. Data from a language variety which was last spoken a thousand years ago and whose speakers were not generally literate will always be scarce, which means that it can be useful to aggregate multiple linguistic variables to discover patterns. This very aggregation however leads to a loss of information about the behavior of the individual input variables. In this paper the authors present a method which avoids this problem – Individual Differences Scaling (INDSCAL). With this technique a researcher can use the aggregate of the various variables to identify variable groupings even where the data is thin or offers only weak contrasts. The values of the aggregated variables are partly preserved in the model, so that if a particular variable does not fit the majority pattern, this fact remains visible. The authors apply the method to data from Old High German in their illustrative study, but it is equally as applicable to synchronic dialectometric studies. Speyer takes a diachronic look at the relative ordering of complements in different stages of High German from Old High German to Early New High German, but also considering Old Saxon. He finds that the base argument order in Old High German, and most likely Old Saxon, is Accusative > Dative and attributes this to the Thematic Hierarchy. He identifies a number of additional factors which can affect surface order, such as noun phrase type and animacy, but argues that these play a greater role in the modern language than in earlier stages, supporting these claims with data from a questionnaire study. This paper is the archetypical example of the approach that we discussed above in which gradient data



Introduction 

 7

from historical sources and current elicited acceptability are related to produce a merged diachronic/psycholinguistic picture of variation in sentence grammar. This issue is sufficiently fundamental to this volume to deserve clarification in a separate paragraph. The underlying assumption is radical but entirely convincing: degrees of perceived well-formedness drive frequency in output. This means that if one structural variant is easier to process or compute than another, then the easier one will appear in the output and thus in corpus data. If the two are similar in processing effort but not identical, then both will appear but we will observe more of the (slightly) easier one and fewer of the (slightly) less easy one. Occurrence is thus non-linearly correlated with perceived acceptability. The paper by Stein & Trips attempts not only to trace developments across succeeding stages of two languages, but also to account for these developments, a desideratum often likened to the holy grail of historical linguistics. They look at topicalization and left dislocation in the development of Middle English and Old French. By taking into consideration a finer distinction between the types of dislocated phrases and register effects together with the information-structural criteria used by previous work, they provide fresh evidence for the hypothesis that the loss of V2 in French was correlated with the replacement of topicalization with left-dislocation, and also point to the diachronic evidence that the increased occurrence of both topicalization and left dislocation in 1250s Middle English is a contact-induced, rather than a language-internal phenomenon. The study by Volodina & Weiss is one of those that show that diachronic studies can produce results which are not only of relevance to diachronic linguistics. They look at the occurrence of null subjects in stages and varieties of German and the factors that license them. While previously plentiful data had only be collected and analysed for Old High German and present-day German, the authors start the process of filling in this significant gap; in particular they produce new evidence on the dropping of second-person singular subject pronouns in Early New High German. The results contribute to the recent insight that the generalizations previously made are over-simplifications which need far greater specification. Different types of null subjects need to be distinguished and different licensing conditions need to be kept track of. This sort of paper tends to support the claim we noted above that detailed diachronic and variational data can provide evidence for fine differentiations in the mental grammar in the same way that synchronic spoken data can. In the final article Winkler, Radó & Gutscher deliver a classic contribution to current empirically informed linguistics. For example, Müller (2010) analyzes extraction from a subject was für phrase and finds it unacceptable, labelling it ‘freezing’. He notes that the structure is however acceptable if the object is scrambled over the subject, and dubs this ‘melting’. Winkler, Radó & Gutscher

8 

 Sam Featherston and Yannick Versley

show with a rating study that this purely syntactic analysis is unlikely, since the structures are sensitive to context manipulation. They argue that there is no homogeneous phenomenon of freezing but rather the effects attributed to this are dependent upon the interaction of multiple factors from syntax, information structure, and prosody, which combine to increase processing complexity. This last paper is a classic of the current trend to an empirical approach to sentence-level grammar because it focuses on an outstanding question and reports data gathered specifically to address this issue. As so often, the results show that the previous assumptions about the extent and nature of the phenomenon were not wrong, they were rather incomplete. The fuller picture of the structures involved, including multiple lexical variants of the structures, shows that the phenomenon is more complex, involving the interaction of multiple factors. This sounds like bad news, but is in fact very good news for researchers in sentence grammar and the adjoining areas, as they are able to develop far more empirically adequate accounts of the primary data. We thus find the following great advantages of the application of quantitative methods to increasing numbers of linguistic questions: first, we can find definitive answers to questions for which there were previously only speculative accounts, but also second, we can see the problem more clearly. The collection of data often permits us to identify more precisely exactly what we are looking for an answer to: it shows us what we don’t yet know. And when we know that, we can make real progress, as in the articles in this collection.

References Müller, Gereon. 2010. On deriving CED effects from the PIC. Linguistic Inquiry, 41(1): 35–82. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.

Markus Bader

Complex center embedding in German – The effect of sentence position 1 Introduction Clauses with multiple center-embeddings have provided a rich source of information for psycholinguistic investigations of working memory constraints on sentence processing. As illustrated by the famous example in (1) (from Chomsky & Miller 1963, p. 286), two degrees of center-embedding can already be enough to make a sentence quite difficult to process. (1) The rat the cat the dog chased killed ate the malt. Somewhat surprisingly, sentences containing a doubly center-embedded relative clause seem to improve in acceptability if the VP of the higher relative clause is omitted. This effect, which has become known as the missing-VP effect, was first reported in Frazier (1985).1 Frazier’s example of the missing-VP effect is given in (2). (2) The patient the nurse the clinic had hired […] met Jack. While the missing-VP effect is well established for SVO languages like English (Gibson & Thomas 1999, Christiansen & MacDonald 2009, Vasishth et al. 2010) or French (Gimenes et al. 2009), it is controversial whether a similar effect also occurs in SOV languages. The only language for which relevant experimental evidence is available seems to be German, which is an SOV language although not a strict one. For example, in contrast to strict SOV languages, relative clauses in German follow their head NP and the relative pronoun is located in a left-­ peripheral position.2 The first experimental investigation of the missing-VP effect in German was reported by Bader et al. (2003), who tested sentences as in (3), among others, using the method of speeded grammaticality judgments.

1 In Frazier (1985), the original observation of the missing-VP effect is attributed to Janet Fodor. 2 It is an open question of whether results of the sort reported in this chapter for German can also be obtained in strictly head-final languages. As far as I can see, there is no obvious reason why the missing-VP effect should be absent from strictly head-final languages.

10  (3)

 Markus Bader Heute morgen ist das Programm, das den Programmierer, today morning is the program that the programmer der die Dokumentation erstellen musste, [geärgert hat,] abgestürzt. who the documentation compile must bothered has crashed ‘This morning, the program that had bothered the programmer who had to compile the documentation crashed.’

Whereas grammatical sentences with all three VPs present were judged as grammatical 76% of the time, ungrammatical sentences in which the second VP [marked in (3)] was missing were still accepted as grammatical in 57% of all cases. Note that this is clearly different from cases where an ungrammaticality is easily detected. In such cases, sentences are rejected reliably most of the time. For example, in an experiment reported in Bader (submitted), sentences similar to (3) but with the relative clauses in extraposed position were accepted in only 13% of the time when the second VP was missing. With less complex ungrammatical sentences, acceptance rates fall well below 10% or even 5% (see e.g., Bader & Bayer 2006, Bader & Schmid 2009). From the finding of relatively high acceptance rates for missing-VP sentences as in (3), Bader et al. (2003) concluded that the missing-VP effect also occurs in German.3 More recently, Vasishth et al. (2010) presented a series of reading time experiments investigating the missing-VP effect in both English and German. For English, Vasishth et al. (2010) confirmed the missing-VP effect. Reading times on the last verb, that is, the matrix clause verb, and the words following it (not shown in the example above) were faster when the VP of the higher relative clause was missing than when it was present. This finding indicates that participants did not recognize that a VP was missing in missing-VP sentences. They processed such sentences basically as sentences with a single level of center-embedding, which explains why reading times were faster in comparison to complete sentences, which contain a doubly center-embedded relative clause. For German, the reverse pattern was found. In sentences like (4), reading times for the verb überzeugte (‘convinced’) were slower when the higher relative clause missed its verb schnitt (‘avoided’) than when this verb was present.

3 To my knowledge, for English the missing-VP effect was never investigated by an experimental procedure directly asking participants whether such sentences are grammatical or not. It must therefore be considered an open question of whether this effect occurs with a higher rate in English than in German or not.



Complex center embedding in German – The effect of sentence position 

 11

(4) Der Anwalt, den der Zeuge, den der Spion betrachtete, [schnitt,] the lawyer who the witness who the spy watched avoided überzeugte den Richter. convinced the judge ‘The lawyer that the witness that the spy watched avoided convinced the judge.’ The experiments by Bader et al. (2003) differ from the experiments by Vasishth et al. (2010) in several ways. First, as pointed out by a reviewer, the sentences of Bader et al. (2003) contained subject relative clauses while the sentences of Vasishth et al. (2010) contained object relative clauses. For English, it has been shown time and again that object relative clauses are harder to process than subject relative clauses (cf. overview in Gordon & Lowder 2012). For German, however, Häussler & Bader (2012) failed to find a significant reading time ­difference between the two kinds of relative clauses. A second difference concerns the experimental method. Whereas Bader et al. (2003) used an off-line end-of-sentence judgment task, Vasishth et al. (2010) measured on-line reading times. It is unlikely, however, that the different conclusions arrived at by Bader et al. and Vasishth et al. are simply a consequence of methodological differences. First, the initial experimental evidence for the missing-VP effect in English provided by Gibson & Thomas (1999) also made use of an off-line judgment procedure. Second, a recent ­self-paced reading experiment using material similar to that used by Bader et al. (2003) again found evidence for the missing-VP effect in German (­Häussler & Bader 2015). A third difference between Bader et al. (2003) and Vasishth et al. (2010) concerns the syntactic structure of the experimental material. To ease the following discussion, (5) shows the so-called topological model of German main and embedded clauses. (5) Main clause Embedded clause

pref.

C°

midf.

verb cluster

postf.

Max M.

wird will dass that

es it Max es M. it

lesen read lesen wird read  will

... ...

C° at the left edge and the verb cluster at the right edge of a clause define three ‘fields’. The area between C° and the verb cluster is the middle field; in front of C° is the prefield, which can house a single constituent; the postfield following the verb cluster is the typical area for dependent clauses.

12 

 Markus Bader

As a comparison between (3) and (4) shows, the sentences investigated by Bader et al. (2003) differ from the sentences investigated by Vasishth et al. (2010) with regard to the position of the doubly center-embedded relative clause. In (3) from Bader et al., the complex relative clause is contained within the middle field. In (4) from Vasishth et al., the complex relative clause is located within the prefield. As will be discussed in section 2, existing theories of sentence complexity make the prediction that doubly center-embedded relative clauses should be easier to process when they are contained within the prefield than when they are contained within the middle field. Under the common assumption that the missing-VP effect occurs only under conditions of high processing load (Frazier 1985; Gibson & Thomas 1999), this may be among the reasons for why Bader et al. found evidence for the missing-VP effect, but Vasishth et al. (2010) did not. The literature concerned with the missing-VP effect has treated this effect as a phenomenon restricted to language comprehension. A recent corpus study of doubly center-embedded relative clauses in German revealed, however, that the very same incomplete sentences that are erroneously accepted as grammatical during language comprehension are also produced from time to time during language production (see Bader, submitted). This finding provides independent evidence for the assumption that the missing-VP effect occurs not only in SVO languages like English but also in SOV languages like German. So far, this evidence is restricted to only one kind of structure, namely sentences in which the doubly center-embedded relative clause is contained within the middle field of an embedded clause. For sentences in which the complex relative clause is located within the prefield, corpus evidence is lacking. Based on new empirical evidence from an on-going corpus study and a speeded grammaticality judgment experiment, I will argue for two points in this chapter. First, the missing-VP effect is a quite general phenomenon also in German, both in language comprehension and in language production. Second, the conflicting findings of Bader et al. (2003) and Vasishth et al. (2010) can be traced back to structural differences in the sentence materials investigated by these researchers. The organization of this chapter is as follows. The next section discusses the processing complexity associated with the various positions that a relative clause in German can occupy. Section 3 summarizes corpus evidence concerning the position of doubly center-embedded relative clauses, both complete ones and incomplete ones due to a missing VP. Section 4 presents an experiment that has investigated complete and incomplete doubly center-embedded relative clause in three different positions within their superordinate clause. The chapter ends with a general discussion in section 5.



Complex center embedding in German – The effect of sentence position 

 13

2 The position of relative clauses in German As pointed out above, a major difference between the experiments of Vasishth et  al. (2010) and the experiments of Bader et al. (2003) concerns the position of the complex relative clause within the upper matrix clause. The experimental stimuli of Vasishth et al. (2010) had the complex relative clause within the prefield of a main clause. The experimental stimuli of Bader et al. (2003), in contrast, had the relative clause within the middle field of either a main or an embedded clause. To see why this might make a difference, we first have to consider more generally how the position of a relative clause affects its processing complexity. There are competing theories of parsing complexity (e.g., Frazier 1985, Hawkins 1994, 2004, Gibson 1998, 2000, MacDonald & Christiansen 2002). In the following, we will rely on the efficiency theory of Hawkins (1994) because this is the most comprehensive theory relating processing complexity to word order variation. The central principle of this theory is the principle of early immediate constituents (EIC) given below. (6) EIC (Hawkins 1994, p. 77) The human parser prefers linear orders that maximize the IC-to-non-IC ratios of constituent recognition domains. To see the EIC at work, consider the two embedded clauses in (7). Sentence (7a), in which a relative clause occurs intraposed, the VP contains three immediate constituents, the subject, the object and the verb. The constituent recognition domain for the VP spans the complete VP beginning with the first word Max and ending with the last word liest (‘reads’). Since the constituent recognition domain (CRD) comprises three immediate constituents (=ICs) and six words (=non-ICs), this sentence has an IC-to-non-IC ratio of 3/6 = 50%. (7) a. dass Max [np Bücher [cp die ihn faszinieren]] liest that M. books which him fascinate reads 1 2 3 4 5 6 b. dass Max [np Bücher] liest [cp that M. books reads 1 2 3

die which

ihn him

faszinieren] fascinate

When the relative clause is extraposed, as in (7b), the CRD of the VP is only three words long. Since the VP still contains three immediate constituents, the ­IC-to-non-IC ratio is 3/3 = 100%, which is substantially higher than the

14 

 Markus Bader

I­ C-to-non-IC ratio for (7a). The EIC thus predicts that the relative clause is preferentially extraposed in the case of (7).4 We next turn to main clauses. When the head noun of the relative clause is contained within the middle field and the clause final verb position is filled, basically the same considerations apply as in the case of embedded clauses. This is shown in (8). (8) a. Max wird [np Bücher [cp die ihn faszinieren]] lesen M. will books which him fascinate read 1 2 3 4 5 6 b. Max wird [np Bücher] lesen [cp M. will books read 1 2 3

die which

ihn him

faszinieren] fascinate

In (8), the finite verb in the verb-second position (= C°) has been included in the CRD of the VP because ultimately it has to be interpreted within the VP by associating it with its trace in the verb-final position. A variation to (8) arises when a sentence contains no verb other than the finite one and the clause final verb position therefore remains empty. This configuration is shown in (9). (9) Max M.

liest [np reads 1

Bücher [cp die ihn faszinieren]] books which him fascinate 2

Syntactically, sentence (9) is ambiguous because one cannot decide whether the relative clause has been extraposed or not. From a parsing perspective, however, this is of no concern. With regard to the EIC, a sentence like (9) is optimal. The CRD for VP is as short as possible, and the relative clause is adjacent to its head noun. The final configuration to consider is the one in which the relative clause is located within the prefield, as shown in (10).

4 The situation is in fact more complicated because the shorter CRD for the VP when extraposition has applied is offset by a longer CRD for the NP containing the relative clause. However, when the relative clause is sufficiently long in comparison to the material intervening between the head noun and the relative clause, extraposition will be favored. In the sentences that are considered later, this will always be the case. See Hawkins (1994) for discussion of this point and Webelhuth et al. (2013) for an overview of current approaches to extraposition.



Complex center embedding in German – The effect of sentence position 

 15

(10) [np Bücher [cp die ihn faszinieren]] wird Max lesen books which him fascinate will M. read 1 2 3 Here, the CRD of VP is again optimal. We now also have to consider how the phrase in the prefield affects efficiency. This amounts to the question of what the CRD for the top-level CP is. So far, this question could be neglected because the sentences under consideration were all on a par in this regard. It may seem as if the CRD of CP begins at the first word of the sentence initial NP and ends with the verb in clause-second position. However, this ignores the fact that the CP node does not need to be constructed before it is unambiguously signaled by the input, which is the finite verb in C°. After all, in authentic texts, it is not uncommon for non-clausal phrases to be used as complete utterances (see next section for quantitative evidence). Under this assumption, processing a clause-initial phrase in the prefield causes fewer storage costs than processing a phrase within the middle field because no higher structure has to be stored for a phrase in the prefield. For the middle field, in contrast, higher structure has to be stored throughout the whole sentence. The idea that storage costs are an important source of processing complexity goes back at least to Yngve (1960). In the context of the dependency locality theory of Gibson (2000), Chen et al. (2005) have provided direct experimental evidence for this notion. In summary, the EIC of Hawkins (1994) predicts that relative clauses should preferentially occur in a peripheral position, either right peripheral (­ extraposition) or left peripheral (prefield). There is much evidence for a ­preference for extraposition in comparison to intraposition (e.g., Hawkins 1994, ­Uszkoreit et al. 1998). Corresponding evidence for the prefield comes from Bader & Häussler (2010). In this corpus study of the order of subject and object, 103 main clauses contained a relative clause. (11) shows the distribution of relative clauses across the four main clause structures discussed above. (11) prefield (10) 46

middle field (8a) 7

midf./extrap (9) extraposed (8b) 22

26

As shown in (11), 46 relative clause were found in the prefield and thus left-­ peripheral, 48 relative clauses were right peripheral either by means of extraposition or because there was no clause-final verb, and only 7 relative clauses were contained within the middle field.

16 

 Markus Bader

3 Corpus evidence Doubly center-embedded relative clauses are rare, but in the written modality, they have been attested in a range of languages (Karlsson 2007). For German, Bader (submitted) has presented detailed corpus results for doubly center-embedded relative clauses that are contained within the middle field of an embedded clause. The corpus that was analyzed in Bader (submitted) is the deWaC corpus made available by the University of Bologna (http://wacky.sslmit.unibo.it; see Baroni et al. 2009). The deWaC corpus is a huge corpus of written German built by web crawling. It contains about 1,600,000,000 tokens of text in ca. 92,000,000 sentences. Because the deWaC corpus is not a treebank, relevant corpus instances cannot be retrieved by searching for particular tree structures. For the present corpus study, the sentences of interest were extracted by search patterns stated in terms of part of speech tags. For a relative clause containing another relative clause, the pattern in (12) was used [(¬V)* is a possible empty sequence of non-verbs]. (12) RelPro (¬V)* N, RelPro der heute den Freund der krank ist, who today the friend that sick is

besucht visits

The search pattern in (12) specifies only the left part of the complex relative clause, up to the second relative pronoun. It therefore matches both complete and incomplete relative clauses. By imposing further constraints on the left context, the clausal position of the complex relative clause was specified as the prefield of a main clause, the middle field of a main clause, or the middle field of an embedded clause. The corpus instances found in this way were then manually checked to remove false alarms. Eleven percent of the deWaC corpus have been analyzed so far.5 For all three clausal positions, both complete and incomplete doubly center-embedded relative clauses were found in the deWaC corpus. Original corpus examples are shown in Table 1 (complete doubly center-embedded relative clauses) and Table 2 (doubly center-embedded relative clauses missing a VP).6

5 Eleven percent of the deWaC corpus correspond to the first two of 19 files making up the deWaC corpus. In Bader (submitted), the complete deWaC corpus was analyzed. To make the corpus counts comparable, the frequency counts from Bader (submitted) are from the same two deWaC files analyzed here. 6 The main clause/prefield sentence in Table 2 contains an additional error. The sentence-initial NP Die große Herausforderung (‘the big challenge’) does not bear dative case as required by the verb but nominative/accusative case instead. Given the complexity of the sentence-initial NP, an error of this kind is not unexpected (see Bader & Bayer 2006).



Complex center embedding in German – The effect of sentence position 

 17

Table 1: Corpus examples of complete sentences with double center-embedding Main clause: prefield (deWaC-1/27789) M Bücher books C-1 DIE MIR   DER    SCHULLEITER, WHICH ME    THE    HEADMASTER der selbst die     Ausleihe    besorgte, C-2 who himself the    lending    managed VERMUTLICH     NICHT       GEGEBEN      HÄTTE, C-1 PRESUMABLY      NOT      GIVEN      HAD nahm ich    “   für      meinen      Vater  ”     mit. M took I   for      my     father     with ‘I took books that the headmaster, who himself managed the lending, would not have given to me, “for my father”.’ Main clause: middle field (deWaC-2/30873) M Insbesondere sein Bruder  Ludwig   wollte   das  Andenken    seines    Bruders, in particular his brother  L.        wanted   the  memory    of-his   brother WELCHES ER    DURCH      EINIGE     PASSAGEN, C-1 WHICH HE    BY      SOME     PASSAGES die er     als      vulgär  bezeichnete, C-2 that he    as     vulgar  called GEFÄHRDET     SAH, C-1 ENDANGERED    SAW schützen. M protect ‘In particular his brother Ludwig wanted to protect the memory of his brother, which he saw endangered by some passages which he called vulgar.’ Embedded clause: middle field (deWaC-1/25004) M Probleme gab es, problem existed it F-1 weil einige Kurse, because some courses C-2 DIE IN DER SCHÖNEN BROSCHÜRE, THAT IN THE NICE BROCHURE C-3 die man vorher zugeschickt    bekommt, that one before sent        got C-2 AUFGELISTET    WAREN, LISTED       WERE gestrichen worden   waren. F-1 canceled been     were ‘There were problems because some courses which were listed in the nice brochure which was sent out in advance were canceled.’ Note: M = main clause; C = center-embedded clause; F = finally embedded clause

18 

 Markus Bader

Table 2: Corpus examples of incomplete sentences with double center-embedding Main clause: prefield (deWaC-1/70958) Die große  Herausforderung, M the big  challenge die mit der Erhaltung der Kirchenbauten, C-1 which with the preservation the church buildings C-2 die wie keine zweite historische Baugattung that like no second historical building-type unsere Städte und Dörfer prägen    […] our towns and villages shaped C-1 […] M kann nur mit einer konzertierten Aktion begegnet werden. can only with a concerted action met be ‘The big challenge which […] with the preservation of the church buildings that have shaped our towns and villages like no other historical kind of building, can only be met with a concerted action.’ Main clause: middle field (deWaC-2/10970) Ebenso ist der Herr Jesus Christus, M likewise is the lord Jesus Christ C-1 der hier mit vollem Titel, who here with full title C-2 der Seine ganze Größe und Herrlichkeit andeutetet, which His whole grandness and glory indicates C-1 […] M die Quelle von Gnade und Friede. the source of mercy and peace ‘Likewise is the lord Jesus Christ, who here […] with His full title, which indicates His whole grandness and glory, the source of mercy and peace.’ Embedded clause: middle field (deWaC-1/95201) Dieser Typ entsteht, M this type emerges F-1 wenn lin-3 oder ein Gen, when lin-3 or a gene C-2 das für die Induktion, that for the induction C-3 die von der Ankerzelle ausgeht, that from the anchor cell originates C-2  […] F-1 mutiert ist. mutated is ‘This type emerges when lin-3 or a gene that […] for the induction that originates from the anchor cell has mutated.’ Note: M = main clause; C = center-embedded clause; F = finally embedded clause



Complex center embedding in German – The effect of sentence position 

 19

The authentic examples in Tables 1 and 2 show that writers produce both complete and incomplete doubly center-embedded relative clauses in all three syntactic contexts under consideration. Quantitative information for each syntactic context is provided in Table 3. The upper part shows the number of doubly centerembedded relative clauses in each context. In main clauses, the total number of examples was much higher for the prefield than for the middle field. The total number of examples in the middle field of embedded clauses is in between, but because the overall number of main and embedded clauses is not known, a direct comparison is not possible. Table 3: Number of corpus hits

Centerembedded Final

Prefield/ main clause

Middle field/ main clause

Middle field/ embedded clause

Complete Missing VP

87 2

  12 8

42 12

Total

89 51

   20 190

54   –

Middle field/embedded clause data are from Bader (submitted)

When looking at the distribution of complete and missing-VP center-­embedded relative clauses, the three sentence positions differ strikingly. In doubly ­center-embedded relative clauses located within the prefield of a main clause, the VP was missing in only 2 out of 89 examples. For doubly center-embedded relative clauses located within the middle field, the rate of missing-VP examples is much higher. For the middle field of embedded clauses, the VP was missing in 12 out of 52 example, and for the middle field of main clauses, the VP was missing in 8 out of 20 examples. For the two middlefield conditions, the ratios of complete to incomplete sentences did not differ significantly (Fisher’s p = .148), but they differed significantly when comparing prefield to middlefield (Fisher’s p < .001). These data show clearly that the chance of correctly producing a doubly ­center-embedded relative clause depends on the position of the relative clause within its superordinate clause. As discussed in the preceding section, relative clauses in the prefield are advantageous from a processing perspective because they are in a sentence peripheral position. This hypothesis, together with the additional assumption that the missing-VP effect occurs under conditions of high processing load, is in full agreement with the data in Table 3. Doubly center-­ embedded relative clauses in the prefield are least vulnerable to the m ­ issing-VP effect. In the middle field, in contrast, relative clauses lead to enhanced

20 

 Markus Bader

­ rocessing load. Doubly center-embedded relative clauses, for which processing p load is especially high, are accordingly prone to errors. The bottom row (‘Final’) shows the number of examples in which a complex relative clause occurred in a final position without necessarily being extraposed. For the condition middle field/main clause, these are sentences in which the clause final verb cluster remains empty [see example (9)]. As shown in Table 3, examples of this type occur with a frequency that is almost ten times higher than the frequency for corresponding center-embedding clauses. This confirms the complexity predictions derived in the preceding section. For the condition prefield/main clause, examples in the row ‘Final’ are NPs that dominate a complex relative clause but which are not themselves dominated by anything else. That is, these are NPs that are not part of a sentence but function, for example, as answer to a wh-question. These examples are listed in the column ‘prefield/main clause’ because in a left-to-right parse of a sentence, they are indistinguishable from an NP within the prefield until the absence or presence of a finite verb signals the NP’s status. What is interesting about examples of this kind is that they occur quite frequently. Neither examples of this kind nor examples that continue with the verb in verb-second position seem to pose ­problems for the parser. Thus, when processing a potentially clause-initial NP, the parser cannot have strong expectations about what to see next – a finite verb or a final punctuation mark. The corpus data presented in this section support the hypothesis that the conflicting experimental findings of Bader et al. (2003) and Vasishth et al. (2010) are caused by differences concerning the experimental material. Vasishth et al. (2010) investigated sentences with a doubly center-embedded relative clause in the prefield and found no evidence for the missing-VP effect. Bader et al. (2003), in contrast, found evidence for the missing-VP effect when testing sentences with a doubly center-embedded relative clause in the middle field. This corresponds closely, although not completely, with the corpus data presented in this section. As shown in Table 3, the missing-VP effect occurred quite rarely in prefield sentences, but it was not altogether absent even there. Since the position of the relative clause has so far not been varied within a single experiment, an experiment was run to determine whether the position of a doubly center-embedded relative clause affects language comprehension in a way similar to language production.

4 Experiment The experiment investigated sentences containing a doubly center-embedded relative clause using the same speeded grammaticality judgment procedure as



Complex center embedding in German – The effect of sentence position 

 21

the experiments reported in Bader et al. (2003). This complex relative clause either contained both the VP of the higher and the lower relative clause and thus was complete or the VP of the higher relative was missing. The complex relative clause occurred in one of the three positions discussed above: the prefield of a main clause (13), the middle field of a main clause (14), or the middle field of an embedded clause (15). (13) Prefield in main clause Der Polizist, der den Verbrecher, der die Bank überfallen hatte, the policeman who the criminal who the bank robbed has [gefasst hat,] wird eine Auszeichnung erhalten. caught has will a award get ‘The policeman who caught the criminal who had robbed the bank will get an award.’ (14) Middle field in main clause Vermutlich wird der Polizist, der den Verbrecher, der die Bank presumably will the policeman who the criminal who the bank überfallen hatte, [gefasst hat,] eine Auszeichnung erhalten. robbed has caught has a award get ‘Presumably, the policeman who caught the criminal who had robbed the bank will get an award.’ (15)

Middle field in embedded clause In der Zeitung stand, dass der Polizist, der den Verbrecher, in the newspaper stood that the policeman who the criminal der die Bank überfallen hatte, [gefasst hat,] who the bank robbed had caught has eine Auszeichnung erhalten wird. a award get will ‘The newspaper reported that the policeman who caught the criminal who had robbed the bank will get an award.’

4.1 Method Participants: Thirty-four students from the University of Konstanz participated in the experiment. All participants were native speakers of German. Materials: Thirty sentences were constructed following the model of sentences in (13), (14), and (15). Each sentence appeared in six versions according to the two factors structure (complete versus missing-VP) and position (prefield of main clause versus middle field of main clause versus middle field of embedded

22 

 Markus Bader

clause). Both the higher and the lower relative clause consisted of a subject, an accusative object and a verb. The relative pronoun was always the subject, and the relative clauses therefore always had subject-before-object word order. The higher relative clause modified the subject of the superordinate clause, which also was a clause with subject-object word order. The sentences were distributed across six lists according to a Latin square design. Each list contained only a single version of each sentence and an equal number of sentences in each condition. The experimental lists were interspersed in a set of about 90 filler sentences that were mainly from unrelated experiments. Each participant saw only one list. Procedure: The experiment was run using the DMDX software developed by K.I. Forster and J.C. Forster at Monash University and the University of Arizona. Each trial began with the presentation of the words ‘Bitte Leertaste drücken’ (‘Please press spacebar’). After the spacebar was pushed, a fixation point appeared in the center of the screen for 1050 ms. Thereafter, the sentence was presented in a ­word-by-word fashion with each word appearing successively in the center of the screen. The presentation time for each word was 225 ms plus an additional 25 ms per character. There was no interval between words. Immediately after the last word, three question marks appeared on the screen, indicating to participants that they now had to judge the grammaticality of the sentence. Participants had to give their answer by pressing the right shift key to judge a sentence as grammatical and the left shift key to judge a sentence as ungrammatical. Type of response and response time were recorded automatically. If a participant did not respond within 2000 ms, the words ‘zu langsam’ (‘too slow’) appeared on the screen and the trial was aborted. Each participant received 10 practice items before the experimental session started.

4.2 Results The percentages of sentences judged as grammatical are shown in Table 4. The judgment data were analyzed by means of a mixed effect logistic regression model.7 Contrast coding was used for the two factors in order to obtain comparisons between means. Following the advice given in Barr et al. (2013), the random effects (participants and items) contained the full factorial design. The results of the mixed effect model are shown in Table 5. Complete sentences were more often judged as grammatical than incomplete sentences (76% vs. 41%), resulting in a significant main effect of structure. The

7 All statistical analyses were done using the statistics software R (R development core team 2005).

Complex center embedding in German – The effect of sentence position 



 23

Table 4: Mean percentages of grammatical judgments

Complete Missing VP

Prefield/ main clause

Middle field/ main clause

Middle field/ embedded clause

88 (2.5) 30 (6.0)

64 (5.3) 48 (5.3)

76 (3.8) 44 (5.4)

Standard error (by participants) is given in parenthesis. Table 5: Results of mixed-effect model for judgment data Contrast Structure Position (pre/m vs. mid/m) Position (pre/m vs. mid/m) × structure Position (mid/m vs. mid/e) Position (mid/m vs. mid/e) × structure

Estimate

Std. Err.

z value

Pr(>|z|)

–2.37 –0.12 3.68 0.13 –0.79

0.34 0.29 0.63 0.26 0.48

–6.95 –0.44 5.85 0.50 –1.66

< 0.01 0.66 < 0.01 0.62 0.10

factor position, in contrast, had no main effect. Neither the comparison between prefield/main clause and middle field/main clause (59% vs. 56%) nor the comparison between middle field/main clause and middle field/embedded clause (56% vs. 60%) was significant. There was, however, a significant interaction for the comparison between the two main clause conditions. In the condition prefield/ main clause, the difference between complete and missing-VP sentences was significantly larger than in the condition middle field/main clause (58% vs. 16%). The corresponding difference for the two middle field conditions failed to reach significance (16% vs. 32%). For complete sentences, pairwise comparisons revealed that the condition prefield/main clause received higher grammaticality ratings than the condition middle field/embedded clause (88% vs. 76%, p < .01). The latter condition in turn received higher ratings than the condition middle field/main clause (76% vs. 64%; p < .01). For missing-VP sentences, the percentages of grammatical judgments were lower in the condition prefield/main clause than in the condition middle field/embedded clause (30% vs. 44%; p < .01). The parallel contrast for the middle field conditions was not significant (44% vs. 48%; p > .3).

4.3 Discussion The experimental results reported in this section allow us to derive a ranking of sentence complexity based on the position of the doubly center-embedded

24 

 Markus Bader

relative clause. The complexity incurred by a doubly center-embedded relative clause is smallest when the relative clause is contained within the prefield. Processing complexity increases when the complex relative clause is put into the middle field. This holds for the middle field of both main and embedded clauses, with the complexity increase being even greater for the middle field of main clauses than for the middle field of embedded clauses. The position-based complexity variation was visible for complete sentences as well as for missing-VP sentences. For complete sentences, the percentages of sentences judged as grammatical decreased with increasing complexity, from a high 88% for prefield/main clause sentences to just 64% for middle field/main clause sentences. For missing-VP sentences, increasing complexity led to higher rates of incomplete sentences erroneously accepted as grammatical. When the complex relative clause was part of the prefield, sentences with a missing-VP were accepted 30% of the time. When the complex relative clause was contained within the middle field, the acceptance rate for missing-VP sentences increased to about 46%, with only a small difference between main and embedded clauses. In sum, the missing-VP effect occurs in all relative clause positions, but the magnitude of the effect is modulated by position.

5 General discussion Taken together, the corpus data and the experimental results presented in this chapter allow two conclusions. First, the missing-VP effect is a quite general phenomenon also in German, both in language comprehension and in language production. Second, the conflicting findings of Bader et al. (2003) and Vasishth et al. (2010) can be traced back to structural differences in the sentence materials investigated by these researchers. The remainder of this chapter will focus on the question of why sentences with a missing-VP often give rise to an illusion of grammaticality, and why this illusion depends on the position of the doubly center-embedded relative clause within its superordinate clause. For reasons of space, the discussion will necessarily be sketchy. With regard to the source of the missing-VP effect, several broad types of accounts can be distinguished.8 According to a first type of account, the parser

8 A further type of account is the connectionist approach of Christiansen & Chater (1999) and Christiansen & MacDonald (2009). A consideration of this approach is beyond the scope of this chapter.



Complex center embedding in German – The effect of sentence position 

 25

may delete part of a syntactic structure if it is short of processing resources. If, in the case of doubly center-embedded relative clauses, the prediction of a VP for the higher relative clause is among the deleted structure, the missing-VP effect follows. Variants of this limited capacity hypothesis have been proposed by Frazier (1985) and Gibson and Thomas (1999). A major challenge facing this kind of account is how to restrict deletion in such a way that missing-VP sentences may be erroneously accepted, whereas other kind of sentences missing some obligatory part are easily rejected as ungrammatical (see Gibson and Thomas 1999, Vasishth et al. 2010). As an alternative to the limited capacity hypothesis, I propose the discrimination hypothesis in (16), which goes back to an earlier proposal by Bader et al. (2003). (16) The discrimination hypothesis Under conditions of high processing load, the integration of new material becomes error prone when the correct attachment site stands in competition with an incorrect integration site that is similar to the correct integration site. To see the discrimination hypothesis at work, consider again the original ­ issing-VP sentence of Frazier (1985) [repeated in (17)]. m (17) a. The patient the nurse the clinic had hired […] met Jack. IP b.

NP patient

I’ Rel. Clause

I

VP V

nurse [NP clinic hired] V At the point where the verb met has to be integrated next, the parser has computed the phrase structure tree in (17b) (details omitted). This tree contains two empty V nodes [the boxed nodes in (17b)] that have to be filled in order to arrive at a complete and thereby grammatical representation. The V node that has to be filled first is the V node of the higher relative clause. Thus, met should be attached to this V node, which will result in a structure that is ungrammatical because there is no further verb which could fill the verb slot of the main clause and there is no licit attachment site for the final NP Jack. Given that the tree in (17b) contains two obligatory verb positions, the verb met could as well be attached to the second empty verb position, the verb slot of the main clause. The

26 

 Markus Bader

resulting tree would then be ill-formed because the higher relative clause remains without a verb. Empirical evidence suggests that the missing-VP effect is caused by attaching the verb met to the empty verb slot of the main clause and not to the empty verb slot of the higher relative clause. In both cases, the resulting phrase-structure tree contains a violation because of a missing verb, either in the higher relative clause or in the main clause. First, Gibson and Thomas (1999) found that sentences with a doubly center-embedded relative clause were rated significantly worse when the VP of the main clause was omitted. Furthermore, experimental data and corpus results presented in Bader (submitted) show that sentences lacking a verb are erroneously accepted or produced only when the missing verb is the verb of the higher relative clause in a sentence with a doubly center-embedded relative clause. In the experiment reported in Bader (submitted), participants had to judge sentences in which a complex relative clause either occurred centerembedded as in (18a) or extraposed as in (18b). (18) a. Center-embedded Ich weiß, dass der Lehrer, der das Fest, das der Referendar I know that the teacher who the party that the trainee-teacher organisiert hat, [kritisiert hat,] verärgert ist. organized has criticized has disgusted is b. Extraposed Ich weiß dass der Lehrer verärgert ist, der das Fest, I know that the teacher disgusted is who the party das der Referendar organisiert hat, [kritisiert hat.] that the trainee-teacher organized has criticized has ‘I know that the teacher who criticized the party that the trainee-teacher organized is disgusted.’ Whereas sentences as in (18a) were often accepted as grammatical, sentences as in (18b) were rejected most of the time. Corpus data reveal a similar picture. In particular, verbs in less deeply embedded position were missing rarely if at all, especially when the VP was in a final position. Since the main clause VP in (17a) is in final position, it seems unlikely that a syntactic structure with an empty main clause verb gives rise to the missing-VP effect. In sum, with regard to the processing of missing-VP sentences as in (17a) or (18a), the Discrimination Hypothesis advances two claims. First, when the final verb in such sentences is processed, two verbs are still needed, the verb of the higher relative clause and the verb of the superordinate clause. In many cases, the final verb is attached to the verb slot of the superordinate clause, effectively skipping the verb slot of the higher relative clause. Second, the parser often fails



Complex center embedding in German – The effect of sentence position 

 27

to detect that the higher relative clause remains incomplete because it is missing a verb (phrase). The net result is an illusion of grammaticality. In accordance with much of the literature on sentence parsing and memory (see overview in van Dyke & Johns 2012), the discrimination hypothesis sees the missing-VP effect as a failure due to interference. Interference comes about because at the point where the verb of the higher relative clause has to be attached to the ongoing syntactic representation, the verb slot of the matrix clause offers a competing attachment site that is similar to the correct attachment site. Of course, the availability of two similar attachment sites alone does not necessarily lead to performance errors. This is the reason why the statement of the discrimination hypothesis in (16) contains the clause “under conditions of high processing load”. A closer look at what these conditions might be provides further evidence for interference as the source of the missing-VP effect. First, experimental results (Gibson and Thomas 1999) as well as corpus evidence (Bader, submitted) show that the verb of the most deeply embedded relative clause is not erroneously attached at a higher position, despite the existence of even two competing attachment sites for this verb (the verb slot of the higher relative clause and the verb-slot of the matrix clause). Consider furthermore sentence (19), which differs from the double center-embedding example in (18a) only insofar as the most deeply embedded relative clause has been extraposed behind the higher relative clause, thereby reducing the degree of center-embedding from two to one. (19) Ich glaube, dass der Lehrer, der I believe that the teacher who das der Referendar organisiert that the trainee-teacher organized

das Fest kritisiert hat, the party criticized has hat, verärgert ist. has disgusted is

In the corpus study of Bader (submitted), sentences like (19) occurred with a rate that was about twice as high as the rate for sentences exhibiting double centerembedding, but the bold-marked verb was never missing in corpus examples of this type. Experimental evidence does not exist for such sentences, but intuitions suggest that a missing-VP would be easily detected in this context. This indicates that the mis-attachment of a verb presupposes that the verb has been separated from the prior part of its clause by a further clause. When this is not the case, as for the verb of the lower relative clause in doubly centerembedded relative clauses and in examples like (19), the verb is firmly attached to its correct attachment site. This makes perfect sense from an interference perspective, as illustrated in (20) for the case of a doubly center-embedded relative clause.

28 

 Markus Bader

(20) [cp1 … [cp2 … [cp3 … V3] V2] V1]. When encountering V3, the verb of the most deeply embedded clause, the parser has just been engaged in processing V3’s clause, which will accordingly be in a highly active state, thereby preventing any interference from material that has been introduced at an earlier time. When encountering V2, in contrast, CP2 must be retrieved from working memory in order to make an attachment site available. Due to the intervening clause CP3, the initial processing of CP2 was already some time ago. Since the same is true for CP1, there are now two competing attachment sites for V2, enhancing the probability of attaching V2 to an incorrect verb slot. This leaves us with the question of why V2 is preferentially attached into CP1, the superordinate clause, although CP1 is further back in time than CP2. While a detailed answer to this question is beyond the scope of this chapter, it can at least be noted that the phrase-structure configuration under consideration is found in a similar way in an example like (21) (from Gibson et al. 1996). (21) The [lamp(s)] N1 near the [painting(s)] N2 of the [house(s)] N3 [that was damaged during the flood]. The relative clause in (21) can be attached to any of the three nouns. The experimental results of Gibson et al. (1996) show a strong preference for attachment to N3, the most recent attachment site. The least preferred attachment site is N2, the middle site. This matches the findings for missing-VP sentences. V3 is always correctly attached into CP3, the most recent clause. V2, however, is often not attached into CP2, the middle clause, but into CP1, the highest clause. Gibson et al. (1996) propose two principles – recency preference and predicate proximity – to account for their findings concerning (21). To determine whether and how these principles can be integrated into the discrimination hypothesis must be left as a task for future research.9 Let us turn next to the question of how the results presented in this chapter are explained by the Discrimination Hypothesis. In particular, the missing-VP effect is weaker when a doubly center-embedded relative clause is contained within the prefield than when it is contained within the middle field. To ease the discussion, the respective structural configurations are shown in (22) and (23).

9 Gibson et al. (1996) propose the principle of predicate proximity as a generalization of the relativized relevance principle of Frazier (1990). Relativized relevance claims a preference for attachments to the ‘main assertion’ of the current clause and therefore directly predicts a preference for attachment to CP1. However, because Gibson et al. (1996) present evidence against the relativized relevance principle, it must be considered an open question of how best to account for the CP1 preference in missing-VP sentences.

Complex center embedding in German – The effect of sentence position 



(22)

 29

CP NP N

C’ C

RC . . . [NP RC] V

(23)

VP

V

CP

C

VP

dass ‘that’ NP N

V/XP RC

. . . [NP RC] V/XP There are two reasons why detecting a missing-VP should be easier when a doubly center-embedded relative clause is contained within the prefield [= (22)] than when it is contained within the middle field [= (23)]. First, the two possible attachment sites are easier to discriminate in the prefield structure (22) than in the middle field structure (23). In the middle field structure, the choice is between two completely parallel verb positions – both are the head of a headfinal VP. In the prefield structure, in contrast, the choice is between a clausefinal verb and a verb in verb-second position, that is, a verb moved to C°. It will therefore be easier for the parser to discriminate between the two verb positions, which in turn will ease the detection of a missing verb. As pointed out by a reviewer, English missing-VP sentences contain a configuration that is similar to German sentences in which the doubly center-embedded relative clause is in the middle field. Both the verb in the matrix clause and the verb of the relative clause are in V° and thus the potential attachment sites for the verb are highly similar. In addition to interference, storage cost has been hypothesized as a second source of sentence complexity (Yngve 1960, Gibson 2000). In section 1.2, the notion of storage cost was invoked to account for the finding that relative clauses occur much more often in the prefield than in the middle field. As argued there, storage cost is lower in a left-peripheral clausal position than in a clause-internal position.

30 

 Markus Bader

This implies that there is less processing load when the NP containing the doubly center-embedded relative clause is in the prefield than when it is located within the middle field. Because less processing load means more resources for correctly discriminating between competing attachment sites, the probability of making an error will be reduced when the doubly center-embedded relative clause is housed by the prefield. In sum, the discrimination hypothesis claims that at least two reasons increase the chance of the parser to spot a missing-VP when it is contained within the prefield. First, the competing attachment sites for the verb are less similar, and second, processing load due to storage cost is lower. Whether both reasons are in fact necessary for explaining the observed empirical findings is a further issue for future research. A final point to consider is the relationship between language comprehension and language production. As the data presented in this chapter as well as the data from Bader (submitted) show, the very same missing-VP sentences that give rise to the illusion of grammaticality during language comprehension are also produced by writers from time to time. The strongest hypothesis would be that the processes underlying language production and the processes underlying language comprehension share the same working memory resources, as recently argued for by Kempen et al. (2012). However, even if language production and language comprehension were completely separate, it would still be reasonable to assume that they are governed by the same processing and storage principles – after all, they are subserved by the same cognitive architecture.

References Bader, Markus. submitted. Complex center embedded relative clauses in German. Manuscript submitted for publication. Bader, Markus & Josef Bayer. 2006. Case and linking in language comprehension: Evidence from German. Dordrecht: Springer. Bader, Markus, Josef Bayer & Jana Häussler. 2003. Explorations of center embedding and missing VPs. Poster, 16th CUNY Conference on Sentence Processing, MIT, Cambridge/MA. Bader, Markus & Jana Häussler. 2010. Word order in German: A corpus study. Lingua 120(3): 717–762. Bader, Markus & Tanja Schmid. 2009. Verb clusters in Colloquial German. The Journal of Comparative Germanic Linguistics, 12(3): 175–228. Baroni, Marco, Silvia Bernardini, Adriano Ferraresi & Eros Zanchetta. 2009. The WaCky Wide Web: A collection of very large linguistically processed web-crawled corpora. Language Resources and Evaluation Journal, 23(3): 209–226. doi:10.1007/s10579-009-9081-4. Barr, Dale J., Roger Levy, Christoph Scheepers & Harry J. Tily. 2013. Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68: 255–278.



Complex center embedding in German – The effect of sentence position 

 31

Chen, Evan, Edward Gibson & Florian Wolf. 2005. Online syntactic storage costs in sentence comprehension. Journal of Memory and Language, 52: 144–169. Chomsky, Noam & George A. Miller. 1963. Introduction to the formal analysis of natural languages. In: R. Duncan Luce, Robert R. Bush and Eugene Galanter (Eds.), Handbook of mathematical psychology. Vol. II, 269–321. New York, London and Sidney: Wiley. Christiansen, Morton H. & Nick Chater. 1999. Toward a connectionist model of recursion in human linguistic performance. Cognitive Science, 23(2): 157–205. Christiansen, Morten H. & Maryellen C. MacDonald. 2009. A usage-based approach to recursion in sentence processing. Language Learning, 59: 126–161. van Dyke, Julie A. & Clinton L. Johns. 2012. Memory interference as a determinant of language comprehension. Language and Linguistics Compass, 6(4): 193–211. Frazier, Lyn. 1985. Syntactic complexity. In: David R. Dowty, Laurie Karttunen and Arnold Zwicky (Eds.), Natural language parsing. Psychological, computational and theoretical perspectives. 129–189. Cambridge: Cambridge University Press. Frazier, Lyn. 1990. Parsing modifiers: Special purpose routines in the human sentence parsing mechanism. In: David A. Balota, Giovanni B. Flores d’Arcais and Keith Rayner (Eds.), Comprehension processes in reading, 303–330. Hillsdale, NJ: Lawrence Erlbaum Associates. Gibson, Edward. 1998. Linguistic complexity: locality of syntactic dependencies. Cognition, 68: 1–76. Gibson, Edward. 2000. The dependency locality theory: A distance-based theory of linguistic complexity. In: Alec Marantz, Yasushi Miyashita and Wayne O’Neil (Eds.), Image, language, brain. Papers from the first Mind Articulation Project Symposium. 95–126. Cambridge, MA: MIT Press. Gibson, Edward, Neal Pearlmutter, Enriqueta Caseco-Gonzales & Gregory Hickok. 1996. Recency preference in the human sentence processing mechanism. Cognition, 59: 23–59. Gibson, Edward & James Thomas. 1999. Memory limitations and structural forgetting: the perception of complex ungrammatical sentences as grammatical. Language and Cognitive Processes, 14: 225–248. Gimenes, Manuel, François Rigalleau & Daniel Gaonac’h. 2009. When a missing verb makes a French sentence more acceptable. Language and Cognitive Processes, 24(3): 440–449. Gordon, Peter C. & Matthew W. Lowder. 2012. Complex sentence processing: A review of theoretical perspectives on the comprehension of relative clauses. Language and Linguistics Compass, 6(7): 403–415. Hawkins, John A. 1994. A performance theory of order and constituency. Cambridge: Cambridge University Press. Hawkins, John A. 2004. Efficiency and complexity in grammars. Oxford: Oxford University Press. Häussler, Jana & Markus Bader. 2012. Locality and anti-locality effects in German: insights from relative clauses. Talk, 25th CUNY Conference on Sentence Processing, CUNY, New York City, NY. Häussler, Jana & Markus Bader. 2015. An interference account of the missing-VP effect. Frontiers in Psychology, 6: 1–16. Karlsson, Fred. 2007. Constraints on multiple center-embedding of clauses. Journal of Linguistics, 43(2): 365–392. Kempen, Gerard, Nomi Olsthoorn & Simone Sprenger. 2012. Grammatical workspace sharing during language production and language comprehension: Evidence from grammatical multitasking. Language and Cognitive Processes, 27(3): 345–380.

32 

 Markus Bader

MacDonald, Maryellen C. & Morten H. Christiansen. 2002. Reassessing working memory: comment on Just and Carpenter (1992) and Waters and Caplan (1996). Psychological Review, 109: 35–54. R development core team, . 2005. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing, http://www.R-project.org. Uszkoreit, Hans, Thorsten Brants, Denys Duchier, Brigitte Krenn, Lars Konieczny, Stephan Oepen & Wojciech Skut. 1998. Studien zur performanzorientierten Linguistik Aspekte der Relativsatzextraposition im Deutschen. Kognitionswissenschaft, 7: 129–133. Vasishth, Shravan, Katja Suckow, Richard L. Lewis & Sabine Kern. 2010. Short-term forgetting in sentence comprehension: Crosslinguistic evidence from verb-final structures. Language and Cognitive Processes 25(4): 533–567. Webelhuth, Gert, Manfred Sailer & Heike Walker (Eds.). 2013. Rightward movement from a cross-linguistic perspective. Amsterdam: Benjamins. Yngve, Victor H. A. 1960. A model and a hypothesis for language structure. Proceedings of the American Philosophical Society, 104: 444–466.

Gisbert Fanselow, Jana Häussler and Thomas Weskott1

Constituent order in German multiple questions: Normal order and (apparent) anti-superiority effects 1 Introduction In many languages of the world, in particular in those with a clause-final positioning of the verb, the order of the constituents of a clause is fairly free. Neverthe­ less, clauses have an ‘unmarked’ or ‘normal’ arrangement of their constituents in most of these free constituent order languages – polysynthetic languages such as Mohawk are a notable exception (Baker 1996). This chapter addresses the factors that determine whether a given ­constituent order is unmarked or not. In particular, we report a series of judgment experiments concerned with constituent order preferences in German multiple questions. The results show that multiple questions are a further, hitherto unknown, argument for the claim that normal order is not just determined by (semantic) role but also by noun phrase (NP) type: normal word order for wh-phrases differs from normal word order in simple declaratives. We will offer an attempt of an explanation for this difference in terms of a hierarchy of case-assigning heads in the final section of the chapter. The chapter is organized as follows. We first give a short overview of ordering preferences in German declarative sentences, identifying the class of constructions that we will focus on here, viz. constructions with an unmarked object-before-subject order in declarative clauses. Section 3 briefly summarizes ­previous findings concerning constituent order in multiple questions. In section 4, we present five experiments examining ordering preferences in multiple questions. The implications of the experimental findings will be ­discussed in section 5.

1 The research reported here was supported by grant FA 255/6-2 of the Deutsche Forschungsgemeinschaft to the first author. We want to thank Andreas Schmidt and Ümmühan Yildiri for their hard work during data collection. We thank the editors and two anonymous reviewers for their valuable comments.

34 

 Gisbert Fanselow, Jana Häussler and Thomas Weskott

2 Linearization preferences in German declarative sentences German is a language with an underlying clause-final placement of the verb. Main clauses are subject to a ‘finite second’ constraint, which forces the movement of the inflected main verb or auxiliary to second position in the clause, i.e., to some head in the complementizer phrase (CP) of the sentential projection. Like most other languages with an underlying object-before-verb serialization, German shows a relatively high degree of freedom of constituent order. The canonical ordering is subject-beforeobject (SO), but the reverse, object-before-subject (OS) is also possible. However, OS is, at least in standard agentive transitive clauses, less preferred: OS sentences are rare (Hoberg 1981; Kempen & Harbusch 2005; Bader & Häussler 2010), they receive lower acceptability ratings (at least when presented in isolation) (e.g., Pechmann, Uszkoreit, Engelkamp & Zerbst 1994; Keller 2000) and they cause processing difficulties even when ambiguity is not involved (Krems 1984; Fanselow, Kliegl & Schlesewsky 1999; Fiebach, Schlesewsky & Friederici, 2002; Felser, Münte & Clahsen 2003).2 The seminal work of Lenerz (1977) and the research building on its insights identified various factors that favor the choice of marked OS order in agentive clauses, such as definiteness, pronominality, animacy, information structure (e.g., Lenerz 1977; Uszkoreit 1987; Müller 1999). Outside the realm of agentive constructions, OS order can be unmarked or even preferred. Passivized ditransitive verbs are a major source for this type of OS structure, as already noted by Lenerz (1977). In addition, several classes of active non-agentive verbs license OS as an unmarked order, see the classification in (1) adapted from Eisenberg (2004: 80). (1) a. Object (accusative or dative) = experiencer (i) accusative: anekeln (‘disgust’), begeistern (‘inspire’), … (ii) dative: auffallen (‘strike’), gefallen (‘please’), ... b. Object (dative) = cause: gelingen (‘succeed’), passieren (‘happen’) c. Object (dative) = possessor: gehören (‘belong to’), zustehen (‘be entitled to’), … Experiencer-object verbs and the other two groups do not only have peculiar linearization preferences. In addition, they exhibit specific restrictions on passivization, nominalization, etc. (Belletti & Rizzi 1988; Grimshaw 1990; Pesetsky 1995; Bayer 2004; Landau 2010).

2 For overviews of evidence for a strong SO-preference in locally ambigous sentences, see Bader & Bayer (2006) and Bornkessel & Schlesewsky (2006).

Constituent order in German multiple questions 



 35

Corpus data match the claims made in theoretical work. The majority of sentences exhibit SO order; OS sentences are very rare. Closer inspection reveals an impact of case: accusative objects predominate in SO sentences, but dative objects predominate in OS sentences (Hoberg 1981; Kempen & Harbusch 2005). Bader & Häussler (2010) report differences between constituent order in the so-called middle field and constituent order involving the prefield.3 For the former, their data confirm the flip of the case bias. When both the subject and the object occur in the middle field, the object is most of the time an accusative object in SO sentences but mainly a dative object in OS sentences. When one of the arguments occupies the prefield, the object is most of the time an accusative object – both in SO sentences (subject in prefield) and OS sentences (object in prefield). However, the proportion of dative is still higher in OS sentences compared to SO sentences. Verbs in OS sentences are mainly unaccusative verbs, psychological predicates, and passivized ditransitive verbs, that is, the classes of verbs that are identified as OS verbs in the literature.

3 Some general remarks on multiple questions In German wh-questions, just as in English, exactly one wh-phrase is fronted to the left periphery of the clause (preceding the finite verb or auxiliary). In multiple questions, there is more than one wh-phrase that could be moved to the left, but the selection of the wh-phrase that actually appears in the left periphery is not random in English and German. In an English construction such as (2a,b), the fronting of the

3 The terms middle field and prefield are taken from the topological model of German syntax. Since Drach (1937), the topological model has provided a means to describe important aspects of the German syntax in a theory-neutral way (Askedal 1986; Engel 1972; Höhle 1986). The prefield corresponds to SpecCP in generative work, the middle field spans the part to the right of C up to the verb cluster.

(i) (ii) (iii)

Prefield SpecCP

Left bracket C0

Middle field

Right bracket verb cluster

Peter P. Heute Today

hat has hat has dass that

den Vater the father Peter den Vater P. the father Peter den Vater P. the father

beobachtet watched beobachtet watched beobachtet hat watched has

36 

 Gisbert Fanselow, Jana Häussler and Thomas Weskott

structurally higher (‘superior’) wh-phrase is mandatory, while the lower wh-phrase must be left in situ. The reverse constellation is not acceptable, as shown in (2b). (2) a. Who recommended who? b. *Who did who recommend? c. Wer hat wen empfohlen who.nom has who.acc recommended? ‛Who recommended who?’ d. Wen hat wer empfohlen? This superiority effect – a lower wh-phrase cannot be moved across a higher wh-phrase – is very robust in English and has been investigated for more than four decades. However, there is still no consensus as to whether the effect is due to a principle of syntax or not (see, e.g., Chomsky 2008 and Hofmeister & Sag 2010 for negative answers to this question). Nor is there consensus what such a principle of syntax would look like (Chomsky 1973, 1981; Pesetsky 2000; Haider 1997 for various alternatives). Fortunately, we do not have to side with a particular proposal at the present moment. German also shows a superiority effect in its multiple questions, at least when one confines one’s attention to questions with two animate wh-phrases such as (2c,d) (Featherston 2005; Fanselow et al. 2011; Häussler et al. 2015). Fronting the wh-subject as in (2c) is preferred compared to fronting the wh-object as in (2d). The effect is, however, much smaller than in English, i.e., the decrease in acceptability found with crossing wh-phrases is larger in English than in German. If we want to avoid the assumption that different syntactic constraints are at work in English and German or that grammatical principles may intrinsically have different strengths in English or German, the following analysis seems plausible. There is a strong grammatical constraint against crossing A-bar movement of an operator phrase penalizing it by a large drop in acceptability (see Pesetsky, 1982). This constraint is responsible for the contrast between (2a) and (2b). The constraint is at work in German too and blocks the fronting of a wh-object across a wh-subject in German SO sentences, that is, (2d) cannot be derived from an underlying structure hat wer (S) wen (O) empfohlen. However, German allows OS structures as a marked option. If (2d) is derived from an underlying hat wen (O) wer (S) empfohlen (as suggested by Haider 1993, cf. also Fanselow 2001), no crossing A-bar movement is necessary for the derivation of (2d), so the strong superiority constraint is not violated. However, the structure underlying (2d) is a marked one, and this markedness may be blamed for the small yet significant drop in acceptability in German object-initial ­multiple questions.



Constituent order in German multiple questions 

 37

This line of reasoning makes a clear prediction: Constructions involving verbs with an unmarked OS order should show the same ordering preference in multiple questions: the object-initial multiple question involves no crossing movement violating superiority when derived from an underlying OS serialization; nor does it incur any penalty for using an OS clause, since OS is unmarked for these verbs. Below we present a series of experiments contradicting this prediction. The experiments attest an SO preference in multiple questions regardless of ordering preferences in declarative counterparts.

4 Multiple questions with predicates with unmarked OS order We constructed the material for our experiments with OS verbs selecting dative rather than accusative objects because many such accusative governing verbs also possess an additional agentive interpretation that might interfere with the participants’ judgments in an unpredictable way. Differences in animacy between the wh-phrases might further blur the already small ordering effects in structures such as (2c,d) (see, e.g., Fanselow et al. 2011). Therefore, the wh-phrases were all animate in our experiments. This has a further advantage: a possible confounding effect of a further serialization strategy of German can be avoided. German prefers the placement of animate noun phrases before inanimate ones. For a sentence such as dass dem Kind das Buch gefällt ‘that the.dat child the.nom book pleases’, it is therefore difficult to tell apart the serialization effects of the animacy preference from the serialization effects triggered by verb class. By confining ourselves to pairs of animate wh-phrases, this difficulty can be avoided. In three separate experiments, we examined ordering preferences in multiple questions for three types of verbs: dative selecting agentive verbs, experiencerobject verbs selecting haben (‘have’) as perfect auxiliary, and experiencer-object verbs selecting sein (‘be’). A fourth experiment examined passive ditransitive sentences, and a fifth experiment investigated their active counterparts.

4.1 Multiple questions with agentive verbs Experiment 1 serves as a baseline: we need to establish that word order in multiple questions is independent of the case of the object. Agentive verbs with a dative object preferentially occur with SO order, just like agentive verbs with

38 

 Gisbert Fanselow, Jana Häussler and Thomas Weskott

an accusative object. Discovering the small superiority-related effect of (2d) also in corresponding multiple questions with dative objects would not conflict with the basic constituent order preference. Experiment 1 employed eight sentence pairs as in (3), containing the following verbs: widersprechen (‘to contradict so.’), zujubeln (‘to cheer for so.’), beistehen (‘to back so. up’), zuprosten (‘to raise one’s glass to so.’), winken (‘to wave to so.’), zuzwinkern (‘to wink at so.’), zürnen (‘to be angry at so.’),4 absagen (‘to call off’). All sentences were root questions containing two wh-phrases – one asking for the subject and the other one asking for the dative object. The two wh-phrases were both unambiguous with respect to case, and they both referred to animate entities. The members of each sentence pair differed in the order of the two wh-phrases. In one version, the wh-subject precedes the wh-object (SO), in the other version the wh-object precedes the wh-subject (OS). hat nach dem Vortrag wem widersprochen? (3) a. Wer Who.nom has after the talk who.dat contradicted ‘Who contradicted who after the talk?’ b. Wem hat nach dem Vortrag wer widersprochen? Who.dat has after the talk who.nom contradicted ‘Who did who contradict after the talk?’ The sentences were distributed across two lists according to a Latin square design. The items within each list were randomized and interspersed in a larger questionnaire containing 100 sentences in total. Sixteen students from the University of Potsdam, all monolingual native speakers of German, judged the acceptability of the sentences on a 7-point scale ranging from 1 (absolutely unacceptable) to 7 (perfectly acceptable). Table 1 gives the mean ratings. Table 1: Mean ratings in Experiment 1 (agentive verbs) SO (Wer ... wem) 5.59

OS (Wem ... wer) 4.48

Multiple questions in which the subject-wh is fronted received higher ratings than their counterparts with opposite ordering. Analyses of variance confirmed this

4 This verb is actually not an agentive verb but a psychological predicate with an e­ xperiencer ­argument. Note, however, that the experiencer is the subject. In any case, excluding the ­respective item from the analysis did not change the results.



Constituent order in German multiple questions 

 39

impression; they revealed a weak but significant main effect of Order [F1(1,15) = 12.1, p < 0.01; F2(1,7) = 9.5, p < 0.05]. We conclude that agentive verbs with a dative wh-object exhibit the same SO preference that has been attested for multiple questions with an accusative wh-object (for the same pattern in ditransitive constructions, see Featherston 2005). Apparently, dative wh-objects and accusative wh-objects behave similarly compared to wh-subjects. Fronting of the wh-subject is preferred regardless of the case of the wh-object.

4.2 Multiple questions with experiencer-object verbs We now turn to constructions with an unmarked OS order in declarative sentences. Experiments 2 and 3 examined multiple questions with an experiencer verb as in (4) and (5). (4) a. Wer hat auf der Tagung wem Who.nom has at the conference who.dat ‘Who impressed who at the conference?’ b. Wem hat auf der Tagung wer Who.dat has at the conference who.nom

imponiert? impressed

ist auf der Tagung wem (5) a. Wer Who.nom is at the conference who.dat ‘Who struck whom at the conference?’ b. Wem ist auf der Tagung wer Who.dat is at the conference who.nom

aufgefallen? struck

imponiert? impressed

aufgefallen? struck

All sentences were root questions containing a wh-subject and a dative wh-object. The factor order varied the order of the two wh-phrases: either the wh-subject was fronted or the wh-object was fronted. Different verb classes based on perfect auxiliary selection were tested in Experiments 2 and 3 because of claims in the literature that differences in auxiliary selection come with differences in underlying syntactic representations in German (see ­Grewendorf 1989 and subsequent work). Experiment 2 tested four (possibly unergative) experiencer-object verbs selecting haben (‘have’) as perfect auxiliary: imponieren (‘to impress so.’), missfallen (‘to displease so.’), gefallen (‘to please so.’), leidtun (‘to feel sorry for’); each verb occurred in two sentences. Experiment 3 employed four unaccusative verbs (with sein ‘be’ as their perfect auxiliary), each in two sentences: auffallen (‘to catch so.’s attention’), einfallen (‘to occur to so.’), entfallen (‘to slip so.’s mind’) and verfallen (‘to fall under so.’s spell’).

40 

 Gisbert Fanselow, Jana Häussler and Thomas Weskott

What we have said so far about multiple questions in German predicts that there is either no difference in acceptability between the a and b examples or that the object-initial examples are even better than their subject-initial counterparts because of an OS preference characteristic of these verbs. For each experiment, we constructed eight sentence pairs, distributed them across two lists each, mixed them with a variety of filler items and randomized the resulting lists. The resulting questionnaires contained 100 items each and were completed by 16 participants each (all monolingual native speakers of German and not participants in one of the other experiments). The participants rated the items’ acceptability on a 7-point scale as in Experiment 1. Table 2 gives the mean ratings for each verb type broken down by order. We see a preference for the wh-subject to precede the wh-object. The factor order reached significance both in Experiment 2 [F1(1,15) = 14.0, p < 0.01; F2(1,7) = 21.0, p < 0.01] and in Experiment 3 [F1(1,15) = 10.7, p < 0.01; F2(1,7) = 5.7, p < 0.05]. Table 2: Mean ratings in Experiments 2 and 3 (experiencer object verbs)

SO (Wer ... wem) OS (Wem ... wer)

Experiment 2 (haben Auxiliary)

Experiment 3 (sein Auxiliary)

4.67 3.47

3.81 2.77

Note that the observed ordering preference contrasts with the OS preference claimed for declarative sentences (e.g., Eisenberg 2004) and therefore disconfirms our expectations. On the assumption that OS is the base-generated order with object-experiencer verbs (see Fanselow 2001; Haider 1993), wh-phrases appear to prefer crossing movement in sentences with such ­predicates. To ensure that the OS linearization preference in declarative clauses indeed holds for the predicates we used in Experiments 2 and 3, we conducted two control experiments. The items for these experiments were constructed from the ones in Experiment 2 and 3 by replacing the wh-phrases with definite NPs as shown in (6) and (7) below. (6) a. Der Doktorand hat auf der Tagung dem the.nom PhD student has at the conference the.dat Professor imponiert professor impressed ‘The PhD student impressed the professor at the conference.’

Constituent order in German multiple questions 



 41

b. Dem Professor hat auf der Tagung der the.dat professor has at the conference the.nom Doktorand imponiert. PhD student impressed Doktorand war auf der Tagung dem (7) a. Der the.nom PhD student was at the conference the.dat Professor aufgefallen. professor struck ‘The PhD student struck the professor at the conference.’ b. Dem Professor war auf der Tagung der the.dat professor was at the conference the.nom Doktorand aufgefallen. PhD student struck The two control experiments were part of larger questionnaires. The questionnaire containing the control experiment for Experiment 2 contained 92 sentences in total and was assessed by 48 participants; the other questionnaire study had 78 items in total and 40 participants. Ratings were obtained on a 7-point scale as in Experiments 1-3. The results are shown in Table 3. Table 3: Mean ratings in the two control experiments (declarative sentences)

SO OS

Control for Experiment 2 (haben Auxiliary)

Control for Experiment 3 (sein Auxiliary)

5.72 5.69

5.09 5.64

The two control experiments confirm that experiencer-object verbs exhibit no SO-preference. The verbs selecting haben (‘have’) as their perfect tense auxiliary used in Experiment 2 show no ordering preference at all; the verbs selecting sein (‘be’) of Experiment 3 also show no strong preference for either order but a trend toward an inverted ordering preference: sentences with OS order received higher ratings than SO sentences; the main effect of Order is significant in this control experiment [F1(1,39) = 12.4, p < 0.01; F2(1,7) = 7.6, p < 0.05], but the effect size is very small (eta-squared = 0.02). Note that the reversed preference is visible even though the OS sentences come with a potential disadvantage due the positioning of the adverbial above the subject.

42 

 Gisbert Fanselow, Jana Häussler and Thomas Weskott

Taken together, the findings in Experiments 2 and 3 and the two control experiments show that the preferred order in German multiple questions cannot be simply equated with the unmarked word order of declaratives. Descriptively speaking we have observed an anti-superiority effect that favors the reversal of normal word order in multiple questions, at least for objectexperiencer verbs selecting sein as perfect auxiliary; for verbs selecting haben, we found no preference in declarative sentences but an SO preference in multiple questions.

4.3 Multiple questions with passive verbs Passive sentences formed with a ditransitive verb often exhibit either no ordering preference or an OS preference. Experiment 4 examined questions in the passive voice containing a wh-subject and a dative wh-object. For comparison, the experiment also included declarative counterparts that use definite NPs instead of whphrases. (8) a. Wer wurde gestern wem vorgestellt? Who.nom was yesterday who.dat introduced-to ‘Who was introduced to whom yesterday?’ b. Wem wurde gestern wer vorgestellt? Who.dat was yesterday who.nom introduced-to c. Der neue Mitbewohner wurde gestern the.nom new flat-mate was yesterday dem Vermieter vorgestellt the.dat landlord introduced-to ‘The new flat-mate was introduced to the landlord yesterday.’ d. Dem Vermieter wurde gestern der neue the.dat landlord was yesterday the.nom new Mitbewohner vorgestellt flat-mate introduced-to We constructed 16 sentences as in (8), each in four versions according to the four conditions resulting from fully crossing the two factors Order (SO vs. OS) and Sentence Type (multiple question vs. declarative sentences). The items were distributed across four lists, randomized and mixed with a variety of filler items. The questionnaire contained 123 sentences in total. Thirty-six students from the University of Potsdam (all monolingual native speakers of German) rated the acceptability of the sentences on a 7-point scale. The mean ratings are given in Table 4.

Constituent order in German multiple questions 



 43

Table 4: Mean ratings in Experiment 4 broken down by sentence type and order.

SO OS

Declarative sentences

Multiple questions

6.02 5.91

5.85 4.91

As expected, declarative sentences exhibit no ordering preference (t1 = 0.78, t2 = 0.63). Multiple questions, in contrast, again received a penalty when the wh-object is fronted (t1 = 5.06, t2 = 6.41; both p-values dat. It is a version of the syntactic function hierarchy of Keenan and Comrie (1977).6 Our experiments show that a wh-phrase low on the Case hierarchy should not cross a wh-phrase higher on the hierarchy when it moves to the left periphery of the clause. 6 Keenan & Comrie (1997) proposed the hierarchy orginally under the label ‘NP accessibility ­hierarchy’ to account for patterns of relative clause formation. Later, the hierarchy was adapted for various phenomena including argument linearization (for an overview, see Croft 2003).

46 

 Gisbert Fanselow, Jana Häussler and Thomas Weskott

The thematic hierarchy is a bit more complex. The agent always occupies the highest position in the verbal projection. Experiencers are higher than themes and patients. The relative hierarchical position of goals and themes is variable. In a German declarative clause, the noun phrase with a higher thematic role preferentially precedes phrases with a lower role, quite irrespective of case. Thus, experiencers precede themes in normal order independently of whether they bear nominative, dative, or accusative case. The two hierarchies completely coincide only for standard agentive transitive constructions (Bornkessel et al. 2003). We believe that our findings on word order preferences can be made sense of along these lines with the help of a few reasonable assumptions. Let us begin with declaratives. Arguments and adjuncts are merged in the verbal projection domain, above which we postulate a set of functional projections from the tense phrase (TP) domain. Some of these functional heads are also responsible for the assignment/checking of case. Tense assigns nominative case, and we work with two further heads that govern accusative and dative case, respectively, and which, for reasons of simplicity, we will simply label as F-Acc and F-Dat, so that we do not have to commit ourselves here as to whether these heads can be identified with categories such as Aspect, etc. This yields a sentence structure such as (10). (10) [ a Tns [ b F-Acc [ c F-Dat [vP α v [VP β [V γ]]]]]] Arguments have to be merged/inserted within vP, in positions determined by their thematic role. Thus, the specifier of vP is the position of agents, and the specifier of VP, the position for experiencers, whereas themes are inserted into the complement position of the verb. Arguments are thus merged in the hierarchy/ order agent > experiencer > theme, which corresponds to their normal order in declaratives. This follows if (unlike what holds for English) arguments normally do not have to leave their merge position in German. In particular, German case assignment does not presuppose that noun phrases move overtly to the specifier position of the case-assigning heads, so that they can stay in vP (as already argued for by den Besten 1985). Fanselow (2001) proposes that the movement to the specifiers of the case-assigning heads is a matter of the covert component of grammar in German, so that it is invisible at surface structure. This is different for wh-phrases as we will explain below. The (marked) alternative constituent order in declaratives can be derived in various ways, e.g., by A-scrambling. A-scrambling is licensed by semantic-­ pragmatic features and implies a certain degree of markedness in grammatical status. We can make A-scrambling responsible for the lower acceptability of marked word order.



Constituent order in German multiple questions 

 47

Let us now turn to multiple questions. We propose that their different ­behavior with respect to normal order can be attributed to the syntactic properties of wh-phrases. Wh-phrases move to an operator position (Spec,CP) in the CP domain of the clause in overt syntax. Since Spec,CP is the highest position in the clause, and given that there is no syntactic lowering, the wh-phrase cannot move covertly from the Spec,CP position to the specifier position of the caseas­signing head. Rather, the wh-phrase already has to pass the specifier of the case-assigning head in overt syntax on its way up to Spec,CP. Consider now the ­following sluicing data: (11) Ich habe bemerkt, dass auf der Feier I have noted that at the party a. jemand jemandem besonders gefallen hat, someone.nom someone.dat particularly pleased had und ich weiß auch wer wem and I know also who.nom who.dat b. *jemand jemandem besonders gefallen hat, someone.nom someone.dat particularly pleased had und ich weiß auch wem wer and I know also who.dat who.nom c. jemandem jemand besonders gefallen hat, someone.dat someone.nom particularly pleased had und ich weiß auch wer wem and I know also who.nom who.dat d. *jemandem jemand besonders gefallen hat, someone.dat someone.nom particularly pleased had und ich weiß auch wem wer and I know also who.dat who.nom The antecedent clause of the sluicing construction can appear with both SO and OS order without any markedness effect because gefallen ‘please’ is not an agentive predicate. The sluiced clause, however, must have SO order, despite the fact that gefallen tolerates OS order, and even when the antecedent clause itself exhibits OS order (11c,d). The sluicing construction thus resembles multiple questions in that normal order is not derivable from the word order preferences of the corresponding declarative. The pattern in (11) follows from our assumption that wh-phrases must move through the specifier positions of the corresponding case phrases in overt syntax. Sluicing constructions are derived by Ā-moving all wh-phrases to the left periphery and subsequently deleting the other material in the sluiced clause as ­illustrated in (12).

48  (12)

 Gisbert Fanselow, Jana Häussler and Thomas Weskott John likes someone and I know [John likes who] → John likes someone and I know [who [John likes who]] → John likes someone and I know [who [John likes who]]

Since there is no lowering, the wh-phrases in (11) must move to the respective case-assigning heads before they undergo Ā-movement to Spec,CP. Thus, the experiencer dative moves to the specifier of F-Dat before it undergoes operator Ā-movement, and the theme nominative goes to Spec,Tense. They reverse their relative order in this movement step. If the superiority condition is formulated in such a way that it forces relative order to be kept constant when more than one phrase moves (as in Müller 2001), and if the superiority condition affects Ā-movement, then the wh-phrases are predicted to appear at the left periphery in the order they have in the specifier positions of the case-assigning heads, viz. the positions before Ā-movement. The normal order facts of sluicing are thus ­accounted for. In contrast to sluicing constructions, only one wh-phrase moves in the overt syntax in multiple questions. This wh-phrase must pass through the specifier of its case-assigning head. We need to show that the preferred order is determined by the case hierarchy. This follows if the wh-phrase left in situ also moves to Spec,CP in the covert component, passing through the specifier of its case as­signer. Thereby, the constellation characteristic of sluicing constructions arises, and we derive the preferred word order if the superiority condition demands that the hierarchies between the highest pre-operator movement positions must be respected by Ā-movement. Deviations from the preferred word order can again be explained in terms of scrambling, now in the domain of case-assigning heads. We have suggested that the differences in canonical word order between declaratives and interrogatives follow from the fact that wh-phrases are forced to move to operator positions through case-assigning heads that have a hierarchy different from the one characterizing thematic roles.

References Askedahl, J.O. 1986. Über ‘Stellungsfelder’ und ‘Satztypen’ im Deutschen. Deutsche Sprache, 14: 193–223. Bader, M., & Bayer, J. 2006. Case and linking in language comprehension: Evidence from German. Heidelberg: Springer. Bader, M. & Häussler, J. 2010. Word order in German: A corpus study. Lingua, 120: 717–762. Baker, M. 1996. The Polysynthesis Parameter. New York: Oxford UP. Bayer, J. 2004. Non-nominative subjects in comparison. In: P. Bhaskararao & K. V. Subbarao (Eds.), Non-nominative Subjects. Vol. 1, 49–76. Amsterdam: Benjamins.



Constituent order in German multiple questions 

 49

Belletti, A., & L. Rizzi 1988. Psych-verbs and theta-theory. Natural Language and Linguistic Theory, 6: 291–352. Den Besten, H. 1985. The ergative hypothesis and free word order in Dutch and German. In J. Toman (ed.) Studies in German Grammar. 23 – 64 Dordrecht: Foris. Bornkessel, I., Schlesewsky, M. & Friederici, A.D., 2003. Eliciting thematic reanalysis effects: the role of syntax-independent information during parsing. Language and Cognitive Processes, 18, 269–298. Bornkessel, I., & Schlesewsky, M. 2006. The extended argument dependency model: A neurocognitive approach to sentence comprehension across languages. Psychological Review, 113, 787–821. Chomsky, N. 1973. Conditions on transformations. In: S. R. Anderson & P. Kiparsky (Eds.), A Festschrift for Morris Halle, 232–286. New York: Holt Rinehart and Winston. Chomsky, N. 1981. Lectures on government and binding. Dordrecht: Foris. Chomsky, N. 2008. On phases. In R. Freidin, C. Otero, & M. L. Zubizarreta (Eds.) Foundational Issues in Liguistic Theory. Essays in Honor of Jean-Roger Vergnaud. 133–166. Cambridge, Mass.: MIT Press. Croft, W. 2003. Typology and Universals, 2nd ed. Cambridge: Cambridge University Press. Drach, E. 1937. Grundgedanken der deutschen Satzlehre, 4th edition. Frankfurt a. M.: Diesterweg. Eisenberg, P. 2004. Grundriß der deutschen Grammatik. Band 2: Der Satz, 2nd edition. Stuttgart: Metzler. Engel, U. 1972. Regeln zur ‘‘Satzgliedfolge’’: Zur Stellung der Satzglieder im einfachen Verbalsatz. Linguistische Studien,1: 17–75. Fanselow, G. 2001. Features, theta-roles, and free constituent order. Linguistic Inquiry, 32: 405–437. Fanselow, G., M. Schlesewsky, R. Vogel & T. Weskott 2011. Animacy effects on crossing wh-movement in German. Linguistics, 49: 657–683. Fanselow, G., R. Kliegl, & M. Schlesewsky 1999. Processing difficulty and principles of grammar. In: S. Kemper & R. Kliegl (eds), Constraints on Language: Aging, Grammar and Memory. 171–202. Dordrecht: Kluwer. Featherston, S. 2005. Universals and grammaticality: Wh-constraints in German and English. Linguistics, 43: 667–711. Felser, C., T. Münte, & H. Clahsen. 2003. Storage and integration in the processing of filler-gap dependencies: An ERP study of topicalization and Wh-movement in German. Brain & Language, 87: 345–354. Fiebach, M. Schlesewsky, & A. D. Friederici. 2002. Separating syntactic memory costs and syntactic integration costs during parsing: The processing of German Wh-questions. Journal of Memory and Language, 47: 250–272. Grewendorf, G. 1989. Ergativity in German. Dordrecht: Foris. Grimshaw, J. 1990. Argument Structure. Cambridge, MA: MIT Press. Haider, H. 1993. Deutsche Syntax, generativ – Vorstudien zur Theorie einer projektiven Grammatik. Tübingen: Narr. Haider, H. 1997. Economy in syntax is projective economy. In C. Wilder, H.-M. Gärtner & M. Bierwisch (eds.), The Role of Economy Principles in Linguistic Theory. 205–226. Berlin: Akademie Verlag. Häussler, J., M. Grant, G. Fanselow & L. Frazier 2015. Superiority in English and German: ­Cross-language grammatical differences? Syntax 18(3), 235–265.

50 

 Gisbert Fanselow, Jana Häussler and Thomas Weskott

Hoberg, U. 1981. Die Wortstellung in der geschriebenen deutschen Gegenwarts-sprache. München: Hueber. Hofmeister, P. & Sag, I.A. 2010. Cognitive constraints and island effects. Language, 86: 366–415. Höhle, T. N. 1986. Der Begriff ‘Mittelfeld’. Anmerkungen über die Theorie der topologischen Felder. In A. Schöne (ed.), Kontroversen, alte und neue. Akten des VII. internationalen Germanisten-Kongresses, Göttingen 1985. Band 3, 329–340. Tübingen: Niemeyer. Keenan, E. L. & B. Comrie 1977. Noun phrase accessibility and universal grammar. Linguistic Inquiry, 8: 63–99. Keller, F. 2000. Gradience in Grammar: Experimental and Computational Aspects of Degrees of Grammaticality. PhD thesis, University of Edinburgh. Kempen, G., & Harbusch, K. 2005. The relationship between grammaticality ratings and corpus frequencies: A case study into word-order variability in the midfield of German clauses. In M. Reis & S. Kepser (Eds.), Linguistic Evidence. 329–349. Berlin: De Gruyter. Krems, J. 1984. Erwartungsgeleitete Sprachverarbeitung. Frankfurt: Lang. Landau, I. 2010. The Locative Syntax of Experiencers. Cambridge, MA: MIT Press. Lenerz, J. 1977. Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: Narr. Meinunger, A. 1996. Discourse Dependent DP (de-) Placement. Groningen: University of Groningen, Center for Language and Cognition. Müller, G. 1999. Optimality, markedness, and word order in German. Linguistics, 37, 777–818. Müller, G. 2001. Order preservation, parallel movement, and the emergence of the unmarked. In G. Legendre, J. Grimshaw, & S. Vikner (eds.) Optimality-Theoretic Syntax, 279–313. Cambridge, MA.: MIT Press. Pechmann, T., H. Uszkoreit, J. Engelkamp, & D. Zerbst 1994. Word Order in the German Middle Field: Linguistic Theory and Psycholinguistic Evidence. Report No. 43, Computational Linguistics at the University of the Saarland. Pesetsky, D. 1982. Paths and Categories. Doctoral dissertation, Cambridge, MA: MIT. Pesetsky, D. 1995. Zero Syntax: Experiencer and Cascades. Cambridge, MA: MIT Press. Pesetsky, D. 2000. Phrasal Movement and its Kin. Cambridge, MA: MIT Press. Uszkoreit, J. 1987. Word Order and Constituent Structure in German. Stanford: CSLI.

Jutta Hartmann, Andreas Konietzko and Martin Salzmann

On the Limits of Non-Parallelism in ATB Movement: Experimental Evidence for Strict Syntactic Identity1 1 Introduction

Both the theoretical and the empirical syntax literature contain examples where ambiguous (i.e., syncretic) morphological forms can resolve morphosyntactic feature conflicts. To mention two examples: 1. Disjunctive coordination of conjuncts with different person values gives rise to conflicting agreement requirements on the verb. In the following German example, the syncretic present tense ending -t is more acceptable than the ­non-syncretic form of the past tense (-t vs. -Ø), see Fanselow & Frisch (2006: 302ff.): (1) a. Er oder ihr kommt verspätet he or you(pl) come.3sg/2pl.prs late ‘He or you come late to the meeting.’ b. Er oder ihr kamt verspätet he or you(pl) come.2pl.pst late ‘He or you came late to the meeting.’

zu dem Treffen. to the meeting zu dem Treffen. to the meeting

1 Earlier versions of this research were reported at Linguistic Evidence (Tübingen 2012) and at Workshare (Nantes 2012). We thank the audiences at these occasions, in particular Asaf ­Bachrach, Peter Culicover, Sam Featherston, Kyle Johnson, Luis Vicente, Thomas Weskott and Susanne Winkler. We are also grateful to Gisbert Fanselow for comments on an earlier version of the chapter, to Nicolas Heizmann for help with the materials and running the study, and to Robin Hörnig for help with the statistical analysis and graphics. Finally, the chapter has greatly benefited from the comments of the two reviewers. This research was supported by the DFG SFB 833, Project A7 and B5 (Hartmann, Konietzko), and by grant PA00P1_136379/1 from the Swiss National Science Foundation (Salzmann).

52 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

2. Comparable case conflicts arise with free relatives in German, where the relative pronoun has to match both the case assigned within the relative clause as well as the case assigned by the matrix verb. Again, while syncretic forms like was ‘what’ can resolve mismatches, non-syncretic forms like wer are much less acceptable, see Vogel et al. (2006: 363ff.): (2) a. Ich verteidigeacc, wer __nom mich ergreift. I defend who.nom me moves ‘I defend who moves me.’ b. Ich glaubeacc, was __nom mich ergreift. I believe what.nom/acc me moves ‘I believe what moves me.’ These two examples clearly show that syncretism, and thus morphological case, plays an important role in (certain domains of) syntax. This is not only an interesting fact in itself, it also constitutes an important clue for the proper syntactic analysis of a given phenomenon.2 Across-the-board movement (ATB movement) as in (3), where a filler seems to be related to two gaps simultaneously, can also give rise to conflicting requirements (see Ross 1967). (3) [Which book] did [John like __] and [Mary dislike __]? It has often been reported in the literature on ATB movement (e.g., Dyla 1984; Franks 1995: 61ff.; Citko 2005: 487) that case mismatches are tolerated as long as the filler bears a syncretic, i.e. underspecified, form that is compatible with the conflicting requirements of both verbs. This is illustrated by the following example from Polish where the wh-word kogo is compatible with both accusative and genitive:3

2 Note that the two examples of mismatches in (1) and (2) are not fully parallel. The person mismatches lead to decreased acceptability even with the syncretic form, whereas the case mismatches resolved by was are judged on a par with fully acceptable sentences where the verbs impose the same case value. This suggests that systematic syncretisms like that in the case of what, which can be analyzed as involving underspecification, are treated differently from accidental syncretisms like that between third singular and second plural, which do not lend themselves so easily to an underspecification analysis (but cf. Müller 2006). 3 An exception to this generalization is found in Bondaruk (2003: 230f.). She points out that non-syncretic mismatches between genitive and accusative are possible if the genitive is a genitive of negation.



On the Limits of Non-Parallelism in ATB Movement 

 53

(4) Kogoacc/gen [Jan nienawidzi  __gen] a [Maria lubi  __acc]?    who.acc/gen John hates and Mary likes ‘Who does John hate and Mary like?’ (Polish) If instead the wh-word is compatible with only one of the verbs, the result is ungrammatical, see Citko (2005: 485):4 (5) *Kogoacc/Komudat [Jan lubi __acc] a [Maria ufa __dat]?  who.acc/who.dat John likes and Mary trusts  ‘Who does John like and Mary trust?’ (Polish) In this chapter we will focus on German ATB movement, where – according to the literature – syncretic mismatches are also tolerated:5 (6) [Käse] acc/nom [mag ich nicht __acc] und cheese.nom/acc like I not and [ist  __nom auch nicht gut für mich]. is also not good for me Lit.: ‘Cheese I don’t like and is also not good for me.’  (Standard German, te Velde 2005: 229f.) So far, the tendency seems fairly clear. Mismatches in ATB movement are tolerated only if there is a syncretic/underspecified form that is compatible with the conflicting requirements of both verbs.

4 Surprisingly, the very same example with kogo is given as acceptable in Kluck (2009: 150), who cites personal communication with Barbara Citko. 5 Interestingly, in right-node-raising, another sharing construction, even non-syncretic case mismatches have been claimed to be acceptable in German as long as the filler matches the requirements of the adjacent verb, see Citko (2008: 24): (i) ?Marie vertrauteDAT __ und Marie trusted and Johannes kannteACC [den Mann]/* [dem Mann] Johannes knew the.acc man the.dat man ‘Marie trusted, and Johannes knew, the man.’ We have no indication that such mismatches are possible in ATB movement. The empirical verification of this contrast will be subject to future research.

54 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

However, morphological identity is not sufficient to restrict ATB movement. Consider the following example (Dyla 1984: 704): (7)

*Dziewczyna, której [Janek dał swoją marynarkę __dat]  girl       who.dat John gave refl jacket  a [mimo tego __dat było zimno …]  and in.spite it was cold  ‘The girl who John gave his jacket and in spite of it was cold.’ (Polish)

The example is ungrammatical although both gaps are assigned the same morphological case, viz. dative. Franks (1995: 64–77) argues (against Dylas’ analysis requiring identity in both abstract and morphological case) that next to morphological identity ATB movement is subject to an additional requirement, viz. that arguments must be identical with respect to thematic prominence. He proposes the following constraint (Franks 1995: 67): (8)  I n any ATB construction, the gaps must pertain either to most prominent or to not most prominent arguments, consistently across the conjuncts. This accounts for the grammaticality of mismatches like (4) and (6), where the gaps are not most prominent in both conjuncts (note that the experiencer in (6) is taken to be higher on the thematic hierarchy than the theme). It also correctly rules out (7) because the first gap is not most prominent whereas the second one is most prominent (as it is the only argument). Furthermore, it can also account for mismatches in abstract case in English where a matrix direct object is combined with an embedded subject (Munn 1993: 43 and Williams 1978: 34): (9) a. Who did [John support __] and [Mary say __ would win]? b. I know the man who [John likes __] and [we hope __ will win]. The subject in the second clause does not count as most prominent because of the matrix subject. The importance of thematic relationships is also seen in the following pair involving an experiencer verb (Franks 1995: 76): (10) a. the boy who [frightened Sue __] and [she hit __] b. the boy who [__ frightened Sue] and [__ hit her]

(−agentive) (+agentive)

Verbs with accusative experiencers usually allow for both a non-agentive reading where the nominative subject functions as a source and is generated as an internal argument (e.g., the mere presence of the boy frightens the girl) and an agentive reading where the subject is volitional (e.g., the boy frightens Mary actively and intentionally by pointing a gun at her) and is generated as an external a ­ rgument.



On the Limits of Non-Parallelism in ATB Movement 

 55

Crucially, (10a) only allows for the non-agentive reading whereas (10b) requires the agentive reading. This is in accordance with (8): The gaps are not most prominent in (10a) but most prominent in (10b). However, the literature contains a number of examples that are not ­compatible with Franks’ generalization (as he notes himself). They all involve combinations of a not-most prominent gap in the first conjunct with a most prominent gap in the second (see Franks 1995: 83, Munn 1993: 65, Goodall 1987: 75):6 (11) a.  ?a book which [I haven’t read__] but [__ was recommended by several professors] b.  ?the man who [John suspected __] but [__ hadn’t committed the crime] c.  We went to see a movie which [the critics praised __] but [__ was too violent for my taste]. What is important for the ensuing discussion is the generalization that mismatches in ATB movement are restricted by a morphological identity requirement and some sort of thematic compatibility requirement (even though the exact nature of this requirement is presently ill-understood, we include theta role as a factor in our experiments below to control for this). But mismatches in abstract case, differences in the internal structure of the conjuncts (types of verbs, differences in embedding) and gaps in non-parallel position are in principle tolerated. Since the previous literature on mismatches in ATB movement is only based on individuals’ judgments and since there is some disagreement about the possible mismatches, this chapter intends to explore the limits of

6 Another counterexample to Franks’ generalization is found in Dyla (1984: 701), where the combination of an accusative direct object gap with the gap of a nominative subject of an unaccusative experiencer (acc-nom) verb is judged ungrammatical despite the fact that both arguments are not most prominent and the filler bears a syncretic form. Furthermore, Bondaruk (2003: 236) shows that the combination of a dative experiencer argument with the sole genitive theme argument of an unaccusative verb leads to ungrammaticality even though both roles are most prominent. Still another counterexample is the following (from Munn 2001: 391, fn. 4), where a matrix subject gap is combined with an embedded subject gap: i. the man who [__ read the paper] and [Bob said __ understood it]. Franks suggests that it could be accommodated if the arguments of the matrix verb in the second conjunct only count optionally. This surely weakens the generalization. Bradley Larson (p.c.) suggested an alternative explanation to us: (i) may involve a parenthetical – Bob said – so that the gaps would actually both be most prominent.

56 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

non-parallelism in ATB movement in a systematic empirical fashion. We will focus on German in our experiment because it has morphological case but as opposed to Slavic languages is much less studied in this area. Our chapter is organized as follows: In section 2, we will provide some background information about previous approaches to ATB movement and their predictions with respect to case mismatches. Section 3 introduces two rating studies that investigate to what extent case mismatches are tolerated in German ATB movement and whether it is constrained by a syntactic or a morphological identity requirement. Section 4 reports a self-paced reading experiment based on the rating studies that aims at relating the results of the ratings to processing. Section 5 concludes the chapter.

2 Approaches to ATB movement and their predictions with respect to case mismatches The major current approaches to ATB movement can be largely divided into two groups: approaches where the filler bears a privileged relationship to the first conjunct, as in asymmetric extraction approaches, and sharing ­approaches, where the filler has a symmetrical relationship with both conjuncts. The ­sideward movement approach can be considered a compromise between the two as it involves copying and thus identity as well as asymmetric extraction.

2.1 Asymmetric extraction approaches Asymmetric extraction accounts all share the assumption that the filler originates in the first conjunct and that the gap in the second conjunct comes about in a different way. In the parasitic gap approach to ATB movement (Munn 1993; Franks 1995; Bošković & Franks 2000), coordinating conjunctions are functional heads that project a BP (= Boolean phrase); the BP is adjoined to the first conjunct. In ATB movement there is asymmetric extraction from the first conjunct while empty operator movement in the second conjunct leads to a parasitic gap. The two chains are then combined via chain formation: (12) Which book1 did [tp [tp John like __1] [BP Op2 [B’ and [tp Mary dislike __2]]]]?



On the Limits of Non-Parallelism in ATB Movement 

 57

As far as we can tell, the issue of case mismatches has not been addressed in these accounts; but it seems to us that since the two operators receive case from two different case probes, nothing in principle seems to rule out conflicting cases on the two operators. To rule out such mismatches the chain composition mechanism would have to be extended by an explicit constraint. Zhang (2010) proposes an account based on asymmetric extraction and variable binding: The filler extracts asymmetrically from the first conjunct and binds a pro-DP in the second conjunct (an instance of variable binding): (13) Which book1 did [&P [tp John like __1] and [TP Mary dislike pro-DP1]]? Nothing in this approach seems to require identity in case as there is no direct movement relationship; furthermore, variable binding allows for case mismatches, as in I told every studenti that hei should go. Another variant of this type of approach are analyses based on asymmetric extraction from the first conjunct plus deletion in the second (see George 1980, Wilder 1994, te Velde 2005, An 2006, Salzmann 2012a, 2012b). The constituents of the second conjunct are deleted under identity with the extracted constituents in the first conjunct (i.e., they undergo some form of ellipsis); in Salzmann (2012a: 408), this is executed as follows (strikethrough = regular PF-deletion, angled brackets = deletion under identity): (14) [cp[Which book]1 did3 [&p [tp John did3 [vP [which book]1 like [which book]1]] & [tp Mary [vP dislike [which book]2 ]]]]? The extracted wh-phrase binds into the second conjunct at LF: (15) [cp [Which x] [&p [tp John did [vp like [x book]]] & [tp Mary did [vp dislike [x book]]]]]? Since there is no direct movement relationship between the extractee and the gap in the second conjunct, nothing in principle rules out mismatches in case – to the extent that they are acceptable in ellipsis (which is to be expected given that ellipsis is famous for tolerating morphological mismatches and mismatches between pronouns and R-expressions, see Fiengo & May 1994).

2.2 The sideward movement approach (Nunes 2004) In this approach the filler is merged in the second conjunct. It is then copied to an independent phrase marker (i) from which the first conjunct is built.

58 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

After both conjuncts are complete, they are merged under &. Then there is ­asymmetric extraction from the first conjunct (ii). At PF, the extracted operator forms a chain both with the copy in the first conjunct and the one in the second. Chain reduction leads to the deletion of the lower copies in both conjuncts and thus derives the illusion of simultaneous movement from both conjuncts: (16) a. [Mary dislike [which book1]] b. [like [which book1]]                  (i) (17) Which book1 did [&p [John like which book1] and [Mary dislike which book1]]?   (ii) Whether case mismatches are possible depends on the precise implementation of sideward movement. If, as in the original version that is cast in the checking theory of early minimalism, DPs have pre-specified case values that need to be checked, the two DPs will necessarily bear the same case value since copying is involved. As a consequence, no mismatches, not even in abstract case, are expected; the issue is briefly mentioned in Nunes (2004: 176, fn. 12) where he seems to suggest an account in terms of underspecification, but how that interacts with chain formation remains unclear. If, on the other hand, case checking involves assignment of case values under Agree, the two DPs can differ in case features when they are affected by two different case probes; for this to be possible, sideward movement must take place before Agree. Consequently, case mismatches seem possible (but such derivations may crash since conflicting case values might prevent chain formation and chain reduction – the copies may no longer count as non-distinct).

2.3 Multi-dimensional/sharing-approaches Such approaches have been proposed in various guises, see Williams (1978), Goodall (1987), Moltmann (1992), Citko (2005); they are the default in HPSG, cf. Pollard & Sag (1994), Levine et al. (2001). For simplicity’s sake, we will only discuss Citko (2005), but as far as we can tell, our conclusions carry over to other approaches. In her proposal, constituents can undergo Parallel Merge so that they are dominated by two mothers. For reasons of linearization, such shared constituents have to undergo movement. In ATB movement, this leads to one chain with one deleted copy.

On the Limits of Non-Parallelism in ATB Movement 



 59

CP

(18)

[which book]

C'

did

&P

TP John like

and VP

TP

Mary

[which book]

VP dislike

Since the ATB-moved DP is present only once in the structure, we do not expect case mismatches. However, Citko (2005: 487) claims that mismatches in case, i.e., the assignment of conflicting case values to a single DP, are tolerated as long as there is a syncretic (i.e., underspecified) morphological form as in (4) above that is compatible with both requirements. This implies a late-insertion approach to morphology and a morphological identity requirement (but see Hein & Murphy 2016 for critical assessment).7, 8

2.4 Interim summary Simplifying somewhat, current approaches to ATB movement either predict ­mismatches to be readily available or to be very restricted. Since according to the literature only syncretic mismatches are allowed, the facts tend to support sharing approaches. But since these claims have not been verified empirically, one should not rush to conclusions. In the following two sections, we will report three experiments that put these claims under close scrutiny and explore the consequences of the results for theories of ATB movement.

7 Note that different implementations of multidominance may allow for certain mismatches, see Moltmann (1992: 107ff.) for the domain of reconstruction. 8 In HPSG the nonlocal feature principle (Pollard & Sag 1994: 164) allows for the percolation of slash specifications from more than one daughter; they merge by set union. This predicts strict identity between fillers and gaps as slash specifications can be unified only if they are identical. As a consequence, case mismatches are unexpected. Mismatches with syncretic forms have been addressed in Levine et al. (2001) by revising the case type hierarchy: a syncretic form is both nom and acc so that it can satisfy the unification of the requirements of both gaps, i.e., of being both nominative and accusative.

60 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

3 Rating studies 3.1 Introduction As shown in the first section, the literature suggests that case mismatches are allowed in ATB movement if the shared antecedent bears a syncretic/ underspecified case form. We will therefore investigate the following hypothesis in our experiments: (19) Hypothesis: ATB movement allows for case mismatches with syncretic forms only. Before we introduce the experiments, we need to address a methodological difficulty that obtains when trying to construct an experiment that investigates the role of syncretic mismatches in German ATB movement: When we look at inflectional paradigms in the nominal domain in German, we find the most systematic syncretism between nominative and accusative case with singular feminine and neuter nouns and all nouns in the plural. It thus seems natural to capitalize on this syncretism. However, this requires some care: If we combine two transitive verbs in an ATB construction with the gaps in different case positions, the gaps will usually also differ in thematic role, i.e., agent vs. theme. Since compatibility of thematic role was shown to be important for the acceptability of mismatches, see section 1, transitive verbs are not the ideal test case for German. It differs from Slavic languages, where we find syncretism of cases that can be assigned to internal arguments that bear the same thematic role, i.e., theme (as in (4)). To control for the influence of theta roles, our experiments make heavy use of experiencer verbs because their arguments (Exp, Th) can occur with nominative or accusative.9

9 There are two other syncretisms that one could, in principle, have tested, viz. genitive-dative singular feminine and accusative-dative with bare plurals. We did not rely on the dative-genitive syncretism because genitive is only assigned by very few verbs in contemporary German (essentially just two or three) and is usually perceived as stylistically marked; this makes it unsuitable for experimental testing. We also refrained from using accusative-dative syncretisms of bare plurals because when used with transitive verbs, these will usually involve a mismatch in theta-role (theme-recipient). Gisbert Fanselow suggested that we test accusative-dative mismatches with dative- and accusative-experiencer verbs. Even though dative-experiencer verbs are rare, we consider investigating the effects of this syncretism in future research.



On the Limits of Non-Parallelism in ATB Movement 

 61

Experiencers can be nominative (as in psych-verbs I) or accusative (as in psych-verbs II). If we combine the two types of experiencer verbs, we can construct examples with a syncretic case mismatch without a mismatch in theta role (this is the basis for the first rating study): Nom/Exp Acc/Th e.g., (20) a. psych-V I: b. psych-V II: Acc/Exp Nom/Th e.g.,

like, hate interest, surprise, worry, irritate10

Similarly, the theme can associate with both accusative (with transitive verbs and with psych-verbs I) and nominative (with psych-verbs II). Again, no mismatch in theta role obtains when we combine the two verb types (this is the basis for the second rating study). (21) a. transitive verb: b. psych-V II:

Nom/Ag Acc/Exp

Acc/Th Nom/Th

We have carried out two rating experiments that are based on the same design. They differ in the case and thematic role of the extractee: While it is a nominative experiencer subject in the first study, it is an accusative theme object in the second. We decided to test both because there is reason to believe that whichever choice we make (nominative/accusative) may lead to lower acceptability for independent reasons: First, subject gaps in the first conjunct are less acceptable with mismatches (cf. Munn 1993: 43f.). Second, accusative/object gaps may lead to markedness since we employ topicalization in our ­experiments (a choice we will justify below). While subject topicalization is neutral, object topicalization is information-structurally marked and may therefore affect

10 We follow Belletti & Rizzi (1988) in treating such verbs as unaccusatives with acc>nom base order. The ordering facts discussed in den Besten (1985) provide independent support for this in languages like German. Note that we thus differ from Landau (2010), where non-stative verbs of this class are treated as transitive with the experiencer being base-generated as the (oblique) object and the nominative, which is treated as a causer, as the external argument. Our items include both statives and non-statives; in the latter case, our items are designed such that an agentive interpretation is unlikely.

62 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

acceptability, see Fanselow et al. (2008). By testing both options, we intend to make sure that these effects are neutralized. The following table gives an overview of both rating studies: Table 1: Overview rating experiments Material extractee

gap1

Conditions gap2

Experiment 1

SUNom/Acc SUNom

syn(cretic)

tNom

Experiment 1

DONom/Acc

DOAcc

tAcc

tNom

SUNom

DOAcc

Case form

diff(erent)

same

syn(cretic) tNom tAcc

diff(erent)

tAcc

same

Theta role +θ match

A

–θ match

B

+θ match

C

–θ match

D

+θ match

E

–θ match

F

+θ match

A

–θ match

B

+θ match

C

–θ match

D

+θ match

E

–θ match

F

In both experiments, we have three levels pertaining to form. In two levels, there is a case mismatch between the two gaps.11 In one, and this is the one we are particularly interested in, the filler is syncretic (in bold-face); in the other, it is unambiguously specified for case. In the third level, the filler is clearly specified for case and matches the requirements of both conjuncts. All three levels are tested with matching and non-matching theta roles.

11 Note that the mismatch always obtains in the second conjunct while there never is a mismatch in the first conjunct. We refrained from testing the reverse configuration because it has been shown that configurations where the filler is not compatible with the requirements of the adjacent/closest conjunct are strongly unacceptable; see, e.g., Vogel et al. (2006: 379) for an experiment with coordinated verbs assigning conflicting cases to one single DP; see also Kluck (2009) for similar observations for Right Node Raising.



On the Limits of Non-Parallelism in ATB Movement 

 63

3.2 Experiment 1: Rating study 1 3.2.1 Factors and conditions We investigated the following two factors: (22) Factor 1: Case form: syncretic – different – same Factor 2: Theta role: match – mismatch The factor case form defines the form of the initial DP (the extractee) and its relation to the two gaps. It has three levels: (a) the initial DP bears a syncretic case form and is linked to gaps that have conflicting case-requirements (syncretic); (b) it bears a non-syncretic case-form and is linked to gaps that have conflicting case-requirements (different); (c) it bears a non-syncretic form that is compatible with the case requirements of both gaps (same). The factor theta role is included to control for the influence of the theta role. Crossing the two factors results in the six conditions illustrated in Table 1 above. 3.2.2 Materials We created 24 lexical variants and distributed the test sentences across six lists in a Latin Square design. Additionally, the experiment contained 80 filler sentences. Importantly, we used topicalization instead of wh-movement as ­ extraction type so that we could include comprehension questions after each item to ensure that participants read and process the sentences correctly. We used TP coordination with ‘and’.12 The extractee was a nominative experiencer that matched the first gap. To vary mismatch in case and theta roles we used three different types of verbs altogether: (i) psych-verbs I with a nominative experiencer and an accusative theme (first gap in all conditions, second gap in condition E); (ii) psych-verbs II with accusative experiencer and nominative theme (second gap in condition A, C, F); and (iii) regular transitive verbs with nominative agent and accusative theme (second gap in condition B, D). The material is schematically represented in the following table (gaps are highlighted with gray, underline indicates mismatch).

12 Note that C'-coordination as in (6) may actually involve coordination of two full CPs with the subject in the second CP undergoing topic drop. TP-coordination in our experiments rules out this alternative analysis.

64 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

Table 2: Materials experiment 1 has

Arg1

Arg2

VERB psych-V I

&

Arg1

Arg2

VERB psych-V II

NP

—

NP

—

NP

nom/acc

nom

acc

acc

nom

exp

theme

exp

theme

NP

—

NP

NP

—

nom/acc

nom

acc

nom

acc

exp

theme

agent

theme

NP

—

NP

—

NP

nom

nom

acc

acc

nom

exp

theme

exp

theme

NP

—

NP

NP

—

nom

nom

acc

nom

acc

exp

theme

agent

theme

NP

—

NP

—

NP

nom

acc

nom

acc

exp

theme

exp

theme

psych-V I

psych-V I

psych-V I

samematch

psych-V I

nom

samemismatch

diffmismatch

diffmatch

synmismatch

synmatch

NP

NP

—

NP

NP

—

nom

nom

acc

acc

nom

exp

theme

exp

theme

psych-V I

trans. V

psych-V II

trans. V

psych-V I

psych-V II

A sample item illustrating all conditions is given in (23): (23) A. syn-match: Diese Athletin hat __ den W. respektiert und __ der  R. beunruhigt. This athleteNom/Acc has theAcc   W. respected and __ theNom R. worried Lit.: ‘This female athlete respected Werner and Robert worried.’ B. syn-mismatch: Diese Athletin hat __ den W. respektiert und der  R. __ bestochen. This athleteNom/Acc has theAcc  W. respected and theNom R.  bribed Lit.: ‘This female athlete respected Hans and Robert bribed.’ C.  diff-match: Dieser Athlet hat __ den W. respektiert und __ der R. beunruhigt. This athleteNom has theAcc W. respected and theNom R. worried Lit.: ‘This male athlete respected Werner and Robert worried.’

On the Limits of Non-Parallelism in ATB Movement 



 65

D. diff-mismatch: Dieser Athlet hat __ den W. respektiert und der R. __ bestochen. This athleteNom has theAcc W. respected and theNom R.    bribed Lit.: ‘This male athlete respected Werner and Robert bribed.’ E. same-match: Dieser Athlet hat __ den W. respektiert und __ den R. herbeigesehnt. This athleteNom has theAcc W. respected and theAcc R. longed-for ‘This male athlete respected Werner and longed for Robert.’ F.  same-mismatch: Dieser Athlet hat __ den W. respektiert und den R. __ beunruhigt. This athleteNom has theAcc W. respected and theAcc  R.   worried ‘This male athlete respected Werner and worried Robert.’ The TP coordination structure we assume is illustrated in the following simplified tree structure for the sample item in condition A (syn-match): (24) Tree structure for sample item in condition A (syn-match) CP C'

[this fem. athlete]Nom/Acc C has

&P TP

&

TP

VP ____Nom

the W.Acc

VP V'

____Acc

V'

respected the R.Nom V worried

Thus, we coordinate constituents of the same syntactic category (and of the same semantic type). The internal structure of the two conjuncts differs somewhat, but this is usually unproblematic for coordination (e.g., He laughed and went to the bathroom).13 Since we assume an unaccusative analysis for accusative-experiencer verbs (i.e., type II), the syn-match condition (and also the different-match condition) involves extraction from two structurally (almost fully) parallel gaps

13 According to Frazier et al. (2000), conjuncts with identical structure are processed more quickly, but this does not affect the acceptability of coordinations involving conjuncts that differ in internal structure.

66 

 Jutta Hartmann, Andreas Konietzko and Martin Salzmann

(note that the nominative experiencer originates in Spec,vP while the accusative experiencer originates in Spec, VP; we abstract away from this difference in the tree diagrams). The extraction is certainly parallel in the sense that in each case the structurally highest argument is extracted. Assuming, as is standard, that structural prominence corresponds to thematic prominence, these extractions also satisfy Franks’ constraint in (8): the extractee is the most prominent argument in both conjuncts. In the examples with theta-mismatch, the gaps are located in structurally non-parallel positions (external vs. internal argument or highest vs. non-highest gap/theta role). 3.2.3 Method and procedure We used the method of thermometer judgments (see Featherston 2008) with the help of the WebExp2 software (Keller et al. 1998; Keller et al. 2009); the experiment was carried out under supervision in the Tübingen computer lab. 36 participants from the Tübingen area (all non-linguists, mostly students) rated each sentence and answered a control question after each sentence. All participants answered more than 70% of the control questions correctly. The average accuracy of the 36 participants was 91% (on average 21.4 out of 24 questions). 3.2.4 Results experiment 1 Figure 1 displays the normalized mean ratings of the F1 analysis. While the conditions E/F (form: same) that involve case-matching received comparably high ratings, all the other conditions received much lower ratings.

Figure 1: Normalized mean ratings (z-scores) per condition in experiment 1 (error bars indicate standard errors)



On the Limits of Non-Parallelism in ATB Movement 

 67

The normalized ratings were analyzed with repeated measures ANOVAs. The ­statistical analysis revealed a significant main effect for the factor case form: F1(2,70) = 145, p < 0.001; F2(2,46) = 269, p < 0.001. There is no significant main effect for the factor theta role (F1, F2 < 1) and no interaction of the two factors F1(2,70) = 2.8, p = 0.07; F2(2,46) = 2.1, p = 0.13. We specified two orthogonal contrasts known as reverse Helmert contrasts or difference contrasts, which compare each level of a factor with the mean of the previous levels of the factor; hence we compared, in a first step, morphologically different case forms (diff) with syncretic forms (syn), and, in a second step, we tested averaged different and syncretic forms against case identical forms (same). This reveals a highly significant contrast between identical forms (same) vs. the two other levels (syn,diff): F1(1,35) = 209, p < 0.001; F2(1,23) = 450, p Dat orders is not surprising, as we would actually expect more license in a poetic text. In fact, all instances of Acc>Dat order

Relative Object Order in High and Low German 



 147

produce a rhyme; turning to Dat>Acc would destroy the rhyme. A comparison with another, later poem, the Nibelungenlied (Nibelungenlied 2005), is instructive in that respect. In Speyer (2013) I tried to identify the rate of Acc>Dat in several MHG prose texts and did the same counts also for a poetic text, just for comparison reasons. There, the rate of Acc>Dat is 20%, rhyme is a motivating factor in all instances of Acc>Dat.3 In comparison, in MHG prose as represented by Berthold von Regensburg’s sermons (Berthold v. Regensburg 1965), a negligible rate of 2% could be identified. So the state of affairs is presumably rather similar in OHG and in MHG. Turning again to Otfrid: if we look closer at the examples for Acc>Dat order, we find that they are exclusively governed by one verb geben ‘to give’. Moreover, they are instances of the verbal idiom antwurti geben ‘give answer to s.o.’ (3). So the Acc>Dat order here might be idiosyncratic rather than being a general option in the language (note that in Modern German, this idiom follows the usual Dat>Acc pattern). All other double object verbs in Otfrid show exclusively Dat>Acc order. (3)

Ér gab tho ántwurti \ then líutin mit giwúrti he gave then [answer].acc [the people].dat with gladness ‘He gladly gave the people the [following] answer’ (Otfrid Ev. 3,20,109)

2.3 Old Saxon In OS, that is, the medieval predecessor of Low German, the database is somewhat comparable to the Otfrid part of the OHG data. OS is mostly attested by two poetic texts, the Heliand and the Genesis (Heliand 1996). The Heliand, probably

3 This shows that Dat>Acc is the preferred order in that, if no poetic reasons speak against it, this order is produced. I do not think that it shows that Dat>Acc is the underlying order, although one could argue that this is a clear external criterion that can turn a putative Dat>Acc order to an Acc>Dat order. But consider another example of poetic license: in German poetry, verb final declarative or even imperative main clauses (VFMC) are not uncommon up to the 19th century. VFMCs went out of use in German prose during the MHG period; in fact, it is questionable whether verb final main clauses were an option at all (cf. Axel 2007) or an option serving a certain information structural purpose (cf. Schlachter 2012). An example (not from the 19th, but the 17th century) is the first two lines of Friedrich von Spee’s advent carol ‘O Heiland reiß die Himmel auf/ herab, herab vom Himmel lauf ’ (Spee 2007). VFMCs are usually formed for reasons of rhythm or rhyme. They are, however, not derived from underlying verb second or verb first clauses, but the other way round (following the usual account of, e.g., den Besten 1977): Verb second/first clauses are derived from underlying verb final clauses. So the poetic license is here not an extra derivation, but the abdication of an otherwise standard derivation. This is exactly the same case as the surfacing Acc>Dat orders in Otfrid.

148 

 Augustin Speyer

also the Genesis, was presumably composed before A.D. 840 in the Southwest of the Low German area, that is in Southern Westphalia, by a poet or poets that had relations to Fulda and Mainz (cf. Haubrichs 1999:303f.). Additionally, there are a few juridical prose texts, but these offer no examples of double object constructions involving full NP objects. All OS texts are available in the TITUS database. I used a list of verbs governing a dative and an accusative object, compiled from Sehrt (1966),4 and searched the TITUS database for all occurrences of these verbs, exporting the examples that matched the criteria outlined above (most crucially full NPs, position in the middle field). If we run this search, we see a completely different picture, as is apparent from Table 2. Table 2: Order of objects in OS texts

Order dat>acc Order acc>dat Total

Number

Ratio (%)

27 16 43

62,79 37.21

Although the numbers appear to be not that different from Otfrid, which is the suitable point of comparison, both texts being poetry, there is a clear difference: While in Otfrid only one verb, geban ‘to give’ allows for Acc>Dat order, and that predominantly in the verbal idiom geban antwurti ‘to give an answer’, in the Heliand several verbs allow for Acc>Dat, namely bifelhan ‘to hand over’, dernian ‘to conceal’, geƀan ‘to give’, gilestian ‘to make s.th. for s.o.’, makon ‘to do s.th. to s.o.; to add’, marian ‘to announce’, seggian ‘to tell’, uuirkian ‘to do s.th. to s.o.’ (4a). There is no semantic pattern: seggian constructs both with Dat>Acc and Acc>Dat (4b,c); fremmian has the same meaning as gilestian, but it is attested with Dat>Acc order. So we see on the whole greater freedom in object order in the OS of the Heliand than in OHG. (4) a. than thu mid thînum handun bifeleas | then you with your hands hand over thîna alamosna themu armon manne [your alms].acc [the poor man].dat ‘then you give alms to the poor man with your hands’ (Heliand 18, 1555)

4 The list was compiled by my research assistant Leonie Zitzmann.



Relative Object Order in High and Low German 

 149

b. sagde uuord manag | firiho barnum said [words many].acc mens’ children.dat ‘he said many words to the children of men’ (Heliand 51,4230) c. thar uualdand Crist | […] Iudeo liudeon | […] there ruling Christ Jews’ people.dat lêra sagde teaching.acc said ‘there Christ the ruler said his teachings to the Jewish people’ (Heliand 24, 2080)

3 The thematic hierarchy as the main factor for argument order in High German 3.1 The concept of the thematic hierarchy The OHG data, and, as a matter of fact, also the MHG and ENHG data (see below, Table 5), suggests a strong association between Dat>Acc order and the animacy condition, in that there are no examples that do not follow the presumably rigid Dat>Acc order and at the same time violate the animacy condition. But what kind of an association is it? Primus (2004; 2011) suggests that in many languages word order is influenced by the thematic hierarchy. Following Dowty (1991), thematic roles are conceptualized as points on a scale of agentivity. Depending on what agent-like or patient-like properties the participants have, they can be ordered on the scale, the high-point of which is the maximum of agent-like properties (proto-agent), the low-point of which is the maximum of patient-like properties (proto-patient). Primus (2011) proposes a system of linking between the Thematic Hierarchy and a case hierarchy, which in accusative languages like German has the form Nom>Acc>Dat. The participant highest on the thematic hierarchy links with the highest case on the case hierarchy. The second highest case, the accusative, does not link with the next highest element on the thematic hierarchy, but with the other extreme, the lowest element on the thematic hierarchy, while participants that lie in-between the two extreme points link with whatever case is left on the case hierarchy, in this case the dative as lowest case on the case hierarchy. If in a language the word order is governed by the thematic hierarchy, and if the case linking works in the way Primus (2011) envisages, the order Nom>Dat>Acc in trivalent verbs will arise automatically as a direct result of the linking ­mechanism.

150 

 Augustin Speyer

Similarly, the animacy condition falls out for free. Animacy is not a sufficient condition on agenthood, it is, strictly speaking, not even a necessary condition, but it is a typical property of agenthood nevertheless, because it is a necessary condition on agent-like properties such as control of situation or physical activity. This is not to say that patients are typically inanimate, but lack of animacy is no precondition for any prototypical patient-like property. So, if the participants in a verbal event are ordered according to the thematic hierarchy, chances are high that in the end animated participants are positioned before inanimate participants in the sentence. And what also falls out for free is the intrinsic connection between object order and animacy. If both the object condition and the animacy condition emerge automatically from an ordering following the thematic hierarchy, we may conclude that the object order in OHG, MHG, and early ENHG was governed by the thematic hierarchy, which in turn was closely connected to the case hierarchy and followed the linking mechanism described by Primus (2011).

3.2 Further evidence for the thematic hierarchy: Acc-Acc-verbs If this was true, we would expect the thematic hierarchy to affect object order also independently of case. To test this, one could look at verbs that govern two objects bearing the same case, such as Modern German lehren ‘to teach’, which governs two accusative objects. The OHG version of this verb, lêren ‘to teach’ is not attested with two full NP objects in the middle field. Things are different in MHG. Here, we find quite a number of attestations (Table 3).5 Table 3: Order of double accusative objects according to animateness in MHG

Order + anim > − anim Order − anim > + anim Total

Number

Ratio (%)

47 3 50

94     6

There is a very strong bias toward an ordering of more animated referents before less animated referents. So we see here the thematic hierarchy in the guise of the animacy condition at work.

5 The data was gathered using the Mittelhochdeutsche Begriffsdatenbank (Springeth, Schmidt & Pütz 1992).



Relative Object Order in High and Low German 

 151

3.3 The thematic hierarchy as motivator for short scrambling A question that has to be posed in this context is what the status of this word order is. As we will see in section 4, there is evidence that the order Dat>Acc that conforms to the thematic hierarchy is not the base-generated order. This has certain consequences. First, the thematic hierarchy does not determine the basic word order and by that the structure of the VP, but instead serves as a motivating force for a short scrambling operation. The base order (which is Acc>Dat) thus is not determined by thematic roles, as commonly assumed (see the references in Jackendoff 1990:46ff., who argues against this view), but has, at least in the stage of development represented by OHG, a life of its own, so to speak, being independent from semantic concepts. Second, the thematic hierarchy is such a strong motivator that short scrambling takes place most of the time, leading to a situation in which the base order is mostly obliterated. This leads to the question why no language change takes place that results in an inversion of the base order. In a situation as outlined just now, we would expect language learners to interpret the data (consisting mostly of Dat>Acc structures) such that Dat>Acc is the base order, and the few instances of Acc>Dat are the result of a moving operation, because this is the easiest representation that fits the data (cf. Clark & Roberts 1993). But then there would be the problem what would motivate the Acc>Dat order since it is obviously not correlated to a semantic property. We may assume that in such a situation – two orders A>B and B>A, A>B common and correlated to a semantic concept, B>A rare and chaotic – the language learner would opt for the derivation of A>B out of B>A via the semantic concept correlated with it and put up with assigning base status to the rarer variant, because it would be impossible to find a generalization under which A>B should be scrambled to B>A. For A>B, on the other hand, there is the thematic hierarchy as a motivating concept, playing a role in other parts of the language faculty anyway. Another point is that the order Dat>Acc is violated more often if two pronouns are involved (in Modern standard German, Acc>Dat is the usual order, see Hoberg 1997, in dialects and in earlier stages of German, Acc>Dat is at least an option, see, e.g., Fleischer 2005), so this variant is not so rare, after all.

3.4 The mismatch between the thematic hierarchy and the base order What remains to be explained is why the semantic structure does not match the syntactic structure here. This is a problem that essentially goes back to the era

152 

 Augustin Speyer

of Proto-Indoeuropean, if we assume that the base order Acc>Dat originally dates from that time. As a matter of fact, it is this property that might offer a cue to this problem. A double object verb basically has two possible lexical-semantic representations, depending on what kind of verb is in the inner VP shell (adopting Larson’s (1988) analysis of double object verbs). For a verb like to give, there are basically two options (5; see also Rappaport & Levin 1988). (5) a. [[to give]]1 = λzλyλx.CAUSE (x, BECOME (HAVE y,z))

b. [[to give]]2 = λyλzλx.CAUSE (x, BECOME (BE-AT z,y))

I chose the variables in (5) such that x is the highest on the thematic hierarchy and z the lowest. In the version give² (5b), the linking mechanism of Primus (2004; 2011) would produce a base order Acc>Dat, as the innermost argument is the goal argument that is in this case higher on the Thematic hierarchy and thus is assigned dative case. Consequently, we might assume that Proto-Indoeuropean standardly constructed double object verbs of the semantic type of to give (most double object verbs belong to this type) in the version give². Is there evidence for that? There is, albeit only of an indirect sort. Proto-Indoeuropean presumably had no verb of possession like the modern English to have, German haben, Latin habere / possidere etc. At least it is not possible to reconstruct a verb with that meaning for all of Indo-European. The verbs meaning to have in the daughter languages of Proto-Indoeuropean are all products of semantic change of verbs which originally meant something like to grasp (like Germanic *haben, predecessor of German haben and English have, which is related to Latin capere ‘catch’) or to win (like Ancient Greek échein, that is related to German siegen ‘to win’). What we can reconstruct as an expression denoting possession is a construction in which the possessor is in the dative, the possessum in the nominative, and the verb is the copula, as in (6).6 (6) magnum hoc vitium vino ‘st big that fault wine.dat is ‘the wine has this big fault.’ (Plautus, Pseud. 1250, cited from Hofmann & Szantyr 1965:90) So we might assume that Proto-Indoeuropean had no possession verb of a setup like HAVE (y,z) available, but rather a possessive construction like

6 A small clause analysis of possession (e.g., Uriagereka 2002) thus was not available in Proto-­ Indoeuropean; there was only a structure such as, e.g., Larson & Cho (2003) proposed that ­surfaced as a construction with nomninative and oblique case (here the dative).



Relative Object Order in High and Low German 

 153

­ E-IN-POSSESSION-OF (z,y). Consequently, it is this predicate that was embedB ded in double object constructions of the give-type, if the inner verb predicate is a predicate of possession. So the choice for a Proto-Indoeuropean speaker when expressing a double object verb of the give-type was actually not between the representations in (5), but those in (7). (7) a. [[to give]]3 = λyλzλx.CAUSE (x, BECOME (BE-I-POS-OF z,y)) b. [[to give]]2 = λyλzλx.CAUSE (x, BECOME (BE-AT z,y)) And whichever is chosen, we always get the more animate object as the internal argument of the inner predicate. So the base order Acc>Dat in Proto-Indoeuropean at least can be argued to be a direct consequence of this lexical-semantic setup. The mismatch between syntactic and semantic form appeared only later, in the daughter languages, once a transitive verb of possession was developed and consequently give 1 came to be an option. We might assume that this is a rather marked stage of language development, so we should not be surprised to find processes of syntactic or lexical change that restore the syntax-semantic match. In English, such a change obviously happened at some time, as here the base order is Recipient>Patient (Larson 1988). In German, this change has not (yet) happened.

3.5 The role of the thematic hierarchy in Old Saxon If German object order at least from OHG to early ENHG times followed the thematic hierarchy, and if we have seen that the impact of this factor changes in later ENHG, we might ask whether the crucial role of the thematic hierarchy as a factor governing object order came about by some change as well, and if so, when. By investigating other closely related languages we can attempt an answer to this question. Let us go back to the OS data (section 2.2, Table 5). We might ask what the reason is for the relatively high frequency of Acc>Dat order in OS compared to the OHG of Otfrid. One might assume, judging from the OHG data, that the strong link between the thematic hierarchy, of which the animacy condition is a symptom, and the case hierarchy, which is visible as the object condition, is perhaps not as prominent in OS. On this assumption, two patterns of data distribution are possible: either the OS examples are ordered following the thematic hierarchy about as restrictively as is the case in OHG. If this were so, we could say that the thematic hierarchy plays a role for constituent order, but that the linking to the case hierarchy is not as developed as in OHG. Or the OS data shows violations of the animacy condition, and by that the underlying thematic hierarchy, to a considerable extent. In this case, we could say that the thematic hierarchy is not a relevant factor for constituent order at all. Table 4 shows the actual distribution in the OS data.

154 

 Augustin Speyer

Table 4: Animateness and case as factors in object order in OS

+ anim. > − anim. − anim. > + anim. sum Ratio of + anim. > − anim. (%)

Dat>Acc

Acc>Dat

    27 0     27 100

   1 15 16 6.25

Animateness, and underlying that the thematic hierarchy, are strongly asso­ ciated with case assignment, in that less animate referents are assigned ­accusative case, while more animate referents are assigned dative case. This information can be drawn from the numbers in Table 4: All cases of +anim > − anim are at the same time cases of Dat>Acc, and almost all of −anim > +anim are cases of Acc>Dat. The only counterexample is (8). Animateness detached from case, however, does not play a crucial role for the constituent order. There are numerous examples in which a less animated object precedes a more animated object (9). (8) Thar sia that godes barn | te iro landuuîsu, lîco hêlgost, | There they [that God’s child].acc to their custom body holiest foldu bifulhun field.dat entrust ‘There they entrusted God’s child, the holiest body, to the field, according to their custom.’ (Heliand 68, 5740) (9) uuelda heƀenrîki | liudiun gilêstean would heaven-kingdom.acc people.dat fulfil ‘He wanted to attain the heavenly kingdom for the people.’ (Heliand 13, 1052) If we try to speculate on what the governing factor in the ordering is, we see quite easily that it is the alliterating rhyme in which the poem is written. In (9), liudiun is in the second half-line simply because it produces an alliterating rhyme with gilêstean (note that for alliterating rhyme prefixes are irrelevant; it is the first segment of the stem that counts); likewise, in (8) foldu is the last constituent of the clause so it can stand in the same half line as bifulhun, producing alliteration. The fact that alliterating rhyme is a relevant factor for object order in the Heliand shows one thing: The relative order of dative and accusative objects was relatively free in general so the poet could utilize this to produce ­alliteration



Relative Object Order in High and Low German 

 155

rhymes as he wished. If the object order had been fixed in OS, the poet of the Heliand would not have been able to utilize word order variation, but rather would have been reduced to choosing some other verb that would produce an alliterating rhyme with the accusative object. For a capable poet, this should not cause any trouble. Apart from that, there are several examples in which we get an order Acc>Dat, although an order Dat>Acc would not have deteriorated the (often rather bad) quality of the alliterating rhyme (10). (10) thie ira alamosnie armun mannun | gerno gâƀun who [their alms].acc [poor men].dat gladly give ‘who gave their alms gladly to poor men’ (Heliand 15, 1226) The order in (9) could be attributed to a condition “definite before indefinite” that plays an important role in Modern German. In Old Saxon, the data does not show that it had a decisive role because there are quite a number of c­ ounterexamples.7 So we can confidently say that Acc>Dat was a possible word order option in OS. The animacy condition was not a factor governing word order, neither were factors such as givenness or definiteness relevant. If this is so, OS is already one step further than OHG, as argument order is somewhat emancipated from the thematic hierarchy. This step was taken by German only around 1500 AD when pragmatic factors began to oust the thematic hierarchy as the dominant factor affecting argument order.

4 Arguments for a base order accusative before dative 4.1 The Middle High German evidence As hinted at in Speyer (2015), the evidence of the older stages of German points more in the direction that the base order was accusative before dative. The argument that can be derived from that runs as follows:

7 The role of the definiteness condition (definite > indefinite) in older High and Low German is currently under investigation. Although definiteness is not yet constantly marked by an article, NPs are definite or indefinite insemantic terms, so this condition could in principle play a role in older stages of Germanic.

156 

 Augustin Speyer

In MHG and ENHG before 1500, the data is strongly biased toward Dat>Acc, but it is not as exclusively Dat>Acc as it appears to be the case in OHG. That means we have some Acc>Dat orders that can be interpreted as orders that do not conform to the thematic hierarchy. If we look at the numbers (Table 5), we see the strong link between Dat>Acc order and animacy as established by the thematic hierarchy at work: There are no examples that are at the same time Dat>Acc and violate the animacy condition. On the other hand, there are no examples of the Acc>Dat order that conform to the animacy condition.8 Table 5: Animated (a) and inanimate (i) referents in MHG and ENHG prose texts Dat > Acc a>i Berthold Lancelot ENHG sermons ENHG narrative total

69 27     4 24 124

Acc > Dat

i>a

a>a

i>i

a>i

i>a

a>a

i>i

– – – – –

17 7 2 6 32

8 – 1 – 9

– – – – –

2 3 3 7 15

– 4 1 1 6

– – 1 – 1

This can be interpreted in such a way that we can identify a factor that derives Dat>Acc, viz. the relative agentivity in line with the thematic hierarchy in the guise of the animacy constraint, but we cannot identify a factor that derives Acc>Dat. Other factors have been tested, and none of them derives Acc>Dat as neatly as animacy does in the case of Dat>Acc. This can be demonstrated, e.g., by the numbers of instances for the order following the given>new condition (Table 6). Table 6: Given (g) and new (n) referents in MHG and ENHG prose texts Dat > Acc

Berthold Lancelot ENHG sermons ENHG narrative total

Acc > Dat

g>n

n>g

g>g

n>n

g>n

n>g

g>g

n>n

29     9     1 16 55

13     2     –     – 15

30 16     4     9 59

16     4     2     4 26

    2     1     1     1     5

    –     2     –     3     5

    –     3     2     3     8

    –     1     2     1     4

8 As a reviewer pointed out, it is instructive to record also the ratios of matching to non-matching of the factor pairings. This is 6.5: 1 in animacy + first position, 7.5:1 in case + first condition, and 17:1 in case + animacy. This shows that the association between case and animacy is strongest.



Relative Object Order in High and Low German 

 157

Here we see that there is some effect of givenness, but we do not see that the given>new condition can override the object condition: The order Acc>Dat is not consequently or even preferentially associated with given>new. With animacy, it is different, as we have seen. As we can find no factor that derives Acc>Dat, we are forced to conclude that this is the base order. This, however, means, if cast in terms of hierarchical structure, that the accusative should be higher in the tree than the dative.

4.2 Wide focus and small scrambling Can we find other arguments in favor of this? Before we do that, we have to establish what kinds of arguments can provide convincing evidence. One line to follow would be to look for what is possible, acceptable, or more common in language usage. But there is a hazardous pitfall looming. It is not immediately clear that what we observe here is really the basic constituent order; it could be rather the unmarked order. Already Höhle (1982) made the point that we have to distinguish between the base generated order and what he calls the stylistically unmarked order. As a matter of fact, following this line of argumentation would almost unanimously point in the direction that Dat>Acc is the more basic order. Lenerz’ (1977) conditions on the testing of the unmarked order led him to identify it with Dat>Acc. He actually even demonstrated that Acc>Dat is not possible if it violates testable conditions such as given>new or definite>indefinite. But does that mean that Dat>Acc is the basic order? Following the same line of argumentation as above one could say that this is actually rather an argument for Acc>Dat as base order, as Dat>Acc could be derived by the givenness condition or the definiteness condition, whereas Acc>Dat could not. More of a challenge are data like that in (11), where we have identical definiteness status but arguably cannot derive Acc>Dat under preservation of wide focus (Helmut Weiß, p.c.). (11) a. Gestern hat sie einer Bekannten ein Kind gebracht. yesterday has she [an aquaintance].dat [a child].acc brought #Gestern b. hat sie ein Kind einer Bekannten gebracht. yesterday has she [a child].acc [an aquaintance].dat brought ‘Yesterday she brought a child to an acquaintance.’ c. Auf der Party möchte ich der Maria den Sepp vorstellen. on the party like I [the Mary].dat [the John].acc introduce #Auf der Party möchte ich den Sepp d. der Maria vorstellen. on the party like I [the John].acc [the Mary].dat introduce ‘I would like to introduce John to Mary at the party.’

158 

 Augustin Speyer

Problematic in the first place is that the wide focus test really applies to the unmarked order rather than the base order. One could, however, say that movement automatically leads to narrow focus. But: Is this the case? While this is a question that demands closer investigation, it would be good to have at least a rough idea of what is going on. To that end, I did a small questionnaire study among some B.A. students in Göttingen (n=43). They were at an early point of their courses and had not been exposed to questions of constituent order yet, so they can count as linguistically naïve.9 I read them 7 sentences (4 targets, 3 fillers), and they had to evaluate the naturalness of the sentences using a scale from 1 (completely natural) to 4 (completely unnatural). The parameters in the target sentences were verb (1 verb with natural Dat>Acc order: aushändigen ‘hand over s.th to s.o.’, 1 verb with natural Acc>Dat order: aussetzen ‘to subject s.o. to s.th.’) and object order (1 sentence with order Dat>Acc, 1 sentence with order Acc>Dat, respectively). All of the sentences were preceded by a small dialogue that followed the following pattern: participant A gives some context in which he mentioned all referents of the target sentence (thus producing an identical givenness status of all referents in the target sentence), participant B asks a question eliciting wide focus, participant A answers by using the target sentence. The target sentences all had a narrow focus on the verb (motivated by contrastive focus to a verb in the following clause, in example (12) ausgehändigt vs. gefragt), so the stress of the dative and the accusative object are identical. The filler sentences were built in a similar way, also embedded in dialogues in which a wide focus was elicited etc.; the verbs in those cases were verbs with an accusative object and a locative or directional argument. An example for a target sentence is given in (12). The possible pairings of the parameters are given in Table 7, together with target sentence numbers for reference. (12)

A: Also, das kann mir niemand erzählen, dass der Richter  well that can me nobody tell that the judge Beugemann unparteiisch sei. Da war gestern doch Beugemann impartial be There was yesterday (modal part.) der Prozess, der Beschuldigte war der Bauunternehmer Grube, the trial the defendant was the building contractor Grube und die Anklage in der zehnseitigen Anklageschrift lautete and the charge in the 10-page bill of indictment called

9 This was actually disguised as a methodological introduction to doing questionnaire studies. After the conclusion of the experiment I asked them what they thought this was about. Only one guessed that it might have something to do with object order – but his judgments matched the others’, so I did not exclude it. The others usually guessed that it had something to do with emphasis.

Relative Object Order in High and Low German 



 159

auf Unterschlagung. on embezzlement B: Und was ist dann passiert? And what is then happened [Target sentence:] A: Interessanterweise hat der Richter die Anklageschrift   Interestingly has the judge [the bill of indictment]acc dem Beschuldigten nicht nur ausgehändigt, sondern ihn sogar [the defendant]dat not only handed over but him even gefragt, ob er mit den Anklagepunkten einverstanden sei. asked whether he with the charges in agreement be ‘A: Well, nobody can tell me that judge Beugemann is impartial. You know, yesterday was this trial against building contractor Grube, in the ten-page bill of indictment he was charged with embezzlement. – B: And, what happened next? – A: Interestingly, the judge did not only hand over the bill of indictment to the defendant, but he even asked him if he was happy with the charges.’ Table 7: Design of target sentences sent. #

verb

actual order

T1 T2 T3 T4

Dat>Acc (aushändigen) Dat>Acc (aushändigen) Acc>Dat (aussetzen) Acc>Dat (aussetzen)

Dat>Acc Acc>Dat Dat>Acc Acc>Dat

There are two variables in our calculation, namely whether wide focus can be preserved under movement of an argument, and whether the base order is Dat>Acc or Acc>Dat. Depending on how these variables are filled, we arrive at different expectations (E1-4): E1:  wide focus cannot be preserved under movement, and the base order is Dat>Acc: T1 should be good, T2 should be bad (movement!), likewise should T4 be bad, as it violates the movement constraint. T3 should be intermediate, as here the animateness constraint and the non-movement-constraint compete with each other. wide focus cannot be preserved under movement, and the base order is Acc>Dat: E2:  T4 should be good, T2 should be good, as well, T1 should be bad (movement!), T3 should be extremely bad (movement and animateness constraint violated).

160 

 Augustin Speyer

E3:  wide focus can be preserved under movement, and the base order is Dat>Acc: Here, only the animateness constraint should play a role, so T1 and T4 in which the animateness constraint is observed, should be good, whereas T2 and T3 should be worse. E4: wide focus can be preserved under movement, and the base order is Acc>Dat: The same considerations apply as those in E3. The design of the four target sentences, the expectation values and the mean evaluation of the questionnaire study is given in Table (8). Table 8: Mean evaluations of targete sentences Sentence Verb no.

Actual order E1

E2

E3/E4

T1 T2 T3 T4

Dat>Acc Acc>Dat Dat>Acc Acc>Dat

bad good very bad good

good medium medium good

Dat>Acc (aushändigen) Dat>Acc (aushändigen) Acc>Dat (aussetzen) Acc>Dat (aussetzen)

good bad medium bad

Mean evaluation 1,56 1,79 2,74 2,12

We see that none of the expectations is fully borne out, but that the fit is actually worse for the two expectations E1 and E2, which are based upon the assumption that movement destroys wide focus. So we can say that wide focus reading is a property not of the base-generated constituent order, but of the unmarked word order, no matter what its generation history is (as Höhle 1982 observed). However, the fact that the distance between animateness-observing T4 and animatenessviolating T3 is bigger than between animateness-observing T1 and animatenessviolating T2 requires some explanation. If we assume Acc>Dat as base order, a possible avenue of explanation opens up: in T2, the violation of the animateness constraint results from the fact that no movement occurred that would produce a constituent order in line with the animateness constraint. In T3, on the other hand, the violation comes about just because movement occurred that produced a word order violating the animateness constraint. If we assume that a high animateness status is a triggering feature for upward movement, short scrambling in this case, T2 and T4 should be okay (no movement means no cost), T1 should be okay as well (movement occurred, but it is warranted by the appropriate feature), but T3 should be bad, as here movement occurs that is not triggered by the appropriate feature. It is this expectation that is borne out.



Relative Object Order in High and Low German 

 161

4.3 The Binding argument I conclude that arguments from the unmarked order have no explanatory power for the base order. A strong syntactic argument can be derived from Binding Theory, more precisely, principle A (Chomsky 1981). The argument is elaborated in Müller (1999) and applied to the problem at hand in Speyer (2015), so I summarize it here only briefly. Consider the data in (13). The accusative object can bind the reciprocal einander ‘each other’, which represents the dative object. A dative object could not bind a reciprocal; the result is an ungrammatical sentence (13b). So this is clear evidence that the accusative is hierarchically higher than the dative and consequently the base order is Acc>Dat. (13) a. …dass Jörg [die Gäste]i einanderi vorstellt. …that Jörg.nom [the guests].acc each-other introduces ‘[I see] that Jörg introduces the guests to each other’ b. *… dass Jörg [den Gästen]i einanderi vorstellt. … that Jörg.nom [the guests].dat each-other introduces ‘[I see] that Jörg introduces the guests to each other’ This argument has come under attack by Rothmayr (2006). In Speyer (2015) I show that her arguments against Müller’s (1999) binding argument do not hold. An easier line of attack would be to say that this effect might hinge on the particular verb vorstellen ‘to introduce’ and/or the reciprocal. But if other verbs are substituted for it, and the reciprocal changed to a simple reflexive, it is difficult to come up with a naturally sounding scenario, but if one succeeds one sees that the same grammaticality judgments prevail (14, 15).10 (14) a. Der Vater hat [die Eheleute]i {sich/einander}i versprochen the father has [the spouses].acc {REFL/REC} promised b. *Der Vater hat [den Eheleuten]i {sich/einand.}i versprochen the father has [the spouses].dat {REFL/REC} promised ‘the father has promised the spouses to each other’ (15) a. Jörgj Jörg

hat [den Mann]i sich i,*j im Spiegel gezeigt. has [the man].acc himself in-the mirror shown

10 There are probably other factors involved, and the judgments are not as clear-cut as one might wish (see, e.g., Featherston 2002). Obviously, there is a demand for further research.

162 

 Augustin Speyer

b. Jörgj hat [dem Mann]i sich *i, j im Spiegel gezeigt. Jörg has [the man].dat himself in-the mirror shown ‘Jörg showed the man to himself in the mirror’

5 Conclusions The base argument order in OHG was Acc>Dat (as it is up to the present day in Modern German). The same presumably holds for OS. This order is reversed to Dat>Acc if two full NPs are involved because the arguments are ordered according to the thematic hierarchy, which is easily detectable by the animacy condition. The difference between OHG on the one hand and Modern German and OS on the other hand is that in OHG there is a very strong tendency to use only the thematic hierarchy for the ordering of arguments (a tendency that prevails throughout MHG and ENHG up to around A.D. 1500), whereas other factors play (almost) no role. In OS and Modern German, on the other hand, the thematic hierarchy is only one factor among others (albeit a strong one) and can be violated by other conditions of, e.g., a pragmatic nature. One very important factor for the object order in Modern German is the definiteness condition (Lenerz 1977; 2002; Reis 1987), which disallows indefinite NPs before definite NPs. In some dialects, such as Bavarian, it is the decisive condition (s. Weiß 2004). This condition obviously can play a role only if definiteness is coded consistently in the language, if one believes it is morphological definiteness that is responsible. I leave this here as an issue of further research.

References Sources: Berthold von Regensburg. 1965. Vollständige Ausgabe seiner Predigten. Ed. Franz Pfeiffer & Kurt Ruh. Bd. 1. Berlin: de Gruyter (reprint; original: Wien: Braumüller, 1862). Gippert, Jost et al. 1987. Thesaurus indogermanischer Text- und Sprachmaterialien (TITUS). URL: http://titus.uni-frankfurt.de/indexd.htm. Last access: Feb 24, 2013. Heliand. 1996. Heliand und Genesis. Ed. by Otto Behaghel & Burkhard Taeger. 10th impression. Tübingen: Niemeyer. Nibelungenlied. 2005. Das Nibelungenlied. Nach der St. Galler Handschrift herausgegeben und erläutert von Hermann Reichert. Berlin/New York: de Gruyter. Notker. 1979. Die Werke Notkers des Deutschen. Ed. by Edward H. Sehrt, Taylor Starck, James C. King & Petrus W. Tax. Bd.8: Der Psalter. Ps. 1–50. Tübingen: Niemeyer. Notker. 1983. Die Werke Notkers des Deutschen. Ed. by Edward H. Sehrt, Taylor Starck, James C. King, & Petrus W. Tax. Bd.10: Der Psalter. Ps. 101–150. Tübingen: Niemeyer.



Relative Object Order in High and Low German 

 163

Otfrid von Weißenburg. 1973. Otfrids Evangelienbuch. Ed. by Oskar Erdmann & Ludwig Wolff. 6th impression. Tübingen: Niemeyer. Spee, Friedrich. 2007. Geistliche Lieder. Ed. by Theo G. Oorschot. Tübingen: Francke. Springeth, Margarete, Klaus M. Schmidt & Horst P. Pütz. 1992. Mittelhochdeutsche ­Begriffsdatenbank. URL: http://www.mhdbdb.sbg.ac.at:8000/. Last access: Feb 24, 2013. Tatian. 1994. Die lateinisch-althochdeutsche Tatianbilingue Stiftsbibliothek St. Gallen Cod. 56. Ed. by Achim Masser. Göttingen: Vandenhoeck & Ruprecht. Aids: Greule, Albrecht. 1999. Syntaktisches Verbwörterbuch zu den althochdeutschen Texten des 9. Jahrhunderts. Frankfurt/M.: Lang. Köbler, Gerhard. 2006. Neuhochdeutsch-althochdeutsches Wörterbuch. URL: http://www. koeblergerhard.de/germanistischewoerterbuecher/althochdeutscheswoerterbuch/ nhd-ahd.pdf, last access: Feb 24, 2013. Sehrt, Edward H. 1966. Vollständiges Wörterbuch zum Heliand und zur altsächsischen Genesis, 2nd edition. Göttingen: Vandenhoeck & Ruprecht. Research literature: Axel, Katrin. 2007. Studies on Old High German Syntax. Amsterdam/ Philadelphia: Benjamins. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Clark, Robin & Ian Roberts. 1993. A computational model of language learnability and language change. Linguistic Inquiry, 24: 299–345. den Besten, Hans. 1977. On the interaction of root transformations and lexical deletive rules. In: Abraham, Werner (Hg.): On the formal syntax of the Westgermania. 47–131. Amsterdam/Philadelphia: John Benjamins. Dowty, David R. 1991. Thematic proto-roles and argument selection. Language, 67: 547–619. Featherston, Sam. 2002. Coreferential objects in German: experimental evidence on reflexivity. Linguistische Berichte, 192: 457–484. Fleischer, Jürg. 2005. Zur Abfolge akkusativischer und dativischer Personalpronomen im Althochdeutschen und Altniederdeutschen (8./9. Jh.). In: Simmler, Franz, Claudia Wich-Reif & Yvon Desportes (Eds.): Syntax Althochdeutsch-Mittelhochdeutsch: Eine Gegenüberstellung von Metrik und Prosa. 9–48. Berlin: Weidler. Haubrichs, Wolfgang. 1999. Heliand und Altsächsische Genesis. In: Hoops, Johannes et al. (Eds.): Reallexikon der Germanischen Altertumskunde. Bd. 14. 297–308. Berlin/New York: de Gruyter. Haubrichs, Wolfgang. 2003. Otfrid von Weißenburg. In: Hoops, Johannes et al. (Eds.): Reallexikon der Germanischen Altertumskunde. Bd. 22. 381–387. Berlin/New York: de Gruyter. Hoberg, Ursula. 1997. Die Linearstruktur des Satzes. In: Zifonun, Gisela, Ludger Hoffmann & Bruno Strecker (Eds.): Grammatik der deutschen Sprache. Band 2. Schriften des Instituts für deutsche Sprache 7.2. 1496–1680. Berlin: de Gruyter. Höhle, Tilman. 1982. Explikationen für ‘normale Betonung’ und ‘normale Wortstellung’. In: Abraham, Werner (Ed.): Satzglieder im Deutschen. 75–153. Tübingen: Narr. Hofmann, Johann Baptist & Anton Szantyr. 1965. Lateinische Syntax und Stilistik. München: Beck. Jackendoff, Ray. 1990. Semantic Structures. Cambridge, Ma.: MIT Press. Larson, Richard K. 1988. On the double object construction. Linguistic Inquiry 19, 335–391. Larson, Richard K. & Sungeun Cho. 2003. Temporal adjectives and the structure of possessive DPs. Natural Language Semantics, 11: 217–247. Lenerz, Jürgen. 1977. Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: Narr. Lenerz, Jürgen. 2002. Scrambling and Reference in German. In: Abraham, Werner & C. Jan-Wouter Zwart (Eds.): Issues in Formal German(ic) Typology. 179–192. Amsterdam/ Philadelphia: Benjamins.

164 

 Augustin Speyer

Lötscher, Andreas. 1981. Abfolgeregeln für Ergänzungen im Mittelfeld. Deutsche Sprache, 9: 44–60. Müller, Gereon. 1999. Optimality, markedness, and word order in German. Linguistics, 37: 777–818. Primus, Beatrice. 2004. Division of labour: The role-semantic function of basic order and case. In: Willems, Dominique et al. (Eds.): Contrastive analysis in language: Identifying linguistic units of comparison. 89–136. Palgrave: Macmillan. Primus, Beatrice. 2011. Animacy, generalized semantic roles, and different object marking. In: Lamers, Monique & Peter de Swart (Eds.): Case, word order, and prominence. Interacting cues in language production and comprehension. 65–90. Dordrecht: Springer. Rappaport, Malka & Beth Levin. 1988. What to do with theta-roles. In: Wilkins, Wendy (Ed.): Thematic Relations. Syntax and Semantics 21. 7–36. New York: Academic Press. Reis, Marga. 1987. Die Stellung der Verbargumente im Deutschen – Stilübungen zum Grammatik: Pragmatik-Verhältnis. Lunder germanistische Forschungen, 55: 139–177. Rothmayr, Antonia. 2006. The binding paradox in German double object constructions. Linguistische Berichte, 206: 195–215. Schlachter, Eva. 2012. Syntax und Informationsstruktur im Althochdeutschen. Untersuchungen am Beispiel der Isidor-Gruppe. Heidelberg: Winter. Speyer, Augustin. 2011. Die Freiheit der Mittelfeldabfolge im Deutschen – ein modernes Phänomen. Beiträge zur Geschichte der deutschen Sprache und Literatur, 133: 14–31. Speyer, Augustin. 2013. Performative Mündlichkeitsnähe als Faktor für die Objektstellung im Mittel- und Frühneuhochdeutschen. Beiträge zur Geschichte der deutschen Sprache und Literatur, 135: 1–36. Speyer, Augustin. 2015. Object order and the Thematic Hierarchy in Older German. In: Gippert, Jost & Ralf Gehrke (eds.): Historical Corpora: Challenges and Perspectives. Corpus Linguistics and Interdisciplinary Perspectives on Language 5. Tübingen: Narr, 101–124. Uriagereka, Juan. 2002. Derivations. London: Routledge. Weiß, Helmut. 2004. Information structure meets minimalist syntax. On argument order and case morphology in Bavarian. In: ter Meulen, Alice & Werner Abraham (Eds.): The Composition of Meaning: From Lexeme to Discourse. 139–165. Amsterdam/Philadelphia: Benjamins.

Achim Stein and Carola Trips

Modeling language contact with diachronic crosslinguistic data 1 Introduction This chapter provides new insights* into the language contact situation between Old French (OF) and Middle English (ME) by investigating instances of structural borrowing (Thomason and Kaufman 1988, Matras 2009), sometimes also called grammatical replication (Heine & Kuteva 2005, 2007). At least since Weinreich’s seminal work in 1953, language contact has been a central topic in linguistics, and manifestations of which can be found in a plethora of domains like language acquisition, language processing, and production, discourse and language change. Many theoretical and methodological insights gained in synchronic studies can be transferred to the diachrony of a language that has been in contact with another language in its earlier stages, for example, the ­classification and distinction between lexical and structural borrowing (for a discussion of the term borrowing Matras 2009, Hickey 2010) or the assumption that the degree and intensity of borrowing can be measured by means of borrowing scales. Whereas some approaches to borrowing hierarchies are more descriptive and pertain to the differences between grammatical categories in borrowing situations (e.g., Haugen 1950; Muysken 1981; Matras & Sakel 2007), others try to provide structural and socio-linguistic explanations (or even constraints) for borrowing phenomena (e.g., Weinreich 1953; Thomason & Kaufman 1988; Campbell 1993, for a survey see Muysken 2010). So far, only few papers have investigated structural borrowing between OF and ME from a theoretical and/or methodological perspective (cf. Haeberli 2010; Ingham 2010; Cinque & Kayne 2008). We seek to fill this gap by providing a ­comparative study of information-structural properties of syntactic constructions like topicalization (in the sense of movement into sentence-initial position, leading to verbsecond ­structures in OF, OE, and ME) and left dislocation (LD), as well as prosody. The outline of the chapter is as follows. Section 2.1 introduces the properties of LD in the languages under investigation, i.e., Modern French (ModF), PresentDay English (PDE), OF, Old English (OE) and ME. It further introduces the main ­ ypothesis is based on, and provides assumption that the so-called compensation h definitions of terminology we are going to use in terms of information structure. In

* This contribution was submitted in August 2013. It does not take into account literature and corpora published after that date.

166 

 Achim Stein and Carola Trips

section 3, the historical background of the language contact situation between OF and ME is discussed and pieces of evidence for intense contact are provided which speak in favor of structural borrowing. In section 4, we describe methodological issues of investigating the borrowing of structures like LDs and discuss our findings from a number of corpus-based studies. Section 5 concludes with a summary of our findings and a further outlook.

2 Marking infomation structure 2.1 The phenomenon of left dislocation Both in topicalization and LD structures, objects are fronted to the first position in the sentence by attributing special prominence to this constituent.1 Whereas in PDE both types of structures occur, in ModF topicalization is possible only if the fronted constituent is resumed by a pronoun (compare the examples of ­topicalization (1) and LD (2 below): (1) ModF: *Les haricots je mange. PDE: Beans I eat. (2) ModF: Les haricots, je les mange. PDE: Beans, I eat them. In earlier stages of French and English, topicalization of an object resulted in verb-second, the finite verb occurred in second position and was inverted with the subject of the sentence. Instances of object topicalization in verb-second ­sentences in OF and OE are given in (3) and (4): (3) Treis escheles ad l’emperere Carles. three divisions has the-king Charles ‘King Charles has three divisions.’ (Roland,219.3025) (4) þæt hus hæfdon Romane to ðæm anum acne geworht ... that house had Romans to the one sign made ‘The Romans had made that house to their sole sign.’ (Or_3:5.59.3.1042)

1 In the following, we will restrict our study to object LD because we will compare it with object topicalization.



Modeling language contact with diachronic crosslinguistic data 

 167

In OF and ME two types of LD occur: one type that expresses topic continuity, and one type which expresses contrastive focus. The examples in (5) for OF, and (6) for ME, are instances of topic continuity: (5) Ceste bataille, veirement la ferum This battle really it [we]-will-fight ‘We will really fight this battle.’ (Roland,70.846) (6) and slonked her in so hungerly that he lefte neyther flessh ne bone / nomore but a fewe fethers the smal fethers he slange them in wyth the flessh the small feathers he swallowed them in with the flesh ‘He swallowed the small feathers with the flesh.’ (REYNAR,53.337–8 ) In both the OF and the ME context, mention was made before of ceste bataille and the smal fethers (in the French example topicness is marked by the demonstrative ceste). In (7) for OF, and (8) for ME, examples are given for LD with contrastive focus: (7) Mais vos barons en sa ballie S’ il les trovout... But your barons in his realm if he them found... ‘But if he found your barons in his realm’ (Béroul,1106ss) (8) and of on þei put oute his eyne, þe oþir þei broke his bak and of one they put out his eye, the other they broke his back ‘and of the one, they put out an eye, and of the other, they broke his back.’ (CAPCHR,249.4166–7)

2.2 The compensation hypothesis applied to Old French In his chapter, Kroch (1989) discusses the loss of verb-second in OF by correlating the function(s) of LD structures to the function of object topicalization with the aim of finding further proof for the Constant Rate Effect. More precisely, he claims that “loss of verb-second word order in French took place via the replacement of topicalization by left-dislocation.” (Kroch 1989, 214)

An explanation for the loss of verb-second was suggested by Adams (1987, 7), who pointed out that it was associated with a change in the phrasal accent pattern of

168 

 Achim Stein and Carola Trips

French. The change from OF lexical and phrasal accent to the Middle French (MF) (and ModF) phrasal accent could have caused the word order change. Based on these assumptions, Kroch claimed that phrasal accent forced preposed constituents to occur in a new, left-dislocated position. This in turn led to structures without subject-verb inversion and to a rise in frequency of resumptive pronouns. From a quantitative, statistical perspective, Kroch predicts that the increase in the rate of LD correlates with the decrease of the rate of subject-verb inversion, based on Priestley’s (1955) data. In the following, we are going to use these observations as a test case for the modeling of language change with diachronic crosslinguistic data. More ­precisely, our corpus-based diachronic syntactic study, and the problems arising from it, will allow us to discuss, at a more general level, the following methodological questions: 1. What kind of assumptions are necessary to interpret quantitative developments as language change? 2. Which pieces of information do we have to include to make statements about linguistic levels for which we have no direct evidence, like prosody? 3. What kind of indicators allow us to distinguish between internal change and contact-induced change?

2.3 Information structure For our analysis, we adopt the definitions of topic and focus as resumed by Krifka (2007, 18): “Focus indicates the presence of alternatives that are relevant for the interpretation of linguistic expression.” Hence, “focusness” is a scalar property (see also Firbas’s 1971 notion of communicative dynamism, Chafe’s 1976 notion of contrastiveness and Rooth’s 1985 alternative semantics). Other criteria, such as “highlighting, importance, newness or presupposition of existence [...] are statistical correlatives, but not definitional features, of focus” (Krifka 2007, 30). This definition accounts for apparent contradictions, such as foci referring to “old” content which has been previously mentioned, or topics referring to “new” information (see also Prévost’s remarks on OF in section 3.2). The topic is the point of departure of a sentence (the logical subject) that is complemented by the comment (the logical predicate). The focus must not be identical with the comment, as shown by (Krifka 2007, 42): (9) A: When did [Aristotle Onassis]Topic marry Jacqueline Kennedy? B: [He]Topic [married her [in 1968]Focus]Comment



Modeling language contact with diachronic crosslinguistic data 

 169

Krifka (2007, 41) defines topic from the perspective of the hearer as the “constituent which identifies the entity [...] under which the information expressed in the comment constituent should be stored in the C[ommon] G[round] content.” Concerning the use of further terminology dealing with information-structural properties of constituents in certain syntactic constructions, we would like to point out that we use ‘sentence-initial elements’ for both topicalized and left-dislocated elements. We use topicalize(d) (and topicalization) as syntactic terms, meaning that a topic constituent (Krifka 2007, 18) is moved in sentence-initial position. Dislocated elements (DE) are, according to Lambrecht (2001), extra-propositional while alternating with an intra-propositional position, coindexed with a pronoun, and prosodically marked.2 Transferring these definitions to the description and analysis of older stages of a language creates a number of theoretical and methodological problems: It is difficult or even impossible to verify necessary conditions for the felicitous use of non-canonical sentence structures, contrary to modern language (e.g., Birner & Ward 1998; Birner 2004). But even in modern language, these conditions depend on text type, code (written vs. spoken), individual preference, and other factors. Thus, these problems are not entirely due to the nature of diachronic data. More­ over, some discourse-related criteria (cf. Prince 1981, 1992) can be and have been applied to older stages of languages, e.g., to OF by Prévost 2003. Prince’s criteria ‘discourse old’ vs. ‘dicourse new’, for example, can indeed be verified in ancient texts. On the other hand, we normally have no evidence for the ‘hearer old’ vs. ‘hearer new’ criterion and can only hypothesize about the hearers’ common ground knowledge. We thus conclude that discourse-related properties can be investigated diachronically if the problems mentioned are taken into account and if the data is interpreted with caution.

3 Language contact 3.1 The historical background Thanks to the invaluable work by Rothwell (e.g., 1983; 1993) today it is generally acknowledged that the Norman Conquest of 1066 had a greater effect on English

2 It is not clear how strong the prosodic criterion is: in a recent investigation of spoken French (Avanzi et al. 2010) could not show a prosodic difference between left-dislocated NPs (mon mari, il est instituteur) and heavy NPs (mon mari est instituteur).

170 

 Achim Stein and Carola Trips

than any other political event in its history. The massive integration of French lexemes are proof of this on the level of the lexicon. During at least the first two centuries after the Conquest, the vocabulary of English underwent a considerable transformation, especially in the fields of administration, government, law, and trade. Moreover, the multilingual situation of post-Conquest medieval England, which lasted from 1066 to the 15th century, led to linguistic phenomena like codeswitching between Anglo-Latin and/or Anglo-French,3 ME and its many varieties, typical for intense language contact situations (cf. Rothwell & Trotter 2005; Schendl & Wright 2011). During that time, Anglo-French gained the role of an official written language over Anglo-Latin, whereas ME was the predominantly spoken language. By the end of the 15th century, it had taken on the role of the acknowledged national language, both spoken and written. These historical facts must lead to the methodological consequence of ­including sources of multilingualism in any linguistic analysis that seeks to draw as clear a picture of this language contact situation as possible. Unfortunately, the annotated corpora for English and French available at present do not include these texts. This means that an analysis must combine several different sources like annotated corpora on the one hand, and full text analyses on the other hand. These are glosses in dictionaries, didactic texts (e.g., Walter of Bibbesworth’s Tretiz from the 13th century, and texts discussed in Hunt 1991), mixed texts like business writings, inventories (see the work by Wright 1995, 2003; Trotter 2000; Ingham 2010), as well as direct translations like the Ayenbite of Inwyt (Morris 1866), translated from Somme le roi (Laurent 2008). Palsgrave’s grammar from 1530 is a further interesting source. Written for learners of English (actually especially for Mary Tudor), it includes descriptions of the French language on all levels. As to sounds and accent, Palsgrave remarks: “The hole reason of theyr accent [French speakers, CT, AS] is grounded chefely upon thre poyntes: fyrst, there is no worde of one syllable whiche with them hath any accent, or that they use to pause upon, and that is one great cause why theyr tong semeth to us so brefe and sodayn and so harde to be understanded whan it is spoken, especially of theyr paysantes or commen people; for, thoughe there come never so many wordes of one syllable together, they pronounce them nat distinctly a sonder as the latines do, but sounde them all under one voyce and tenour, and never rest nor pause upon any of them, except the commyng next unto a poynt be the cause therof; seconde, every worde of many syllables

3 A number of different terms have been proposed to refer to the French spoken and written on the island after 1066 like Anlgo-Norman, stressing its Norman origin, or Anglo-French, stressing its French origin. In our chapter, we will use the more neutral term ‘Old French’, to refer to a ­certain time span without adding further connotations.



Modeling language contact with diachronic crosslinguistic data 

 171

hath his accent upon the last syllable, but yet that nat withstandynge they use upon no suche worde to pause, except the commyng next unto a poynt be the causer therof: and this is one great thyng whiche inclineth the frenchemen so moche to pronounce that latin tong amysse, which contrary never gyve theyr accent on the last syllable. [...]” (Palsgrave 1530, L’esclaircissement de la langue françoyse, quoted from the edition F. Génin 1852, XXI)

In this paragraph, he clearly refers to phrasal stress (see the parts highlighted) which is a distinguishing feature of French, contrary to Latin (the prosodic unit is called mot phonétique). So by investigating this source, we have “direct” proof that at the beginning of the 16th century the change from the system of lexical and phrasal stress to the system of phrasal stress was accomplished, which might have had consequences for the development of syntactic constructions like verb-second and LD, as discussed above (for further work on the transition from OF word to phrasal accent, see Adams 1987; Marchello-Nizia 1995; Noyer 2002; Mathieu 2012). In the following section, we will take an overall look at the main properties of OF.

3.2 Early French In the literature, it is fairly uncontroversial that OF (until 1300) exhibited the following properties (for more elaborate summaries see Bouchard et al. 2007 or Rinke & Meisel 2009): 1. OF had a topic position for sentence-initial elements. 2. OF had V2 sentences, mostly in main clauses. 3. OF had null subjects. 4. OF had two cases: cas sujet and cas régime. 5. OF had two accent systems: phrasal and lexical. 6. OF had clitic object pronouns, subject pronouns were initially pronouns and gradually became clitics in later OF. A more controversial point is how strict verb-second actually was in OF, i.e., triggered in all contexts or not. Further, there is debate about the structural position of the topicalized element. It is also not unanimously agreed upon whether OF had sentence-initial focus elements. Diachronically, verb-second structures decrease in the course of the OF period, but MF (ca. 1300–1500) still had both patterns, XVS and XSV (see, among many others, Roberts 1993; Vance 1997; Muller 2009). Concerning the phenomenon of dislocation, OF had both left and right ­dislocation. However, as opposed to ModF, OF could also have dislocated focus elements (Marchello-Nizia 1998; Prévost 2003; Bouchard et al. 2007; Prévost

172 

 Achim Stein and Carola Trips

2009), as in (10), where Prévost considers Voici que... (‘Here is...’) as the adequate ModF translation: (10) Li nies Marsilie, il est venuz avant sur un mulet... the nephew of-Marsilius, he is come forward on a mule The nephew of Marsilius has come forward on a mule. (Chanson de Roland, late 11c., quoted from Prévost 2003) Similarly, the fronted object in OVS structures can be either topic or focus, but here the difference between discourse topic and sentence topic comes into play, since Prévost (2003, 114) observes that (cf. Krifka’s remarks on focus quoted in section 2.3) Although the cognitive status of the object must not be neglected, it is nevertheless important to distinguish between cognitive and informative status: the object being “thematic” does not mean that it has the role of a topic, on the contrary, it can be in a focus relation to the rest of the proposition. [our translation]

With respect to the information status, in OF, the dislocated element is normally the topic, exceptionally the focus. Topicalized objects are normally the focus, exceptionally the topic. Diachronically, OVS structures were initially unmarked, as in (11), but become marked along with the gradual increase of VO word order in the 13c. (11) Li empereres est par matin levet /Messe e matines  the emperor is by morning awoken /mass and morning prayers ad li reis escultet has the king listened The emperor is up in the morning, the king has listened to mass and morning prayers (Roland,11.138–9)

3.3 Early English In OE (850–1150), the phenomenon of verb-second can occur with any ­constituent [e.g., an object NP as in (12)], similarly to Present-Day German and other modern Germanic languages. It is obligatorily triggered by wh-elements and negations in sentence-initial position, as can be seen in (13) and (14). Following standard generative assumptions (cf. also, e.g., Kemenade 1987, Pintzuk 1999), in these cases, the element in SpecCP is focused.



Modeling language contact with diachronic crosslinguistic data 

 173

(12) Object-Vfin-subject: þæt hus hæfdon Romane to ðæm anum tacne geworht ... that house had Romans to the one sign made ‘The Romans had made that house to their sole sign.’ (Or_3:5.59.3.1042) (13) Wh-Vfin-subject: Hwæt sculon we þæs nu ma secgan? what shall we afterwards now more say ‘What shall we afterwards say now more?’ (Bede_2:9.132.1.1253) (14) Neg-Vfin-subject: ne cræwþ se hana todæg ær þu me ætsæcst. neg crows the cock today before you me deny ‘The cock will not crow today until you deny me.’ (wsgosp,Lk_[WSCp]:22.34.5465) In ME (1150–1500), verb-second still occurs in contexts with non-initial subject, wh-element, and negation, but during the late ME period it gradually declines. This can be observed by variation between dialects where verb-second starts to disappear, for example in sentences with sentence-initial temporal adverbs (cf. e.g., Kroch & Taylor 1997; Trips 2002). As OF, OE exhibits two types of dislocation constructions, left and right dislocation. Traugott (2007, 243) claims that LD can involve discourse-old and discourse-new referents. An example of topic dislocation is given in (15), an example for focus dislocation is given in (16): (15) ða fuglas þa we hie ne onweg flegdon these birds then we them not away drove ‘These birds, we didn’t drive them away.’ (Alex:21.11.258) In the example above, ða fuglas ‘these birds’ is the topic in the sentence, because it clearly is marked as discourse-old by the demonstrative ða. (16) & [Perseuse Mæacedonia cyninge] him wæron on fultume ealle and Perseus Macedonia’s king him were in help all Thraci & Ilirice Tracians and Ilirians ‘And Perseus, the king of Macedonia, was helped by all Tracians and ­Ilirians.’ (Or_4:11.110.12.2305)

174 

 Achim Stein and Carola Trips

In this example, Perseuse Mæacedonia cyninge is discourse-new information, the referent has not been introduced before.

4 Empirical study 4.1 Corpora In this section, we are going to present our corpus study. First, we will briefly describe the methodological issues before we discuss our findings. The data for the diachronic study were drawn from the following syntactically annotated corpora: the York-Toronto-Helsinki Parsed Corpus of Old English Prose (Taylor et al., 2003) for OE, the Penn-Helsinki Parsed Corpus of Middle English 2 (Kroch & Taylor, 2000) for ME, and the MCVF corpus (Modéliser le changement: les voies du français, Martineau 2009) for Medieval French. The English corpora are large collections of written texts (samples), based on the diachronic part of the ­Helsinki corpus, representative of various text types, levels of style, modes of ­expression, and geographical and social varieties. The French corpus is structured along the dialectal, social, and historical dimension and mainly consists of unabridged texts. It covers four historical periods. OF, MF, sixteenth-century French, and ­Classical French. The extraction and evaluation of the results was done using ­CorpusSearch (UPenn), TIGERSearch (IMS, Stuttgart) and self-made scripts.

4.2 Early French: periods and global data The MCVF corpus contains nineteen OF and MF texts. We grouped these texts as represented in Table 1. We preferred defining the periods using the gaps between the composition dates (rather than a strict century definition), thus putting Joinville (1306) in the 13th century and Froissart (1374) in the 14th century, not only because of the gap of at least sixty years between Joinville’s Mémoires and the following texts (Froissart’s chronicles and Prise d’Alexandrie), but also because the number of sentences per period is more equally distributed. Table 2 provides our findings concerning the occurrence of null subjects, verb-second, and LD. “Object-V2” means that the query was restricted to sentenceinitial object and dative NPs (annotated NP-ACC, NP-DTV), to have a direct comparison with LD of the same categories (“Object-LD”). Of course, these verb-second structures represent only a small fraction of all verb-second structures: the relative frequency of verb-second in all the French texts is 0.031744, which is very similar to the frequency of verb-second in the ME corpus (0.038352).

Modeling language contact with diachronic crosslinguistic data 



 175

Table 1: Definition of four periods in the MCVF corpus Period

Approximate date (first text)

Approximate date (last text)

12c 13c 14c 15c

1080 (Roland) 1200 (Sermon sainte Agnes) 1371 (Prise Alexandrie) 1425 (Morchesne)

1194 (Chièvres) 1306 (Joinville) 1375 (Froissart) 1491 (Commynes)

Sentences 16841 12294 27684 15036

Table 2: Diachronic variation of OF null subject, object V2, and object LD Period

12c 13c 14c 15c total

Null subject

Object-V2

Object-LD

fabs

frel

fabs

frel

fabs

frel

3067 861 1995 1489

0.1821151 0.0700342 0.0720633 0.0990290

300 76 90 17

0.0178137 0.0061819 0.0032510 0.0011306

53 30 47 23

0.0031471 0.0024402 0.0016977 0.0015297

7412

0.103152

428

0.006722

153

0.002129

Figure 1 visualizes our findings from Table 2 for the three patterns defined above. The blue curve describes a rather steep decline of null subjects from the 12th to the 13th century, and a slight increase from the 13th to the 15th century. Since the values for Object-V2 and Object-LD are small compared to the null-subject values, almost no change is visible here. For a clearer picture, see Figure 3 on p. 16.

Figure 1: Diachronic variation of OF null subject, object V2, and object LD

176 

 Achim Stein and Carola Trips

4.3 Early English: periods and global data Next, we will turn to our findings for OE and ME. For both periods, we distinguished subperiods as defined by the YCOE corpus, and the four ME periods of the PPCME2 corpus (see Table 3).4 Table 3: Periods in the OE/ME corpus Period

Composition date (first text)

Composition date (last text)

Sentences

OE ME-m1 ME-m2 ME-m3 ME-m4

before 850 1150 1250 1350 1420

1150 1250 1350 1420 1500

110136 17745 13870 32969 20080

Table 4 presents our findings for null subjects, Object-V2 and Object-LD in OE and ME, and in Figure 2, the numbers are translated into curves. Recall that according to Traugott, in OE LD can involve discourse-old and discourse-new referents. This is confirmed by our data, but topic dislocations are much more frequent than focus dislocations (note that the distinction is not given in the table). Concerning the development of the three phenomena under investigation, three observations can be made: First, there is a drastic decline of null subjects from OE to ME-m2, and a further but more moderate decline from ME-m2 to ME-m4. Second, Object-V2 increases from OE to ME-m1 and then drops down again to ME-m2. After ME-m2, no big changes can be observed. Third, Object-LD also increases from OE to ME-m1 and then drops down again from ME-m1 to ME-m2. After ME-m2 there is a slight d ­ ecrease toward ME-m4. The two curves for Object-V2 and Object-LD seem to run parallel in their development, whereas the developmental pattern of null subjects does not.

4 Periods are based on the composition date (or the manuscript date if the composition date was unknown).



Modeling language contact with diachronic crosslinguistic data 

 177

Table 4: Diachronic variation of OE/ME null subject, object-V2, and object-LD Period

Null subject

Object-V2

Object-LD

fabs

frel

fabs

frel

fabs

frel

OE ME-m1 ME-m2 ME-m3 ME-m4

2073 257 45 56 27

0.018822 0.014483 0.003244 0.001699 0.001345

522 119 45 98 60

0.004740 0.006706 0.003244 0.002972 0.002988

372 109 38 73 38

0.003378 0.006143 0.002740 0.002214 0.001892

total

2458

0.012618

844

0.004333

630

0.003234

Figure 2: Diachronic variation of OE/ME null subject, object-V2, and object-LD

How can these findings be interpreted? If we take a look at the figures for LD, we see that this construction does not occur very frequently in OE and ME. Similarly, in PDE, LD is quite rare. According to a study by Creswell (2004, 12), dislocations are almost absent from written language (4 tokens in 1 mio. words in the Wall Street Journal 1989), but more frequent than clefts and topicalizations in the oral Switchboard corpus (406 tokens, frel p.sent ~0.008). Thus, ­dislocations are about 500 times more frequent in speech than in writing. In our historical corpora, dislocation frequencies are very similar to the frequencies in the oral Switchboard corpus. In his work, Biber (1988, 1991) established a method to define distinctive linguistic characteristics of spoken and written ­registers, one of them being the occurrence of pied piping constructions in

178 

 Achim Stein and Carola Trips

‘elaborated vs. situation-dependent reference’ (dimension B). We tentatively suggest that dislocation constructions could be seen as one (of many) distinctive linguistic feature to characterize orality in written texts. Further research is needed, of course, to verify this assumption. Figure 1 for French and 2 for English show that null subjects in both languages decrease as expected, except for a slight increase in 15th century French. Note, however, that the frequency of null subjects in the 12th century is about ten times higher in French than in English, where the omission of the subject is restricted to some grammatical persons.

4.4 Comparison of French and English data Figure 3 (p. 16) visualizes the data for Object-V2 (dotted lines) and LD (solid lines) for both languages. Since the English and French subperiod boundaries do not match, intermediary values were added to bridge the gaps. The added graph in the top right corner is Priestley’s (1955) curve for LD as quoted by Kroch (1989).

Figure 3: English and French variation of object V2 and object LD

In French, Object-V2 frequencies develop as predicted, decreasing from 1.8% to 0.6% until 1300, but our data for French LD match Priestley’s only in the OF period (0.3%–0.4%). According to Priestley, in MF LD occurrences more than double their frequency (>1%), but in our corpus, LD frequencies decrease steadily



Modeling language contact with diachronic crosslinguistic data 

 179

between 1300 and 1500.5 This implies (contra Kroch) that the developments of V2 and LD are not correlated. However, these studies as well as the data we considered so far consider LD structures in a very general way, whereas they are in fact syntactically quite heterogeneous. A closer analysis which considers the criterion of heaviness6 yields at least three different types of dislocated elements (DE): 1. free wh-clauses as in (17) and (18); 2. “heavy” NPs, modified by a relative or adverbial clause as in (19) and (20), where the resumptive pronoun is triggered by the distance of the NP; 3. unmodified NPs as in (21) and (22), where the resumptive pronoun is an indicator of topichood (rather than triggered by the distance). mult te sert, malvais luer l’ en dunes. (17) Ki Who often youACC serves bad reward him for that [you]give ‘Whoever serves you, you reward him badly for it.’ (Roland,187.2585) (18) Swa hwilc mon swa me timbreð gebedhus, sele þu mede him on whosoever me builds chapel give you reward him in heofonum. heaven ‘Whoever builds a chapel for me will be rewarded in heaven.’ (Mart-5-[Kotzor]:Jy15,A.14.1166) (19) Tuz lur amis qu’ il i unt morz truvet, Ad un carner  all their friends whom they there have dead found to a grave sempres les unt portet. always them [they]have carried ‘All their friends whom they found dead, they brought them to a grave’ (Roland,212.2943) (20) ðæt, þæt ic eow secge, þæt ic secge eallum mannum... that that I you say that I say all men ‘That that I tell you, I tell to all men.’ (ÆLet-6-[Wulfgeat]:112.42)

5 Following Hilpert & Gries’s 2009 method for assessing frequency changes in multistage diachronic corpora we applied Kendall’s τ: our data exhibit a clear downward trend (τ = –1, P = 0.04167). 6 Traugott (2007) retains the following criteria to classify dislocations: animacy, quantification, type of resumptive, heaviness (modifying finite clause after the dislocated element), occurrence in main or subordinate clauses, antecedence, contrast, and topic continuity.

180 

 Achim Stein and Carola Trips

(21) Mes escheles, tutes les guiereiz; my divisions, all them [you] will lead ‘You will lead all my divisions.’ (Roland,236.3266) (22) ða fuglas þa we hie ne onweg flegdon these birds then we them not away drove ‘These birds, we didn’t drive them away.’ (Alex:21.11.258) Whereas most diachronic studies (as well as our global data presented in Tables 2 and 4) associate a given construction with a default information structural value (e.g., a topicalized element with focus, or a DE with topic), our classification of LD on the grounds of the proximity between the DE and the resumptive pronoun allows for a more specific assumption: only in the case of unmodified NPs is dislocation triggered by the desire to mark the information structural status of the DE (rather than triggered by the distance of the dislocated NP). This DE is more often a topic than a focus. Note that most dislocated topics in spoken ModF are also topics, and rarely modified by subordinate clauses (Les haricots, je les mange ‘The beans, I eat them’). In the light of this distinction, the English data (Table 4, p. 14) and the assumption that French may have had a structural impact on English should be re-considered. At first sight, the overall frequencies of Object-LD in OE and ME look similar (0.2–0.3%). The distinction between the four ME periods, however, shows that the frequency of both Object-V2 and Object-LD structures increases at the beginning of the ME period (‘ME-m1’ in Fig. 3). On these grounds alone, one could be tempted to argue in favor of French influence in early ME, interrupting the otherwise gradual decrease of LD structures in the history of English (0.3% in OE, 0.1% in EME, 0.05% in PDE; compare also Creswell’s figures mentioned in section 4.3). But the motivation is stronger if we account for the distinction between modified and unmodified DEs. According to Traugott (2007), in OE, only 5.8% of DEs are unmodified. Our count yields a higher percentage, but the scores confirm her observation (Table 5). Table 5: Types of dislocated elements: free wh-clause (CP-FRL), NP-dependent relative clause (CP-REL), unmodified NPs (other)

OE (Traugott 2007) OE ME OF

Wh-clause (CP-FRL)

Heavy NP (CP-REL)

NP (other)

36.6% 28.3% 22.9%

53.5% 53.5% 27.5%

5.8% 9.9% 18.2% 49.6%



Modeling language contact with diachronic crosslinguistic data 

 181

Figure 4: Types of dislocated elements: free wh-clause (CP-FRL), NP-dependent relative clause (CP-REL), unmodified NPs (other)

It is obvious that OF dislocation is different from OE dislocation: Figure 4 shows that in ME-m1, the percentage of unmodified NPs increases significantly7 (narrow dotted line with square dots). In ModF, the DE is normally not followed by a modifying clause, and Table 5 shows that already in OF half of the DEs are unmodified NPs (bottom line, 49.6%). Thus, the increase in ME-m1 could be attributed to French influence: although clitic LD has not been borne out in the history of English, the “French type” has steadily increased, reaching 26.3% of all the LD structures in ME-m4. At the end of this empirical section, let us reconsider two information structural issues. First, it is (usually tacitly) assumed that the need for information structural marking is constant over time. That some of these markers may be phonetic and hence empirically not directly accessible is a particular problem for diachronic studies. Second, in a diachronic perspective, it is almost impossible to postulate that a given construction (LD or topicalization) marks a constant information value (topic only or focus only). In OF, topicalization develops from an unmarked toward a focus-marking construction because V2 loses ground to the (S)VO structures, becomes less frequent, and hence more marked. In OF, LD was a topic marker, but not exclusively. If we follow the assumption of constant IS marking, the decrease of V2 structures therefore raises the question of how focus marking was compensated.

7 χ2 = 26.7, P = 0.000023.

182 

 Achim Stein and Carola Trips

The study of verb-second and LD alone can obviously not provide an answer to this question, but it is tempting to cast a quick glance on cleft sentences, since they are indisputably8 focus markers: the final position of the cleft main clause: It is X that... clearly is a syntactic and a prosodic focus. This assumption is corroborated by Marchello-Nizia (1999) and Combettes (1999), who maintain that the development of clefts starts in the 13c., favored by the loss of V2, and accelerates in the 14c.9 Our findings presented in Table 6 support their assumptions (for comparison, we repeat the columns for Object-V2 and Object-LD from Table 2, p. 12). Table 6: Cleft sentences (CP-CLF...) in French Period

12c 13c 14c 15c

Clefts(CP-CLF)

Object-V2

Object-LD

fabs

frel

fabs

frel

fabs

frel

21 11 12 40

0.0012470 0.0009131 0.0004846 0.0021974

300 76 90 17

0.0178137 0.0061819 0.0032510 0.0011306

53 30 47 23

0.0031471 0.0024402 0.0016977 0.0015297

Cleft constructions (i.e., the structures annotated as “CP-CLF...” in the MCVF corpus) are too infrequent to compensate for the decrease of Object-V2 in OF, and frequencies increase too late to fit the compensation theory (for a critical and more detailed analysis see Dufter 2008 & Stein Trips 2012).

5 Conclusion This chapter has provided new data for the discussion of the relation between object verb-second and object LD structures, based on Kroch’s (1989) compensation hypothesis. For the questions posed in section 2.2, we would like to provide some tentative answers. As to the first question, i.e., the kind of assumptions which are necessary to interpret quantitative developments as language change, we think that it is necessary to assume that topic and focus marking is constant

8 For a more detailed view and a classification of cleft constructions see Dufter (2007) and Dufter (2008). 9 But note that Bouchard et al. (2007, 4ff) show that clefts (i.e., ce subjects) occur as early as in the 12c., but then, ce is often attributive, and therefore not a ‘real’ cleft construction.



Modeling language contact with diachronic crosslinguistic data 

 183

over time, at least in speech. We argued that relatively high frequencies of a clearly ‘spoken phenomenon’ like LD (compared to PDE and ModF) indicates that our corpora, or at least parts of them (some texts more than others), can be taken to reflect spoken language. The second question pertained to the kind of information we have to include to make statements about linguistic levels for which we have no direct evidence, like prosody. We have shown that by including meta-­ linguistic information, for example, Palsgrave’s grammar, we can gain insights into the syntax-prosody-interface even in diachrony. Concerning our third question of the kind of indicators that allow us to distinguish between internal change and contact-induced change, our answer is that significant quantitative changes (see the peak in Fig. 4) are not compatible with the internal changes, or would at least require exceptional explanations. We suppose that these findings would gain additional support if texts directly displaying the multilingual situation are taken into account as well, in addition to the existing syntactically annotated corpora used in this study.

References Adams, Marianne: From Old French to the theory of pro-drop. In: Natural Language and Linguistic Theory 5 (1987), No. 1, p. 1–32. Avanzi, Mathieu; Gendrot, Cédric; Lacheret, Anne: Is there a prosodic difference between left-dislocated and heavy subjects? Evidence from spontaneous speech. In: Every Language, Every Style. Speech Prosody, 5th International Conference, Chicago, Illinois, May 2010. Chicago, 2010. – URL http://www.lacheret.com/Xinha/UPLOAD/ Analor-prom.pdf. Biber, Douglas: Variation across Speech and Writing. Cambridge: Cambridge University Press, 1988. Biber, Douglas: Variation across Speech and Writing. 1. paperback ed. Cambridge: Cambridge Univ. Press, 1991. – ISBN 0521425565. Birner, Betty: Discourse functions at the periphery: Noncanonical word order in English. In: Shaer, Benjamin (ed.); Frey, Werner (ed.); Maienborn, Claudia (ed.): Proceedings of the Dislocated Elements Workshop, ZAS Berlin, November 2003 Vol. 35. Berlin: ZAS, 2004, p. 41–62. Birner, Betty; Ward, Gregory: Studies in language Companion series. Vol. 40: Information status and noncanonical word order in English. Amsterdam: Benjamins, 1998. – ISBN 9027230439. Bouchard, Jacynthe; Dupuis, Fernande; Dufresne, Monique: Un processus de focalisation en ancien français: le développement des clivées. In: Actes du congrès annuel de l’Association canadienne de linguistique 2007. Proceedings of the 2007 annual conference of the Canadian Linguistic Association, 2007. Campbell, Lyle: On proposed universals of grammatical borrowing. In: Aertsen, Henk (ed.); Jeffers, Robert (ed.): Historical linguistics 1989: papers from the 9th International

184 

 Achim Stein and Carola Trips

Conference on Historical Linguistics, Rutgers University, 14–18 August 1989 Vol. 106. Amsterdam: Benjamins, 1993, p. 91–110. – ISBN 1556195605. Chafe, Wallace: Givenness, contrastiveness, definiteness, subjects, topics, and point of view. In: Li, C. N. (ed.): Subject and Topic. New York: Academic Press, 1976, p. 25–56. Cinque, Guglielmo (ed.); Kayne, Richard (ed.): The Oxford Handbook of Comparative Syntax. Oxford: Oxford University Press, 2008. Combettes, Bernard: Thématisation et topicalisation: leur rôle respectif dans l’évolution du français. In: Guimier, Claude (ed.): La thématisation dans les langues. Actes du colloque de Caen, 9–11 octobre 1997. Paris: Peter Lang, 1999, p. 231–245. Creswell, Cassandre: Syntactic Form and Discourse Function in Natural Language Generation. New York: Routledge, 2004 (Outstanding dissertations in linguistics). – ISBN 0415971047. Dufter, Andreas: Clefting and discourse organization. Comparing Germanic and Romance. In: Dufter, Andreas (ed.); Jacob, Daniel (ed.): Focus and Background in Romance Languages. Amsterdam, Philadelphia: Benjamins, 2007 (Studies in Language Companion Series), p. 83–121. Dufter, Andreas: On explaining the rise of c’est-clefts in French. In: Detges, Ulrich (ed.); Waltereit, Richard (ed.): The Paradox of Grammatical Change. Amsterdam, Philadelphia: Benjamins, 2008, p. 31–56. Firbas, Jan: On the concept of communicative dynamism in the theory of functional sentence perspective. In: Sbornik praci filosoficke fakulty brnenske University 19 (1971), p. 135ff. Haeberli, Eric: Investigating Anglo-Norman Influence on Late Middle English Syntax. In: Ingham, Richard (ed.): The Anglo-Norman Language and Its Contexts. Woodbridge: Boydell and Brewer, 2010, p. 143–163. Haugen, Einar: The analysis of linguistic borrowing. In: Language 26 (1950), p. 210–231. Heine, Bernd (ed.); Kuteva, Tania (ed.): Language Contact and Grammatical Change. Cambridge: Cambridge University Press, 2005 (Cambridge approaches to language contact). – ISBN 0521608287. Heine, Bernd; Kuteva, Tania: Identifying instances of contact-induced grammatical replication. 2007. Hickey, Raymond (ed.): The Handbook of Language Contact. Malden, Mass.: Wiley-Blackwell, 2010. – ISBN 9781405175807. Hilpert, Martin; Gries, Stefan: Assessing frequency changes in multistage diachronic corpora: Applications for historical corpus linguistics and the study of language acquisition. In: Literary and Linguistic Computing 24 (2009), No. 4, p. 385–401. Hunt, Tony: Teaching and Learning Latin in Thirteenth-Century England. Volume I: Texts. Cambridge: Brewer, 1991. Ingham, Richard (ed.): The Anglo-Norman Language and its Contexts. Boydell and Brewer, 2010. Kemenade, Ans v.: Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris, 1987. Krifka, Manfred: Basic notions of information structure. In: Féry, Caroline (ed.); Fanselow, Gisbert (ed.); Krifka, Manfred (ed.): The Notions of Information Structure. Potsdam: ­Universitätsverlag Potsdam, 2007, p. 13–55. Kroch, Anthony: Reflexes of grammar in patterns of language change. In: Language Variation and Change (1989), No. 1, p. 199–244. Kroch, Anthony; Taylor, Ann: Verb movement in Old and Middle English: Dialect variation and language contact. In: Kemenade, Ans v. (ed.); Vincent, Nigel (ed.): Parameters of ­Morphosyntactic Change. Cambridge: Cambridge University Press, 1997, p. 297–325.



Modeling language contact with diachronic crosslinguistic data 

 185

Kroch, Anthony (ed.); Taylor, Ann (ed.): The Penn-Helsinki Parsed Corpus of Middle English, Second Edition (PPCME2). Philadelphia: University of Pennsylvania, 2000. Lambrecht, Knud: A framework for the analysis of cleft constructions. In: Linguistics (2001), No. 39, p. 463–516. Laurent, Frère; Brayer, Anne-Françoise (ed.): La Somme le roi. Paris, Abbeville: Société des anciens textes français, Paillart, 2008. Marchello-Nizia, Christiane: L’évolution du français. Ordre des mots, démonstratifs, accent tonique. Paris: Colin, 1995. Marchello-Nizia, Christiane: Dislocations en ancien français: thématisation ou rhématisation? In: Cahiers de Praxématique 30 (1998), p. 162–178. Marchello-Nizia, Christiane: Le français en diachronie: douze siècles d’évolution. Paris: Ophrys, 1999 (Collection L’essentiel français). – ISBN 2708009133. Martineau, France (ed.): Le corpus MCVF. Modéliser le changement: les voies du français. Ottawa: Université d’Ottawa, 2009. – URL http://www.voies.uottawa.ca/corpus_pg_ fr.html. Mathieu, Eric: The left periphery in Old French. In: o appear in a volume on Old French. Springer, 2012. Matras, Yaron: Language contact. Cambridge: Cambridge University Press, 2009 (Cambridge textbooks in linguistics). – ISBN 9780521532211. Matras, Yaron; Sakel, Jeanette: Grammatical Borrowing in Cross-Linguistic Perspective. Berlin: Mouton de Gruyter, 2007. – ISBN 9783110196283. Morris, Richard (ed.): The Early English Text Society 23. Vol. I: Dan Michel’s Ayenbite of Inwyt or Remorse of Conscience. London: N. Trübner und Co., 1866. Muller, Claude: Major constituent order, information packaging, and narrative structure. In: Dufter, Andreas (ed.); Jacob, Daniel (ed.): Focus and Background in Romance Languages. Amsterdam, Philadelphia: Benjamins, 2009 (Studies in Language Companion Series), p. 239–279. Muysken, Pieter: Generative Studies on Creole Languages. Dordrecht: Foris, 1981. – ISBN 9070176106. Muysken, Pieter: Scenarios for Lanugage Contact. In: Hickey, Raymond (ed.): The Handbook of Language Contact. Malden, Mass.: Wiley-Blackwell, 2010, p. 265–281. – URL http://books. google.de/books?id=xGZedef70zAC. – ISBN 9781405175807. Noyer, Rolf: Generative metrics and Old French octosyllabic verse. In: Language Variation and Change (2002), p. 119–171. Palsgrave, Jean: L’éclaircissement de la langue française. Suivi de la grammaire de Giles de Guez. Publiés pour la première fois en France, par F. Génin. Paris: Imprimerie nationale, 1852. Pintzuk, Susan: Phrase Structures in Competition: Variation and Change in Old English. New York: Garland, 1999. Priestley, Lawrence: Reprise constructions in French. In: Archivum Linguisticum 7(1955), p. 1–28. Prince, Ellen: Toward a taxonomy of given/new information. In: Cole, Peter (ed.): Radical pragmatics. New York: Academic Press, 1981, p. 223–254. – ISBN 0121796604. Prince, Ellen: The ZPG letter: Subjects, definiteness, and information-status. In: Mann, William (ed.); Thompson, Sandra (ed.): Discourse Description: Diverse Linguistic Analyses of a Fund-Raising Text. Amsterdam: Benjamins, 1992, p. 295–325. – ISBN 1556192827. Prévost, Sophie: Détachement et topicalisation: des niveaux d’analyse différents. In: Cahiers de praxématique 40 (2003), p. 97–126.

186 

 Achim Stein and Carola Trips

Prévost, Sophie: Topicalisation, focalisation et constructions syntaxiques en français médiéval: des relations complexes. In: D. Apothéloz, B. Combettes et F. N. (ed.): Les linguistiques du détachement, actes du colloque international de Nancy. Bern: Peter Lang, 2009, p. 427–439. Rinke, Esther; Meisel, Jürgen: Subject-inversion in Old French: Syntax and information structure. In: Kaiser, Georg & Remberger, Eva-Maria (ed.): Proceedings of the Workshop ‘Null-subjects, expletives, and locatives in Romance’. Konstanz: Fachbereich ­Sprachwissenschaft, 2009 (Arbeitspapiere Fachbereich Sprachwissenschaft 123), p. 93–130. – URL http:// nbn-resolving.de/urn:nbn:de:bsz:352-opus-78604. Roberts, Ian: Verbs and Diachronic Syntax. A Comparative History of English and French. Dordrecht: Kluwer, 1993. Rooth, Mats: Association with Focus. Amherst: PhD Thesis, UMass, 1985. Rothwell, William: Language and government in medieval England. In: Zeitschrift für französische Sprache und Literatur (1983), No. 93, p. 258–90. Rothwell, William: The ‘Faus franceis d’Angleterre’: later Anglo-Norman. In: Short, Ian (ed.): Anglo-Norman Anniverary Essays. 1993 (Anglo-Norman Text Society Occasional Publications), p. 310. Rothwell, William (ed.); Trotter, David (ed.): Anglo-Norman Dictionary 2. Online Version. London: MHR, 2005. Schendl, Herbert (ed.); Wright, Laura (ed.): Code-switching in early English. Berlin: Mouton de Gruyter, 2011. – ISBN 9783110253351. Stein, Achim; Trips, Carola: Cleft Sentences in the History of French and English: A Critical Review of Compensation and Contact Theories. Conference given at Going Romance, Leuven, December 8, 2012. – URL http://www.uni-stuttgart.de/lingrom/stein/downloads/ stein-trips2012_goingromance-slides.pdf. Taylor, Ann et al.: The York-Toronto-Helsinki Parsed Corpus of Old English Prose (YCOE). Heslington, York: University of York, 2003. Thomason, Sarah G.; Kaufman, Terrence: Language Contact, Creolization, and Genetic Linguistics. Berkeley: University of California Press, 1988. Traugott, Elizabeth C.: Old English left-dislocations: Their structure and information status. In: Folia Linguistica (2007), No. 41/3-4, p. 405–441. Trips, Carola: From OV to VO in Early Middle English. Amsterdam/Philadelphia: Benjamins, 2002. Trotter, David (ed.): Multilingualism in Later Medieval Britain. Cambridge: D.S. Brewer, 2000. Vance, Barbara: Syntactic Change in Medieval French. Verb-second and null subjects. Dordrecht: Kluwer, 1997. Weinreich, Uriel: Languages in Contact. Findings and Problems. Reprint 1968. The Hague, Paris, New York: Mouton, 1953. Wright, Laura: A hypothesis on the structure of macaronic business writing. In: Fisiak, Jacek (ed.): Medieval Dialectology. Berlin: Mouton de Gruyter, 1995 (Trends in Linguistics Studies and Monographs 79), p. 309–321. Wright, Laura: Models of language mixing: Code-switching versus semicommunication in medieval Latin and Middle English accounts. In: Kastovsky, Dieter (ed.); Mettinger, Arthur (ed.): Language contact in the history of English Vol. 1. Frankfurt/Main: Peter Lang, 2003, p. 363–377.

Anna Volodina and Helmut Weiß

Diachronic Development of Null Subjects in German1 1 Introduction Null subjects have been a central research topic in generative syntax from the 1980s (e.g., Rizzi 1982, 1986; Huang 1984; Jaeggli & Safir 1989) until nowadays (e.g., Biberauer et al. 2010; Camacho 2013). In many languages, pronominal subjects can be phonological null. However there are different forms of null subjects, so that not each occurrence of a null subject (NS) is the same. A traditional distinction is the one between pro drop and topic drop. Null subjects in a narrow sense (i.e., pro) are related to agreement and occur in languages like Italian or Greek, which possess rich verbal inflection. In these languages, pro is licensed structurally (e.g., via c-­command by I° in consistent null subject languages, Camacho 2013: 68, or by C° in certain German dialects, Weiß 2005a), although there is much debate on the precise mechanisms (Camacho 2013: ch. 4). Topic drop languages, on the other hand, do not require verbal agreement to identify the content of null subjects. In a somewhat loose sense, null subjects in the sense of null topics have to be identified by an antecedent that is present in the discourse. Prototypically, the antecedent is the topic of the preceding sentence (topic drop in a narrow sense). Another form of context-linked subject drop appears in certain coordinated sentences where both conjuncts share an identical subject which has to be a null subject in the second conjunct.2 We will call this type antecedentlinked subject drop. Diary drop (i.e., the drop of first or second personal pronouns) may form a third type of context-linked null subjects. In German, there is a very clear syntactic difference between ­context-linked null subjects and pro drop, because the first type of null subjects is restricted to the prefield position

1 We would like to thank the conference audience for providing useful comments and the organizers for organizing this highly interesting and stimulating workshop. We are grateful to two anonymous reviewers, who provided insightful feedback. Many thanks to Maria Theresa Distler, Bianca Hufnagel, Anja Schenk, and Saskia Schmadel for their help. 2 Note that we will only consider coordinative structures that are no instances of conjunction ­reduction, i.e., coordinated sentences where the subject of the first conjunct is not in SpecCP (as the null subject in the second conjunct is).

188 

 Anna Volodina and Helmut Weiß

(i.e., precedes C°), whereas pro always follows C°. For that reason, we will strictly distinguish between these two types of NS (although pro is topic oriented as well, Camacho 2013). There is a discussion in literature as to whether old Germanic languages allow (referential) pro drop or not. For Old Icelandic, Sigurðsson (1993) argues that pronominal null subjects are generally licensed through coindexing with some discourse antecedent, i.e., null subjects are licensed at the level of ­discourse semantics. Similar views are found for Old High German (OHG) in traditional literature (Eggenberger 1961, Schrodt 2004, Schlachter 2010). In contrast, Axel (2007) argues for the view that genuine pro drop existed in OHG and pro was licensed syntactically. Based on new corpus research, we reevaluate the diachronic development of null subjects in German. In particular, we will discuss the time span from MHG to early NHG, which has never before been analyzed. Our hypothesis is that null subjects in German cannot be totally accounted for by the notion of topic drop. Instead, we argue that German is diachronically (and to a certain degree even synchronically) also a pro drop language. This discussion requires a detailed analysis of texts from different historical stages of German. Of all historical stages, it is only OHG in which null subjects have been studied extensively. For this stage, there are some recent theoretical and empirical studies (Axel 2005, 2007; Axel & Weiß 2010, 2011; Schlachter 2010). For the analysis of MHG and ENHG, however, reliable data is still missing. The only detailed empirical work that concerns null subjects is Held (1903); however, this study does not correspond to current methodical and theoretical standards. At the Institute for Historical Linguistics at the Goethe University in Frankfurt, in cooperation with the IDS Mannheim, a new, large-scale corpus study is being prepared that should shed light on the distribution of null subjects in MHG and ENHG. The first pilot studies from this project have validated the hypotheses of Axel (2007) and Axel & Weiß (2010) (Volodina 2009, 2011). Thereby, it has been observed that for the study of null subjects in ENHG both prose (Denkwürdigkeiten der Helene Kottanerin: Bonner Frühneuhochdeutschkorpus, http://www.korpora.org/fnhd/, Text 113) and verse texts (Fastnachtspiele von Hans Sachs) are highly useful for the study of null subjects (see the corpus results presented in sections 3.1 and 3.2). The chapter is organized as follows: section 2 introduces the types of null subjects attested in the history of German; section 3 presents and analyzes new corpus data concerning antecedent-linked null subjects (section 3.1) and pro drop (section 3.2); in the concluding section (section 4), we will address some more questions regarding the status of German as a (non-)null-subject language in general.



Diachronic Development of Null Subjects in German 

 189

2 Types of null subjects Our historical data shows that in German, there were at least two types of null ­subjects at all times: context-linked null subjects (topic drop, diary drop and ­antecedent-linked subject drop in coordinative structures) and pro drop in the narrower sense. Apart from these cases (a-b), there are also further types of subject omissions (c-e) that are hard to analyze either as pro drop or as contextlinked null subjects. a. context-linked null subjects: In sentence initial position (SpecCP), topical elements in question-answer pairs (topic drop, 1a) and subjects in coordinative structures (1b) can be elided. Such structures are already found in OHG (1a, b): thu … thiu minu wórt ellu?”OHG (1) a. “Gilóubist Believe you … the my words all? “[e] sint druhtin” quad si, [e] are Lord says she, “festi in mines hérzen brusti …” fast in my heart’s breast  (O III 24,33, from Axel 2007: 308) ‘Do you believe all of my words?’ ‘(They) are, Lord’, says she, ‘fast in my heart.’ b. denne uuarant engila uper dio marha,OHG then go angels over the country [e] uuechant deota, [e] uuissant ze dinge [e] wake up people, [e] point to court  (Muspilli 79f.) ‘Then the angels go out over the country, wake up the people and point to the place of judgement.’ Diary drop differs from topic drop and antecedent-linked subject drop in coordinative structures in that it is constrained to person ellipsis without any requirement for anaphoric antecedents, which explains the morphological constraints (1/2 SG/PL). (2) Wir-1PL haben auch 12 Schwein We have also twelve pigs In der Mast gehabt. In the fattening had. [e]-1PL haben ein Schwein davon [e] have a pig thereof

ENHG

190 

 Anna Volodina and Helmut Weiß

Geschlacht den 25ten Novemberis. Slaughtered the 25th November.  (Preis 1636–1667, from Volodina 2009: 60) ‘We have also fattened 12 pigs. (We) have slaughtered one of these pigs on the 25th of November.’ b. pro drop: Apart from sentence initial null subjects, sentence internal null subjects (= pro drop) are also found. pro is licensed in the middle field (Mittelfeld) with a c-command relation to AGR. This is the result of verb movement to C°. pro drop can be found from OHG to early NHG (Axel 2007; Axel & Weiß 2010, 2011; Volodina 2009, 2011), but it is no longer possible in present-day NHG. an uns beiden (3) daz koufest [e] That purchase [e] at us both ‘(You) purchase that at both of us.’

MHG (AH662)

In both cases (a and b), there is a surprising diachronic continuity, but at the same time, striking changes can also be observed (see section 3 for more details). c. null subjects in embedded V2 clauses: In the case of context-linked null ­s ubjects, the prefield is empty, which is probably the place of the null subject. (4) er chot [e] wollti sizzin nordinMHG He said [e] wanted to live north  (Summa Theolog. 49) ‘He said (he) wanted to reside in the north’ d. null subjects in embedded dass- (5a) or ob-clauses (5b) and ­verb-final-clauses without dass (6): the null subject can only be clause internal in clauses introduced by a complementizer, which probably extends to structures without complementizer as well. This makes these cases candidates for pro, despite the absence of an AGR-in-C licenser in these cases. deme sin gewissede daz sagete MHG (5) a. Whom his conscience that said daz [e] gotes hulde niene habet that [e] God’s grace not had  (Ava, JG9,7) ‘whose conscience told him, that (he) didn’t have God’s grace’ b. ich solt versuechen, ENHG I should try,



Diachronic Development of Null Subjects in German 

 191

ob [e] ier kran mocht hinab zu ier bringen if [e] her crown could down to her bring  (Kottanerin, 13,13, from Volodina 2011: 281) ‘I should try, if (I) could bring her crown down to her.’ (6) sie sprachen [e] iz gerne taetenMHG They said [e] it gladly would do  (Diemer 133,11) ‘They said (they) would do it gladly’ e. null subjects in asymmetrical coordination: verb-final-clauses can also appear as second conjuncts in matrix-clause-coordination, exhibiting a null subject at the same time. (7) datzu hab ich allzeyt die weysse gehabt ENHG To.this have I always the disposition had Und [e] fortan haben will and [e] henceforth have will  (Luther, from Ebert et al.1993: 346) ‘I have always had the disposition for this and (I) will have it henceforth.’ In the next section we will discuss only two of the six types of null subjects presented above in greater detail, namely antecedent-linked null subjects and pro drop. The other types, although they are highly relevant for an investigation of the development of null subjects in German, will not be the focus of this chapter because they would require extensive corpus studies which we have not yet done. The data available at the moment are derived from Held (1903) and would not be sufficient for a quantitative and/or qualitative analysis which fulfills modern standards.

3 Continuity and change The first question to be discussed in this section is whether instances of antecendent-linked subject drop, which we find in OHG/MHG/ENHG, are really subject to the same conditions as in NHG. The second question to be discussed in section 3.2 is: What happened in terms of pro drop, or, put differently, has pro drop actually disappeared?3

3 Due to the lack of systematic corpus research we cannot say much about the special cases c and e; with reference to d, we will present new material from ENHG in section 3.2.

192 

 Anna Volodina and Helmut Weiß

3.1 Continuity and change of antecedent-linked subject drop Antecedent-linked subject drop adheres to the following constraint: as anaphors (in the original sense), they undergo an anaphoric relationship with the ­structurally highest referential argument of the preceding clause, which establishes their referent (Reich 2009a, 2009b). Therefore, a complete match of the φ-features between antecedent and null subject in NHG is a strict condition for subject drop in the second conjunct (te Velde 1999). According to te Velde (1999), this restriction goes back to the Coordinate Feature Matching (CFM) principle, which states that given a certain similarity of features between the subjects (or objects) of two conjuncts, the second one can be ‘erased’. In older stages of German, this matching requirement was obviously much weaker. For instance, in OHG, the contrary condition seemed to hold, since object control was the usual case. This is at least the result of Held‘s ­investigation: „Größer ist die Freiheit der Auslassung, wenn die mit einander verbundenen Sätze nicht das gleiche Subjekt haben. In der Regel ist dann das Subjekt des zweiten Satzes in einem casus obliquus des ersten enthalten“



(Held 1903: 33) There is a greater freedom of omission whenever the two sentences have different subjects. Generally, the subject of the second sentence is in the oblique case in the first sentence.

This is confirmed by a small empirical investigation which we have conducted. The text we analyzed is the autobiography of Helene Kottanerin, the maid of ­Elisabeth, the queen of Hungaria and Bohemia in the first half of the 15th century. In this text, we found 41 subject gaps whose antecedent is the subject of the preceding sentence. In this condition, subject drop is also possible in present-day German. In 19 cases, we observe a mismatch in number between the two subjects. In a further 13 cases of subject gaps, the case feature does not match. Both of these latter cases would lead to ungrammaticality in present day NHG as will be shown below.4

4 Results based on the text: “Helene Kottanerin, Denkwürdigkeiten, Wien 1445–1452” (Bonner Frühneuhochdeutschkorpus Text 113).

Diachronic Development of Null Subjects in German 



 193

45 40 35 30 25 20 15 10 5 0

Case mismatch

Number mismatch

Without change

Figure 1: Three types of antecedent-linked subject drop

Number mismatch: In historical examples, a mismatch of the number ­features is documented very often. Number mismatch occurs in two different forms: (i) with split antecedent and (ii) partial antecedent. (i) is the case when either subject and object of the first conjunct (8a) or the subjects of the two preceding sentences (8b) constitute the combined antecedent for the subject gap. (8) a. same tet adami-sg   joch sîn wîbj-sg lussam, [ei+j-PL] muosen mit So did3sg Adam   und his wife agreeably, (they)   had.to with armuote lîden ire nôt poverty suffer their misery  (Gen 597f) ‘So did Adam and his wife agreeably, (they) had to suffer their misery in poverty’ b. do stuend meiner fraun gnadi-sg auf, vnd ichj-sg nam ain Then stood my mistress’ grace up, and I took a Windtliecht, vnd [ei+j-PL] giengen mitenander auf dy maur lantern and went together up the wall  (Kottanerin 23,12) ‘Then my mistress stood up and I took a lantern and we went together onto the wall.’

194 

 Anna Volodina and Helmut Weiß

A type of partial antecedent occurs in terms of the so-called constructio ad sensum, i.e., when a subject which is grammatically singular but plural in meaning constitutes the antecedent for a plural null subject (8c).5 This becomes even clearer in the example (8d): the antecedent is grammatically a singular noun (volk) that denotes a plurality – and the subject gap in the second conjunct agrees with a verb in the plural – that is, it realizes the plural meaning: (8) c. dô saz dar manich nakit mani-sg There sat there many a naked man unde vil Sêre [ei-PL] schametin sich and were ashamed of themselves very much  (König Rother 1362) ‘There sat many naked men and (they) were very much ashamed.’ d. do gieng [...] ein michel volkchi-sg mit von fraun vnd von There went a large crowd with of women and of manen, dem edelen kung entgegen vnd [ei-PL] enphiengen In men the noble king toward and received him  (Kottanerin 34, 39) ‘There a large group of men and women went to meet the noble king, and they received him.’ Case mismatch: As mentioned already, not (only) structural aspects are important for the choice of the antecedent with a null subject up until ENHG because it is not only the subject of the first conjunct that can function as antecedent. Whether the object of the first conjunct can license the null subject in the second conjunct probably also depends on the type of verb, which is indicated in (9): (9) a. des-GEN nam die herreni-ACC wunder vnd [ei-SUBJ] rieten [...] That took the Lords wonder and advised  (Kottanerin 11, 24) ‘That made the Lords wonder and they counselled’ b. der rat-NOM der geviel iren gnadeni-DAT wol, vnd [ei-SUBJ] sprach. The advice that pleased her ladyship well, and said.  (Kottanerin 30, 19) ‘The advice pleased her ladyship very much and she said...’

5 An anonymous reviewer pointed out to us that such constructions are more often found in less restricted spoken texts than in written language, which may be an effect of processing issues (Steiner 2009). The fact that the texts under discussion seem closer to spoken texts may explain the wider distribution of constructio ad sensum structures in ENHG.



Diachronic Development of Null Subjects in German 

 195

The ergative verbs in (9) occur with two arguments: a thematic subject (des in (9a) or der rat in (9b)) as well as an accusative or dative experiencer argument. Experiencer arguments like these are referred to as oblique subjects (Barðal 2004; Barðal & Eythórsson 2008; Eythórsson & Barðal 2005). Since objects can also serve as antecedent with transitive verbs – like ‘to bring s.o.’ (9c), where the direct object In ‘him’ serves as antecedent of the null subject in the second conjunct – only corpus research with statistical analysis can indicate a possible diacronic change. (9) c. Da liesz In-ACC der truchsesse-NOM auf fueren also There let him (= the king) the steward up bring as krankchen vnd [e = the king]-SUBJ kam zu der Plintenpurg invalid and (the king) came to the Plintenburg  (Kottanerin 10, 10-11) ‘Then the steward had the ill king brought and (the king) came to the Plintenburg.’ Summing up, we arrive at the following conclusion: as was shown in the examples with mismatch in number and case, a complete match of features between the dropped subject and the antecedent was not necessary until ENHG. As opposed to NHG (10a), for earlier stages of German, we assume the structure in (10b), i.e., the anaphoric relationship between subject gap and antecedent is not yet strictly structural. This is the reason why a subject gap was not blocked when the subjects differed between the conjuncts: (10) a. [ … αi … βj … ] [ ei/*j [C’… ]] b. [ … αi … βj … ] [ ei/j [C’… ]]

NHG OHG/MHG/ENHG

One question yet to be answered concerns the extent of the possible incongruence of φ-features. The working hypothesis is that some features allow or tolerate incongruence more than others. This must be analyzed using larger diachronic corpora in further research.

3.2 Continuity and change in pro drop As Axel (2005, 2007) has shown, OHG was a partial null subject language (NSL). New corpus data from ENHG (Volodina 2009, 2011) show striking parallels to OHG, in particular, that null subjects are constrained to matrix clauses, i.e., the structural licensing conditions are the same as in OHG. In contrast to that, at least referential pro drop is no more possible in NHG. Thus, NHG is commonly held to be a non-NSL (Roberts 2007).

196 

 Anna Volodina and Helmut Weiß

However, that holds only for the standard variety of German. In German dialects, the situation is different: referential pro drop is attested in many dialects at least for the 2.SG, so it is not really the case that pro drop has completely disappeared from German. Bavarian (see Bayer 1984; Weiß 1998, 2005a) is an example of such a pro drop dialect where, in contrast to OHG, pro also occurs in non-root sentences (11b). (11) a. morng bist [pro] wieda gsund(Bavarian) tomorrow are-2sg again healthy ‘Tomorrow you’ll be healthy again.’ b. wennst [pro]   moanst if-2sg           mean-2sg ‘if you think so’ There is a second difference compared to OHG: the modern dialects do not allow pro drop in all persons. It is most widespread in the 2.SG: According to Fleischer and Schallert (2011: 204f.) the Wenker sentence in (12a) was often translated without the subject pronoun by speakers of High German dialects:6 (12) a. Wo gehst du hin? Where go you to? b. Wo gasch hi? c. wo göscht hin? ‘Where do you go to?’

(High Alemannic) (South Bavarian)

In many dialects, pro is only allowed in the 2.SG, but not in all (see below). Modern German dialects are thus partial pro drop languages, but in a different sense to OHG: in OHG, pro and (weak) pronouns were more or less pragmatically equivalent, but pro could appear in all persons (Axel 2005, 2007), whereas in modern dialects pro is restricted to some persons and it contrasts with its nonnull pendant on the pragmatic level (since the latter has an emphatic interpretation). OHG and modern German dialects are thus both untypical NSL, but in different ways: that null and overt pronouns obviously were in free alternation in OHG is as untypical for canonical pro drop languages of the Italian type as the restriction of pro to certain persons in modern German dialects.7

6 The Wenker material (http://www.diwa.info/Geschichte/Fragebogen.aspx) represents the ­dialect situation in the German ‘Kaiserreich’ at the end of the 19th and the beginning of the 20th century. 7 There are several kinds of non-canonical, non-consistent or semi-NSL, see Biberauer et al. (2010) for extensive documentation and discussion.



Diachronic Development of Null Subjects in German 

 197

So we can observe that pro drop is attested throughout the history of German, if we take into account the regional varieties of German as well. One crucial ­question is whether pro drop in present-day dialects is a direct continuation of OHG pro drop or whether it represents a new development. Axel & Weiß (2010, 2011) tried to reconstruct the development as a continuous one (for a discontinuous account, cf. Rosenkvist 2009). In particular, the account of Axel & Weiß attempts to explain why the possibility of pro drop became at the same time confined and expanded in German dialects. They explain the expansion of pro drop to non-root sentences with the independent development of an additional property, namely complementizer agreement: in (13), the complementizer wenn is inflected (in addition to the verb). The consequence is that pro is c-commanded by Agr-in-C both in root and embedded sentences. (13) wennst [pro] moanst if-2sg      mean-2sg Axel & Weiß (2010, 2011) then conclude that with reference to the structural licensing condition for pro, nothing has changed since OHG. The restriction to certain persons (mostly the 2.SG) is explained as a kind of strengthening of the morphological requirement: not every kind of Agr-in-C can now license pro. Compare the contrast in (14a vs. b), as reported by Weise (1900) for Thuringian: (14) a. schreib mir einmal den Brief, write-2sg me once the letter kriegst [pro] auch einen Groschen get-2sg too a penny ‘Write the letter to me sometime, (you) will get a penny.’ b. schreibt mir einmal den Brief, write-2pl me once the letter *kriegt [pro] auch einen Groschen get-2pl too a penny ‘Write the letter to me sometime, (you) will get a penny.’  Thuringian (Altenburg) Thuringian allows pro drop in the 2.SG, but not in the 2.PL. Why is that so? The relevant factor seems to be the difference in agreement morphology, since only the inflectional marker of the 2.SG is derived from the respective subject pronoun, which was re-analyzed as part of the verbal inflection (Weiß 2005a). In many German dialects, there is a strict correlation between this kind of ‘pronominal’

198 

 Anna Volodina and Helmut Weiß

agreement and the possibility of pro drop.8 Central Bavarian may be an extreme example in that it displays null subjects in three persons (i.e., 2.SG, 1.PL, 2.PL) (15, 16), but it is definitely not exceptional (Weiß 2005a). partial pro drop (15) a. *morng bin [pro] wieda gsund tomorrow am-1sg again healthy ‘Tomorrow I will be healthy again.’ b. morng bist [pro] wieda gsund tomorrow are-2sg again healthy c. *morng is [pro] wieda gsund tomorrow is-3sg again healthy d. morng sama [pro] wieda gsund tomorrow are-1pl again healthy e. morng sads [pro] wieda gsund tomorrow are-2pl again healthy f. *morng san [pro] wieda gsund tomorrow are-3pl again healthy Central Bavarian ü (CA is pronominal) (16) 2SG – 1PL – 2PL 1SG – 3SG – 3PL  (CA is non-pronominal) The licensing conditions are thus morphologically strengthened: pro drop is possible, if pronominal Agr-in-C c-commands pro. Based on Axel & Weiß’s account, we can draw the following preliminary conclusions: – From OHG to the recent dialects, the licensing conditions for pro became more specific – despite the expansion to embedded sentences. – This restriction had presumably already emerged in OHG times, somewhere between Otfrid and Notker (Axel & Weiß 2010, 2011) with the consequence that pro drop first significantly decreased. – The trigger for this development was presumably the emergence of double agreement (special inflection on verbs in C) (Axel & Weiß 2010, 2011).

8 Alemannic varieties may be an exception from this generalization in that they allow null subjects in non-root sentences without any complementizer agreement, but see Weiß (2005a) for an analysis that attempts to explain these cases along the lines of the correlation.



Diachronic Development of Null Subjects in German 

 199

Axel & Weiß (2010, 2011) based their account almost exclusively on data from OHG and from modern dialects, since there is no recent systematic research for the periods in between. The knowledge of null subjects in MHG and ENHG is still based on the empirical study done by Held (1903) more than one hundred years ago (Fleischer & Schallert 2011: 202). Due to the lack of more recent research, it is hard to say whether any null subjects occur at all in MHG and ENHG, and if so, whether they pattern more with the OHG type or with the dialectal one. In the following, we will report first findings from a research project that we intend to carry out in the future, but for which we have already conducted some preliminary studies. We will restrict ourselves to the 2.SG. Given that the crucial development (i.e., from non-pronominal to pronominal Agr-in-C) can be dated back to late OHG times, one would expect that MHG and ENHG (i) were already partial pro drop languages (ii) in which pro was licensed by pronominal Agr-in-C If this is correct, we can take this expectation as a working hypothesis for investigations of MHG and ENHG texts. Additionally, we can exploit the observation that pro drop occurs in the 2SG in many (if not most) German dialects (goups): (17) a. wennsd [pro] mogsd(Bavarian) if-2g will-2sg b. kriegst [pro] auch einen Groschen(Thuringian) get-2sg too a penny c. datst [pro] jûn komst (Frisian) that-2sg tomorrow come-2sg d. dat maakst [pro] recht (Low German, from Appel 2007: 49) that make-2sg right The working hypothesis would then be: – There is an increased probability for the occurrence in 2SG. Before we report our findings, a methodological caveat is indicated: since historical linguists only have access to written documents and we know that written languages never reflect the native competence of the spoken language in a one-to-one fashion (Weiß 2005b), we should not be too surprised if we observe changes in the written language only with some delay. This means that MHG texts may still pattern with OHG with reference to null subjects, although the spoken language at that time already displayed the ‘modern’ correlation between pronominal agreement and pro drop. This is one reason why we decided to start with ENHG texts – and additionally with texts that were ‘conceptually spoken’, i.e., to represent spoken

200 

 Anna Volodina and Helmut Weiß

language. This is for example the case with Hans Sachs, a Nurembergian poet from the 16th century. In his carnival play, Der Schwangere Bauer (‘The pregnant farmer’) we find many instances of pro drop with the 2.SG. This is predominantly the case with main clauses where the verb is in second position and the prefield, as in (18), is filled with a non-subject or empty. (18) a. Daran must etlich unkost wenden that-on must-2sg some expenses apply ‘You have to spend some money on this.’ b. mein Kargas! wie sichst so schmal my K.! How look-sg so thin ‘My K.! How thin you look!’ c. Hast im sein grosse würst helffen essen? Have him his big sausages help eat ‘Have you helped him to eat his big sausages?’ Interestingly, the most frequent option Hans Sachs selects in these inversion contexts with the 2SG seems to be pro drop, which occurs 60 times – this is the same number as for the use of full (41 times) and clitic pronouns (19 times) taken together.9

60 50 40 30 20 10 0

bist du

bistu

bist

Figure 2: Distribution of full/clitic/null pronouns

9 The numbers in the table are taken from an investigation by a student of ours (B. Hufnagel), who has evaluated 11 carnival plays of Hans Sachs.



Diachronic Development of Null Subjects in German 

 201

Therefore, in the 2.SG pro drop is well attested for root sentences. But did it already occur in complementizer-introduced embedded sentences? We have seen above that in modern dialects the possibility of pro drop in embedded sentences depends on the presence of pronominal inflection on the complementizer. According to Held (1903: 62), null subjects in that-clauses (as in 5a above) occurred frequently in MHG without any restrictions with respect to person. In ENHG, null subjects also occurred in other kinds of complementizer-introduced clauses (e.g., in relative clauses), but mostly with 2.SG subjects (Held 1903: 132). This restriction to the 2.SG is further evidence that pro was probably licensed by pronominal agreement in ENHG, but what is lacking so far is clear evidence for inflected complementizers at that time. However, Hans Sachs uses forms of the complementizer – e.g., dast ‘that-t’ or weilt ‘because-t’ – which look like inflected complementizers. (19) a. Dast uns so weit fuerst aus der stat that-2sg us so far lead out the town ‘That you lead us so far out of the town.’ (Sachs, FNS 244 V.152) b. dast in habst verlorn that-2sg him have lost ‘That you have lost him’ (Sachs, FNS 172 V.127) c. dast ein zygeuner seist that-2sg a gipsy are ‘That you are a gipsy’ (Sachs, ML 126 V.211) d. weilt mir warst versprochen because-2sg me were promissed ‘because you were promised to me’ (Sachs, ML 109 V.126) Note that the combination of the complementizer dass with the enclitic form of du often results in dastu, which is very frequently attested at that time.10 The oldest unambiguous example of an inflected complementizer known to us is from 1618 and can be found in the carnival play Ein kurtzweilig Faßnachtspill (Weiß 2005), (20): the form wanst ‘if-2.SG’ is completely identical with the inflected complementizer used in present-day Bavarian.

10 Although obviously not in the investigated texts of Hans Sachs, but in texts of contemporary writers like Ulrich von Hutten, (i). ( i) Du Tyrann dencke dastu mir mehr gebest dan die andern You tyrant think that-you me more give than the others  ‘You tyrant think that you give more to me than the others.’

202 

 Anna Volodina and Helmut Weiß

(20) Ja! wanst mir wolst var liegn [vorlügen] Yes, if-2sg me want-2sg for-lie ‘Yes, if you want to lie to me’ Interestingly, Hans Sachs also applies pro drop in that-sentences without complementizer inflection. (21) a. wie, das so fröhlich bist how, that so cheerful are-2sg b. wie, das so bald … heut kommest aus dem wald how, that so soon … today come-2sg out the wood Whether similar forms were also used by other authors and to what extent must be left to future research. It is reminiscent of pro drop in Alemannic varieties, where the subject pronoun of the 2.SG can be dropped in cases without complemetizer agreement (although in Alemannic, this is clearly a recent development, Weiß 2005a).

3.3 Conclusion In the preceding subsections, we presented new corpus data concerning antecedent-linked subject drop and pro drop. Concerning antecedent-linked null subjects, we have seen some differences to modern German, because it seems to have been the case that in earlier periods of German it was not necessary for antecedent and null subject to share all morpho-syntactic features. With regard to pro drop, our new findings confirm the hypothesis presented by Axel & Weiß (2010, 2011) that pro drop is a constant property of German and the differences between OHG and the modern dialects are the result of minor and/ or independent changes (as, for instance, the development of complementizer agreement which enabled pro to occur also in complementizer-introduced sentences).

4 Conclusion and further issues In this concluding section, we will address some issues that arose from the reconsideration of the known data and from the new data we presented in the previous section and that we think to be open questions that have to be answered by future research. We will discuss one general theoretical



Diachronic Development of Null Subjects in German 

 203

­ uestion (I) and several null-subject specific (II) and construction-specifc q (III) questions: (I)  What is the status of German with regard to the null-subject property? The general question is to what extent and in what sense German is/was a NSL (pro drop, semi pro drop, topic drop). To answer this question, we need an empirically solid and theoretically sound classification of null subjects, since one can assume that null subjects are not a homogenous phenomenon. It is beyond doubt that null subjects have very different syntactic/structural, referential, and pragmatic properties (e.g., Sprouse & Vance 1999; Fuß & Wratil 2013). Moreover, it is plausible to assume that even in one and the same language there are heterogenic properties both synchronically and also diachronically. (II) What are the diachronic changes in the syntactic/structural and pragmatic licensing conditions for different types of null subjects? The specific questions relate to the development of pro drop in the narrow sense and its relation to the V2 property, but also about the factors that play a role in the selection of the proper antecedent for topic drop. We would also like to discuss whether different types of null subjects can have one unique licensing condition – such as the empty left edge condition (ELEC) of Sigurðsson & Maling (2008), which says that an empty left periphery guarantees context-linking, which can be interpreted either as topic-linking or antecendent-linking (and speaker-hearer linking). German dialects obviously pose a problem for such accounts: pro is sentenceinternal and the left periphery need not be empty (see 11a above). (III) What can be said about the special cases of null subjects (+/–pro, +/– antecendent-linked etc.) and their licensing conditions? In special cases, i.e., in embedded verb-final clauses with or without complementizers or in V/E verb-final matrix clauses, the question is whether we have instances of pro drop or antecendent-linked null subjects, and what the exact licensing conditions are.

References Appel, Heinz-Wilfried. 2007. Untersuchungen zur Syntax niederdeutscher Dialekte. Frankfurt a.M. u.a.: Lang. Axel, Katrin. 2005. Null subjects and verb placement in Old High German. In: Stefan Kepser, and Marga Reis (Eds.), Linguistic Evidence. Empirical, Theoretical and Computational Perspective, 27–48. Berlin/New York: Mouton de Gruyter.

204 

 Anna Volodina and Helmut Weiß

Axel, Katrin. 2007. Studies on Old High German Syntax. Left sentence periphery, verb placement and verb-second. Amsterdam/Philadelphia: John Benjamins. Axel, Katrin, & Helmut Weiß. 2010. What changed where? A plea for the re-evaluation of dialectal evidence. In: Anne Breitbarth, Christopher Lucas, Sheila Watts, and David Willis (Eds.), Continuity and Change in Grammar, 13–34. Amsterdam/Philadelphia: Benjamins. Axel, Katrin, & Helmut Weiß. 2011. Pro-drop in the History of German. From Old High German to the modern dialects. In: Peter Gallmann and Melani Wratil (Eds.), Empty Pronouns, 21–52. (= Studies in Generative Grammar 106.) Berlin/New York: Mouton de Gruyter. Barðal, Jóhanna. 2004. The semantics of the impersonal construction in Icelandic, German and Faroese: beyond thematic roles. In: Werner Abraham (Eds.), Topics in Germanic typology, 105–137. Berlin: Akademie Verlag. Barðal, Johanna, & Thórhallur Eythórsson. 2008. The origin of the oblique subject construction: An Indo-European Comparison. In: Vit Bubenik, John Hewson, and Sarah Rose (Eds.), Grammatical Change in Indo-European Languages, 179–193. Amsterdam/Philadelphia: Benjamins. Bayer, Josef. 1984. COMP in Bavarian syntax. The Linguistic Review, 3: 209–274. Biberauer, Theresa, Anders Holmberg, Ian Roberts, & Michelle Sheehan. 2010. Parametric variation: Null subjects in minimalist theory. Cambridge: Cambridge University Press. Camacho, José. 2013. Null Subjects. Cambridge University Press. Ebert, Robert Peter, Oskar Reichmann, and Klaus-Peter Wegera. 1993. Frühneuhochdeutsche Grammatik. Tübingen: Niemeyer. Eggenberger, Jakob. 1961. Das Subjektpronomen im Althochdeutschen. Ein syntaktischer Beitrag zur Frühgeschichte des deutschen Schrifttums. Grabs: Selbstverlag. Eythórsson, Thórhallur, & Jóhanna Barðal. 2005. Oblique subjects: a common Germanic inheritance. Language, 81(4): 824–881. Fleischer, Jürg, & Oliver Schallert. 2011. Historische Syntax des Deutschen. Eine Einführung. Tübingen: Narr. Fuß, Eric, & Melani Wratil. 2013. Der Nullsubjektzyklus. In: Jürg Fleischer, and Horst Simon (Eds.), Sprachwandelvergleich – Comparing Diachronies, 163–196. (Linguistische Arbeiten 550.) Tübingen: Niemeyer. Held, Karl. 1903. Das Verbum ohne pronominales Subjekt in der älteren deutschen Sprache. Berlin (Palaestra 31), Zugl.: Univ. Göttingen, Diss. 1902. Huang, Cheng-Teh James. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry, 15: 531–574. Jaeggli, Osvaldo, & Kenneth Safir. 1989. The null-subject parameter and parametric theory. In: Oswaldo Jaeggli, and Kenneth Safir (Eds.), The Null Subject Parameter, 1–44. Dordrecht: Kluwer. Reich, Ingo. 2009a. Asymmetrische Koordination im Deutschen. Tübingen: Stauffenberg Verlag. Reich, Ingo. 2009b. Ist asymmetrische Koordination wirklich ein Fall asymmetrischer Koordination? In: Veronika Ehrich, Christian Fortmann, Ingo Reich, and Marga Reis, (Eds.), Koordination und Subordination im Deutschen, 203–222. Hamburg: Buske. Rizzi, Luigi. 1982. Issues in Italian Syntax. Dordrecht: Foris. Rizzi, Luigi. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry, 17: 501–557. Roberts, Ian. 2007. Diachronic Syntax. Oxford: OUP. Rosenkvist, Henrik. 2009. Referential null subjects in Germanic languages – an overview. Working Papers in Scandinavian Syntax, 84: 151–180.



Diachronic Development of Null Subjects in German 

 205

Schlachter, Eva. 2010. Syntax und Informationsstruktur im Althochdeutschen. Untersuchungen am Beispiel der Isidor-Gruppe. Ph.D. diss., Humboldt-Universität Berlin. Schrodt, Richard. 2004. Althochdeutsche Grammatik II: Syntax. Tübingen: Niemeyer. Sigurðsson, Halldór Ármann. 1993. Argument-drop in Old Icelandic. Lingua, 89: 247–280. Sigurðsson, Halldór Ármann, & Joan Maling. 2008. Argument drop and the empty left edge condition (ELEC). Working Papers in Scandinavian Syntax, 81: 1–27. Sprouse, Rex, & Barbara Vance. 1999. An explanation for the decline of null pronouns in certain Germanic and Romance languages. In: Michel DeGraff (Ed.), Language Creation and Language Change: Creolization, Diachrony and Development, 256–284. Cambridge, MA: MIT Press. Steiner, Ilona. 2009. Partial agreement in German: A processing issue? In: Susanne Winkler, and Sam Featherston (Eds.), Fruits of Empirical Linguistics II, 239–260. Berlin: de Gruyter. te Velde, John. 1999. Asymmetrische Kordination? Jain! Another look at subject and object gaps in coordinate structures. Gronninger Arbeiten zur germanistischen Linguistik, 43: 155–172. Volodina, Anna. 2009. pro-drop im frühen Neuhochdeutschen. In: Gisela Brandt and Rainer Hünecke (Eds.), Historische Soziolinguistik des Deutschen IX. 51–66. (S.A.G. 449.) Stuttgart. Volodina, Anna. 2011. Null ist nicht gleich Null: Zur diachronen Entwicklung von Nullsubjekten im Deutschen. In: Jörg Riecke (Ed.), Historische Semantik, 269–283. (= Jahrbuch für Germanistische Sprachgeschichte 2.) Berlin/New York: de Gruyter. Weise, Oskar. 1900. Syntax der Altenburger Mundart. Leipzig: Breitkopf und Härtel. Weiß, Helmut. 1998. Syntax des Bairischen. Studien zur Grammatik einer natürlichen Sprache. Tübingen: Niemeyer (Linguistische Arbeiten 391). Weiß, Helmut. 2005a. Inflected Complementizers in Continental West Germanic Dialects. Zeitschrift für Dialektologie und Linguistik, 72(2): 148–166. Weiß, Helmut. 2005b. The double competence hypothesis. On diachronic evidence. In: Stephan Kepser and Marga Reis (Eds.), Linguistic Evidence: Empirical, Theoretical, and Computational Perspectives, 557–575. Berlin/New York: Mouton de Gruyter.

Susanne Winkler, Janina Radó and Marian Gutscher

What Determines ‘Freezing’ Effects in was-für Split Constructions?1

1 Introduction to the freezing-melting puzzle The term was-für (‘what for’) construction refers to the sequence was-für plus N(oun) in German and is usually translated as what kind of DP as in (1). An interesting feature of this construction, first observed by Bennis (1983) and den Besten (1985), is that the wh-element was (‘what’) can be subextracted out of the DP, as in (2). This is called was-für split. (1) [DP1 Was für Bücher] hat er t1 what for books has he ‘What kind of books did he criticize?’

kritisiert? criticized

(2) Was1 hat er[DP t1 what has he

kritisiert? criticized

für for

Bücher] books

In some cases, however, extraction out of was-für constructions leads to marked sequences, as discussed by den Besten (1985), Corver (1991), Diesing (1992), Pafel (1996), Bayer (2004), and Müller (2010), and many others. Müller (2010) provides the German data in (3) and (4) and calls them instances of freezing and melting, respectively. (3) *Was1 haben [DP3 t1 für Bücher][DP2 den Fritz] beeindruckt? what have for booksnom the Fritzacc impressed ‘What kind of books impressed Fritz?’ (4) Was1 haben[DP2 den Fritz][DP3 t1 für Bücher]t2 beeindruckt? what have the Fritzacc for booksnom impressed ‘What kind of books impressed Fritz?’ (Müller 2010: 61 (36))

1 We thank the editors of this volume and the reviewers for their comments and advice. We also thank the organizers and the participants of the Linguistic Evidence conference 2012 for the fruitful discussion. Particular thanks go to Peter Culicover, Jutta Hartmann and Andreas Konietzko for their feedback and discussion. We thank Sara Holler, who contributed to the preparation of the rating study. Any remaining errors are ours. This research was supported by the DFG (SFB833 Bedeutungskonstitution, Projekt A7 Focus Constructions and Freezing).

208 

 Susanne Winkler, Janina Radó and Marian Gutscher

According to Müller, (3) is marked, and in his view ungrammatical, since was-für extraction takes place from a frozen structure, here the last-merged subject in vP. However, in (4), local scrambling removes the freezing configuration. Müller calls this case melting and considers it grammatical.2 Basically, two types of accounts deal with constraints on extractability in was-für constructions. Syntactic approaches investigate the structural configurations that lead to the markedness (for example, den Besten 1985, Diesing 1992, Müller 2010, Jurka 2010), whereas Bayer’s (2004) information-­structural account bars A-bar extraction from topics or topic-like constituents. In this chapter, we adopt an alternative view. We explain the markedness of the putatively frozen structure, such as (3), in terms of a constraint satisfaction processing account. We will argue that the freezing vs. melting opposition is not caused by genuinely syntactical factors, as suggested by Müller (2010), but rather by processing problems. Müller’s classical syntactic account of (3) assumes that a certain grammatical configuration causes the markedness. He updates the freezing principle by Wexler & Culicover (1980: 119) by formulating a general constraint in terms of configuration, rather than extraction, as in (5):3 (5) Freezing Generalization A trace t may not be included in a moved XP (i.e., an XP that binds a trace) if the antecedent of t c-commands XP (Müller 2010: 56). Under (5), instances of was-für split from a last-merged subject in German as in (3) are not allowed: the last-merged specifier in its projection (the vP, see (6) below) forms an island for extraction (Müller 2010: 61). Müller proposes that

2 The limited amount of solid empirical data (see Jurka 2010 and Melchiors 2007 for rating studies) has led to considerable disagreement about the acceptability of different was-für split constructions (cf. Jurka 2010 for an overview). 3 Freezing principle (Wexler & Culicover 1980: 119): a. If the immediate structure of a node in a phrase marker is nonbase, that node is frozen. b. If a node A of a phrase marker is frozen, no node dominated by A may be analyzed by a transformation. The basic idea is that a transformationally created structure that is not compatible with the base phrase structure rules of a language is frozen.



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 209

transitive constructions like (6) are marked in just the same way as are psych verb constructions like (3).4 (6) *[CP Was1 haben [TP [vP [DP3 t1 für Ärzte] [VP [DP2 den Minister] what have for doctorsnom the ministeracc kritisiert?]]]] criticized ‘What kind of doctors criticized the minister?’ Müller’s proposal also predicts that if the definite DP den Minister is scrambled over the subject and adjoined to the highest vP, grammaticality is restored. This is the process Müller calls melting, defined as in (7). (7) Definition of Melting Local scrambling in front of what would otherwise qualify as a last-merged specifier renders the specifier transparent for extraction (Müller 2010: 35). An example and the structural representation of melting is given in (8). (8) [CP Was1 haben [TP [vP [DP2 den Minister]2 [DP3 t1 what have the ministeracc [VP t2 kritisiert?]]]] criticized ‘What kind of doctors criticized the minister?’

für for

Ärzte] doctorsnom

Considerable disagreement in the literature (cf. Diesing 1992, who judges examples like Müller’s freezing case acceptable, and Frey 2004) and limited amount of solid empirical data cast some doubt on Müller’s (2010) description of the freezing vs. melting opposition in terms of grammatical vs. ungrammatical structures. It seems that the judgment disagreements stem from the variability of contextualization,

4 Müller (2010: 61, FN 35) suggests that freezing effects occur not only with psych verb subject extractions like beeindrucken (impress), but also with regular transitive verbs, such as kritisieren (criticize) as in (8). He provides the following grammaticality judgments for the transitive cases: a. Was1 haben [DP2 den Fritz] [DP3t1 für Leute] getroffen? what have the Fritzacc for peoplenom met b. *Was1 haben [DP3t1 für Leute] [DP2 den Fritz] getroffen? what have for peoplenom the Fritzacc met ‘What kind of people met Fritz?’

210 

 Susanne Winkler, Janina Radó and Marian Gutscher

which needs to be accounted for. As we will show, a suitable context can repair the apparent unacceptability of Müller’s freezing structure. This suggests that the rendering of the freezing-melting opposition in terms of different syntactic configurations might serve as a description but not as an explanation of the acceptability differences. Following Culicover & Winkler’s (2011) hypothesis that no grammatical principle in itself yields frozen structures, we put forth the hypothesis in (9): (9) Freezing in was-für split constructions is an artifact of processing complexity which results from the competition between different possible information-structural and prosodic analyses of the structure. Conflicting requirements of constraints render the parser’s choice of one or the other analysis more difficult. Under hypothesis (9), this difficulty increases processing complexity and leads to the impression of unacceptability. To test this hypothesis, we will examine the role of IS, focus and implicit prosody in the processing of was-für structures. Following initial observations by Culicover & Winkler (2011), we conducted a rating study. The results indicate that Müller’s judgments are indeed correct, but the data cannot be fully explained by a grammatical principle like freezing or melting. Rather, conflicting requirements of implicit prosody (among others Fodor 2002a, b) and IS need to be taken into consideration as well. In section 2, we will introduce the constraint satisfaction approach and its application to the was-für sentences. In section 3, we will present the results of our rating study. Section 4 concludes.

2 A constraint satisfaction processing approach to the freezing-melting puzzle Constraint satisfaction theories of sentence processing assume that all possible analyses of a sentence structure are activated at the same time, but with different levels of activation. Activation depends on the support that the different analyses receive from the constraints relevant to processing the sentence. More strongly activated alternatives can lower the activation of weaker interpretations. Eventually, one analysis wins the competition for activation (cf. MacDonald 1994). Constraint satisfaction theories usually assume that processing is easy when one analysis gets much more support than all the alternative analyses, i.e., when most constraints clearly ‘point’ to a certain analysis. When this preference is less clear, processing load is assumed to increase (cf. Pickering & van Gompel 2006: 460). Our proposal is that in the freezing examples in (3) and (6), information-structural



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 211

constraints and requirements on prosodic phrasing lend support to different analyses, rendering the parser’s choice for the correct analysis more difficult and thereby increasing processing load. Following Fodor’s (2002a) implicit prosody hypothesis in (10) we assume that prosody is projected onto a sentence in silent reading. According to Fodor, the human parsing mechanism selects the default prosodic pattern for extraction constructions with unspecified contexts. (10) Implicit prosody hypothesis In silent reading, a default prosodic contour is projected onto the stimulus. Other things being equal, the parser favors the syntactic analysis associated with the most natural (default) prosodic contour for the construction (Fodor 2002a: 113). What does the implicit prosody hypothesis imply for incremental processing of sentences? A constraint-based parser is assumed to process the structure word by word, left to right, and activate different possible structures for the elements parsed so far, depending on possible continuations for the sentence. These possible continuations are determined by the rules of the grammar. Since in constraint satisfaction theories of processing, projection of prosody and syntactic processing take place simultaneously or ‘interactively’ (cf. Pickering & van Gompel 2006: 460), we assume that the prosodic contour is built up incrementally as well. This means that, while processing the syntactic structure word by word, different possible prosodic contours are activated (depending on how the sentence structure and the prosodic contour could continue). Each possible syntactic continuation is paired with the appropriate default prosodic contour. The contour that receives the highest activation, i.e., the most support from all other relevant sources of information, wins. Since processing is interactive, the level of activation of a particular prosodic contour also influences the level of activation of competing syntactic and information-structural analyses. Therefore, it can help determine the informationstructural status of a word that is being processed. In short, syntactic, prosodic, and information-structural sources of information influence sentence processing and they do so interactively. Section 2.1.2 illustrates this process. A problem occurs when the most highly activated prosodic contour later on conflicts with requirements of an information-structural constraint whose application is demanded by syntactic properties of the written input. We believe this to be the case in putatively frozen constructions like (3), repeated as (11) for convenience: (11) *Was1 haben [DP3 t1 für Bücher] [DP2 den Fritz] beeindruckt? what have for booksnom the Fritzacc impressed ‘What kind of books impressed Fritz?’ (Müller 2010: 61 (36))

212 

 Susanne Winkler, Janina Radó and Marian Gutscher

We propose that context manipulation can induce the selection of a non-default contour. More specifically, contrastive focus can repair the construction by supporting one of the contesting analyses. We will elaborate our assumptions in the following sections.

2.1 Parsing was-für constructions 2.1.1 Constraints In this section, we will consider the relevant information-structural constraints. There are in particular three general information-structural default rules which apply to was-für extraction in German. Constraint 1 – PreV-Foc(us): Assign default focus to preverbal position – was first formulated by Höhle (1982) for declarative sentences and subsequently used as a diagnostic for wide focus readings (cf. Reis 1993, Haider & Rosengren 2003, Truckenbrodt 1995, Selkirk 2011, among others). A wide focus reading is found in a contextually unmarked situation, for example, as an answer to What happened? We refer to this type of reading as default prosody and assume that it is projected onto sentences without context presented for silent reading. Default intonation in German and English canonical sentences is determined by the nuclear stress rule (e.g. Chomsky & Halle 1968, Chomsky 1972, Cinque 1993: 245, Halle & Vergnaud 1987: 265, among others). In an unmarked context, default nuclear accent is realized on the most deeply embedded DP in a sentence with canonical word order. There are different renderings of this observation. In the German V-final clause structure the most deeply embedded NP in VP receives nuclear stress. A schematic representation of the default intonational pattern in German declaratives is given in (12), where we distinguish the different fields according to their information-structural contribution. (12) Illustration of default intonation pattern in German (Pre-V-Foc): H*L [CP … [TP ... < topic field>...[vP (denn) [vP ... ...] Vfin]]]



In (13), we show how Pre-V-Foc applies to regular V-final patterns in German. The same constraint accounts for the default intonation of the V2 was-für split construction in (14). In both examples, the DP Bücher (books) receives the default focus, typically associated with an H*L accent. (13) weil er (dann) because he PRT ‘because he then read books’

BÜCHER books

gelesen read

hat has



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 213

(14)  Was hat er (denn) für BÜCHER gelesen? what has he PRT for books read ‘What kind of books did he read?’ Constraint 2 – DestressGIV(en) – is a general information structural constraint on the deaccentuation of old information in discourse explicitly proposed by Féry & Samek-Lodovici (2006), but also observed by Chomsky (1972), Ladd (1980), Culicover & Rochemont (1983), Schwarzschild (1999), and Rochemont (2013 a, b) among others. Constraint 3 – TraceInFOC(us): A-bar extract only from a focused constituent – is a reformulation of Bayer’s (2004) generalization in (15). (15) In einer Topic/Fokus Struktur [[TOP X] [FOC Y]] darf die A’-Bewegung X nicht affizieren. (‘In a topic/focus structure [[TOP X] [FOC Y]], A’-movement is not allowed to affect X.’) Bayer (2004) observes that apparent subject/object asymmetries in long extractions are better captured as a difference in information structural status. He distinguishes the topic domain and the focus domain and identifies the topic, which is defined in terms of Reinhart’s (1981) aboutness topic, with the left middle field and focus with the right middle field, or VP domain. On the basis of the examples in (16), he argues that no extraction is allowed from the topic domain (cf. Bayer 2004: 238). (16) a. Was1 glaubst du, daß die Anitanom [t1 für Leutendat] vertrauen würde? what believe you that the Anitanom for peopledat trust would ‘What kind of people do you think Anita would trust?’ a’. ?*Was1 glaubst du, daß [t1 für Leutendat] die Anitanom vertrauen würde? b. Was1 glaubst du, daß der Anitadat [t1 für Leutenom] vertrauen würden? what believe you that the Anitadat [t1 for peoplenom] trust would ‘What kind of people do you think would trust Anita?’ b’. ?*Was1 glaubst du, daß [t1 für Leutenom] der Anitadat vertrauen würden? More specifically, he proposes that the ungrammaticality of (16a’, b’) is due to the topic status of the extraction site für Leute in each case. Related information structural constraints can be found in Erteschik-Shir’s and Goldberg’s work (cf. Erteschik-Shir 1973, Goldberg 2006). The three information-structural constraints introduced so far are widely accepted in the literature. We assume that they are not ranked and they apply whenever the relevant conditions are met. As was-für split involves A’-movement,

214 

 Susanne Winkler, Janina Radó and Marian Gutscher

we also need a constraint, DropFIL(LER), which causes the parser to search for gaps and assign as yet unassigned fillers: (17) DropFIL When the parser has a filler in its working memory, it must actively search for gaps. If a gap can be posited at a position that is syntactically suitable for the filler in the working memory, the filler must be assigned immediately.5 According to (17), a filler can no longer be kept in working memory once a suitable gap has been found. In terms of a constraint satisfaction (interactive) processing account: if a gap can be posited that is syntactically suitable for the filler in working memory, then all analyses that involve positing this gap and assigning the filler receive more activation than analyses in which no gap is posited and the filler is not assigned.

2.1.2 Applying the constraints to was-für patterns In this section, we illustrate how the three IS constraints and DropFIL in German, introduced above, apply to the was-für constructions at hand. The core cases of freezing, melting, and extraction from a moved subject are provided in (18a–c). They are based on some of Müller’s (2010: 68) examples, which unlike (3) and (4) above, involve the particle denn. We chose these examples as denn marks the boundary between the topic and the focus field (cf. Bayer 2012, Bayer et al. 2011, Bayer & Obenauer 2011), thus we can clearly identify the position of the subject remnant; otherwise, the freezing case in (18a) and the subject movement case in (18c) would be indistinguishable. Since Müller reports similar judgments as for the examples without the particle, the use of denn does not affect our argument concerning the freezing vs. melting puzzle. We will discuss implications of the presence of denn for the processing of (18c) in section 2.1.2.4. (18) No movement (nm) = ‘nm case’ ⇒ Müller’s freezing a. Was1 haben denn [DP3 t1 für Ärzte] [DP2 den Minister] kritisiert? what have PRT for doctorsnom the ministeracc criticized ‘What kind of doctors criticized the minister?’

5 As the reader might have noticed, DropFIL is essentially a rendering of the active filler strategy (e.g., Frazier 1987, Frazier & Flores D’Arcais 1989). We deliberately avoid the term ‘active filler strategy’, to stay neutral to the debate in the literature as to how ‘active’ the filler is in its search for suitable gaps, which is not crucial for our hypotheses.



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 215

Object movement (om) = ‘om case’ ⇒ Melting b. Was1 haben [DP2 den Minister] denn [DP3 t1 für Ärzte] t2 kritisiert? what have the ministeracc PRT for doctorsnom criticized ‘What kind of doctors criticized the minister?’ Subject movement (sm) = ‘sm case’ ⇒ Control Condition; Freezing c. Was1 haben [DP3 t1 für Ärzte] denn [DP2 den Minister] kritisiert? for doctorsnom PRT what have the ministeracc criticized ‘What kind of doctors criticized the minister?’ The three different was-für patterns are given schematically in (19). Since all three patterns involve wh-extraction from the subject position, we represent the word order patterns prior to this operation for simplicity. We will refer back to (19) in our analysis below. (19) [cp …wh [tp ......[vp (denn) [vp ......] V ]]] i. nm case [cp [tp (denn) [vp Subj Obj] V ]] ii. om case [cp [tp Obj1 (denn) [vp Subj t1] V ]] iii. sm case [cp [tp Subj1 (denn) [vp t1 Obj] V ]] The wh-extraction in (19i) is from the last-merged subject in SpecvP. We refer to it as the nm case since both the subject and the object are first merged in the focus field and not moved. This is the configuration Müller calls freezing. The wh-extraction in (19ii) is again from the subject in vP, but here the object is scrambled to the left of the particle denn. We refer to this case as om. Note that the subject is the only argument left in the vP and linearly, it is immediately preverbal. This is the melting configuration. In (19iii), the subject moves out of the focus field into the topic field prior to was-extraction. This case is called sm. For our experiment, this configuration serves as a control. It is a structure that is characterized as showing a genuinely grammatical freezing effect by Culicover & Winkler (2011, submitted). The next step is the application of the constraints to the different was-für patterns. As already noted, we assume a constraint satisfaction parser that simultaneously uses different sources of information relevant to processing the sentence. These include implicit prosody, cf. section 2, and the constraints we introduced in section 2.1.1. All relevant sources of information give weight to one or several possible analyses. If one analysis receives more support from all the interacting constraints and processed information than the competing ones, then it becomes the most highly activated one. In other words, the more constraints an analysis can satisfy, the higher its activation. However, the picture can change when the parser encounters the next incoming element. If the favored analysis turns out to conflict with information provided by the next constituent, a different analysis might become the most highly activated one. In a nutshell, at each point of analysis, the parser must reassess whether all relevant sources of

216 

 Susanne Winkler, Janina Radó and Marian Gutscher

information still support the analysis favored so far. At the end of the sentence, the most highly activated analysis wins. 2.1.2.1 Processing the ‘freezing’ example (nm) We will first consider the nm case (nm/19i) and focus on two positions in the sentence: the subject remnant and the preverbal direct object. These ‘points of analysis’ (PoAs) are given in (20). The role of the particle denn will be discussed in section 2.1.2.4. (20) Was haben denn [für Ärzte]DP1 [den Minister]DP2 kritisiert PoA 1 PoA 2 First point of analysis: DP 1 In processing the sentence, the constraint satisfaction parser first encounters the subject remnant, PoA 1. The constituents parsed at this point are given in (21). At this position, several different sentence continuations are possible. The relevant ones are given in brackets in (21i–iii). The subject remnant für Ärzte could turn out to be the preverbal element if the sentence ends with an intransitive verb (21i). Alternatively, the subject remnant could be followed by an object – possibly a complex DP – before ending with a transitive verb (e.g., Müller’s freezing example, 21ii). At PoA 1, even a completely different analysis is imaginable, where für Ärzte is not a subject remnant but a PP (21iii). This last analysis would need a suitable context, though, one which facilitates contrastive focus on the DP. (21) Was1 haben what have

denn t1 PRT

für for

Ärzte doctors

i. nm-a Was1 haben denn t1 für Ärzte what have PRT for doctorsnom ‘What kind of doctors were having a discussion?’ ii. nm-b Was1

haben

denn t1

für

Ärzte

[diskutiert?] discussed

[den Minister kritisiert]

nm-c iii. Dass die Reformen für Patienten einiges verändert haben, weiß ich schon, aber ‘I already know that the reforms brought about many changes for patients, but was haben denn [für Ärzte]PP [die Reformen verändert?] what have PRT for doctors the reforms changed what changes did the reforms bring about for doctors?’ We hypothesize that an analysis corresponding to the structure in (21i) (= analysis nm-a) receives the highest level of activation at PoA 1. There are three processing arguments for this assumption.



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 217

First, application of both DropFIL and TraceInFOC is highly probable. The parser can posit a gap and treat für Ärzte as the remnant of a wh-DP. According to DropFIL, the parser checks if this gap is suitable for the wh-filler was in working memory and finds that it is. The filler is assigned to the gap inside the subject remnant, and ‘_ für Ärzte’ combines into a syntactically complete wh-phrase ‘was für Ärzte.’ As a consequence of applying DropFIL, TraceInFOC (A-bar extract only from a focused constituent) has to be applied at PoA 1, too. So the subject remnant für Ärzte must be focused. The interdependence of DropFIL and TraceInFOC lends support to any analysis in which both constraints are satisfied. Second, since (21i) is a possible sentence continuation, PreV-FOC must be applied, i.e., default focus must be assigned at PoA 1. Otherwise, the parser runs the risk of reaching the sentence-final verb without having assigned default focus, and failing to apply a relevant constraint would imply a constraint violation in this case. Third, as now becomes evident, the demands of PreV-FOC and TraceInFOC ­coincide at PoA 1: both require focus on the subject remnant für Ärzte. Thus, analysis nm-a allows satisfaction of both constraints as well as DropFIL. DestressGIV is not relevant at PoA 1, since there is no context or any other cues marking the subject remnant as given. Its non-application therefore does not cause a violation. We think the competing analyses nm-b and nm-c receive less activation at PoA 1 because they do not allow satisfaction of all constraints that need to be checked. A continuation like in Müller’s freezing case (analysis nm-b) would be possible, but it minimally involves violation of TraceInFOC. In analysis nm-b, PreV-FOC is not applied to the subject remnant für Ärzte because it is not the preverbal element (den Minister would be immediately preverbal). However, since analysis nm-b assumes that DropFIL is applied and was and für Ärzte combine into a was-für wh-phrase, the extracted wh-operator would still be assigned to the gap. But if DropFIL is applied at PoA 1, while PreV-FOC is not, then TraceInFOC is violated. A structure like (21iii), i.e., analysis nm-c, in which für Ärzte is a PP, receives very low activation because of DropFIL. If a gap can be posited, the parser has to make use of this possibility. But choosing a PP-analysis would render positing a gap highly unlikely. Second point of analysis: DP 2 At PoA 2, the preverbal element den Minister, the picture changes. Obviously, the subject remnant was mistakenly identified as the preverbal element, thus PreV-FOC now demands the nuclear accent on the direct object den Minister. DestressGIV, on the other hand, requires den Minister to be unfocused, since the definite article indicates givenness. Due to these conflicting demands, the two analyses receive comparable levels of activation. (22) shows the constituents parsed at this position.

218 

 Susanne Winkler, Janina Radó and Marian Gutscher

(22) Was haben denn für Ärzte den Minister what have PRT for doctors the minister In analysis nm-b1 (23a), the nuclear accent remains on the subject remnant and the direct object is destressed. This causes a violation of PreV-FOC, but satisfies DestressGIV. In analysis nm-b2 (23b), nuclear focus is shifted from the subject remnant to the direct object. In this analysis, PreV-FOC is satisfied, but DestressGIV is violated. In addition, TraceInFOC is violated at für Ärzte, since now extraction takes place from an unfocused element. The two competing analyses are given in (23a) and (23b). (23) a. analysis nm-b1 (focus on DP 1 ‘für Ärzte’) [Was]1 haben denn [t1 für ÄRZTE]foc den Minister kritisiert PoA 2 b. analysis nm-b2 (focus on DP 2 ‘den Minister’) [Was]1 haben denn t1 für Ärzte [den MINISTER]foc kritisiert PoA 2 Other analyses would also be theoretically conceivable, but they receive even less activation. Both analyses nm-b1 (subject focus, 23a) and nm-b2 (object focus, 23b) involve a constraint violation at PoA 2. There is no clear preference for either analysis, hence the comparable amounts of activation. That some constraints are not applied at a given PoA does not matter, since non-application simply means the constraint is not relevant at that point. Following a constraint satisfaction approach, processing load increases when the competition for activation between analyses is undecided (cf. MacDonald 1994). This increased processing complexity could explain why the nm structure (19i) is marked in comparison to the om case (19ii), which we will analyze next. 2.1.2.2 Processing the ‘melting’ example (om) In processing the nm example (19i, Müller’s freezing case), there was not a clear preference between two competing analyses. We believe that in Müller’s melting example – here the om case (19ii) – this problem does not arise. We will again look at the constituents den Minister and für Ärzte, introducing these points of analysis in (24): (24) Was haben [den Minister]DP1 denn [für Ärzte]DP2 kritisiert PoA 1 PoA 2 First point of analysis (DP 1) On reaching the direct object, PoA 1, the parser assumes that this cannot be the preverbal element. If it were, it would have to be analyzed as the subject of the



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 219

sentence. But the inflected auxiliary requires a plural noun, whereas den Minister is singular. Furthermore, it is marked as accusative, resulting in a second mismatch. Therefore, an analysis in which PreV-FOC is applied and DestressGIV violated (which would be the consequence of stressing den Minister) does not receive high activation – it seems very improbable. We call this only barely activated alternative analysis om-b (see (25b) below). On the other hand, the alternative where a DP is still missing – analysis om-a (25a) – is very highly activated: PreV-FOC is not applied; DestressGIV is applied (definite article indicates givenness). DropFIL is not applied, since no gap is encountered. Consequently, TraceInFOC is not applied, either.

Second point of analysis (DP 2) Moving on to PoA 2, the subject remnant, analysis om-a (25a), is still the most highly activated one: having applied DestressGIV to the direct object, the parser now encounters a gap, and applies DropFIL and TraceInFOC. PreV-FOC is applied, too, anticipating the subject remnant für Ärzte to be the preverbal element. The same rationale applies as in the freezing case: DropFIL, PreV-FOC and TraceInFOC are interdependent, so applying all of them at the same time reduces the risk of violating a constraint and results in a highly probable analysis. (25) a. analysis om-a (subject-focus) Was haben den Minister denn [t für ÄRZTE]foc kritisiert PoA 2 b. analysis om-b (object-focus) Was haben [den MINISTER]foc denn [t für Ärzte] kritisiert PoA 2 In the alternative om-b (25b), PreV-FOC is not applied at für Ärzte, hence TraceInFOC is violated, assuming that the filler is still assigned to the gap. This analysis receives less activation than analysis om-a (25a) for two reasons: für Ärzte can be the preverbal element, so the constraint-based parser wants to apply PreV-FOC to guarantee that it has been applied by the end of the sentence. In addition, om-b causes a constraint violation, whereas analysis om-a (25a) does not. The preference is thus clear. We can now explain why the nm structure (19i/20) seems more marked than its om counterpart (19ii/24). In the latter case, where the object has been scrambled across the subject remnant, one analysis (25a, om-a) is clearly preferred: it involves no constraint violation. In the nm case (19i/20), where both subject and direct object remain inside the vP, preferences for (23a) or (23b) are less clear. This results in higher processing load, which in turn leads to the impression of markedness.

220 

 Susanne Winkler, Janina Radó and Marian Gutscher

2.1.2.3 Processing the sm cases The third case is the sm case (sm case) as schematically given in (19iii), repeated as (26). (26) Was1 haben [DP3 t1 für Ärzte] denn [DP2 den Minister] kritisiert? PoA 1 PoA 2 When the parser reaches PoA 1 in (26/19iii), the subject remnant für Ärzte, we face a situation similar to the nm example (19i/20). The parser does not know if this DP is the preverbal element or if more constituents will follow. Both scenarios are possible [Was haben für Ärzte diskutiert? or a structure like example (26)]. So the parser must assume that the following element will be the sentence-final one and apply PreV-FOC. An analysis similar to analysis nm-a in the freezing case (cf. 23a) is the most highly activated one: PreV-FOC is applied and satisfied, DropFIL is applied, and TraceInFOC is applied and satisfied (cf. 27a/ analysis sm-a). (27) a. analysis sm-a (subject-focus) [Was]1 haben [ t1 für ÄRZTE]foc denn den Minister b. analysis sm-b (object-focus) [Was]1 haben t1 für Ärzte denn [den MINISTER]foc At PoA 2, the preverbal DP-object den Minister, we find the same conflict as in the freezing case (19i/20). Two analyses receive comparably high activation – analysis sm-a, in which PreV-FOC is violated and DestressGIV is applied and satisfied, and analysis sm-b, in which PreV-FOC is applied and satisfied, therefore violating DestressGIV at the subject remnant. Judging from this configuration, we would expect the same markedness for the sm (19iii/26) as for the nm construction (19i/20). But we also have to consider the particle denn. 2.1.2.4 The role of the particle denn As already noted, denn acts as a signaling device: it marks the boundary between the topic and the focus field as shown in (12) (cf. Bayer 2012, Bayer et al. 2011, Bayer & Obernauer 2011). In the most highly activated analysis for the sm case (19iii/26) at PoA 1, analysis sm-a (27a), PreV-FOC is applied to für Ärzte. But once the parser reaches the particle, this analysis runs into trouble: focused elements should occur to the right of denn, but in (27a), an element to its left (für Ärzte) has been focused. Analysis sm-a (27a) loses support, but the alternative in which für Ärzte is not focused still seems unlikely. It would involve violating TraceInFOC. At this point, the parser could choose between violating TraceInFOC and ignoring



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 221

the ­information-structural function of denn. Consequently, the different analyses receive comparable amounts of activation, resulting in increased processing load. Reaching den Minister, PoA 2 (see 27a), the same problems occur as in parsing the nm example (see 23a and 23b): if focus stays on für Ärzte, PreV-FOC is violated, since now the direct object den Minister turns out to be the preverbal element. Ignoring denn’s information-structural function and violating PreV-FOC would thus imply two violations in analysis sm-a (27a), just like in the alternative analysis, sm-b (27b): changing focus assignment to the direct object leads to a violation of both DestressGIV (at the direct object) and TraceInFOC (at the subject remnant). Compared to the nm case (19i/20), the parser has more difficulty processing the sm case (19iii/26). There are two positions in the sentence where preference is far from clear, denn and PoA 2. From the moment that the parser comes across denn, it has to carry the additional burden associated with unclear processing preferences. Reaching the preverbal element, there is still no most highly activated analysis. In the nm case (19i), problems only occur upon reaching the preverbal element. In other words, processing the sm structure (19iii/26) means dealing with unclear processing preferences for a longer time than in the nm case (19i/20). We assume that this ‘longer competition’ results in more complexity and thus extra processing burden for our subjectextraction case.

2.1.3 Interim conclusion It appears that Müller’s (2010) freezing vs. melting opposition can be cast in terms of a constraint satisfaction processing account. Following this perspective, the nm structure (19i/20) is not frozen, but difficult to process, and is therefore marked relative to the om case (Müller’s melting, 19ii/25). The core problem is that information-structural, prosodic and syntactic constraints lend support to different possible ways of analyzing the sentence structure. Only in the om case (19ii/25) do all relevant constraints point to the same processing alternative throughout the sentence. In the other structures – nm (19i/20) and sm (19iii/26) – competing analyses receive similar amounts of activation and leave the parser with a higher processing burden. Thus far, we have only considered examples presented without context, where the parser’s processing choices were based solely on the relevant constraints and the structural input. If more information is provided, the picture can change. In particular, our account predicts that the degraded status of Müller’s freezing case should be improved by a suitable context.

222 

 Susanne Winkler, Janina Radó and Marian Gutscher

2.2 Influence of context In a constraint satisfaction account, context constitutes an additional factor that gives weight to one or the other analysis (e.g., MacDonald et al. 1994: 686). This also holds for our constraint satisfaction information-structural approach. In this section, we will discuss the role of context and show that systematic contextual manipulation influences the IS and silent prosody of the target sentence. In particular, we will examine the combination of the context in (28i) with the three syntactic was-für constructions in (19i–iii). The context clause (28i) constitutes a topicalized CP in German and is constructed in such a way that it introduces every constituent in the target sentence, except Ärzte. This forces the reader to put contrastive focus on the only new constituent, für Ärzte, in the target sentence in (28ii). First, consider (28), which represents the structure of the nm case (19i/20): (28) i.  Dass den Minister Journalisten kritisiert haben, weiß ich schon, aber ‘I already know that journalists criticized the minister, but ii.  was1 haben denn [DP3 t1 FÜR ÄRZTE]foc [DP2 den Minister] kritisiert? what have PRT for doctorsnom the ministeracc criticized what kind of doctors criticized the minister?’ The discourse centers on the implicit question ‘Who criticized the minister?’ We designed the context clause (28i) using basic insights of IS theory, and here in particular contrastive focus, such as discussed in Krifka (2008) and Krifka & Musan (2012). The context establishes that the speaker already knows that the journalists criticized the minister, and now he wants to know which doctors criticized the minister. Thus, the context influences the IS of the target sentence in a systematic way. The idea that IS, and in particular information focus, influences perceivers’ processing of the target sentence is not new (cf. Frazier 2009 for an overview). The contribution of the present chapter is using context to precisely control prosody and interpretation by manipulating contrastive focus. The central question is: why should contrastive focus on the subject in (28) facilitate processing? To answer this question, we will reconsider the two realizations in (23a, b) (subject vs. object focus). Here, subject focus excludes parsing alternatives: in the competition for activation, there is now one more reason to favor analysis nm-b1 (23a), since contrastive focus on the subject specifies the relevant reading. When the parser reaches the preverbal DP, the competition between analyses nm-b1 (23a) and nm-b2 (23b) is already decided: with subject focus, (23a) is the more probable analysis and receives higher activation. Thus, the processing load associated with unclear preferences should be reduced and the nm case (19i) should improve significantly if presented in this kind of context.



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 223

The situation, however, is different in the second case, Müller’s ‘melting’ (our om case, 19ii/25). The context in (28) induces contrastive focus on für Ärzte, which is preverbal since the given DP den Minister is scrambled. Thus, focus on the DP supports the analysis that is the most probable one even without context because it does not violate any constraints. Given that, adding context should not improve the melting example (19ii/25). Turning to the third relevant structure (19iii/26), we see that our subject-focus inducing context supports the sm-a analysis in (27a). The parser thus has a reason to favor this analysis. However, since it still violates two IS constraints, it should still be marked even if presented in context. The prediction is that acceptability should improve if context precedes the target sentence, but to a lesser degree than with the nm case (19i/20), which involves only one violation in the favored analysis nm-b1. A purely syntactic account of the freezing vs. melting puzzle would not predict improvement in a context that controls the IS of the target sentence. We conducted a rating study varying both word order and context to assess the acceptability of the examples that Müller (2010) used to illustrate the freezing vs. melting opposition and to test our predictions concerning the influence of context. In our analysis, Müller’s melting example is the only one that should not improve in a context like (28).

3 Rating study: Implicit prosody and default vs. contrastive focus The goal of the experiment was to find out whether IS-based processing factors can explain the gradedness of the freezing-melting phenomena. Without an information-structurally informative context (cf.(29) below), we expected om (om, 29b/19ii) to be rated best, since the competition for activation has a clear winner. sm (29c/19iii) should be the worst, as two analyses receive equal support from the relevant constraints and each one implies the violation of two constraints. Furthermore, the uncertainty about the correct analysis appears earlier in the sentence than in the no movement condition (nm, 29a/19i) and persists throughout the rest of the sentence. The nm condition should be intermediate: two analyses compete for activation, but in contrast to the sm condition, processing difficulty only occurs at the preverbal element. When context makes the object given and the subject contrastively focused, no change is expected in the om condition (30b), since context supports the analysis that was most highly activated anyway. The nm and sm conditions should improve significantly: in both, one of the competing analyses gets

224 

 Susanne Winkler, Janina Radó and Marian Gutscher

­ dditional support from the context and wins the competition for activation. a The sm condition (30c) is still expected to be more degraded, however, since two constraints are violated as opposed to one in the nm condition (30a). These predictions were tested in a rating study.6

Materials We manipulated word order (nm out of the vP vs. om vs. sm) and default vs. contrastive focus on the subject induced by a preceding context (default focus vs. contrastive focus), yielding a total of six conditions as illustrated in (29a)–(30c). (29) (default focus, DF) Sag mal: ‘Say’ a. Was haben denn für Ärzte den Minister kritisiert? b. Was haben den Minister denn für Ärzte kritisiert? c. Was haben für Ärzte denn den Minister kritisiert? ‘what have for doctorsnom PRT the ministeracc criticized?’ (30) (contrastive focus on subject, CF) Dass den Minister Journalisten kritisiert haben, weiß ich schon, aber ‘I already know that journalists criticized the minister, but’ a. Was haben denn für Ärzte den Minister kritisiert? b. Was haben den Minister denn für Ärzte kritisiert? denn den Minister kritisiert? c. Was haben für Ärzte ‘what have for doctorsnom PRT the ministeracc criticized?’ Eighteen experimental items like (29)–(30) were constructed.7 The particle denn was used to mark the boundary between the topic and the focus field (cf. Bayer 2012, Bayer et al. 2011, Bayer & Obernauer 2011). We selected transitive verbs to avoid potential problems with establishing the base position of the direct object. The was-für phrase was always the subject DP. Both the subject and the object DP were animate to make the sentences more natural. In the CF context conditions, the target sentences were preceded by a subordinate clause that introduced the direct object of the matrix clause and an alternative to the matrix subject [‘journalists’ in the example in (30)]. This manipulation was intended to place

6 Strictly speaking, our predictions concern not only the overall acceptability of the sentences, but also particular points where difficulties are expected to arise during processing. This online aspect of our hypothesis will be tested in future experiments. 7 All items appear in the Appendix. Each item is given in the nm condition with CF-context.



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 225

c­ ontrastive focus on the matrix subject while making the direct object ‘given’ (see the discussion in 2.2). In the no-context conditions, the sentences were introduced by the phrase Sag mal, (‘Say,’), as complex sentences such as those used in our experiment may sound less natural without any context, especially in comparison to the contrastive context versions. In addition, 36 fillers were constructed that superficially resembled the experimental items. Most of the fillers either included was-für phrases as direct objects, which is more common than the subject was-für phrases used in the items, or they contained the words was and für as parts of different constituents, as in (31). Based on our intuitions, we divided the fillers into five categories ranging from quite acceptable to rather unacceptable, roughly corresponding to the range of ratings we expected for the items. All fillers included a modal particle such as denn, bloß, nur, or wohl; half of them had an initial subordinate clause similar to the items, while the other half were introduced by a phrase similar to Sag mal. (31) Ich überlege gerade: Was sollte ich denn nochmal einkaufen für diese Woche? ‘(I am wondering) What should I PRT again buy for this week?’

Procedures Items and fillers were combined and the resulting 54 sentences were distributed on six presentation lists according to a Latin square design. To ensure careful reading of the sentences, a yes-no question was written for each sentence. Each participant saw every item in exactly one condition; across participants, each item in each condition was rated equally often. The study was conducted in a computer pool at Tübingen University as an internet questionnaire using the WebExp2 software (Mayo et al. 2006). Subjects received written instructions asking them to rate the naturalness of the sentences from 1 (completely unnatural) to 15 (completely natural) and answer the comprehension questions. We chose a 15-point rating scale as we expected relatively fine distinctions between conditions. The scale corresponds to the grades commonly used in German high schools, and participants were told to refer to this grading system in their responses. The task was formulated in terms of naturalness rather than acceptability to avoid normative judgments, as was-für split is much more common in informal registers. Subjects were instructed to judge whether the sentences are natural examples of spoken German. Four examples were provided to illustrate the notion of (linguistic) naturalness and distinguish it from the naturalness of the situation described in the sentence. After reading the instructions, participants proceeded through the experiment at their own pace. The first five sentences in the experiment were fillers presented in a fixed order, to allow participants to get used

226 

 Susanne Winkler, Janina Radó and Marian Gutscher

to the judgment scale. The remaining 49 sentences appeared in an individually randomized order. An experimental session lasted ca. 25 minutes. Forty-eight native German speakers (students at Tübingen University) participated in the study for a payment of 5 euros. They were all naive as to the purpose of the experiment.

Results The overall question-answering accuracy was 86% and did not differ across conditions. Ratings were normalized using z-transformation. Figure 1 shows the mean normalized ratings per condition. The scores were subjected to 2 × 3 repeated measures ANOVAs with the within factors context (default vs. contrastive focus) and word order (nm vs. om vs. sm), using participants (F1) and items (F2) as random factors. There was a significant main effect of context [F1(1,47) = 14.047, p < 0.01, F2(1,17) = 26.817, p < 0.01]: the contrastive focus conditions were rated significantly higher than the default focus conditions (mean ratings: 0.0459 vs. ‒0.0568, respectively). The main effect of word order was significant as well [F1(2,94) = 44.387, p < 0.01, F2(2,34) = 69.448, p < 0.01], the om condition being rated highest, followed by the nm condition, while sm received the lowest ratings (mean ratings: 0.3064, 0.1532, and ‒0.476, respectively). Crucially,

Figure 1: Mean normalized ratings per condition. Error bars show the 95% confidence interval



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 227

there was a significant interaction of context and word order [F1(2,94) = 6.141, p < 0.01, F2(2,34) = 4.67, p < 0.03] due to the fact that contrastive focus led to higher ratings in the nm and sm conditions, but not in the om condition. Planned comparisons showed that this improvement in the nm and the sm conditions was significant: DF nm vs. CF nm t1(47) = ‒2.947, p < 0.01, t2(17) = ‒3.415, p < 0.01; DF sm vs. CF sm t1(47) = ‒5.187, p < 0.01, t2(17) = ‒5.192, p < 0.01. However, even with context, the nm condition remained significantly worse than the om condition [t1(47) = ‒2.169, p < 0.04, t2(17) = ‒2.223, p < 0.04], and the sm con­ dition was still significantly worse than the nm condition [t1(47) = 3.977, p < 0 .01, t2(17) = ‒5.118, p < 0.01]. Discussion The results confirm our predictions: Without context, the construction with the direct object outside the vP and the split subject still in the focus field (DF_om/29b/19ii) was significantly better than the one with both subject and object inside the vP (DF_/nm29a/19i). Focus assignment in the latter case was ambiguous due to conflicting information-structural constraints (cf. 20–23). The worst case, however, was the one where the subject containing the trace of the was-für element appeared outside the vP in the topic field (DF_sm/29c/19iii). Here, besides conflicts between IS constraints, the occurrence of denn to the right of the subject phrase further complicated processing. Moreover, as predicted, a contrastive context that places focus on the subject was able to guide interpretation in (CF_nm/30a/19i). Sm (CF_sm/30c/19iii) was also improved by context, but it still remained suboptimal as parsing the structure involves two violations in any case. Note that the best case, om (CF_om/30b/19ii), did not benefit from the contrastive context at all. This is predicted by our account since the construction can already be parsed unambiguously without guiding context: the analysis in which focus is assigned to the preverbal subject remnant is clearly the most probable one. The lack of improvement cannot be explained by a ceiling effect: the om conditions were still rated considerably worse than the best fillers (mean ratings DF om: 0.3062, CF om: 0.3067, fillers category ‘best’: 1.0264, category ‘good’: 0.7609). Interestingly, the contrastive context did not suffice to make the nm condition as good as the om condition. Although it clearly supported the analysis where the subject remnant is focused (analysis nm-b1 = 23a) and increased its activation level, the competing analyses were apparently still available. This is because the contrastive context is just one of the relevant factors (constraints in our view) that jointly determine the activation of a given analysis. The high level of activation that analysis nm-b1 received made the alternative analyses dispreferred, but their (lower) level of activation still resulted in some additional processing load, which

228 

 Susanne Winkler, Janina Radó and Marian Gutscher

led to the impression of markedness. Processing of the om sentence did not face these problems. Concerning the particle denn, we already noted that Müller reports similar judgments for the examples without the particle. In his view, the freezing vs. melting opposition is independent of the presence of the particle (Müller 2010: 68). Still, one might object that the effect of context on freezing that we observed has to do with an interaction between denn and contrastive focus, rather than with IS and silent prosody in general, as we claim here. We are not sure what such an interaction between denn and focus would look like. To exclude this possibility, however, we are currently conducting follow-up experiments to tease apart the influence of denn and that of IS. Preliminary results show that our context manipulation leads to the same effect regardless of whether the sentences contain a particle marking the boundary between the topic and the focus field or not.

4 Conclusion We have presented evidence for an alternative approach to Müller’s (2010) freezing vs. melting opposition in was-für extraction (19ii vs. 19i). Our results show that a syntax-only account cannot describe the phenomenon properly. As we have demonstrated, the freezing structure (19i) improves relative to its melting counterpart (19ii) when presented to the reader in a suitable context. A purely syntactic explanation does not predict this. Following a constraint satisfaction approach to processing, we presented an explanation that takes this context effect into account. In short, conflicting requirements of information-structural, prosodic, and syntactic constraints lend support to different analyses of the sentence structure. The competition among them leads to higher processing complexity and the impression of ungrammaticality. A suitable context can aid processing by reducing the competition among alternative analyses. In our study, we showed how this effect is facilitated using systematic context manipulation, which requires contrastive focus in the target sentences and thereby influences the prosody in silent reading in a very precise way.

References Bayer, Josef. 2004. Was beschränkt die Extraktion? Subject – Objekt vs. Topik – Focus. In: Deutsche Syntax: Empirie und Theorie, Franz J. dʼAvis (ed.), 233–57. Göteborg: Acta Universitatis Gothoburgensis.



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 229

Bayer, Josef. 2012. From modal particle to interrogative marker: A study of German denn. In: Functional Heads, Laura Brugè, Anna Cardinaletti, Giuliana Giusti, Nicola Munaro, and Cecilia Poletto (eds.), 13–28. (The cartography of syntactic structures 7) New York: Oxford University Press [date of publication: spring 2012]. Bayer, Josef, Jana Häusler, Markus Bader & Simon Hopp. 2011. Connecting to illocutionary force: A theoretical and experimental study of the German discourse particle denn. talk at GLOW, Vienna 28–30 April 2011. Bayer, Josef & Hans-Georg Obenauer. 2011. Discourse particles, clause structure, and question types. The Linguistic Review (TLR) 28(4): 449–491. Bennis, Hans. 1983. A case of restructuring. In Linguistics in the Netherlands, Hans Benni, and Wus van Lessen Kloeke, 9–19. Dordrecht: Foris. den Besten, Hans. 1985. The ergative hypothesis and free word order in Dutch and German. In: Studies in German Grammar, Jindrich Toman (ed), 23–64. Dordrecht: Foris. Chomsky, Noam. 1972. Deep structure, surface structure and semantic interpretation. In: Semantics, Danny Steinberg, and Leon Jakobovits (eds), 183–216. London: Cambridge University Press. Chomsky, Noam & Morris Halle. 1968. The sound pattern of English. New York: Harper and Row. Cinque, Guglielmo. 1993. A null theory of phrase and compound stress. Linguistic Inquiry 24(2): 239–297. Corver, Norbert. 1991. The internal syntax and movement behaviour of the Dutch “wat voor”construction. Linguistische Berichte 133: 190–228. Culicover, Peter W. & Michael S. Rochemont. 1983. Stress and focus in English. Language 59: 123–65. Culicover, Peter W. & Susanne Winkler. 2011. Freezing – a conspiracy. Unpublished manuscript. University of Ohio, University of Tuebingen. Culicover, Peter & Susanne Winkler. (submitted). Freezing: Between Grammar and Processing. In: Hartmann, Jutta M., Marion Knecht, Andreas Konietzko, Susanne Winkler (eds.), Freezing: Theoretical Approaches and Empirical Domains. Studies in Generative Grammar. Berlin, New York: Mouton de Gruyter. Diesing, Molly. 1992. Indefinites. Cambridge, MA: MIT Press. Erteschik-Shir, Nomi. 1973. On the nature of island constraints. Ph.D. diss. MIT. Erteschik-Shir, Nomi. 2007. Information structure: the syntax-discourse interface. New York: Oxford University Press. Féry, Caroline & Vieri Samek-Lodovici. 2006. Focus projection and prosodic prominence in nested foci. Language 82(1): 131–150. Fodor, Janet D. 2002a. Prosodic disambiguation in silent reading. In: Proceedings of the North East Linguistic Society 32, Masako Hirotani, 113–132. Amherst, MA: GLSA. Fodor, Janet D. 2002b. Psycholinguistics cannot escape prosody. Proceedings of the Speech Prosody 2002 Conference, Bernard Bel & Isabelle Marlin, 83–88. Aix-en-Provence: Laboratoire Parole et Langage. Frazier, Lyn. 1987. Syntactic processing: Evidence from Dutch. Natural Language and Linguistic Theory 5: 519–559. Frazier, Lyn. 2009. Explorations in ellipsis: The grammar and processing of silence. In: The fruits of empirical linguistics. Vol 2: Product, Susanne Winkler & Sam Featherston (eds.), 75–102. Berlin: de Gruyter. Frazier, Lyn & Flores d’Arcais Giovanni B. 1989. Filler-driven parsing: A study of gap filling in Dutch. Journal of Memory and Language 28: 331–344.

230 

 Susanne Winkler, Janina Radó and Marian Gutscher

Frey, Werner. 2004. The grammar-pragmatics interface and the German prefield. Sprache & Pragmatik 52: 1–39. Goldberg, Adele E. 2006. Constructions at Work. Oxford: Oxford University Press. Haider, Hubert & Inger Rosengren. 2003. Scrambling - non-triggered chain formation in OV languages. Journal of German Linguistics 15(3): 203–267. Halle, Morris & Jean-Roger Vergnaud. 1987. An essay on stress. (Current Studies in Linguistics 15) Cambridge, Mass.: MIT Press. Höhle, Tilmann N. 1982. Explikation für ‘normale Betonung’ und ‘normale’ Wortstellung. In: Satz­glieder im Deutschen, Werner Abraham (ed.), 75–153. Tübingen: Narr. Jurka, Johannes. 2010. The importance of being a complement. Ph.D. diss. University of Maryland. Krifka, Manfred. 2008. Basic notions of information structure. Acta Linguistica Hungarica 55(3–4): 243–276. Krifka, Manfred & Renate Musan. 2012. Information structure: Overview and linguistic issues. In: The Expression of Information Structure, Manfred Krifka & Renate Musan (eds.), 1–45. Berlin: De Gruyter. Ladd, D. Robert. 1980. The structure of intonational meaning. Bloomington: Indiana University Press. MacDonald, Maryellen C. 1994. Probabilistic constraints and syntactic ambiguity resolution. Language and Cognitive Processes 9(2): 157–201. MacDonald, Maryellen C., Neal J. Pearlmutter & Mark S. Seidenberg. 1994. Lexical Nature of Syntactic Ambiguity Resolution. Psychological Review 101(4): 676–703. Mayo, Neil, Martin Corley & Frank Keller. 2006. WebExp2: Experimenter’s manual. See their webpage at “http://www.webexp.info/”. Melchiors, Maria. 2007. Die syntaktische Analyse des deutschen Dativs – ein komplizierter Fall. Ph.D. diss., University of Tübingen. Müller, Gereon. 2010. On deriving CED effects from the PIC. Linguistic Inquiry 41(1): 35–82. Pafel, Jürgen. 1996. Die syntaktische und semantische Struktur von was-für Phrasen. Linguistische Berichte 16: 137–67. Pickering, Martin J & Roger P. G. van Gompel. 2006. Syntactic parsing. In: Handbook of Psycholinguistics, 2nd ed, Matthew J. Traxler & Morton A. Gernsbacher (eds.), 455–503. Amsterdam: Elsevier. Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27: 53–94. Reis, Marga. 1993. Wortstellung und Informationsstruktur. Tübingen: Niemeyer. Rochemont, Michael. 2013a. Discourse new, F-marking, and information structure triggers. Lingua 136: 38–62. [Special issue on information structure triggers, Hartmann, Jutta, Janina Radó & Susanne Winkler (eds)]. Rochemont, Michael S. 2013b. Discourse new, focused and given. Approaches to Hungarian, vol. 13. In: Johan Brandtler, Valeria Molnár, and Christer Platzack (eds.), 199–228. Amsterdam: John Benjamins. Schwarzschild, Roger. 1999. GIVENness, avoidF and other constraints on the placement of accent. Natural Language Semantics 7: 141–177. Selkirk, Elisabeth. 2011. The syntax-phonology interface. In: The Handbook of Phonological Theory, 2nd ed., John A. Goldsmith, Jason Riggle & Alan C. Yu, (eds.), 435–484. Oxford: Blackwell. Truckenbrodt, Hubert. 1995. Phonological phrases: Their relation to syntax, focus, and prominence. Ph.D. diss. Cambridge, MA: MIT dissertation. Wexler, Kenneth & Peter W. Culicover. 1980. Formal principles of language acquisition. Cambridge, MA: MIT Press.



What Determines ‘Freezing’ Effects in was-für Split Constructions? 

 231

Appendix: Materials The list below contains the nm condition of every experimental items in the contrastive focus conditions. The default focus conditions used the same target sentences, the context was invariably Sag mal. 1. Dass den Förster Zecken gebissen haben, weiß ich schon, aber was haben denn für Ameisen den Förster gebissen? 2. Dass den Jäger Hunde angefallen haben, weiß ich schon, aber was haben denn für Wildtiere den Jäger angefallen? 3. Dass den Bademeister Insekten gezwickt haben, weiß ich schon, aber was haben denn für Krebse den Bademeister gezwickt? 4. Dass den Zoodirektor Löwen verletzt haben, weiß ich schon, aber was haben denn für Affen den Zoodirektor verletzt? 5. Dass den Zoohändler Kaninchen gekratzt haben, weiß ich schon, aber was haben denn für Katzen den Zoohändler gekratzt? 6. Dass den Dompteur Tiger angegriffen haben, weiß ich schon, aber was haben denn für Bären den Dompteur angegriffen? 7. Dass den Segelweltmeister Haie gerammt haben, weiß ich schon, aber was haben denn für Wale den Segelweltmeister gerammt? 8. Dass den Hausmeister Ratten bedrängt haben, weiß ich schon, aber was haben denn für Hunde den Hausmeister bedrängt? 9. Dass den Minister Journalisten kritisiert haben, weiß ich schon, aber was haben denn für Ärzte den Minister kritisiert? 10. Dass den Kanzler Politiker beraten haben, weiß ich schon, aber was haben denn für Wissenschaftler den Kanzler beraten? 11. Dass den Museumsleiter Studenten unterstützt haben, weiß ich schon, aber was haben denn für Handwerker den Museumsleiter unterstützt? 12. Dass den Prinzen Dichter unterrichtet haben, weiß ich schon, aber was haben denn für Musiker den Prinzen unterrichtet? 13. Dass den Papst Atheisten gerügt haben, weiß ich schon, aber was haben denn für Theologen den Papst gerügt? 14. Dass den Polizeichef Anwohner gelobt haben, weiß ich schon, aber was haben denn für Demonstranten den Polizeichef gelobt? 15. Dass den Botschafter Wirtschaftsexperten begleitet haben, weiß ich schon, aber was haben denn für Juristen den Botschafter begleitet? 16. Dass den Bahnchef Ingenieure attackiert haben, weiß ich schon, aber a was haben denn für Künstler den Bahnchef attackiert? 17. Dass den Uni-Rektor Studenten beleidigt haben, weiß ich schon, aber was haben denn für Forscher den Uni-Rektor beleidigt? 18. Dass den Bürgermeister Arbeitslose gewählt haben, weiß ich schon, aber was haben denn für Unternehmer den Bürgermeister gewählt?

Index A-scrambling 46 accusative experiencers 54, 63 across-the-board movement/ATB-movement  4, 51ff, 58ff, 72, 80, 82 agentive verbs 34, 37ff aggregation studies 113ff, 117ff, 128 antecedent-linked subject drop 189f, 192f, 202 antonyms, in measure phrase constructions  85, 88f, 92f, 96, 98 asymmetric extraction 56ff, 83 ATB-topicalization 67f attributive measure phrase 95, 107 Bonner Frühneuhochdeutschkorpus 188, 192 borrowing situation 165 c-command 190, 197f, 208 case mismatch 4, 52, 56ff, 62, 67, 72f, 78ff center-embedding, multiple 9f, 17f, 20, 27, 31 configuration weights 119ff, 134f, 137f constituent order 33ff, 38, 45f, 49, 144f, 153f, 157, 160, 185 constituent recognition domain 13 Coordinate Feature Matching (CFM) principle 192 coordination 51, 63, 65, 69, 72, 81ff, 191 corpus study 12, 15, 27, 30, 48, 174 cross-linguistic variation 90, 95 data type 2 data-driven linguistics 1 deWaC 16ff differential measure phrase 107 discrimination hypothesis 4, 25ff, 30 disjunctive coordination 51 dislocated elements 169, 179f, 183 ditransitive verbs 34f, 42f DMDX 22 doubly center-embedded relative clause 10, 12, 16, 19f, 25f, 28ff

Early English 172, 176, 185f Early French 171, 174 early immediate constituents 13 empty left edge condition 203 experiencer-object verbs 34, 39, 40f extractability 208 extraposition 14f, 32 feature conflicts 51 free relatives 52 frozen structures 208, 210 gradable adjectives 90, 102f gradable predicate 5, 86, 95ff, S, 100ff, 105, 107 implicit prosody hypothesis 211 individual differences scaling 117f, 123 information structure 3, 6, 34, 163, 165, 168, 184, 186, 229f introspection 1 language acquisition 105, 111, 165, 184, 230 language contact 165f, 169f, 184ff left dislocation 6, 165f linearization preference 34, 40 linguistic variable 113f, 116ff, 123, 138f, 141 MCVF corpus 174f measure phrase constructions 4, 85f, 88, 90, 92, 111 melting 6, 207ff, 214f, 218, 221, 223, 228 memory 9, 17, 27f, 30ff, 49, 140, 214, 217, 229 methodology 117, 122, 126, 141, 199 mismatch, in coordination 193f missing-VP effect 9ff, 19f, 24ff, 31 modular boundaries 2 morphological identity 54ff, 59 multidimensional scaling 115ff, 129f, 132, 136, 140ff multiple questions 33, 35, 36, 37, 38, 39, 40, 42, 43, 44, 45, 47, 48

234 

 Index

Null subjects 6, 174f, 186ff, 195, 198f, 201ff Old French 6, 165, 167, 170, 183, 185f Old High German 5f, 113, 121ff, 130f, 137, 143f 146, 163, 188, 203f Old Saxon 5, 122, 143f, 147, 153, 155 partial pro drop 196, 198f passive verbs 42 Penn-Helsinki Parsed Corpus of Middle English 174, 185 poetry 144, 146ff Polish 52ff predicative measure phrase 91, 107 pro drop 187ff, 195ff, psych verb 61, 63, 68, 73f reanalysis 74ff, 78ff thematic hierarchy 5, 46, 54, 143f, 149ff, 162, 164 reference corpus Old German 121f, 125 relative clause 9ff, 19ff, 24ff relative word order, objects 143 self-paced reading 11, 56, 73, 76, 80 sharing approaches 58f, 72, 80 short scrambling 143, 151, 166 sideward movement 56ff, 72, 80, 83

small scrambling 157 speeded grammaticality judgement 9, 12, 20 strict SOV 9 strong evaluativity 97, 99, 101, 106 subject focus 218, 222 syncretism/syncretic 51ff, 55, 59, 60ff, 67, 71f, 74ff, 78ff thematic prominence 54, 66f, 70, 72, 80 theta roles 60ff, 67ff, 70ff theta roles, reanalysis 74ff right-node-raising 53 topicalization 6, 49, 61, 63, 67f, 72, 81, 165ff, 169, 181 topicalized phrase extraction 68 transparent aggregation 117, 123, 130, 136, 141 underspecified/underspecification 52f, 58ff was-für split 207f, 210, 212f, 225 weak evaluativity 93, 99ff web corpus 16 wide focus 157ff, 212 working memory 9, 28, 30, 32, 214 York-Toronto-Helsinki Parsed Corpus of Old English 174, 186