Sounds and Systems: Studies in Structure and Change. A Festschrift for Theo Vennemann 9783110894653, 9783110175691

Citation preview

Sounds and Systems

W G DE

Trends in Linguistics Studies and Monographs 141

Editors

Werner Winter Walter Bisang

Mouton de Gruyter Berlin · New York

Sounds and Systems Studies in Structure and Change A Festschrift for Theo Vennemann edited by

David Restle Dietmar Zaefferer

Mouton de Gruyter Berlin · New York

2002

M o u t o n de Gruyter (formerly M o u t o n , The Hague) is a Division of Walter de Gruyter G m b H & Co. K G , Berlin.

© Printed on acid-free paper which falls within the guidelines of the A N S I to ensure permanence and durability.

Die Deutsche Bibliothek — Cataloging-in-Publication

Data

Sounds and systems : studies in structure and change ; a festschrift for Theo Vennemann / ed. by David Restle ; Dietmar ZaefTerer. - Berlin ; New York : Mouton de Gruyter, 2002 (Trends in linguistics : Studies and monographs ; 141) ISBN 3-11-017569-X

© Copyright 2002 by Walter de Gruyter G m b H & Co. K G , D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Preface When we began planning this volume with contributions by over twenty different authors, we did not expect that the editing would always be a pure pleasure. Just because it was, we would first like to thank all the linguists who gladly agreed to contribute to this special collection of articles. Although the volume remained a secret until today, the 65th birthday of the person we wish to honor here, Theo Vennemann, we also would like to thank him, because in the overwhelming reaction to our invitation to the participants, we could feel the sincere appreciation for his achievements; the acknowledgement he has earned, indirectly and not known to him, facilitated our task, making it more than a pleasant enterprise. We wish to apologize to those colleagues who may have aspired to be invited but were not, due to lack of space. We are grateful to Anke Beck from Mouton de Gruyter for readily supporting our project, the series-editor for his quick and pertinent comments, and especially Birgit Sievert for her kind, untiring, and patient effort to make this volume possible. Last but not least, we would like to thank Stephen Laker for his kind assistance and especially Gabriele Hollmann for her spontaneous and invaluable help in correcting the final proofs.

München, May 27, 2002 David Restle and Dietmar Zaefferer

Contents Introduction David Restle and Dietmar Zaefferer

1

I. Historical studies on sounds, words, and systems A. Sound change Preglottalization in English and a North Germanic bifurcation Henning Andersen

15

Zur neuhochdeutschen Dehnung in offener Tonsilbe Thomas Becker

35

Lexical diffusion in regular sound change Joan By bee

59

Unveiling a masked change: behind vowel harmony in the dialect of Claro Michele Loporcaro

75

Accents and medieval English phonologists Robert W. Murray

91

Retraction and rounding in Old English breaking Robert P. Stockwell

121

Β. Language change Latin ipse, Continental Celtic -xsi: a tentative proposal Philip Baldi

141

Der Name al-Andalus: neue Überlegungen zu einem alten Problem Georg Bossong 149 Fein gehackte Pinienkerne zugeben! Zum Infinitiv in Kochrezepten Elvira Glaser 165 Language change in early Britain: the convergence account Raymond Hie key

18 5

Irish ainder, Welsh anner, Breton annoar, Basque andere Peter Schrijver

205

viii

Contents

II. Synchronic studies on sounds, words, and uses Cyclicity and base non-identity Larry M. Hyman

223

The meter of nursery rhymes: universal versus language-specific patterns Patrizia Noel Aziz Η anna, Katrin Lindner, and Andreas Dufter

241

Unreine Reime und phonologische Theorie Beatrice Primus

269

Choctaw intensives and syllable theory David Restle

299

Imperatives: the relation between meaning and form Renate Raffelsiefen

321

Struktur und Akzent komplexer Komposita Peter Eisenberg

349

Warum wir zusammenschreiben nicht immer zusammenschreiben Präferenzgesetze im Schriftsystem Joachim Jacobs 367 Noun classification and composition in Kilmeri Claudia Gerstner-Link

391

The puzzle of the autoantonymous argument role. Unraveling the polysemy of risk/riskier en Dietmar Zaefferer

413

Be brief and vague! And how Bidirectional Optimality Theory allows for verbosity and precision Manfred Krifka

439

Publications of Theo Vennemann

459

Tabula Gratulatoria

475

Index

479

Introduction David Restle and Dietmar Zaefferer

Several of the contributors to the present volume have known the honoree much longer than the editors have. One of them is Robert P. Stockwell, who in the first note of his article shares the nice anecdote with us about the first term paper he received from his student Theo Vennemann, "with all the assumptions and baggage of traditional English and Germanic philology", which he returned with the recommendation to "expand his horizons from philology to the then exploding theories of generative phonology and generative syntax". As is well known, Theo Vennemann not only heeded this advice and developed, alongside with Joan Bybee Hooper, the paradigm of Natural Generative Phonology, but he also played an important role in categorial syntax and, together with Renate Bartsch, in logical semantics, to mention only those fields of his research in which he adhered to his teacher's words. But this does not imply that he turned his back on his philological roots. On the contrary, looking at both his own work and at what he untiringly conveyed to his students, it is not an exaggeration to say that the hallmark of his approach to linguistic problems is a critical combination of traditional and modern ways of dealing with language, in varying proportions. In response, the contributions in this volume cover a wide range of research work, from traditionally oriented studies on sound change to optimality theoretic approaches to lexical pragmatics. The articles are arranged in a binary branching way. First, the primarily diachronic studies and then the predominantly synchronic studies are presented. In the first group, a subdivision was made between investigations about language sounds and work on larger subsystems. The synchronic group also starts with phonological contributions, but since there are several papers that straddle the boundary, we decided to arrange them differently: Instead of two blocks there is a continuum reaching from phonology via morphology and lexical semantics to pragmatics.

2

David Restle and Dietmar Zaefferer

It is, of course, not a coincidence that almost half of the papers in this volume deal with phonological issues. Theo Vennemann not only started his career with a phonological dissertation, he also first became renowned in this field. From among the six studies in the first group of historical studies concerned with sound change, Joan Bybee's paper on Lexical diffusion in regular sound change most directly builds on Vennemann's early phonological discoveries. Bybee argues that Labov's Salomonic way of settling the dispute between the Neogrammarians and Hugo Schuchardt by claiming that both are partially right, but in different areas, is not correct: It is not feasible to distinguish between regular sound change (gradual and only phonetically motivated) and lexical diffusion change (abrupt and lexically or grammatically conditioned), since gradual lexical diffusion can also be observed in regular sound change, i.e. in gradual, phonetically-conditioned change. In the reported findings on reductive phonetic change it turns out to be highly conditioned by word-frequency. To account for this, a model of phonological representation is required that allows phonetic detail in the cognitive representation of words. Theo Vennemann's proposal, presented in 1974, that all words are listed in their systematic phonetic form, is the starting point for Bybee's solution, the exemplar model: There, Vennemann's systematic phonetic form is replaced by the frequency-weighted cluster of tokens or exemplars, thus allowing for the modeling of gradual changes. Frequency also plays an important role in Michele Loporcaro's study Unveiling a masked change: behind vowel harmony in the dialect of Claro. He points to a so-far overlooked phenomenon in an otherwise well-described and famous sound-system, that of the Italo-Romance dialect of Claro (Canton of Tessin, Switzerland), namely to the blocking in one environment of the vowel harmony Claro is famous for. It turns out that vowel harmony conceals another sound change, the fronting of final /a/ to /e/, which therefore has remained unnoticed by other scholars. Loporcaro's explanation for this reanalysis is based upon the statistical distribution of stressed vowels. Tangible evidence for our observation that Theo Vennemann's approach to linguistics is characterized by a critical combination of traditional and modern ideas are his thoughts about the concept of the syllable. When, in the wake of the most influential Sound Patterns of English, it was fashionable to do away with syllables, he remained unim-

Introduction

3

pressed and relentlessly collected evidence for the usefulness of this concept. This blazed the trail for another pioneering achievement of Theo Vennemann's phonological investigations, later than the above mentioned representation of phonological words in their systematic phonetic form but of similar relevance: the rediscovery of Sievers' syllable cut opposition. One way of conceiving it is as a restriction on syllable structure which ties the quantity of stressed nuclei to the presence of an offset in such a manner that open syllables require long vowels and short vowels require closed syllables. Thomas Becker's Zur neuhochdeutschen Dehnung in offener Tonsilbe ('On open syllable lengthening in New High German') refutes a standard assumption concerning the history of the German sound system, namely that open syllable lengthening in New High German is a phonetically motivated internal sound change. In doing so, he builds on the above mentioned rediscovery of the syllable cut opposition and on Vennemann's finding that the New High German sound system is based upon a Low German substrate. According to Becker, the topical phenomenon is a result of language contact, namely the influence of speakers with a Low German dialect basis whose pronunciation was considered prestigious because it preserved distinctions present in traditional orthography. Another testimony of the impact of Vennemann's rediscovery is Robert Murray's study on Accents and Medieval English phonologists. His analysis of the accentuations in Medieval English manuscripts led him to the conclusion that the evidence of these manuscripts should be "thoroughly reconsidered once again in light of syllable cut theory". Such reconsideration is likely to confirm the notion "that syllable cut had already been phonologized by the mid 11th century". Of course, a central theme in historical phonology is the question: how exactly does sound change take place? In seeking an answer, it is helpful to look not only at historical changes but also at ongoing processes. Joan Bybee does this and so does Henning Andersen in his paper on Preglottalization in English and a North Germanic bifurcation. He finds that the development of North American preglottalization exemplifies an internally motivated sound change: Following the same internal logic, it develops in different places at the same time, with no indication that it is "spreading from place to place by speakers imitating the pronunciation of their neighbors". Andersen thus argues for polygenesis

4

David Restle and Dietmar Zaefferer

and concludes: "This mode of development is probably the most usual one in internally motivated sound change". In his paper Retraction and rounding in Old English breaking, Robert P. Stockwell discusses the characterization of the phonetic process that created the new Old English diphthongs (breaking). He argues that the usual assumption of retraction or backing as the phonetic trigger does not suffice and that it must be complemented by additional rounding. And this rounding is preserved, according to a further claim, in the output of the process, at least in mid and high vowel breaking. The second group of historical studies in this book deals with language change more generally. Sound change, of course, is also covered in the etymological papers, but the other two studies either completely abstract away from or only marginally touch upon sound change. This does not mean that they are not related to the research of the honoree. On the contrary, four out of the five papers in this group address issues that directly or indirectly connect with another area of study that has attracted Theo Vennemann's curiosity during the last decade, that of pre-Indo-European relics in the languages of North-West Europe. Raymond Hickey's paper Language change in early Britain: the convergence account addresses this topic most directly. Vennemann finds that the languages of North-West Europe have been determined by three factors: Indo-European heritage, Vasconic substrate, and Semitidic super- and substrate. Since this scenario heavily relies on language contact, Hickey warns that explanations by contact alone are rather weak and that they become stronger (multicausal) when they are complemented by internal factors, therefore yielding a convergence account. He discusses both the parallels between Semitic and Insular Celtic as well as the latter and English. The connection between Celtic and Basque is the background of Peter Schrijver's study on Irish ainder, Welsh anner, Breton annoar, Basque andere. Vennemann and others have suggested a Celtic borrowing from Basque, others have argued for the opposite, while still others see a common, unknown source behind these forms. To settle the issue, Schrijver analyzes the problematic aspects of the Celtic forms and concludes that the results support the first view. Some problems remain, but the picture of a large Vasconian speaking area in early Europe gains in plausibility.

Introduction

5

The topic of toponymy is central in Theo Vennemann's investigations of early European languages, since toponyms tend to be reliable indicators of substratal languages. So, Georg Bossong's study Der Name al-Andalus: neue Überlegungen zu einem alten Problem ('The name al-Andalus: new thoughts about an old problem') should be a welcome birthday present for the honoree. Without coming to a conclusive result, the article reveals interesting possible connections of the name in question with the concepts 'slope' or 'margin' and 'long'. Philip Baldi's article Latin ipse, Continental Celtic -xsi: a tentative proposal is offered "to one of today's most imaginative practitioners of the art of etymology, Theo Vennemann". Drawing on recent archeological findings, Baldi proposes to interpret the parallelism of the two forms in the title of his paper as evidence for an isogloss joining Continental Celtic and Latin. The last historical study to be introduced here is Elvira Glaser's "Fein gehackte Pinienkerne zugeben!" Zum Infinitiv in Kochrezepten ('"Add finely chopped pine nuts!" On the use of infinitives in cooking recipes'). Glaser discusses the changing preferences for verbal forms in cooking recipes in German, observing that the preference for infinitives is a rather recent innovation, which is shared by a large number of European languages. She supports a multicausal explanation in a similar fashion as Hickey: Pragmatic factors, contact, as well as the grammatical inventory of the language in question seem to play a role. The second section of this volume consists of ten synchronic studies on sounds, words, and uses. Again, the most extensively discussed topic is sounds. Larry Hyman's reflections on Cyclicity and base non-identity begin with the generally accepted notion of base identity, i.e. the assumption that a cyclically derived form closely resembles its base. According to Theo Vennemann, this phenomenon is due to the principle of the dominance of semantically primitive categories. Hyman states that this standard assumption is an over-simplification for two reasons: first, cyclicity does not always come from linear affixation but more complicated cases like interfixation have to be taken into account as well, and second, cyclicity does not always result in increased base identity. In a joint paper, Patrizia Noel Aziz Hanna, Katrin Lindner, and Andreas Dufter investigate The meter of nursery rhymes: universal versus language-specific patterns. Their findings coincide with the Maxim of

6

David Restle and Dietmar Zaefferer

Natural Versification formulated by the honoree of this volume in 1995, according to which natural poetic metrics only stylize linguistic properties of everyday language: Their data, which are drawn from an exploratory sample of fifteen languages, show that the metrics of nursery rhymes are not the same across languages, and they support the proposition that autochthonous verse forms are shaped by language specific prosody. Another paper that links phonological insights and the analysis of literary forms is Beatrice Primus' Unreine Reime und phonologische Theorie ('Off-rhymes and phonological theory'). Like Hickey and Glaser, she favors a multicausal analysis, arguing that in order to explain the asymmetry between the five sound changes that license literary off-rhymes and those that do not, it does not suffice to adduce sociolinguistic factors and that they have to be complemented by socioliterary ones. David Restle's paper on Choctaw intensives and syllable theory takes us back to morphology and phonology proper. The author uses his theory of syllabic oscillation to tackle a notoriously recalcitrant descriptive problem in morphophonology, namely the complex type of ablaut found in Choctaw intensives. Together with the broken plural of Arabic, it has been one of the prime examples used to show the merits of the theoretical framework called Prosodic Morphology. Restle dismisses previous treatments mainly for two reasons: their unwarranted complexity and their failure to reflect the iconic character of the prosodic base extension when it is used to signal intensification. He utilizes a single rule: infixation of a syllabic oscillation. Renate Raffelsiefen's study on Imperatives: the relation between meaning and form shares with Joan Bybee not only the major reference to Vennemann's (1974) statement that the mental lexicon contains only surface word forms but also the functionalist conviction (expressed by Bybee in other places) that form, especially morphological form, is motivated by function. Nevertheless, in an exploratory sample of five languages, she finds evidence not only that imperatives are formed by clipping final segments, especially suffixes, from existing verb forms but also that phonological bases need not be semantic. This impels her to ask a set of intriguing questions that ought to spur further research. Another influence from one of the many inspiring ideas of the honoree is documented in Peter Eisenberg's study of Struktur und Akzent

Introduction

7

komplexer Komposita ('Structure and accent of German compounds'). He starts from Yennemann's proposal that sometimes there may be a mismatch between the prosody and the morphology of German nouns, which is exemplified by polysyllabic simplicia which receive the accentuation pattern of corresponding compounds. Eisenberg suggests accounting for the variation of accent placement found in German complex compounds where the second component is a deverbal nominalization by assuming another kind of mismatch: Whereas grammatical build-up is determined by morphological constraints, the prosody marks focus structure and thus obeys syntactic principles. Again, as with simplicia and compounds, the prosody shows properties of the higher level, ignoring in a way the 'correct' analysis of the formal structure. Since 1983, the notion of linguistic preferences and preference laws has played an important role in Theo Vennemann's publications. Joachim Jacobs investigates this notion in his explanatory account of the orthographic representation of separable verbs Warum wir zusammenschreiben nicht immer zusammenschreiben - Präferenzgesetze im Schriftsystem (,Why zusammenschreiben is sometimes spelled zusammen schreiben. Preference laws in orthography'). Since Optimality Theory also works with preference laws (soft constraints), it is only natural that he also uses this tool for his analysis. He shows that three preference laws/constraints together with a plausible ranking suffice to account for the data. The last three papers deal with lexical morphology, lexical semantics, and lexical pragmatics, respectively, and therefore are not focused on one of the main research fields of the honoree. But since they are written by students of Theo Vennemann's, they document a quality of this scholar as a teacher that deserves to be mentioned: He does not keep his students in leading strings. Therefore, they do not form a close-knit 'school' but have developed into quite different directions. Claudia Gerstner-Link, for instance, has decided to do fieldwork on an endangered language in Papua New Guinea, and her paper about Noun classification and composition in Kilmeri investigates the special kind of classification system that can be observed in the natural kind and place-name terms of this language. Comparing it to other nominal classification systems, she shows that it is closer in type to classification systems found in certain Australian languages than to those of other Papuan languages.

8

David Restle and Dietmar Zaefferer

A finger-exercise in lexical semantics is presented in Dietmar Zaefferer's paper The puzzle of the autoantonymous argument role. Unraveling the polysemy of RISK/RISKIEREN. Drawing mainly on Fillmore and Atkins' and on Pustejovsky's analyses of the paradoxical behavior of this special verb that is able to produce synonymous sentences from antonyms when they are inserted in its direct object position, he studies several alternative analyses before reaching the somewhat surprising conclusion that at a higher level of abstraction there is no polysemy at all and that the preferred interpretations found at lower levels are explainable in terms of the well-known phenomenon of redundancy reduction with cognate argument verbs. As a by-product, a generalization of Talmy's Law on verbs of motion (again a preference law) is proposed which constrains the number of ontological levels conflated in monomorphemic verbs to two. In his study on the lexical pragmatics of measurement terms entitled Be brief and vague! And how Bidirectional Optimality Theory allows for verbosity and precision, Manfred Krifka makes crucial use of preference laws and, like Joachim Jacobs, he embeds them in the context of Optimality Theory, but he works with a different version. His topic is the relation between granularity or precision level and linguistic form in measurement terms; he explains our preference for vague interpretations of brief expressions and for precise interpretations of longer ones by the interaction of the two principles formulated in his title: Bidirectional Optimality Theory yields them as the two superoptimal combinations out of the four possible ones. And the application of this apparatus to the lexical pragmatics of measurement terms turns out to be just a special case of the general principle of iconicity in markedness, which states the preference for matching relative degrees of markedness: Increase in markedness on the expression-side tends to parallel increase in markedness on the side of the coded meanings. Sadly missing from the present collection of papers offered to Theo Vennemann on the occasion of his 65th birthday is a study entitled Langformen versus Kurzformen im Optativ Präsens der althochdeutschen schwachen Verben - Ein Diskussionsbeitrag ('Long vs. short forms in the optative present paradigm of the weak verbs of Old High German - A contribution'). We were deeply moved when we heard that Wolfgang Ullrich (called Gustav) Wurzel, who had been a good friend of the honoree for many years and gladly accepted our invi-

Introduction

9

tation to contribute to this book, suddenly and unexpectedly passed away one day after having celebrated his 61st birthday on August 4, 2001. Although little more than two pages of the abovementioned paper had been finished, Andreas Bittner kindly offered to complete it. To our regret the time frame was too narrow for the present volume, but we sincerely hope that this posthumous work will be available in a different place in the near future. We have mentioned the fact that the research interests of Theo Vennemann's students are not limited to the topics that he prefers. It is true on the other hand that some of his major areas of investigation, perhaps to his occasional regret, have not been further explored by his students, and, partially for that reason, are not represented here. We will only mention the typology of word order and word order change and refer the reader to the bibliography of the honoree. Like other explorers of new territory, he occasionally made the disappointing experience of discovering that others had been there before, but since this was forgotten, there was no reason for him to stop his efforts to make his findings generally known, and so he went on. Like other searchers who avoid trodden ground, he sometimes encountered hostility from those who prefer the general route. But whoever knows him personally will know the energy that helps him overcome all these obstacles: It can be heard in the enthusiastic tone of his voice when he reports on his latest discoveries in unknown territory. This image of the relentless explorer and discoverer seems to have also inspired Georg Bossong in the last paragraph of his contribution. We now want to pass the word on to him since we could not to say it any better: "Theo Vennemann represents the prototype of the researcher who lets no difficulty prevent him from going his way persistently and consequently. In doing so, he has opened entirely new, truly breathtaking perspectives. For him, the happiness of exploring has been accomplished, true to the saying of Hugo Schuchardt, the great Romance and Basque philologist (1925: 33): 'True finding consists in true searching'". In the name of his colleagues, students, and friends near and far, we wish him many more years of true searching and true finding.

I. Historical studies on sounds, words, and systems

A. Sound change

Preglottalization in English and a North Germanic bifurcation Henning Andersen

0. Introduction Several recent publications testify to the continued interest in understanding the historical and typological significance of preaspiration and preglottalization in the languages of northern Europe. Both Eliasson (2000) and Hansson (2001) focus primarily on preaspirated plosives and the remarkable areal distribution of preaspiration in northern Europe, from Iceland and the Outer Hebrides in the west across the north of the Scandinavian peninsula to the Kola peninsula in the east. Like Salmons (1992) and unlike Liberman (1982) neither of these authors discusses preglottalization in detail, but both papers suggest that a confrontation of preaspiration with preglottalization, particularly in English, can add a valuable perspective to the general questions they raise. Preglottalization in British English has been the topic of several publications, including a monograph by Björn Stälhane Andresen (1968; see Laver 1994 for more recent references). Andresen offered a synchronic sketch of the distribution of preglottalized plosives in standard spoken British English. His survey of the relevant literature showed that although this feature of plosives was not part of English Received Pronunciation at the beginning of the 1900s, it has been spreading across England, roughly from north to south, since the mid-1800s and has entered standard speech, apparently as a change 'from below', since the period between the two World Wars. When I read Andresen's study shortly after its publication, I knew from my own experience that in the U.S. and Canada preglottalization is widespread but quite variable. It occurred to me that this impression would be explained if there was currently a development of preglottalization in progress in North America similar to the one in England, and I wondered if this might be the case.

16

Ηenning A ndersen

1. The investigation Accordingly I decided to conduct a small investigation, intended as a pilot study, and collected data from some three dozen speakers who happened to be available and amenable.1 None of them was told ahead of time what I was after, but each accepted the idea that some words could be pronounced with slight, almost inaudible differences by different people, and that it was interesting for a linguist to have examples of this. Each then read aloud a set of randomized words, written on file cards, as if dictating them, while I noted the presence or absence of preglottalization in each item. At the end of each session I tried to satisfy my subject's curiosity by pronouncing a few examples with and without preglottalization - e.g., Pete as [phit] and [phi?t], patch as [phaec] and [phae?c] - and answering whatever questions the matter might give rise to. The data are discussed in Section 2 below.2 Articulatorily the essence of preglottalization is a glottal closure that precedes the buccal articulation of a (tense) obstruent. The closure may be accompanied by a raising of the larynx, which increases supra-glottal pressure (so Laver 1994: 330-333) and/or reduces subglottal pressure (Ringgaard 1974: 88). The glottal closure has no audible release, and preglottalization is heard simply as an absence of sound between the end of a vowel or sonorant and the following obstruent. In the case of unreleased buccal plosives, as commonly in words such as sap, sat, sack, it seems enough of the formant transitions from the vowel to the plosive is present before the glottal closure to allow identification of the strictly speaking inaudible plosive. But in running speech the identification of such plosive tokens may depend more on contextual semantic redundancy than on actual properties of the speech signal. In varieties of British English, a glottal plosive often replaces /t/, as in water [wo?9], bottle [bo?l] (Andresen 1968, Laver 1994). In North America, since there is usually no plosive glottal release, the glottal closure normally does not replace, but accompanies the buccal articulation of obstruents. One common exception is /t/ in the environment [n...n], e.g., fountain [faun?n], Clinton [klin?n], where a glottal plosive is often proxy for the faucally released /t/-allophone that may occur in slightly more distinct diction.3 The glottal closure that implements preglottalization may be articulated more or less crisply, and between the abrupt closure (in the termi-

Preglottalization

in English

17

nology introduced by Sweet (1877), a "stopped vowel ending") and the simple cessation of glottal vibrations (Sweet's "clear vowel ending") the attentive listener can discern a number of intermediate, more or less creaky, types of vowel ending. In evaluating such tokens in my data I judged them against the categorical distinction in my native dialect of Western Jutish, on which more below. There, differences such as [kat][ka?d] are phonemic, cf. WJut. kat 'cat', kat' 'cats', pak [pak] 'riffr a f f , pak' [pa?g] 'package', hjaslp [hjelp] 'help', hjaelp' [hjel'b] 'to help'. 4 See further section 4. Table 1. Sample test words A

Β

la b c d 2a b c d 3a b c d 4a b c la b c d 2a b c d 3a b c d 4a b c

deep cramp harp kelp happen dampen Gilpin deeper camper carper helper grapple dimple purple oomph turf golf muffin symphony orphan dolphin reefer comfort surfer pilfer raffle

boot pant flirt fault beaten wanton curtain molten Peter painter porter halter whittle mantle hurtle

beach pinch birch filch kitchen penchant merchant

tenth forth filth

prince worse else fasten transom worsen balsam mouser cancer courser ulcer pestle council parcel

lengthen Waltham Reuther panther

bleecher rancher orchard vulture satchel

teak tank park milk deacon sunken darken silken packer tanker porker milker buckle twinkle sparkle

marsh Walsh passion tension portion revulsion pressure

bushel provincial partial

18

Henning A ndersen

The segments and environments tested for preglottalization are illustrated in Table 1, which contains one half of the test words used. The table is organized as follows: A - plosives, Β - fricatives; 1 - word final, 2 - before syllabic nasal, 3 - before syllabic /r/, 4 - before syllabic /l/; a - after vowel, b - after nasal, c - after /r/, d - after /l/. This is a larger range of environments than were examined in Andresen's investigation. It is limited to obstruents following stressed vowels in two metrical environments, which were assumed to be the environments where the phenomenon is most common. It is far from exhausting all the phonotactic contexts, even under stress, in which preglottalization occurs. But it provides a sufficient sample to serve as point of departure for a future, fuller study, and it is sufficiently systematic to suggest what would be worth investigating in greater detail. Table 2. Segments and environments with preglottalization, subject JM A

la b c

Vp

Vt

Vc

Vk

mp

nt

nc

0k

rp

rt

rc

rk

It

lc ncn

d

2a b c

Β

Vpn mpn

ηθ

ns

If

1Θ

Is

mfn

ί)θη

nsm

Vtn rcn Ism

d

3a b c d

4a b

nsn

Vcr ncr

mfr

n9r

nsr

rcr lpr

lcr

lfr

lsr

Vcl nsl

C

Table 2 shows the environments with preglottalization in the data collected from one subject, JM (no. 43 in Table 4). Most subjects showed less consistent preglottalization in some environments than in others. To get an idea of the degree of variation in consistency, I had one of the subjects read the test words altogether ten times over several weeks. The cumulative result of these readings, dis-

Preglottalization

in English

19

played in Table 3, shows that there is a wide range of consistency even in the maximally explicit pronunciation one expects in dictation style - from the possible maximum of twenty tokens (represented by just three segments, each in one environment) to one token (five segments in four environments). Compare the two tabulations for informant KM (no. 39 and the last line) in Table 4. Table 3. Subject KM. Segments and environments with preglottalization, first reading (underlined) and number of tokens for ten readings

Β

A la b c d 2a b c d 3a b c d 4a b c

Y e 19

Vt

18

Vc 19

Vk 16

mp 19

nt 20

nc 16

rjk 20

mf 18

ηθ 17

ns

18

19

rt

19

19

rk

14

rf

8

r0

4

rs

4

rs

7

19

It

14

re lc

9

lk

19

If

8

1θ 20

Is

14

Is

7

IE

Vtn 1 ntn

1

ηθη 2

ncn 2

nsn 4

rcn 2 lsn

3

nsl

2

Vcr 7

lEI

1

ncr

1

rcr

7

lcr

2

ηθΓ

1

nsr

2

lsr

2

nsl

5

Vcl 5 rkl

1

2. Analysis Since the subjects come from different parts of North America and are heterogeneous with respect to age, gender, and class, these data do not constitute a proper sample. Still, the study revealed several clear-cut differences with respect to phonological conditioning which can be presumed to correspond grosso modo to the gradual establishment of preglottalization as a regular feature of pronunciation, that is, they show a change in progress.

20

Henning Andersen

Table 4. Environments with preglottalization across subjects Env. Classes No. of env. Informants 1 AG, NY 2 LP, NY RG,NJ 3 4 PF, NY 5 DN, BC BF, HA 6 AI, Man 7 8 RC, NY 9 DM, Man 10 ΜΗ,ΝΥ 11 TR, NY 12 GC, PQ 13 JC, PQ 14 EL, Alta 15 JS, Alta 16 NP, NY 17 DD, NY 18 WB, MA 19 JH, Man 20 VH, CA 21 RS, NC 22 MD, Alta 23 DS, TX 24 MH, CA 25 CO, BC 26 MS, CA 27 RR, Alta 28 AE, NJ 29 KC, CA 30 RA, WA 31 MP, CA 32 MN, BC 33 AS, WA 34 JB, TX 35 JW, MA 36 LS, Alta 37 RM, BC 38 HA, BC 39 KM, BC 40 TN, BC 41 DW, BC 42 NN, BC 43 JM, BC 39, 10 readings

Al 16

B1 11

3 4

A2 14

B2 14

A3 16

B3 11

A4 10

B4 7

1 4

5 6 7 7 7 8 8 6 10 10 6 12 7 9 11 12 13 16 11 12 14 14 12 16 15 14 13 15 13 16 16 16 15 16 16 15 16 15 15 14 16

1 1 1 1 5 2 1

3 1 2 2 2 2 3 1 1 5

1 1 3 2 1 1 2 3 2 1 4 4 3 7 5 4 6 5 4 3 4 4 4 6 8 6 4 5 4 11

3 2 3 1 5 1 3 5 5 2 4 3 2 3 4 2 3 3

1 1 1 1 1 1 1

1 1 3

1

1

1 3 1 1

1

1

1 1 1

2 2 5 1 2 1 3 5 5 4

2 2 1 2 2

1 1 1 2 2 3 2 5 3

1 2 1 1 2 1 1 3 1 5 5 4 5 8 2 5 5 6 5 5

1 1 1 1 1 1 1 4 1 2 4 3 2 5 3

1 3 1 2 2 1 1

1

1 1 2 1 1 2

2 1 1 1 1 2

Total 99 4 4 4 9 9 10 10 10 12 12 13 13 13 15 15 16 16 16 16 19 19 21 21 21 22 23 24 26 26 26 26 27 27 27 28 32 32 32 34 36 36 37 40 46

Preglottalization

in English

21

First of all, preglottalization is more widely attested in plosives and affricates than in fricatives; compare column A l to B l , A2 to B2, A3 to B3, A4 to B4 in Table 4. Secondly, preglottalization is more in evidence before pause than within a word; compare column A l to A2-A and Bl to B2-A. Thirdly, it is more in evidence before syllabic nasal than before syllabic liquid; compare A2 to A 3 ^ · and B2 to B3^l·. Fourthly, it is more in evidence before syllabic Irl than before syllabic /l/; compare A3 to A4 and B3 to B4. Perhaps these steps in the progression can be interpreted in terms of markedness: preglottalization is introduced earlier in unmarked than in corresponding marked environments - in stops earlier than in fricatives, in monosyllables earlier than in polysyllables, before nasals earlier than before liquids, etc.; cf. Andersen (2001: 238-239). Other kinds of phonological conditioning can be discerned in the data, and perhaps also some differences among individual speakers in the progression of the change. But a broader data base is required to identify these parameters of variation with any assurance. In any case, the partial gradation that has been identified here proves significant when it is contrasted with data from other languages; see Section 5. Preglottalization appears to be more prevalent in the speech of subjects from the west (British Columbia, Alberta, Washington, California) than from the east (Massachusetts, New York, New Jersey); cf. Table 4. Three dozen subjects may be a tenuous basis for such an observation, but the correlation between place of residence in the subject's formative years and ranking in Table 4 is plain to see, and it would not be surprising if this change - like any other change - had progressed further in some areas than in others. It would obviously be interesting to have comparable data from a much greater variety of geographical locations. I return to this point in section 8. It is natural to wonder if preglottalization is subject to social conditioning, that is, if it carries social meaning. There may in fact be a sociolinguistic perspective discernible in the data, but one that is rather different from the parameters of variation to which sociolinguists usually direct their attention. As it happens, several of the subjects are close kin. The incidence of preglottalization in the speech of the sisters KM (no. 39) and JN (no. 40) is closer to that of their mother (no. 43) than to that of their father, DM (no. 9). Similarly, JN's husband DN (no. 5) and his

22

Henning Andersen

sister DW (no. 41) resemble their parents of the same sex, the father having no preglottalization, the mother MN ranking as no. 32 on the list. Both pairs of siblings grew up on relatively isolated homesteads, and probably without a coherent peer-group norm, during their adolescence. The fact that preglottalization may be subject to individual variation in nuclear families is very interesting. On one hand, it is clear from the relative inconsistency of the data summarized in Table 4, and especially of the extensive data from subject no. 23 in Table 3, that the development of preglottalization is at a stage where it has not become rulegoverned or automatized in the speech of the individual; it is far from the consistency of, for instance, postaspiration of /p t k/ before a stressed vowel or the perfectly regular abridgement of syllables before a tense obstruent. On the other hand, the gender-based social conditioning within families that has just been described - assuming that this is not just a coincidence - shows the beginning of the formation of interpersonal norms. This is potentially significant as an indication that it may be the social value of an innovation rather than its phonetic motivation that is the primary force driving its 'embedding' in the grammar of a community language: well before an innovated feature is established as part of a community norm, individual speakers interpret it as a sign of social identity.

3. Phonological interpretation From an articulatory or perceptual point of view it is reasonable to compare preglottalization with other kinds of glottal articulation, as does Laver (1994), who examines preglottalization together with prosodic glottalization under the general title "glottalization". But from a phonological point of view it is important to recognize the quite diverse functions of glottalization. In particular when you are dealing with a phonetic change in progress, you must ask what is the function of the phonetic feature that is being introduced. The study shows that in American English, preglottalization can accompany all tense obstruents in syllable-final position. This is the environment in which syllable peaks (simple or complex) are regularly abridged; contrast the full vs. abridged duration of the syllables in, for instance, mow, moan, mode vs. moat; pain, pained vs. paint; fell, felled

Preglottalization

in English

23

vs. felt\ or cur, curd vs. curt. This distribution suggests that preglottalization is a way of implementing - and adding salience to - this abridgement. Preglottalization, then, can be seen as functionally equivalent to the postaspiration of tense plosives in syllable initial position. In both environments tense obstruents are longer than lax obstruents in relation to the contiguous syllabic. In syllable-initial position, the greater duration of tense obstruents is realized as delayed onset of voice in the following syllabic (effected by postaspiration before vowels and by voicelessness in prevocalic sonorants); e.g. pea [p h i] vs. bee [bi], plead [plid] vs. bleed [blid]. In syllable-final position, the greater duration of tense obstruents is effected by an abridgement of the preceding syllabic - that is, an early cessation of voice - reinforced more and more widely by preglottalization; e.g. pleat[pli?t] vs. plead[plid]. The introduction of preglottalization, then, can be understood as part of a gradual elaboration of the distinction between tense obstruents and lax obstruents, in full accord with the definition of protensity proposed for this distinction by Jakobson and Halle (Jakobson 1971: 485, 554, 740-741).

4. Typological perspective 1: postaspiration and preglottalization Phonetically preglottalization in American English is directly comparable to Western Jutish preglottalization (Da. vestjysk sted), which has been investigated instrumentally by Ringgaard (1974: 65-90; see also Laver 1994: 332, Liberman 1982: 80-90, 252-260). In Western Jutland (and in a small area on the island of Funen), there is a phonemic distinction [ph t h k h ] vs. [b d g] before a stressed vowel and [?b ?d ?g] vs. [p t k] after a stressed vowel. In the practical notation used for Jutish dialects, these distinctions are represented in onsets as p, t, k vs. b, d, g and in codas as p\ t\ k' vs. p, t, k, thus WJut. ta-l [tho:l] 'speak', regen), die Kranzmayer (1956: 81) auf das 12. Jahrhundert datiert. An diesen Vokalen gab es nichts zu dehnen. Der Einfluss des Niederdeutschen auf die so genannte neuhochdeutsche Dehnung in offener Tonsilbe ist hier nur angerissen; die wenigen Mosaikstücke lassen ein Bild erkennen, das natürlich ein Trugbild sein kann. Es sollte aber deutlich geworden sein, dass es eine lohnende Aufgabe ist, die These durch eine genauere Analyse des Wortschatzes und der grammatischen Quellen vom 16. bis zum 20. Jahrhundert und ihre Wirkungen auf die Schreibung zu überprüfen.

Anmerkungen 1. Sogar das wichtigste: „lexical stress in German has its best correlate in acoustic duration which is distributed over the designated syllable" (Dogil 1995: 15). 2. Die Behauptung ist wesentlich älter, vgl. Vietor 1893: 9, Siebs 1898: 18f., ferner die Dissertation von Kirch (1952). 3. Sievers 1901: 222f. beschreibt den scharfen Schnitt („Kurzvokal") als ein Abschneiden des Vokals in einem Moment, wo er noch voll und kräftig ertönt, während er bei sanftem Schnitt („Langvokal") bereits deutlich geschwächt ist; Jespersen (1912: 153f.) beschreibt den Gegensatz als losen/festen Anschluss des Vokals an den folgenden Konsonanten. Auch in der heutigen Dialektologie wird es immer wieder etwas unsystematisch in phonetischen Beschreibungen

52

Thomas Becker

herangezogen. Lediglich Trubetzkoy (1939: 196) sieht es klar als phonologisch an („Silbenschnittkorrelation"); seine Sicht hat die Beschreibung der Phonotaktik der deutschen Standardsprache jedoch nicht beeinflusst. 4. Auf die eine der beiden Richtungen der Wechselbeziehung, die Aussprache nach der Schrift, weist Moser 1987 hin, auch auf die Autorität der humanistischen Leseaussprache des Lateinischen nach Erasmus von Rotterdam (1987: 389). 5. An den Bibeldrucken zeigt sich, dass sich die nhd. Norm der Schreibung der Vokalquantität im 18. Jahrhundert festigt (Rieke 1998: 252, 376). 6. Eckermann, Gespräche III. Teil (Friedr. Sorets Gespräche mit Goethe in Eckermanns Bearbeitung). Gegen Peters' (1999: 169) Behauptung geht er hier sehr wohl auch auf die Vokale ein, daher erlaube ich mir, einen umfangreichen Textabschnitt zu zitieren. 7. Heinrich Heine, Memoiren (Sämtliche Schriften in 12 Bänden, hg. v. Klaus Briegleb, München: Hanser, Bd. 11, S. 584f.). Das Vorbild des Ostfälischen („Hannover") gilt bis heute. Nach v. Bahder 1890: 60 war es bereits im Spätmittelalter im niederdeutschen Sprachgebiet vorbildhaft, weil die Bewohner des südostfälischen Raums früh zum Hochdeutschen wechselten (auch in der Alltagssprache) wodurch sie es naturgemäß besser beherrschten als die übrigen niederdeutschen Sprecher. 8. Diese Darstellung vereinfacht stark; die ausfuhrliche Darstellung in Becker 1996, 1998 und i.E. soll hier nicht wiederholt werden. 9. Für Fortiskonsonanten sieht schon Ickelsamer 1534 die Doppelschreibung vor, was sicher auch die Beschreibungen anderer Grammatiker beeinflusst hat. Die unterschiedliche Motivation der Doppelschreibung war den Grammatikern nicht bewusst, wodurch sich z.B. Schwierigkeiten bei Wörtern wie reißen oder reifen ergaben, wo der Fortischarakter des („Lautverschiebungs-") Konsonanten die Annahme von scharfem Schnitt und Doppelschreibung des Konsonanten nahe legte. Der (schriftbedingte?) Zusammenfall des „Lautverschiebungs"/ mit dem germanischen löste das eine Problem, die Erfindung der Ligatur β (< sz) das andere. Freyer 1722: 49ff. unterschiedet die f-Laute noch, sieht aber den Unterschied in der Schärfe des Diphthongs (S. 52). 10. Da es sich bei der nhd. Dehnung weder um Lautreanalyse im Sprachkontakt noch um phonetisch motivierte Dehnung handelt, sondern um die Interpretation von Schrift, ist eine entsprechende Senkung hier nicht zu erwarten. 11. Laschs Annahme, die tonlangen Vokale wären auf dem gesamten niederdeutschen Gebiet diphthongiert gewesen, lässt sich daher mit der Senkung nicht begründen, worauf bereits Frings 1915: 118 hinwies. Wortmann 1970: 348 wiederholt seltsamerweise ohne Diskussion Laschs Auffassung, obwohl er Schmitt 1931 auf derselben Seite (wenn auch in einem anderen Zusammenhang) zitiert. 12. Bei dieser Kürzung können die mittleren Vokale wieder in die zentralisierten hohen Vokale zurückfallen. Dass es sich nicht um unterbliebene Dehnung handelt, zeigen die gekürzten altlangen Vokale: helag > hillich.

Zur neuhochdeutschen Dehnung in offener Tonsilbe

53

13. Dies ist zunächst eine mittelniederdeutsche Schreibkonvention, die später auf das Hochdeutsche übertragen wurde, was wohl auf den Einfluss niederdeutscher Grammatiker zurückzuführen ist. Luther verwendete sie nicht in der ersten, aber in späteren Ausgaben seines NT (v. Bahder 1890: 55). 14. Vermutlich war im Mittelniederdeutschen der Anfangsrand bei Tonsilben obligatorisch, worauf der in Norddeutschland obligatorische glottale Plosiv in The[?]ater hinweist, der in Süddeutschland nicht üblich ist. Der Endrand war wohl stammintern notwendigerweise unbesetzt, worauf die Kürzung in geschlossener Silbe hinweist; mehrfach geschlossene Silbe scheint außer vor Morphem-/Wortgrenze nicht möglich gewesen zu sein. 15. Es wäre die extremste Verletzung des „Contact Laws", vgl. Vennemann 1988: 40fT. 16. Dieser Endrandkonsonant ist, wie in Fn. 12 vermutet, ohnehin stammintern verboten ist, worauf das Verbot von *bälke (als Simplex; mit dem Suffix - k e ist die Struktur erlaubt) hindeutet. 17. Bei emphatischer Betonung heute noch: Wa:::hn, Wan:::d, Wat::.le. Die Dehnung der Konsonanten in Implosionsposition ist heute noch meßbar (Becker 1998: 67), muß aber früher stärker gewesen sein, wenn man den Darstellungen wie der Adelungs glaubt. Es ist aber nicht auszuschließen, dass Adelung und andere ihren Beschreibungen Überlautungen zugrundelegen. 18. Koronale Laute werden mit Zungenspitze und Zungenblatt artikuliert; das ist der beweglichste und schnellste Artikulator, was bedingt, dass diese Laute auch außen in den Silbenrändern artikuliert werden können: auf Kurzvokal können nur zwei nicht-koronale Laute folgen, dann noch drei koronale: du darfst's. Dadurch sind Koronale besonders für Flexionssuffixe geeignet.

Literatur Adelung, Johann Christoph 1781 Deutsche Sprachlehre. Zum Gebrauche der Schulen in den Königl. Preuß. Landen. Berlin: Voß. 1782 Umständliches Lehrgebäude der Deutschen Sprache, zur Erläuterung der Deutschen Sprachlehre für Schulen. Erster Band. Leipzig: Breitkopf. Nachdruck 1971 Hildesheim: Olms. 1790 Vollständige Anweisung zur deutschen Orthographie nebst einem kleinen Wörterbuche fur die Aussprache, Orthographie, Biegung und Ableitung. Zweyte vermehrte und verbesserte Auflage. Leipzig: Weygand. 1793-1801 Grammatisch-kritisches Wörterbuch der Hochdeutschen Mundart mit beständiger Vergleichung der übrigen Mundarten, besonders aber der Oberdeutschen. 4 Bände. 2. vermehrte und verbesserte Aus-

54

Thomas Becker

gäbe Leipzig: Breitkopf und Härtel. 1793-1801. 2. Reprint mit einer Einführung und Bibliographie von Helmut Henne. Hildesheim 1990: Olms. Auer, Peter, Peter Gilles und Helmut Spiekermann (eds.) i.E. Silbenschnitt und Tonakzente. Tübingen: Niemeyer. Bahder, Karl von 1890 Grundlagen des neuhochdeutschen Lautsystems. Beiträge zur Geschichte der deutschen Schriftsprache im 15. und 16. Jh. Straßburg: Trübner. Bannert, Robert 1976 Mittelbairische Phonologie auf akustischer und perzeptorischer Grundlage. Lund: Gleerup und München: Fink (= Travaux de l'Institut de Linguistique de Lund 10). 1977 Quantität im Mittelbairischen: Komplementäre Länge von Vokal und Konsonant. In: Wolfgang Ulrich Dressler und Oskar E. Pfeiffer (eds.). Phonologica 1976. Akten der dritten Internationalen Phonologie-Tagung, Wien 1.-4. September 1976, 261-270. Innsbruck: Institut für Sprachwissenschaft. (= Innsbrucker Beiträge zur Sprachwissenschaft 19). Becker, Thomas 1996 Zur Repräsentation der Vokallänge in der deutschen Standardsprache. Zeitschrift für Sprachwissenschaft 15: 3-21. 1998 Das Vokalsystem der deutschen Standardsprache. Frankfurt am Main: Peter Lang. i.E. Silbenschnitt und Silbenstruktur in der deutschen Standardsprache der Gegenwart. Manuskript. Erscheint in: Peter Auer et al. (eds.). Besch, Werner 2000 Die Rolle Luthers für die deutsche Sprachgeschichte. In: Werner Besch, Anne Betten, Oskar Reichmann und Stefan Sonderegger (eds.): Sprachgeschichte. Ein Handbuch zur Geschichte der deutschen Sprache und ihrer Erforschung. 2. vollständig neu bearbeitete und erweiterte Auflage. Zweiter Teilband, 1713-1745. Berlin: Walter de Gruyter. Bödiker, Johann 1698 Grund-Sätze der Deutschen Sprachen in Reden und Schreiben. Berlin. Dogil, Grzegorz 1995 Phonetic correlates of word stress. Arbeitspapiere des Instituts für Maschinelle Sprachverarbeitung (AIMS) der Universität Stuttgart 2, Heft 2: 1-59. Ewald, Petra 1992 Das „Grundgesetz der Deutschen Orthographie" bei Johann Christoph Adelung: Darstellung und Wertung. In: Dieter Nerius und Jürgen Scharnhorst (eds.). Studien zur Geschichte der deutschen Ortho-

Zur neuhochdeutschen Dehnung in offener Tonsilbe

55

graphie, 61-89. Hildesheim: Olms (= Germanistische Linguistik 108-109). Fischer-Jargensen, Eli und Hans Peter Jergensen 1969 Close and loose contact („Anschluß") with special reference to North-German. Annual Report of the Institute of Phonetics of the University of Copenhagen (AR1PUC) 4: 43-80. Freyer, Hieronymus 1722 Anweisung zur teutschen Orthographie. Halle/Saale. Nachdruck mit einem Vorwort hg. v. Petra Ewald. Hildesheim 1999: Olms. Frings, Theodor 1915 Tonlange Vokale. Beiträge zur Geschichte der deutschen Sprache und Literatur 40: 112-126. Gottsched, Johann Christoph 1762 Vollständigere und Neuerläuterte Deutsche Sprachkunst, Nach den Mustern der besten Schriftsteller des vorigen und itzigen Jahrhunderts abgefasset, und bey dieser fünften Auflage merklich verbessert. Leipzig: Breitkopf. Ickelsamer, Valentin 1534 Teutsche Grammatica: daraus einer von im selbs mag lesen lernen, mit allem dem, so zum Teutschen lesen u. desselben Orthographiam mangel u. uberflus, auch anderem vil mehr zu wyssen gehört. Nürnberg. Zitiert nach: Johannes Müller 1882, Quellenschriften und Geschichte des deutschsprachlichen Unterrichtes bis zur Mitte des 16. Jahrhunderts. Gotha. Nachdruck Hildesheim 1969: Olms. 120-159. Jespersen, Otto 1912 Elementarbuch der Phonetik. Leipzig: Teubner. Josten, Dirk 1976 Sprachvorbild und Sprachnorm im Urteil des 16. und 17. Jahrhunderts: Sprachlandschaftliche Prioritäten, Sprachautoritäten, sprachimmanente Argumentation. Frankfurt am Main: Lang. Kirch, Max S. 1952 Der Einfluß des Niederdeutschen auf die hochdeutsche Schriftsprache. Glessen: Schmitz (= Gießener Beiträge zur deutschen Philologie 99). Kranzmayer, Eberhard 1956 Historische Lautgeographie des gesamtbairischen Dialektraumes. Wien: Böhlau. Lasch, Agathe 1914 Mittelniederdeutsche Grammatik. Halle: Niemeyer. Nachdruck 1974. Tübingen: Niemeyer. Lessiak, Primus 1933 Beiträge zur Geschichte des deutschen Konsonantismus. Mit einem Vorwort und einem Wort- und Sachverzeichnis von Ernst Schwarz. Brünn: Rohrer.

56

Thomas Becker

Morhof, Daniel Georg 1700 Unterricht von der Teutschen Sprache und Poesie. Hg. nach der Ausgabe Lübeck und Frankfurt 1700 von H. Boetius. (= Ars Poetica, Texte Bd. 1, Bad Homburg 1969.) Moser, Hans 1987 Geredete Graphie. Zeitschrift für deutsche Philologie 106: 379-399. Nerger, Karl 1869 Grammatik des meklenburgischen Dialektes älterer und neuerer Zeit: Laut- und Flexionslehre. Leipzig: Brockhaus. Nerius, Dieter 1989 Die Rolle J. Ch. Adelungs in der Geschichte der deutschen Orthographie. Sprachwissenschaft 14: 78-96. Niekerken, Walther 1952 Wechsel der niederdeutschen Vokalart durch Änderung in der Lautdauer (Vokalumstufung). Neuphilologische Mitteilungen 53: 185212.

Paul, Hermann 1884 Beitraege zur geschichte der lautentwicklung und formenassociation. 11. Vokaldehnung und vokal Verkürzung im neuhochdeutschen. Beiträge zur Geschichte der deutschen Sprache und Literatur 9: 101— 134. Peters, Robert 1983 Mittelniederdeutsche Sprache. In: Jan Goossens (ed.). Niederdeutsch: Sprache und Literatur. Bd. 1: Sprache. 2., verbesserte und um einen bibliographischen Nachtrag erweiterte Auflage, 66-115. Neumünster: Wachholtz. 1999 Zur Rolle des Niederdeutschen bei der Entstehung des Neuhochdeutschen. In: Walter Hoffmann, Jürgen Macha, Klaus J. Mattheier, Hans-Joachim Solms und Klaus Peter Wegera, (eds.), Das Frühneuhochdeutsche als sprachgeschichtliche Epoche. Werner Besch zum 70. Geburtstag, 161-173. Frankfurt a.M.: Peter Lang. Restle, David 2001 Silbenschnitt - Quantität - Kopplung. Zur Geschichte, Charakterisierung und Repräsentation der Anschlußprosodie unter dem Blickwinkel einer Oszillationssilbentheorie. Diss. München. Rieke, Ursula 1998 Studien zur Herausbildung der neuhochdeutschen Orthographie. Die Markierung der Vokalquantitäten in deutschsprachigen Bibeldrucken des 16.-18. Jahrhunderts. Heidelberg: Winter. Schirmunski, Viktor M. 1962 Deutsche Mundartkunde. Vergleichende Laut- und Formenlehre der deutschen Mundarten. Aus dem Russischen übersetzt und wissenschaftlich bearbeitet von Wolfgang Fleischer. Berlin: AkademieVerlag.

Zur neuhochdeutschen Dehnung in offener Tonsilbe

57

Schmidt, Oskar und Theo Vennemann 1985 Die niederdeutschen Grundlagen des standarddeutschen Lautsystems. Beiträge zur Geschichte der deutschen Sprache und Literatur 107: 1-20, 157-173. Schmitt, Alfred 1931 Akzent und Diphthongierung. Heidelberg: Winter. Schottel, Justus Georg 1663 Ausführliche Arbeit Von der Teutschen HaubtSprache: Worin enthalten Gemelter dieser HaubtSprache Uhrankunft, Uhraltertuhm, Reinlichkeit, Eigenschaft, Vermögen, Unvergleichlichkeit, Grundrichtigkeit, zumahl die SprachKunst und VersKunst Teutsch und guten theils lateinisch völlig mit eingebracht ; wie nicht weniger die Verdoppelung, Ableitung, die Einleitung, Nahmwörter, Authores vom Teutschen Wesen und Teutscher Sprache, von der verteutschung, Item die Stammwörter der Teutschen Sprache samt der Erklärung und derogleichen viel merkwürdige Sachen. Abgetheilet In Fünf Bücher. Braunschweig: Zilliger. Sendlmeier, Walter F. 1981 Der Einfluß von Qualität und Quantität auf die Perzeption betonter Vokale im Deutschen. Phonetica 38: 291-308. Siebs, Theodor 1898 Deutsche Bühnenaussprache. Ergebnisse der Beratungen zur ausgleichenden Regelung der deutschen Bühnenaussprache, die vom 14. bis 16. April 1898 im Apollosaale des Königlichen Schauspielhauses zu Berlin stattgefunden haben. Berlin: Albert Ahn. Sievers, Eduard 19015 Grundzüge der Phonetik zur Einfuhrung in das Studium der Lautlehre der indogermanischen Sprachen. Leipzig: Breitkopf und Härtel. Nachdruck 1976, Hildesheim: Olms. Spiekermann, Helmut 2000 Silbenschnitt in deutschen Dialekten. Tübingen: Niemeyer. Trubetzkoy, Nikolai Sergej ewitsch 19396 Grundzüge der Phonologie. Göttingen: Vandenhoek und Ruprecht. Vennemann, Theo 1988 Preference Laws for syllable structure and the explanation of sound change. Berlin/New York: Mouton de Gruyter. 1991a Syllable structure and syllable cut prosodies in modern Standard German. In: Pier Marco Bertinetto, Michael Kenstowicz und Michele Loporcaro (eds.), Certamen Phonologicum II. Papers from the 1990 Cortona Phonology Meeting, 211-243. Turin: Rosenberg und Sellier. 1991b Skizze der deutschen Wortprosodie. Zeitschrift für Sprachwissenschaft 10: 86-111.

58

Thomas Becker 1994

Universelle Nuklearphonologie mit epiphänomenaler Silbenstruktur. In: Karl Heinz Ramers, Heinz Vater und Henning Wode (eds), Universale phonologische Strukturen und Prozesse, 1-SA. Tübingen: Niemeyer. 1995 Der Zusammenbruch der Quantität im Spätmittelalter und sein Einfluß auf die Metrik. In: Hans Fix (ed.), Symposion Quantitätsproblematik in den Germanischen Sprachen des Mittelalters und die Metrik 1993, Greifswald, 185-223. (Amsterdamer Beiträge zur älteren Germanistik 42.) Amsterdam: Rodopi. Vietor, Wilhelm 1893 Wie ist die Aussprache des Deutschen zu lehren? Marburg: Elwert. Weithase, Irmgard 1930 Anschauungen über das Wesen der Sprechkunst von 1755-1825. Berlin: Ebering. Wortmann, Felix 1970 Zur Geschichte der kurzen Vokale in offener Silbe. In: Dietrich Hoffmann (ed.), Gedenkschrift für William Foerste, 327-353. Köln: Böhlau.

Lexical diffusion in regular sound change Joan Bybee

1. Introduction Lexical diffusion refers to the way a sound change affects the lexicon: if sound change is lexically abrupt, all the words of a language are affected by the sound change at the same rate. If a sound change is lexically gradual, individual words undergo the change at different rates or different times. Whether sound changes exhibit gradual or abrupt lexical diffusion is a topic that surfaces persistently in historical linguistics, but as yet has not reached resolution. One early contribution to this debate by Schuchardt has been brought to modern attention in the Vennemann and Wilbur (1972) translation in which Schuchardt observes that highfrequency words are affected by sound change earlier and to a greater extent than low-frequency words. In this chapter I document the tendency noted by Schuchardt in several ongoing sound changes that are gradual phonetically and I argue that such changes are best accounted for in an exemplar model of phonological representation, a model that is an elaboration of the model argued for in Vennemann (1974).

2. Regular sound change or lexical diffusion? The hypothesis that sound change is lexically regular seems well supported by the facts of change: when we observe that two languages or dialects exhibit a phonological difference it is very likely that this difference is regular across all the words that have the appropriate phonetic environment. This observation is fundamental to the comparative method; the establishment of genetic relations and the reconstruction of proto-languages are based on the premise that sound change affects all words equally. Schuchardt was one of the detractors from this position. When he observed sound change in progress, he noted that all words did

60

Joan Bybee

not change at the same rate, and the differences were not due to "dialect mixture" as often claimed by the Neogrammarians who supported the regularity position. A major challenge to the regularity position in the twentieth century is expressed in the writings of William Wang (1969, 1977; Wang and Cheng 1977) who documents changes that seem to occur word-by-word over a long period of time. While some of these changes result in lexical regularity, Wang and his colleagues also identify changes that seem to be arrested after affecting only part of the lexicon. William Labov (1981, 1994) also deals with the issue availing himself of the data from his numerous studies of sound change in progress. His proposal is that there are two types of sound change: "regular sound change" which is gradual, phonetically motivated and without lexical or grammatical conditioning and is also not influenced by social awareness; and "lexical diffusion change" as in cases studied by Wang, which are "the result of the abrupt substitution of one phoneme for another in words that contain that phoneme" (1994: 542). He observes this type of change most often "in the late stages of internal change that has been differentiated by lexical and grammatical conditioning" (1994: 542). Labov even goes so far as to propose that certain changes, such as the deletion of glides and schwa, will be regular changes, while the deletion of obstruents will show lexical diffusion. A number of researchers have challenged this position. Phillips (1984) has argued that even low-level sound changes exhibit gradual lexical diffusion. Oliveira (1991) argues also that it is likely that gradual lexical diffusion occurs even in changes that turn out to be regular. Krishnamurti (1998) demonstrates that the change of s > h > 0 in Gondi exhibits gradual lexical diffusion but still goes through to completion in some dialects. In this paper I will review evidence that even gradual, phonetically-conditioned change exhibits gradual lexical diffusion, though it is perhaps of a more subtle nature than the lexical diffusion studied by Wang and Labov. The lexical diffusion presented here for reductive phonetic change is highly conditioned by word-frequency. In Hooper (1976) I identified a lexical diffusion paradox: reductive sound change tends to affect high-frequency words before lowfrequency words, but analogical leveling or regularization tends to affect low-frequency words before high-frequency words. Working from this observation, Phillips (1984, 2001) has studied a number of changes

Lexical diffusion in regular sound change

61

that move in each direction and has attempted to refine a hypothesis that predicts the direction of lexical diffusion. In this paper I will only discuss the first direction of change, the characteristic of reductive sound change, with the intent of demonstrating that word-frequency affects even changes that may have seemed regular to the Neogrammarians.

3. Frequency effects on regular sound change Sound changes that are complete can be identified as regular or not, depending upon whether or not they affected all lexical items existing at the time of the change. Ongoing changes cannot be designated as regular or not, since they are not complete. However, one can reference the typical characteristics of a change to project whether it will be regular or not. That is, a phonetically gradual change that has a clear phonetic conditioning falls into Labov's first type and we can project its regularity. In the following I document lexical diffusion from high-frequency to low-frequency words in ongoing changes that can be expected to be regular, as well as in certain reductive changes that may never be complete because of the nature of the lexicon.

3.1. American English t/d-deletion Consider the deletion of final /t/ and /d/ in American English, which occurs most commonly in words ending in a consonant plus /t/ or /d/, such as just, perfect, child or grand. This much-studied variable process has been shown to be affected by the preceding and following consonant, deleting more if a consonant follows, by grammatical status, deleting less if the /t/ or /d/ is the regular past tense of English, and by social and age factors (with more deletion among younger, lower socioeconomic class speakers) (Labov 1972; Neu 1980). My own study of the deletion of /t/ and /d/, using a corpus of phonological variation in Chicano English speakers in Los Angeles (Santa Ana 1991) focussed on lexical frequency as a factor (Bybee 2000). Using 2000 tokens of final /t/ and /d/ following a consonant and referencing word-frequency from Francis and Kucera (1982), I found that deletion occurred more in high-frequency words. Table 1 demonstrates this

62

Joan Bybee

effect with a cut-off point of 35 per million. This number was chosen because I was also interested in whether or not a frequency effect could be found among regular past tense verbs (see Table 3) and this figure is the median for such forms in Francis and Kucera (1982). Table 1. Rate of t/d-deletion for entire corpus by word-frequency

high-frequency low-frequency

deletion

non-deletion

898 137

752 262

% deletion 54.4% 34.3%

Chi-squared: 41.67, ρ < .001, df = 1

Similar results were obtained by Jurafsky et al. (2001) using 2042 monosyllabic content words ending in Ν or /d/ from the Switchboard corpus, which is a corpus of telephone conversations by monolingual American English speakers. They found a strong effect of wordfrequency (p < .0001). Another study, using both function and content words and polysyllabic words in addition to monosyllabic ones in the Switchboard corpus, found a higher level of significance for the association of word-frequency with final /t/ and /d/ deletion (p < .00005) (Gregory et al. 1999). Will final /t/ and /d/ deletion after a consonant turn out to be a regular sound change? There is certainly precedent for such a change being regular in the end, especially in certain phonetic contexts. Final consonant deletion in French was completely regular when a consonant followed, but less so when a vowel followed (Harris 1988). In English erosion has been working on final consonants for some time. The deletion of a Pol and /g/ after a homorganic nasal, as in bomb and gang, was completely regular and leaves English speakers virtually unable to produce final [mb] or [ng] clusters. Final /nd/ could certainly follow and delete regularly as well. Can the deletion of an obstruent be phonetically gradual? Labov (1994) lists obstruent deletion under "lexical diffusion" changes, presumably because he considers obstruent deletion to involve the phonetically abrupt loss of a phoneme. However, there is evidence that the reduction of final /t/ and /d/ can be gradual. A final /t/ or /dJ may vary in length. This length variation occurs under the same conditions and in the same direction as the deletion variation. Losiewicz (1992) has shown that monomorphemic /t/ or /d/ are shorter in duration than regular

Lexical diffusion in regular sound change

63

past tense /t/ or /d/. As mentioned above, monomorphemic /t/ or /d/ are also more likely to delete. Figures from Bybee (2000) confirm this, as shown in Table 2. Table 2. Rate of deletion for regular past tense compared to all other words of comparable frequency (403 or less) percentage deletion 44.1%

all words: -ed verbs

22.6%

In addition, Losiewicz found that a final past tense Ν or /d/ was longer in low-frequency verbs than in high-frequency verbs. This finding parallels the deletion data presented in Bybee (2000), as shown in Table 3. Regular past tense /t/ and /d/ are more likely to delete in highfrequency verbs than in low-frequency verbs. Table 3. The effects of word-frequency on t/d-deletion in regular past tense verbs (non-prevocalic only)

high-frequency low-frequency

deletion

non-deletion

% deletion

44 11

67 47

39.6% 18.9%

Chi-squared: 5.00313, ρ < .05, df = 1

The data on this obstruent deletion process, then, demonstrates both lexical and phonetic gradualness. It thus cannot be said that obstruent deletion is the abrupt deletion of a phoneme. In fact, these data are problematic for any version of phonemic theory. A model that can accommodate these data will be presented in section 4.

3.2. Spanish

[ö]deletion

Another good candidate for a phonetically gradual change in progress that exhibits lexical diffusion and that could turn out to be regular is the deletion of intervocalic /d/ or [ö] in many dialects of Spanish. D'Introno and Sosa (1986) regard the variants ranging from [d] to [ö] to 0 as a continuum, affirming that the reduction is gradual in this case. My own

64

Joan Bybee

study of 571 tokens of intervocalic Idl in Spanish spoken by native New Mexican speakers reveals that the rate of deletion is higher among highfrequency words. The frequency count used for Table 4 is taken from the 1.1 million word Corpus oral de referenda del espahol contemporäneo (COREC). Past participle tokens were removed because the data also showed that past participles have a higher rate of deletion that other items. Table 4. Rate of deletion according to token frequency for all non-past participle tokens in the small corpus using the COREC as a measure of frequency

Retention Deletion Total

Low (0-99)

High (100+)

Total

243 23 (8.6%) 266

287 78 (21.4%) 365

530 101 (16.0%) 631

Chi square (N=631, df=l)= 17.3 p feo 'ugly'; pedes > pies ' f e e t f i d e > fe 'faith'; audire > oir 'to hear'; limpidu > limpio 'clean'; etc. but a few words maintain the /d/, e.g. sudare > sudar 'to sweat'; vadu > vado 'ford'; crudu > crudo 'raw'; nidu > nido 'nest'; nudu > (des) nudo 'naked'. However, all of these words are attested without the d in Old Spanish, leading Penny (1991) to suggest that the form with the d is influenced by the Latin spelling. Others have argued that the deletion was constrained by whether or not the resulting vowel combination obeyed the phonotactics of Spanish at the time (see Pensado 1984). Thus while the earlier change was not completely regular, it was comparable to the present deletion in being phonetically gradual and it affected a large majority of words with intervocalic Latin d. I suspect the Neogrammarians would have counted it as "regular."

Lexical diffusion in regular sound change

65

3.3. Vowel shifts Labov (1994) searches his data on vowel shifts in American English for evidence of lexical diffusion. What he is looking for are cases in which words with the same phonetic environment for the vowel have distinctly different vowel realizations, as opposed to small differences that could be part of a phonetically gradual continuum. The vowel changes reviewed in his Chapter 16 show detailed phonetic conditioning, gradualness and lexical regularity. Labov even tests to see if homonyms can split given differences in their token frequency (pp. 460-465). This test, on words with very similar phonetic environments, such as two, too, to, do, through, turns up no lexical diffusion by word-frequency. The difficulty with this test is that all of the words used occurred three or more times in the interview, and thus must be considered to be of highfrequency. In addition, taking the very small set of words necessary to get comparable phonetic environments yielded so few words that it would be very difficult to discover any general word-frequency effect in these gradual changes. Indeed, it might be that for vowel shifts the phonetic environment is generally more powerful than any other effects because there are fewer words in each phonetic category. However, some lexical diffusion is found in Labov's data on vowel shifts. The most famous and still unexplained case concerns the raising of short [ae] which affects the adjectives ending in [d] mad, glad and bad, but not sad. In this same shift some evidence for lexical diffusion by frequency is cited: Labov notes (1994: 506-507) that when wordinitial short [ae] "occurs before a voiceless fricative, only the more common, monosyllabic words are tensed: tense ass and ask; lax ascot, aspirin, astronauts, aspect, athletic, after, African, Afghan." Similarly, in Moonwomon's (1992) study of the centralization of /ae/ in San Francisco English, she finds that in the environment before a fricative this vowel is more centralized than before a non-fricative; it is also more centralized after [1]. The most commonly used word with this pair of phonetic environments is class. Class shows more centralization than the other words with these two environments: e.g. glass, laugh and so on. Moonwomon also studies the fronting of /a/ in the same speakers. Here a following /t/ or /d/ conditions more fronting than other consonants. Of the words in the corpus ending in final /t/, got is the most fre-

66

Joan Bybee

quently occurring. Moonwomon also shows that the fronting in got is significantly more advanced than in other words ending in alveolars, such as not, god, body, forgot, pot, and so on. It appears, then, that some evidence that high-frequency words undergo vowel shifts before low-frequency words can be found. As mentioned above, it may be more difficult to discern frequency effects in vowel shifts because of the effects of the preceding and following environments, which narrow each phonetic class to a small number of words.

3.4. Vowel reduction and deletion In addition to consonant reduction, as discussed in sections 3.1 and 3.2, another type of change that shows robust word-frequency effects is vowel reduction and deletion. Fidelholz (1975) demonstrates that the essential difference between words that do reduce a pre-stress vowel, such as astronomy, mistake and abstain, and phonetically similar words that do not, such as gastronomy, mistook and abstemious, is wordfrequency. Van Bergem (1995) finds that reduction of a prestress vowel in Dutch also is highly conditioned by frequency. The high-frequency words minuut 'minute', vakantie 'vacation', andpatat 'chips' are more likely to have a schwa in the first syllable than the phonetically similar low-frequency words, miniem 'marginal', vakante 'vacant', and patent 'patent'. Deletion of reduced vowels is also conditioned by word-frequency. Hooper (1976) asked native speakers of American English for their judgments concerning whether or not a post-stress schwa was usually, sometimes, or rarely deleted in words such as every, memory and family. The results showed that subjects exercised some phonological constraints on schwa deletion, but aside from these, the contexts in which deletion occurred were highly influenced by word-frequency. That is, deletion was more likely in nursery than in cursory, in memory than in mammary, in scenery than in chicanery, and so on. These deletions appear to be phonetically gradual in that the variants range from those in which a schwa is followed by a resonant, in these cases, [r], to variants in which the resonant is syllabic, to cases in which the resonant is not syllabic, i.e. all syllabicity has been lost. Such

Lexical diffusion in regular sound change

67

changes, then, are both phonetically gradual and lexically gradual. However, they may never fall into the category of regular sound change. The reason is that as new words enter the language, or low-frequency words become more frequent, there will always be new schwas developing in the context for this deletion process. Similarly, the vowel reductions discussed by Fidelholz and Van Bergem may never be complete. As new full vowels come into unstressed position there will be new material for the reduction to work on.

3.5. Reasons for word-frequency effects The cases documented so far indicate that high-frequency words tend to change before low-frequency words when the change is the deletion of stops (English t/d-deletion), the deletion of fricatives (Spanish ö-deletion), some vowel shifts (Labov 1994; Moonwomon 1992), the reduction of vowels to schwa (in both Dutch and English), and the deletion of schwa (in American English). One might therefore predict that in general reductive changes tend to occur earlier and to a greater extent in words and phrases of high-frequency. If we take linguistic behavior to be highly practiced neuromotor activity (Anderson 1993; Boyland 1996; Haiman 1994), then we can view reductive sound change as the result of the automation of linguistic production. It is well-known that repeated neuromotor patterns become more efficient as they are practiced - transitions are smoothed by the anticipatory overlap of gestures and unnecessary or extreme gestures decrease in magnitude or are omitted. Recent theories of articulatory change point to precisely these two types of changes - increase in overlap of gestures and decrease in magnitude - as describing all changes that occur in casual speech (Browman and Goldstein 1992) or in sound change (Mowrey and Pagliuca 1995). If casual speech processes and reductive sound change are the natural result of the automation of linguistic productions, then it follows that such change will be more advanced in productions that are more highly practiced, i.e. high-frequency words and phrases. In fact, reductive sound change may be just the more salient aspect of an overall reduction in high-frequency words. Jurafsky et al. (2001) have shown, using the Switchboard corpus, that for 1412 tokens of monosyllabic content

68

Joan Bybee

words ending in /t/ or /d/, high-frequency words (at the 95th percentile) were 18% shorter than low-frequency words (at the 5th percentile). That is, the entire articulatory span of high-frequency words may be reduced compared to low-frequency words, a phenomenon that could give rise to some noticeable and not-so-noticeable articulatory changes. Other findings by Jurafsky and colleagues indicate that articulatory reduction cannot run rampant. Using the same set of tokens discussed above, Jurafsky et al. (2001) find that the predictability of the word given the following word affects duration. That is, the first words in frequently occurring word pairs, such as Grand Canyon, grand piano, or Burt Reynolds were shorter than words used in less predictable contexts. Gregory et al. (1999) found that among 4695 monosyllables ending in Ν or /d/, again from the Switchboard corpus, semantic relatedness to the discourse topic affected word duration: words related to the discourse topic were shorter than words that were not. In addition, in Gregory et al.'s study, words tended to be shorter if they were repeated in the same discourse (see also Fowler and Housum 1987). My interpretation of these results is that the tendency for articulatory reduction due to increased automation is always present and shows itself more prominently in highly practiced, frequent words. Reduction can be inhibited by the speaker's sensitivity to the predictability of words in the context. If the speaker knows that the word will be easily accessed in the context, because it or related words have already been activated, the reductive automating processes will be allowed to advance. If the word is less predictable in discourse, the speaker is likely to suppress the reductive processes and give the word a more explicit articulation.

4. Modeling phonetic and lexical gradualness Both Wang's and Labov's views of lexical diffusion assume that a change that diffuses gradually through the lexicon must be phonetically abrupt. This is a necessary assumption if one accepts a synchronic phonological theory that has phonemic underlying representations. Words can change one by one only if the change is a substitution of phonemes in such a theory. The discovery that sound change can be both phonetically gradual and lexically gradual forces a different view of the mental representation of the phonology of words. If subphonemic detail or ran-

Lexical diffusion in regular sound change

69

ranges of variation can be associated with particular words, an accurate model of phonological representation must allow phonetic detail in the cognitive representation of words. Thus a proposal by Theo Vennemann comes to mind. In the late 1960's and 1970's there was much discussion of how abstract underlying phonological forms should be. Very few researchers were willing to take a definite stand on this admittedly complex issue, and the notion of underlying form careened wildly back and forth between concrete and abstract. In a 1974 Chicago Linguistic Society paper, Vennemann states clearly that the only realistic model and the only one that solves certain problems of interdependencies between patterns (or rules) would be one in which all words are listed in their systematic phonetic form. This proposal is a start towards a solution to the lexical-diffusion problem. One objection that was brought up at the time concerned the problem of the variant phonetic shapes of different tokens of the same word. Vennemann's solution was to appeal to the notion of "systematic phonetic" representation as defined by Chomsky and Halle (1968). However, this level of representation is also an abstraction. A modification of the 1974 model, which maintains its beneficial features, can be made now given advances in the study of categorization by psychologists. A recent proposal is that the cognitive representation of a word can be made up of the set of exemplars of that word that has been experienced by the speaker/hearer. Thus all phonetic variants of a word are stored in memory and organized into a cluster in which exemplars that are more similar are closer to one another than ones that are dissimilar, and moreover, exemplars that are frequently-occurring are stronger than less frequent ones (Johnson 1997; Bybee 2000, 2001; Pierrehumbert 2001). In this model, the exemplar "cloud" or cluster continues to change as language is used and new tokens of words are experienced. Thus the range of phonetic variation of a word can gradually change over time, allowing a phonetically gradual sound change to affect different words at different rates. Given a tendency for online reduction, the phonetic representation of a word will gradually accrue more exemplars that are reduced, and these exemplars will become more likely to be chosen for production where they may undergo further reduction, gradually moving the words of the language in a consistent direction. The more frequent words will have more chances to undergo online reduction and thus will change more rapidly. The more predictable words (which are usually

70

Joan Bybee

also the more frequent ones) will have a greater chance of having their reduced version chosen, given the context, and thus will also advance the reductive change more rapidly. The exemplar clusters are embedded in a network of associations among words that map relations of similarity at all levels. Distinct words with similar phonetic properties are associated, as are words with shared semantic features. In Bybee (1985, 1988) I have shown that in such a network morphemes and morphological relations are emergent from parallel phonetic and semantic associations, and that schemas or abstractions over relations of similarity can be formulated that account for the regularities and patterns evident in language use. Other questions about this model with respect to phonology are answered in Bybee (2001). An important property that the exemplar model shares with the 1974 model of Vennemann is the emphasis on words as storage units. Vennemann (1974) argued that even multimorphemic words have lexical listing, citing the fact that appropriate constraints on syllable structure can only be applied to whole words, not to morphemes. The common objection to this proposal made in the 1970's was that the human brain does not have sufficient storage capacity for all the words of a language, especially a language with large morphological paradigms. This argument has now been finessed with the discovery of the huge amount of detail the brain is capable of recording. Moreover, newer conceptions of the lexicon, not as a list, but as a network with tight interconnections provides the insight that listing two related words such as Bund, Bundes, does not take up as much cognitive space as listing two unrelated words, such as Bund, Auto. In addition, the lexical diffusion data provides empirical evidence that multimorphemic words can have lexical storage: as we saw in Table 3 above, high-frequency regular past tense English verbs are more likely to have their final /t/ or Idl deleted than lowfrequency verbs. In order for a frequency effect to accrue to a word that word must exist in memory storage; if multimorphemic words evince frequency effects, they must be stored in the lexicon.

Lexical diffusion in regular sound change

71

5. Conclusions A careful analysis of phonetically gradual reductive changes reveals lexical diffusion at a level not encompassed by the theories of Wang and Labov. It supports the observation made by Schuchardt that "Rarely used words lag behind; very frequently used ones hurry ahead" (Schuchardt [1885] 1972: 58). Such patterns of diffusion have now been documented for changes involving consonant reduction, vowel reduction and deletion and, to a lesser extent, vowel shifts. Changes that are gradual phonetically and lexically present a challenge to the notion that words are represented phonemically in cognitive storage and suggest rather that a language user's memory representation of words is heavily based on his or her experience with the phonetic tokens of the language input. In this model, as in Vennemann's 1974 model, memory representations always consist of complete syllables and words, and phonological generalizations describe only patterns that are present and accessible in the phonetic form of words.

References Anderson, John R. 1993 Rules of the mind. Hillsdale, NJ: Erlbaum. Boyland, Joyce Tang 1996 Morphosyntactic change in progress: A psycholinguistic approach. Ph.D. dissertation, Department of Psychology, University of California, Berkeley. Browman, Catherine P. and Louis M. Goldstein 1992 Articulatory phonology: An overview. Phonetica 49: 155-180. Bybee, Joan L. 1985 Morphology: A study of the relation between meaning and form. Philadelphia: John Benjamins. 1988 Morphology as lexical organization. In: Michael Hammond and Michael Noonan (eds.), Theoretical morphology, 119-141. San Diego: Academic Press. 2000 The phonology of the lexicon: Evidence from lexical diffusion. To appear in: Michael Barlow and Suzanne Kemmer (eds.), Usagebased models of language, 65-85. Stanford: CSLI. 2001 Phonology and language Use. Cambridge: Cambridge University Press.

72

Joan Bybee

Chomsky, Noam and Morris Halle 1968 The sound pattern of English. New York: Harper and Row. D'Introno, Francisco and Juan Manuel Sosa 1986 Elision de la /d/ en el espanol de Caracas: Aspectos sociolingüisticos e implicaciones teoricas. In: Rafael A. Nünez Cedeno, Iraset Päez Urdaneta, and Jorge Guitart (eds.), Estudios sobre la fonologia del Espanol del Caribe. Caracas: Fundacion Casa de Bello. Fidelholz, James 1975 Word-frequency and vowel reduction in English. Chicago Linguistic Society 11: 200-213. Fowler, Carol A. and Jonathan Housum 1987 Talkers' signaling of "new" and "old" words in speech and listeners' perception and use of the distinction. Journal of Memory and Language 26: 489-504. Francis, W. Nelson and Henry Kucera 1982 Frequency analysis of English usage. Boston: Houghton Mifflin. Gregory, Michelle, William D. Raymond, Alan Bell, Eric Fosler-Lussier, and Daniel Jurafsky 1999 The effects of collocational strength and contextual predictability in lexical production, Chicago Linguistic Society 35: 151-166. Haiman, John 1994 Ritualization and the development of language. In: William Pagliuca (ed.), Perspectives on grammaticalization, 3-28. Amsterdam/Philadelphia: John Benjamins. Harris, Martin 1988 French. In: Martin Harris and Nigel Vincent (eds.), The Romance Languages, 209-245. Oxford: Oxford University Press. Hooper, Joan B. 1976 Word-frequency in lexical diffusion and the source of morphophonological change. In: William Christie (ed.), Current progress in historical linguistics, 96-105. Amsterdam: North Holland. Johnson, Keith 1997 Speech perception without speaker normalization. In: Keith Johnson and John W. Mullennix (eds.), Talker variability in speech processing, 145-165. San Diego: Academic Press. Jurafsky, Daniel, Alan Bell, Michelle Gregory, and William D. Raymond 2001 Probabilistic relations between words: Evidence from reduction in lexical production. In: Joan Bybee and Paul Hopper (eds.), Frequency and the emergence of linguistic structure, 229-254. Amsterdam/Philadelphia: John Benjamins. Krishnamurti, Bh. 1998 Regularity of sound change through lexical diffusion: A study of s > h > 0 in Gondi dialects. Language variation and change 10: 193— 220.

Lexical diffusion in regular sound change

73

Labov, William 1972 Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press. 1994 Principles of linguistic change: Internal factors. Oxford: Basil Blackwell. 1981 Resolving the Neogrammarian controversy. Language 57: 267-308. Losiewicz, Beth L. 1992 The effect of frequency on linguistic morphology. Ph.D. dissertation, Department of Linguistics, University of Texas, Austin. Menendez-Pidal, Ramon [1904] 1968Manual de gramätica histörica Espanola. Madrid: Espasa-Calpe. Moonwomon, Birch 1992 The mechanism of lexical diffusion. Paper presented at the Annual Meeting of the Linguistic Society of America, January 1992, Philadelphia. Mowrey, Richard and William Pagliuca 1995 The reductive character of articulatory evolution. Rivista di Linguistically. 37-124. Neu, Helene 1980 Ranking of constraints on /t,d/ deletion in American English: A statistical analysis. In: William Labov (ed.), Locating language in time and space, 37-54. New York: Academic Press. Oliveira, Marco Antonio de 1991 The Neogrammarian controversy revisited. International Journal of the Sociology of Language 89: 93-105. Penny, Ralph 1991 A history of the Spanish language. Cambridge: Cambridge University Press. Pensado, Carmen 1984 Cronologia relativa del Castellano. Salamanca: Ediciones Universidad de Salamanca. Phillips, Betty S. 1984 Word-frequency and the actuation of sound change. Language 60: 320-342. 2001 Lexical diffusion, lexical frequency and lexical analysis. In: Joan Bybee and Paul Hopper (eds.), Frequency and the emergence of linguistic structure, 123-136. Amsterdam/Philadelphia: John Benjamins. Pierrehumbert, Janet 2001 Exemplar dynamics: Word-frequency, lenition and contrast. In: Joan Bybee and Paul Hopper (eds.), Frequency and the Emergence of Linguistic Structure, 137-157. Amsterdam/Philadelphia: John Benjamins.

74

Joan Bybee

Santa Ana, Otto 1991 Phonetic simplification processes in English of the barrio: A crossgenerational sociolinguistic study of the Chicanos of Los Angeles. Ph. D. dissertation, Department of Linguistics, University of Pennsylvania. Schuchardt, Hugo 1885 Über die Lautgesetze: gegen die Junggrammatiker. Oppenheim, Berlin (Reprinted and translated in Theo Vennemann and Terence Wilbur 1972). Van Bergem, Dick 1995 Acoustic and lexical vowel reduction. (Studies in Language and Language Use 16) Amsterdam: IFOTT. Vennemann, Theo 1974 Words and syllables in natural generative phonology. Parasession on Natural Phonology, 346-374. Chicago: Chicago Linguistic Society. Vennemann, Theo and Terence Wilbur 1972 Schuchardt, the Neogrammarians and the transformational theory of phonological change. Frankfurt: Athenäum. Wang, William S.-Y. 1969 Competing changes as a cause of residue. Language 45: 9-25. Wang, William S.-Y. (ed.) 1977 The lexicon in phonological change. The Hague: Mouton. Wang, William S.-Y. and C.-C. Cheng 1977 Implementation of phonological change: the Shaungfeng Chinese case. In: William S.-Y. Wang (ed.), The lexicon in phonological change, 86-100. The Hague: Mouton.

Unveiling a masked change: behind vowel harmony in the dialect of Claro* Michele Loporcaro

1. Introduction A great part of the pleasure of intellectual work resides in discovering things. All things still to be discovered share an elementary property: at first sight, they do not seem to be there at all. In the realm of language change, new objects can be discovered, in a rather trivial sense, whenever new data are described. In this paper, however, I will discuss an entirely different case. The empirical data we shall deal with have been known, in the scientific community of Romance scholars, for more than one century: they concern a vowel harmony process which is at work in the Italo-Romance dialect of Claro, spoken in Canton Tessin (Switzerland). This vowel harmony (henceforth VH) is in itself the product of a remarkable change, which deservedly attracted considerable attention, since it characterizes this dialect with respect to its parent language (viz. Latin) as well as to its cognates (Romance varieties). The aim of this paper is to demonstrate that another phonological change hides behind VH. The organization of the paper is as follows. §2 briefly sketches the working of VH in the dialect of Claro. §3 deals with the interplay of VH and cliticization. In particular, it turns out that VH is blocked in one environment, and that precisely this blocking - which has been previously overlooked in the descriptive literature - is crucial to unveil another phonological change (viz. the fronting of all final low vowels), which also took place in this variety, but has gone unnoticed so far. The identification of this change is an unexpected by-product of the analysis of the interplay of VH and cliticization provided in §3. In fact, as argued in §4, the reason why the change was overlooked is that VH, by far the most distinctive property in the phonology of this dialect, exerts a masking effect, concealing most surface evidence for final -/a/ fronting. Elaborating on this observation, §5 provides an explanation for

76

Michele Loporcaro

the change, which is reduced to the category of 'abductive change' first defined in Andersen (1973). It is argued that precisely the poverty of the surface evidence available to the learner (given the masking effect of VH) was of paramount importance in causing this abduction. Although the paper primarily focuses on (the theory of) phonological change, other topics in theoretical phonology will be touched upon, such as the prosodic representation of clitics, the domain of VH phenomena as well as the structure of the prosodic categories 'foot' (henceforth F) and 'phonological word' (henceforth PW).

2. Vowel Harmony in the dialect of Claro The dialect of Claro - a village with about 2,000 inhabitants, lying 7 km north of Bellinzona - is one of a handful of varieties of the Italianspeaking part of southern Switzerland in which a VH process is observed, first described by Salvioni (1892-94). In Salvioni's paper, as well as in the other dialectological studies which dealt with the phenomenon since then (cf. Sganzini 1924-26: 38^10, 1933: 41 fn. 2; Urech 1946: 10-12; Vicari 1980: 11, 2001: 20), the VH is described as a total assimilation process applying to final /a/, the only Latin (ProtoRomance) vowel which survived final-vowel deletion in the dialects of the area. This vowel assimilates completely to the stressed vowel, as shown in (la). The Standard Italian counterparts are given in (lb) for comparison, as they quite closely correspond to the Proto-Romance inputs, which the assimilation modified:1 Claro 'lumu 'tere 'ίετε 'ra-na 'ro-bo 'go το ko 'zi'tii

b. Standard Italian luna tela terra rana roba gola cucina

gloss 'moon' 'cloth' 'earth' 'frog' 'stuff 'throat' 'kitchen'

As can be seen in (la), the dialect of Claro has a seven-vowel system /i e ε a ο ο u/. VH, however, concerns exclusively final /a/, since all other Latin/Proto-Romance final vowels were deleted: ['ε·ζεη]
/le/

Note that this assumption introduces a syntactically determined allomorphy, since the proclitic form of the same 3FSG DO clitic still sounds [la] (cf. (3)). Clearly, the assumption is costly and has to be convincingly motivated. Actually, I will argue that this specific change concerning this specific clitic is but one particular manifestation of a more general change, which took place in this dialect. The change turned every final /a/ into /e/, as stated in (13):6 (13) /a/>

/e//_]F]pw

This claim has much further-reaching consequences than the statement in (12), which only concerns the enclitic /la/. Moreover, (13) will surely have struck the reader as rather implausible. In this variety, all (PW-)final low vowels invariably surface as a copy of the stressed vowel, owing to the application of VH. More precisely, as already argued in §2, for VH to apply the target vowel must be within the same foot as the trigger vowel, as shown in (14) (where PW-boundaries are redundantly included):

82

Michele

Loporcaro

(14) Vowel Harmony in the dialect of Claro (stage (i)) a. [ . . . [ ' V i ... a . . . ] F ] P W - > b. [... [ 'Vi ... [Place]

[+low]

[Place]

Vi ...] F ] P W [+low]

The formulation in (14) mirrors the description of VH in §2. Since the etymological vowel was /a/ in Latin and Proto-Romance, as shown by the Italian counterparts in (lb), until proof to the contrary, the input to VH can be thought to have remained unchanged over time. That VH still is a synchronically active rule has to be assumed for reasons of descriptive economy. An alternative analysis assuming restructuring of the input would imply a multiplication by seven of the representations of all the morphological categories involved, such as the feminine ending of the 1st nominal class exemplified in (la) (compare final -a in the Italian counterparts in (lb)). 7 One more remark is in order here concerning the input of VH. As already mentioned in §2, VH applies to final /a/ as this is the only Latin (and Proto-Romance) vowel to have survived the regular final-vowel deletion, which is characteristic for most dialects of Northern Italy. Consequently, the feature specification [+low] could be regarded as redundant. Consider first that VH applies to non-final /a/'s as well (cf. (2)): in this context, different vowels also occur. Furthermore, also word-finally two other vowels are nowadays found, as the product of later changes. As shown in (15a), an epithetic vowel -o was added after final clusters of falling consonantal strength, which would have been otherwise bad codas (cf. Vennemann's 1988: 21-27 Coda Law). For the same reason, in -CjVJpw strings (where V[-low]), after final vowel deletion the glide turned into a new nucleus through a cross-linguistically common process which can be compared with Sanskrit samprasärana (γα > i, να > u,ra> r). This prevented the creation of an otherwise disallowed coda Cj]cj with falling consonantal strength:8 (15) a. 'sempro

'always'

b. bo'te'li

'bottles'

'negro

'black.M'

'o-li

'oil'

'altro

'other.Μ'

'to-ni

'Anthony'

After these elucidations, we can conclude that the statement of VH in (14) is fine, as for both the domain specification and for the necessity of including a feature-specification of the target vowel.

Behind vowel harmony in the dialect of Claro

83

We have a major problem, though. Quite obviously, the formulation of VH in (14) is incompatible with the diachronic change we have postulated in (13). If the input to VH is /a/, then it is absurd to postulate a paradigmatic change -/a/ > -Id. In §4,1 will demonstrate that a) this change did take place; b) the fact that the 3FSG DO clitic surfaces as [le] in (11) provides conclusive evidence for this change; and that c) as a consequence of this change, the VH rule, which originally arose under the form given in (14), was modified into (16): (16) Vowel Harmony in the dialect of Claro (stage (ii)) a. [ . . . [ ' V i ]F]PW -H- b. [... [ 'V,

[Place]

-high back

VI

... ]F]PW

-high [Place]

-back

As a preliminary step, it is important to realize that the assumption of the general change -a > -e in (13) is necessarily entailed by the description of the more specific change in (12) concerning the form of the 3FSG DO clitic (a change which is directly observable in the realization of the clitic as [le] in the VH-blocking environment (11)).9 We have seen in (8)-(9) that the (en)clitic /la/ regularly undergoes VH, whenever this is not blocked by another intervening clitic, and we have concluded that application of VH to this clitic is but one specific instance of the general process of VH. Now, if the underlying form of the enclitic 3FSG DO must be /le/, as proved by (11), it follows necessarily that VH has to be stated as in (16) (input /-e/), not as in (14) (input /-a/). It follows in turn that, in spite of all (surface) appearances, the change in (13) did take place, a change whose consequences were almost entirely masked by VH, and that we were able to uncover only thanks to our inspection of the behavior of clitics with respect to VH. Now, the burden of proof lies with us: we have to provide a plausible motivation for this change.

84

Michele Loporcaro

5. Motivating abduction: if quantity prevails over quality We can start from an observation speakers of neighboring dialects usually make when they hear an inhabitant of Claro speaking his/her own dialect: ['e: ma 1 'e 'pjeq da 'e] 'hey, it's full with e's!'. Note that the dialects spoken in the directly surrounding villages lack VH. To the linguist, VH is by far the most striking feature in this dialect with respect to the Romance landscape. This can be easily realized by comparing any utterance in the dialect of Claro with its Standard Italian counterparts: (17) a. dialect of Claro:

'kuntulu'su'twtu

b. Standard Italian: raccontala

tutta

'tell (me) the whole story' The linguist would expect such a striking fact to be reflected in native speakers' stereotypes about their language. Yet, in order to select a shibboleth for the Claro speakers, neighbors do not focus on VH, but rather build upon the observation that the dialect of Claro is 'full with e's'. This is actually true. While the stressed vowel system (cf. §2) still retains the seven distinct Proto-Romance vowel qualities, several phonological changes have turned many occurrences of other vowels into mid front vowels, as exemplified in (18): (18) a. /a/> /ε/ 'kre: 'Claro' go're: 'to fly' 'ε'ζεη

'donkey'

b. h / > /ε/ 'trep 'too (much)' 'ne:f 'nine' 'fkere

'school'

This led to an increase in token frequency for stressed front mid vowels, a fact that can be illustrated by means of a simple text count. Table (19) reports the total number of stressed vowels occurring in the two pages of transcribed speech in Vicari (1980: 54-55):

Behind vowel harmony in the dialect of Claro

a. b. c. d. e.

3

'VCi a(V) 'VCi 3 o 'V(V)C(C) 'V(V) 'VCi 3 VC total:

/

e

4

32 1 36 15

15 10 1 30

ε

a

0

1

16

7 23

16 2 24 7

—

—

84

37

—

ο

u

9

5

—

—

4 10

9 8

—

—

—

49

30

26

3 7 1 16

85

total: 89 3 98 80 2 272

Out of 272 stressed vowels, 121 (= 44.5%) are front mid vowels. Thus, the type-ratio 2:5 corresponds to a token-ratio of nearly 1:1. This quantitative relation is crucial for our issue, and provides the explanation for the change -a > -e ((13)) we were looking for. This explanation is schematically represented in (20): (20) STAGE

(i)

input phonetic realization frequency a. /'skera/ —> ['Jke-re] 'school' 45% b. /fi'nis # om # la/ —> [fi'nisomla] close to 0% 'finish-it.FSG for me'

REANALYSIS: STAGE

(ii)

a. /'skere/ b. /fi'nis # om # le/

—> ['Jke-re] 'school' —> [fi'nisomle]

45% close to 0%

'finish-it.FSG for me'

Before the change, at stage (i), VH has to be represented as in (14) (input /a/). However, at this stage, the learner was confronted with surface data in which stressed front mid vowels were almost one half of the total. This implied that, due to VH, also in final position front mid vowels were dramatically more frequent than any other vowel. The only surface cue allowing the learner to infer the original /a/-quality of the input final vowel - in the /la/ (en)clitic just like elsewhere in the lexicon - was provided by the blocking context (20i-b), where the clitic was not harmonized. But this piece of information was statistically almost insignificant. Among the 272 examples on which table (19) is based, there is not even one single occurrence of clusters of host + double enclitic. To find some, the textual corpus should be dramatically enlarged: their token-frequency would not depart dramatically from 0%. Consequently, the only positive cue allowing the learner to infer underlying final /a/ was anything but robust.

86

Michele Loporcaro

Suppose now the learner is looking for more reliable cues in order to reconstruct the quality of the input final vowels. Considering their surface realization, the learner receives a contradictory indication. To see why, let us limit the count to (19a) and (19e); these correspond to the CVCV(C) environment, in which the stressed vowel is followed by another vowel and VH applies. Here, the total amount of stressed (and hence, by VH, also final) low vowels was 17.5% (16/91), while the total amount of stressed (and hence, by VH, also final) front mid vowels was 45% (41/91). Given these input data, the learner was statistically justified in guessing that the vowel which served as input for VH should actually be represented as a front mid vowel, rather than /a/. Thus, s/he reanalyzed previous */'skera/ 'school' as /'skere/, as shown in (20), stage (ii); the input modification was reflected in the description of VH, which took at this point the form in (16). The new underlying form of the input was generalized throughout the lexicon, in spite of the variety of different surface vowels. This reanalysis, consisting of an inference from the data the learner was faced with, had no consequence whatsoever for the surface realization of final vowels, with one single exception: in fact, the only environment in which a non-harmonized final vowel occurs at the surface is the one in (11), in which VH is blocked by the intervening clitic. Only in this environment, a new surface [e] was deduced from the new input /e/. In the final analysis, the change /a/ > /e/ in the dialect of Claro and, more specifically, its overt manifestation in (11) (= (20ii-b)) turns out to be a typical instance of what has been termed abductive change by Andersen (1973, 1989).

Notes * I am glad to have been given the opportunity to offer this paper to Theo V^nnemann, whose work in phonology and in the theory of language change iS a landmark in the development of our discipline. The data presented and analyzed in the paper stem from my own field notes (June/September 2001). The fieldwork was supported by a grant of the Philosophische Fakultät of the University of Zürich, which I gratefully acknowledge. I also thank most warmly George Dunkel for his friendly advice as well as my informant, Mr. Giancarlo Bullo, for his invaluable help.

Behind vowel harmony in the dialect of Claro

87

1. Vowel quantity shows a complex distribution: stressed final CV(:)C syllables display contrastive length, while in word-internal open syllables (CV'CV) an intermediate length is found (cf. Hajek 2000 for a recent discussion of vowel length in Northern Italian dialects as well as for further references). 2. This is in line with proposals by Dressier (1985: 34) and Booij (1996). Cf. also Zee (1993: 367), who replaces the notion 'clitic group' (as used in Nespor and Vogel 1986) by 'postlexical phonological word'. A scrutiny of the empirical evidence (e.g. stress stability under cliticization, intervocalic /s/ voicing, etc.) adduced in support of all prosodic analyses of Romance clitics departing from (5) is provided in Loporcaro (1999, 2000) and cannot be repeated here. 3. Just like in Standard Italian, clitics occur in enclitic position if their host is a verb in the imperative ((8)) or in the infinitive ((9)). Gerunds, which also host enclitics in the standard language, are very marginal in this dialect. 4. There is a huge literature on these topics: cf. e.g. Bafile (1994, 1996), who provides arguments in favor of ternary feet in Romance. 5. Note that this blocking effect cannot be simply ascribed to phonological factors such as the intervening vowel or the prosodic shape of the words involved. In fact, the same vowels do not block VH in (10), where the same stress pattern is also observed. The exact definition of the blocking will not be dealt with here, owing to limitations of space. 6. In what follows, I'll refer to (domain-)final position without distinguishing between foot-final and PW-final position, since the two coincide in the cases which are relevant for our present argument. 7. The same would apply to 1st class feminine adjectives, 1st conjugation imperatives and all other morphological classes originally ending in -a. 8. Clusters of decreasing consonantal strength do not get epithetic -o, as they score better with respect to the Coda Law: e.g. ['kern] 'meat', ['pert] 'loses'. The distribution of epithesis in this dialect corresponds to a pattern which is widespread, both in Northern Italian dialects and cross-linguistically. Among the neighboring dialects of Italian speaking Switzerland, that of Bedigliora (Malcantone) displays α-epithesis in the same context: e.g. ladra 'thief, quattra 'four' vs. infant 'meanwhile', dent 'within', attravers 'through' (cf. Keller 1939a-b). A similar distribution of the epithetic vowel is observed in the dialect of Travo (province of Piacenza) (cf. Zörner 1989, Repetti 1995) or in Old French, where final [a] occurs in pedre, arbre, perdre, quatre, but not in chien, chat, nor in quand, pert etc. Other Tessinian dialects, such as the one spoken in Cimadera (Valcolla; cf. Keller 1934: 5-7), are less liberal in that they epithesize all final consonant clusters, of both decreasing (e.g. sempre 'always', nuältre 'we') and increasing (e.g. intorne 'around', tempe 'time', böske 'wood') consonantal strength. The implicational distribution of epithesis in these dialects (a perfect illustration of Vennemann's 1988: 2-3 synchronic and diachronic maxims) is schematically represented in table (i) (where r stands for

88

Michele Loporcaro

any consonant which is outranked in consonantal strength by the other consonant in the cluster): (i) Claro/Bedigliora Cimadera a. -Cr_ -ol-a -e b. -rC 0 -e 9. There does not seem to be any viable alternative to this analysis. Assuming that the 3FSG DO enclitic still has to be represented as underlying /la/ would force us to assume a totally ad hoc rule turning it into [le] in (11).

References Andersen, Henning 1973 Abductive and deductive change. Language 49: 567-593. 1989 Understanding linguistic innovations. In: Leiv Egil Breivik and Ernst Häkon Jahr (eds.), Language change. Contributions to the study of its causes, 5-27. Berlin/New York: Mouton de Gruyter. Bafile, Laura 1993 Fonologia prosodica e teoria metrica: accento, cliticizzazione e innalzamento vocalico nel napoletano, PhD diss., Florence. 1994 La riassegnazione postlessicale dell'accento nel napoletano. Quaderni del dipartimento di Linguistica dell'Universitä di Firenze 5: 1— 23. 1996 Sulla rappresentazione delle strutture metriche ternarie. Quaderni del dipartimento di Linguistica dell'Universitä di Firenze 7: 2-24. Berendsen, Egon 1986 The phonology of cliticization. Dordrecht: Foris. Booij, Geert 1984 Neutral vowels: An autosegmental analysis of vowel harmony. Linguistics 22: 629-641. 1996 Cliticization as prosodic integration: The case of Dutch. The Linguistic Review 13:219-242. Dressier, Wolfgang U. 1985 Morphonology: the dynamics of derivation. Ann Arbor: Karoma. Flechia, Giovanni 1898 Atone finali determinate dalla tonica nel dialetto piveronese. Archivio Glottologico Italiano 14: 111-120. Hajek, John 2000 How many moras? Overlength and maximal moraicity in Italy. In: Repetti (ed.), 111-135. Hargus, Sharon and Ellen M. Kaisse (eds.) 1993 Studies in Lexical Phonology. San Diego: Academic Press.

Behind vowel harmony in the dialect of Claro

89

van der Hulst, Harry and Jeroen van de Weijer 1995 Vowel Harmony. In: John A. Goldsmith (ed.), The Handbook of Phonological Theory, 495-534. Cambridge, Mass./Oxford: Blackwell. Keller, Oscar 1934 Dialetti svizzeri. Registrati in collaborazione con I 'Archivio fonografico dell 'Universitä di Zurigo. Dialetti della Svizzera italiana. Canton Ticino. Cimadera (Val Colla). Leipzig: Otto Harrassowitz. 1939a Dialetti svizzeri. Registrati in collaborazione con I 'Archivio fonografico dell 'Universitä di Zurigo. Dialetti della Svizzera italiana. Canton Ticino. Breno e Bedigliora (Malcantone). Leipzig: Otto Harrassowitz. 1939b Voci della Patria. Dialetti svizzeri italiani pubblicati da O. Keller, ch. 3. In: Stimmen der Heimat. Schweizer Mundarten auf Schallplatten. Zürich: Phonogrammarchiv der Universität Zürich. Loporcaro, Michele 1997 L'importanza del piemontese per gli studi di tipologia linguistica. In: Piemonte: Mille anni di Lingua, di Teatro e di Poesia, Atti del Convegno, Vercelli 11-12 ottobre 1997, 11-29. Vercelli: Vercelliviva. 1999 Teoria fonologica e ricerca empirica sull'italiano ed i suoi dialetti. In: Paola Benincä, Alberto M. Mioni, and Laura Vanelli (eds.), Fonologia e morfologia dell'italiano e dei dialetti d'Italia, Atti del XXXI Congresso della Societä di Linguistica Italiana, Padova, 25 settembre 1997, 117-151. Rome: Bulzoni. 2000 Stress stability under cliticization and the prosodic status of Romance clitics. In: Repetti (ed.), 137-168. Monachesi, Paola 1995 A grammar of Italian clitics. Tilburg: ITK Dissertation Series. Nespor, Marina and Irene Vogel 1986 Prosodic phonology. Dordrecht: Foris. Peperkamp, Sharon 1995 Enclitic stress in Romance. CLS 31.2: The Parasession on Clitics, 234—249. 1997 Prosodic words. The Hague: HIL. Repetti, Lori 1995 Constraints on prosodic structure. A study of the dialect of Coli (PC). Studi italiani di linguistica teorica e applicata 24: 279—288. Repetti, Lori (ed.) 2000 Phonological theory and the dialects of Italy. Amsterdam/Philadelphia: John Benjamins. Salvioni, Carlo 1892-94 L'influenza della tonica nella determinazione dell'atona finale in qualche parlata della valle del Ticino. Archivio Glottologico Italiano 13: 355-360.

90

Michele Loporcaro

Sganzini, Silvio 1924-26 Fonetica dei dialetti della Val Leventina. L 'Italia dialettale 1: 190— 212; 2: 100-155. 1933 Le isole di u da ύ nella Svizzera italiana. L 'Italia dialettale 9: 27-64 [Also in: Sganzini (1993: 19-56)]. 1993 Scritti dialettologici. (Romanica Helvetica 109.) Basel/Tübingen: Francke. Urech, Jakob 1946 Beitrag zur Kenntnis der Mundart der Val Calanca. Biel: Graphische Anstalt Schüler. Vennemann, Theo 1988 Preference laws for syllable structure and the explanation of sound change. Berlin/New York/Amsterdam: Mouton de Gruyter. Vicari, Mario 1980 Dialetti svizzeri. Dischi e testi dialettali editi dall'Archivio fonografico dell'Universitä di Zurigo. III. Dialetti della Svizzera italiana, fasc. 5, con la collaborazione di Sonja Leissing-Giorgetti. Lugano: Mazzucconi. 2001 Nota linguistica. In: Giancarlo Bullo, Na fassini da borbatri, 17-24. Locarno: Dado. Zee, Draga 1993 Rule domains in phonological change. In: Hargus and Kaisse (eds.), 365—405. Zörner, Lotte 1989 Die Dialekte von Travo und Groppallo. Diachrone und synchrone Studien zum Piacentinischen. Wien: Verlag der Österreichischen Akademie der Wissenschaften.

Accents and medieval English phonologists Robert W. Murray

**

1. Introduction In the early English manuscripts the most common accent is the acute (or apex) which was used from the beginning of the eighth century (Hulme 1896: 23), but other diacritics are also found including the curl, breve, circumflex, and macron (Napier 1889, Sisam 1953: 186).1 In some cases the accents have been placed by the original scribe, in others they have been added to the manuscript at a later date. In their potential to reflect the intuitions of the earlier scribes it might be expected that a study of the accents could prove extremely valuable in phonological reconstruction, but this expectation has not been fulfilled. The tendency of placing an acute over a long vowel was recognized early, but it was also apparent that there were many confounding variables. Sweet (1888: §379-80) suggests that in some cases the placement of the acute can only be due to "carelessness" (e.g. 'god', or its fairly frequent occurrence over particular suffixes), and in other cases the accents simply served as 'ornaments, without which the page had a bare look, and were consequently partly written mechanically, partly dashed in almost at random'. According to Sievers (1898: §8), the application of the acute was "sporadisch" and "ohne feste Regel" ('without strict rules'). A similar situation is found in Old High German and Old Saxon studies leading P. Sievers (1909: 121) to reluctantly conclude at the end of his thorough investigation, "dass die fehlerhafte und ornamentale Verwendung des Accents wohl weiter geht, als ich angenommen hatte".3 Consequently, in spite of an optimistic burst of scholarly activity in the late 19th century the accents proved to be a highly problematic and dissatisfying area of study, and subsequent research tended to ignore the topic as reflected in Heald (1965) where a survey of previous research can be presented in less than 25 pages. This decline of interest is obviously

92

Robert W. Murray

correlated with the failure of scholars to determine the systematic underpinnings of the accents. Although there is no doubt that accents were often applied inconsistently and for a variety of purposes, it is also a fact that the earlier studies were predicated on the view that the Old English classical quantity system was maintained, more or less intact, into Middle English. However, this traditional reconstruction is facing a significant challenge emanating from such work as Vennemann (2000), which argues that important sound changes of the period (from Homorganic Cluster Lengthening (HCL) to Open Syllable Lengthening (OSL)) can only be understood as the fallout from a significant prosodic change; namely, a shift from the classical quantity system of Old English to the prosody of syllable cut in Middle English.4 These two prosodic types differ in significant ways (Becker 1998, Vennemann 1995): (la)

Classical quantity i. Vowel length and length of a following consonant are independent. ii. Both long and short vowels occur in open syllable. iii. Vowel and consonant length are independent of stress. iv. The long vowel is the marked member of the length opposition.

(lb)

Syllable cut i. Vowel and consonant length are interdependent. (There are only two types of contrasts per syllable: smooth versus abrupt cut.) ii. Abrupt cut requires a following tautosyllabic consonant. ('Short' vowels do not occur in open syllables, but only in syllables closed by a tautosyllabic consonant or by ambisyllabicity.) iii. Phonetic properties: In smoothly cut syllables, vowels tend to be long/tense/peripheral and consonants short/lenis. In abruptly cut syllables, vowels tend to be short/lax/centralized, and consonants long/fortes. iv. The syllable cut opposition is favored under stress, disfavored in the absence of stress. v. Abrupt cut is the marked member of the cut opposition. ('Short' vowels are marked, 'long' vowels unmarked.)

Accents and medieval English phonologists

93

While there is a general consensus that early Old English approached a prototypical classical quantity language, most studies of historical English phonology have failed to recognize that syllable cut was phonologized by the Middle English period. In some of my recent work, I have attempted to strengthen the syllable cut hypothesis by anchoring the diachronic claims in synchronic reconstructions based on various kinds of external evidence. To date the earliest evidence for the phonologization of syllable cut is found in the orthographic system applied in the Ormulum, a text written in the late 12th century in a Northeast Midlands dialect. The text's author, Orm, consistently writes a double graph as an indicator of abrupt cut; e.g. abruptly cut 'kept', 'desire', 'bed', 'sun' vs. smoothly cut 'child', 'need', 'son' (Murray 2000). However, Orm's dialect is a mixed phonological type constituted of syllable cut and nuclear length, for the metre of the poem indicates that one property of the earlier classical quantity system was still maintained; i.e. nuclear length in open syllable as in 'token' versus 'taken'. 5 Over time, however, even this last vestige of classical quantity was lost and early Modern English can be considered a prototypical syllable cut language, much like Modern German as described for example by Becker (1998) and Vennemann (1991), with all the properties in (lb). In my view, this claim finds direct support in the phonological descriptions of a London dialect provided by John Hart, a 16th century orthoepist (Murray, in press). In the case of the accents, there is no doubt that often we are not dealing with consistent, phonologically-based applications, and even where phonological underpinnings are evident scribes will vary in their training, abilities, goals, and consistency. At the same time, if the earlier studies were based on a misunderstanding of late Old and early Middle English phonology, it would be unreasonable to expect from them a full appreciation of various general properties of the accents or of specific types of accent systems. In the present paper, I explore the possibility that the difficulties confronting earlier studies were exacerbated by their failure to recognize the role of syllable cut. In the following sections, I reassess two earlier studies with an eye to determining whether an interpretation in terms of syllable cut can provide any new insights. The first is a longitudinal study that documents the increasing frequency and the

94

Robert W. Murray

decreasing accuracy in the use of the acute over time, and the second deals with the application of an innovative accent system.

2. More and more Accents Hulme (1896: 20-22) observes: Beginning with the ninth century, accent marks occur with increasing frequency in prose MSS. till about the beginning of the eleventh century. [...] In the later prose MSS. accents continue to occur, - in some like the Blooms MS. in profusion, - but the scribes are no longer so careful to place them over long vowels as they were in the earlier MSS. Short vowels and those of unstressed syllables are frequently accented. In numerous instances the marks even stand over consonants, thus showing general carelessness, haste, or ignorance on the part of the scribes.

Hulme's observation leads us to ponder two puzzles that emerge as we move into the Middle English period: (2)

Hulme's Puzzles (a) why did the use of accents increase, and (b) why did their accuracy of placement decrease?

Although paleographic and other non-linguistic factors cannot be ignored, the question I would like to pursue is whether any phonological factors can be identified that could have contributed to these characteristics of the accents.6 Hulme's puzzles receive explicit confirmation in Heald's (1965) study. Comparing three versions of the St. Martin homily as found in the Vercelli Book, the Bückling Homilies, and Junius 85.86, Heald (1965: 61, 104) finds that there is a remarkable difference in the number of acutes (the only accent used) correlating with the age of the manuscript. While there are only 28 acutes in the oldest manuscript, the middle manuscript shows more than a five-fold increase, and the youngest manuscript an almost thirty-fold increase: (3)

Manuscript number of pages Date number of acutes Vercelli 13 late 10th c. 28 Bückling 18 early l i t h e . 156 Junius 23 mid 11 th c. 743

Accents and medieval English phonologists

95

As Heald observes, the first two manuscripts are the most accurate in reflecting a correlation between an acute and an etymologically long vowel. In the Vercelli all the acutes are placed over etymologically long vowels, most of which are in monosyllables. In the Bückling, 142 acutes are placed over etymologically long vowels, 11 over vowels assumed to have been lengthened through HCL, and three are placed over Latin words and a proper name (Heald 1965: 61). However, in the case of the Junius manuscript we are facing an entirely different situation. Not only are there significantly more acutes, Heald (1965: 62) also observes that almost one third ("some 235") of the total are found over etymologically short vowels (Latin words and proper names excluded). Although the first two manuscripts clearly reflect an accurate application of the acute as an indicator of vowel length, in the case of the Junius manuscript the primary thrust of Heald's (1965: 105) study is to deny the relevance of Hulme's second puzzle while at the same time responding to the first. In other words, she argues that many of the original short vowels had undergone diachronic lengthening processes pre-dating the Junius manuscript, so in her view the vast majority of acutes are indeed correctly placed over long vowels. Her implicit answer to the first puzzle is that the increase in the number of accents is expected simply because there are many more instances of long vowels (due to the lengthening processes) in the Junius manuscript compared to the previous two manuscripts. 7 Heald's interpretation, however, is flawed in two respects. First, an increase in the number of long vowel forms would not necessarily result in an increase in the number of acutes used, since in the English tradition there was no convention that all long vowels must be accented, in contrast for example to Notker's system for Old High German (cf. Hulme 1896: 21, Braune and Eggers 1987: §8, Anm. 8). But even if there were some correlation between the number of occurrences of long vowels and the number of acutes, the relatively modest change in the number of forms containing a long vowel due to such sound changes as HCL could not possibly account for the dramatic increase in the use of the acute in the Junius manuscript, especially given that Closed Syllable Shortening (CSS) actually results in a reduction of the number of forms with long vowels.8 These considerations lead us to the second and most significant weakness of Heald's interpretation; namely, her assumed

96

Robert W. Murray

lengthening processes are based on entirely unacceptable assumptions regarding the environmental conditioning factors. Turning first to monosyllables, in her attempt to explain away individual forms Heald demonstrates a general overreliance on questionable claims made by Sweet (1888; followed by Sievers 1898); e.g. her claim of vowel lengthening in 'but'. Although Sweet (1888: §384) did claim that "there is unmistakable evidence of lengthening before single conss. in subordinate monosyllabic words", the strength of Heald's argument is vitiated by circularity since Sweet's (1888: §379) lengthening claims themselves are based primarily on the occurrence of the acute (occasionally of a double vowel graph), sometimes as found in only a single manuscript (the Lauderdale Orosius) which he believes has "hardly more than two undoubted errors" in this regard. Even a single placement is considered "authoritative". In fact, from a phonological perspective the claim of lengthening in such short, typically unstressed function words is not particularly plausible, nor is it supported by their subsequent history or by Orm's rendering of them: Sweet ac of on set

Orm acc off onn att

'but' 'of 'on' 'at'

Indeed, Keller (1908: 97, 119) had already concluded that Sweet's confidence in the accuracy of the acutes in reflecting vowel length was misplaced, especially in the case of some manuscripts where the acute placed over short monosyllables is an entirely unreliable indicator of vowel length.9 Turning now to forms with an etymologically short vowel in open syllable, Heald (1965: 80) lists 34 forms (50 instances in total) displaying an acute, and concludes: After tracing the subsequent development of the residue words, we find that most of the forms (approximately 24 out of the 34 forms in which the 50 accents occur) were lengthened at some time in the Middle English period. Since all of these residues occur in the latest manuscript (Junius 85.86) in a period of late Old English that might well be considered transitional, the

Accents and medieval English phonologists

97

presence of accents over these forms would seem to indicate that the lengthening process had begun.

In fact, this conclusion is highly suspect since ad-hoc, typically implausible, explanations are provided for almost every case of proposed lengthening. For example, OSL (or at least some incipient version of it) is assumed for a number of forms but this claim implies a chronology that is lacking in all plausibility, especially since forms with high vowels are included. 10 High vowels did not regularly undergo OSL, and certainly not at such an early date. While the traditional dating of OSL to the mid 13th century is debatable, it is certain for example that Orm shows no lengthening more than a century after the Junius manuscript. 11 In this respect, Heald's position is also surprising since the dangers of an uncritical acceptance of the acutes as indicators of vowel length had long been known. For example, Sweet (1888: §379) observed that in the case of at least one manuscript the assumption of a correlation between the acutes and length would imply that the dialect had only long vowels. I note too that long before Heald's study Keller (1908: 119) had been critical of a dissertation in which OSL was dated to the 9th century on the basis of the occurrence of the acutes in a particular manuscript. 12 The remainder of the acutes placed over etymologically short vowels involves closed polysyllables, in which case Heald (1965: 62) identifies 185 instances over vowels preceding "consonant clusters, predominantly of the type liquid or nasal plus consonant". 13 In an apparent reference here to HCL, Heald has chronology on her side since this change is usually assumed for the late Old English/Early Middle English period. However, upon closer scrutiny it turns out that she appears to assume that virtually any cluster, even a geminate, can cause lengthening of a preceding vowel. This generous interpretation of lengthening is once again based on a section from Sweet (1888: §395) which Heald (1965: 3) summarizes as follows: Sweet lists examples and states the conditions of lengthening, e.g., analogy; lengthening before vowellikes + cons, ("group-lengthenings"): "lift), rn, rt, rd, 1:11, I f , 1m, It, Id, nmn, nc, nd, ng, mp, mb; long vowels [are kept] before sc, st, isolated lengthening before sn, fn, ft, fd, λ-".

98

Robert W. Murray

It is a fact that the conditions of HCL are difficult to determine, given the dearth of direct evidence and the possibility that forms could have been lengthened and then subsequently shortened in the later Middle English or early Modern English period. Indeed, given the many exceptions to HCL Minkova and Stockwell (1992) assume a position dramatically opposed to Heald's (and Sweet's); namely, that lengthening occurred in only a single environment (preceding -Id) and consequently they argue that HCL should not even be recognized as a general sound change at all. While their position might be considered extreme, Heald's particularly unconstrained version that would assume lengthening not only before HCL clusters such as nd and mb, but also before nt, nc, nf, nw, and even nn cannot be accepted. Even a generous traditional interpretation (e.g. Ritt 1994: 81) would usually only include the conditioning environments in (5): (5)

-Id, -rd, -mb, -nd, -ng, -rl, -rn, -rd, -rz

With the above discussion in mind, let us attempt to get a clearer perspective by taking a more detailed view of a portion of Heald's data. I simply select as representative the words found on the first page of Heald's (1965: 97) listing of forms involving an acute over an etymologically short vowel (a total of 39 instances).14 First of all, there is only one monosyllable, 'man', which Heald reasonably treats as a mistake (perhaps based on confusion with man 'evil deed'). Second, in the case of short vowels in open syllables there are 12 instances of the acute: (6)

gecoren 'chosen', hwaeöere (2x) 'however', feeder 'father', geogoö(hade) 'youth(hood)', lufode 'loved, 3rd sg.', wyrolde 'world, acc.', Mian 'love, acc.', godes 'god's', dsege 'day, dat.', burhgaete 'citygate, dat.', gyrelan 'mantle, acc.'

As discussed above, OSL cannot be posited for this stage, so unless special factors are introduced it is entirely reasonable to reconstruct short vowels.15 Third, an additional 12 instances involve acutes over vowels occurring before clusters not traditionally considered HCL triggers (cf. Orm's , , etc.): (7)

secgan 'say', wintre (3x) 'winters, gen.', haefde (3x) 'had', earmre '(the) poor, dat.', öaencean 'think', meahte 'was able to', öearfan 'poor man, acc.', naefde 'lacked'

Accents and medieval English phonologists

99

Thus far, then, we have 25 instances of straightforward exceptions to the claim that the acute is used to indicate a long vowel. The final 14 remaining instances can be divided into three groups as follows: (8a) ealdorman 'alderman', yldran 'parents', wyrdedan 'honored, 3rd pi.', hingrendum '(the) hungry, dat.' (8b) sceolde{2x)'should', wolde 'would' (8c) wordum 'words, dat.', gebyrda (2x) 'birth, dat.', öingum 'things, dat.', gemunde 'remembered', gesealde{2x) 'gave up' Although each of the forms in (8) contains an HCL cluster (at least from the traditional perspective summarized in (5) above), lengthening in trisyllabic forms such as those in (8a) is not expected (cf. Orm's ). Indeed, trisyllabicity is generally considered a shortening environment (cf. the later shortening in open syllable of the original long vowel in holiday versus holy). Further, as mentioned above, in forms such as and HCL is traditionally assumed not to have occurred before clusters of three consonants (cf. Orm's 'children' versus 'child'). Lengthening in the (8b) forms is also suspect, since forms typically unstressed tended not to undergo HCL (cf. Orm's 'should'; Wright and Wright 1928: §68). Accordingly, even under a generous interpretation of HCL only the instances in (8c) would be subsumed. In sum, in stark contrast to Heald's view that the acute is correlated with vowel length we can conservatively conclude that the acute was used incorrectly over a short vowel in 32 of these 39 instances (about 82%). Consequently, it is clear we are dealing with a manuscript that not only contains significantly more acutes than previous versions, but also one containing a significant number of errors. But is the Junius manuscript simply more fodder for Hulme's puzzles? In fact, I would say that this manuscript allows us to somewhat refine our understanding Hulme's second puzzle, for if we look at the scribe's practice within a general context we find that it combines accuracy with inaccuracy. The first important observation is that the acute is applied with great accuracy in the case of monosyllables. In our above discussion of the 39 instances of the acute over an etymologically short vowel, there is only one monosyllable (i.e. 'man'). This situation is reflective of the entire manuscript. Based on Heald's (1965: 81-103) complete listing, I count 218 instances of monosyllables with acutes, and according to

100

Robert W. Murray

Heald 198 of these instances involve forms with an etymologically long vowel, 20 with an etymologically short vowel. Of these 20, a lengthened vowel can be assumed on the traditional version of HCL in 12 instances, and a long vowel can be assumed for 'do' (which Heald considers etymologically short).16 Accordingly, there are only seven errors in the treatment of the monosyllables, yielding an accuracy rate of about 97% for the entire manuscript. Similarly, the acute was applied quite accurately over open polysyllables. As we indicated above, there are 50 instances of acutes being applied over etymologically short vowels in open syllable. However, even assuming all these applications are "mistakes", we are still only dealing with less than 7% of the total number of acutes. More importantly, if we consider only the open-syllable environment the accuracy of application still remains very high. Based on Heald's (1965: 97-99) list of instances occurring in the first 10 pages of the manuscript, I count 63 correct applications of the acute in open syllable, and 16 incorrect ones. In other words, it was applied with about 80% accuracy.17 Accordingly, the vast majority of errors is confined to a particular syllable type; namely, closed polysyllables. Based on Heald's (1965: 81^4·; 97-9) listing of polysyllables found in the first 10 pages of manuscript, I count a total of 56 instances of acutes over such syllables, and of these 16 instances involve the HCL environment.18 Accordingly, unless special factors are brought in to support the claim of a long vowel the other 40 can be considered incorrect, so the accuracy of application only approaches 29%. In other words, in the case of this syllable type there are significantly more incorrect applications of the acute than correct ones. The results can be summarized as follows: (9)

Accuracy rates per syllable type Instances19 Correct Incorrect Accuracy Monosyllables 218 211 7 97% Open Polysyllables 79 63 16 80% Closed Polysyllables 56 16 40 29%

Consequently, we are led to the conclusion that even though Heald's particular interpretation does not stand, her denial of the relevance of Hulme's second puzzle for the Junius manuscript is still partially correct. That is, we are dealing with a kind of bipartite system where from

Accents and medieval English phonologists

101

the perspective of the traditional reconstruction of Middle English phonology the acute accurately reflects vowel length in the case of monosyllables and open polysyllables, but intervening factors clearly play a role in the case of the closed polysyllables. Under such conditions we cannot simply dismiss the Junius scribe's practice as careless or incompetent a la Hulme. Nevertheless, from the perspective of a traditional reconstruction it is seemingly impossible to provide a phonological basis for the genesis of a system in which the acute is applied correctly over monosyllables and open polysyllables, but incorrectly over closed polysyllables. For example, assuming the primacy of the acute as an indicator of a long vowel, it seems unreasonable to claim that the failure to extend the convention accurately to closed polysyllables was due to scribal incompetence. This claim would carry the implausible implication that the scribe could distinguish nuclear length in monosyllables and open polysyllables, but not in closed polysyllables.20 However, perhaps at this point we should backtrack somewhat and reconsider the bipartite system itself, for it is quite obvious that the scribe's system is only incorrect from a particular perspective; namely, that the only correct use of the acute is as an indicator of a long vowel. Of course, as the earlier studies recognized many accent "errors" can be interpreted as reflections of various new functions that had been assigned to the acute (e.g. its application over short monosyllables and particular suffixes, its use for distinguishing [u] and [w], etc.).21 In fact, once the possibility of different functions is recognized, it is a straightforward matter to put a positive spin on the Junius scribe's practice: (10) Application of the acute (optional): In monosyllables, apply an acute over a long vowel. In polysyllables, apply an acute over any heavy (stressed) syllable (i.e. a syllable that contains a long vowel and/or is closed). From this perspective, the scribe's application is almost flawless with only 50 instances over etymologically short vowels in open syllable (assuming all these are incorrect), and seven over monosyllables (less than eight per cent of the total). However, the bipartite system itself still remains a puzzle: how would such a system arise? Although paleographically-based explanations can likely be developed, it seems clear that from the perspective of the traditional reconstruction of early Mid-

102

Robert W. Murray

die English it will not be possible to identify phonological factors that would lead to, or even favor, the development of such a system. By contrast, once the role of syllable cut is recognized, phonological factors find a natural role in accounting for the placement of the acutes. Let us first assume the reconstruction of the Junius scribe's phonological system as a complex type along the lines of Orm's; i.e., one that combines the two prosodies of syllable cut and nuclear length, although the latter is restricted to open syllables. We can safely assume the Junius scribe would inherit a tradition where the primary location of the acute would be over an etymologically long vowel either in closed monosyllables or in open polysyllables. For the Junius scribe then, the acute could seamlessly take over the function of indicating smooth cut in the case of monosyllables. In fact, the scribe's application of the acute in this case can be considered a virtually flawless reflection of the bifurcation of monosyllables into smoothly versus abruptly cut: (11)

Monosyllables smooth (optional acute) ^öi/'good'

abrupt (no acute) god'god'

At the same time, in the case of the open polysyllables where the phonological contrast is based on nuclear length, the acute could take on (or continue) the function of indicating a long vowel. In other words, we have a bipartite system where the acute reflects smooth cut in monosyllables, but nuclear length in open polysyllables. Now this situation could lead to a kind of scribal "confusion" in the case of the closed polysyllables if we assume that the scribe's primary association in the case of polysyllables involved a linking of the acute with nuclear length. This association would work fine for open syllables but, according to my reconstruction, closed syllables were contrasted only in terms of syllable cut, not nuclear length, so the scribe faced two possibilities in this case: a) never use the acute since that would be an incorrect application in terms of length, or b) use the acute according to other criteria. The Junius scribe clearly settled for the second approach, and apparently an acute could be optionally placed over any heavy (stressed) syllable of a polysyllable.23 In sum, it is clear that the second part of Hulme's puzzle relating to the increased inaccuracy of the accents must be dealt with carefully, and each manuscript considered individually. In general terms, however, it

Accents and medieval English phonologists

103

is quite natural to expect that a changing (changed) prosodic system would lead to "errors" in the application of an accent system that was originally based on an earlier phonological type. The Junius scribe's practice could represent a case in point. If we assume that his phonological system was the same complex type as reconstructed for Orm, it is a straightforward task to speculate on the breakdown on the original system accent system, and the genesis of this new system. Since the prosody of his language had changed the acute could no longer be used entirely in the old way, and a new tradition evolved. Of course, this proposed interpretation is highly speculative, and suffers from the additional weakness that there is a dearth of knowledge regarding the intuitions of speakers of mixed length/syllable cut languages. It is also the case that this type of evidence in support of the reconstruction of syllable cut is very weak in the sense that the acutes in the Junius manuscript do not represent a direct mapping of syllable cut properties. The observed phenomena are simply compatible with the assumption of syllable cut. However, this situation should not be surprising. We cannot expect all scribes to be budding phonologists involved in successful attempts to diacritically mark the new phonology. At the same time, if my assumption of this early date for the phonologization of syllable cut is justified, it would not be surprising to find new accent systems being devised to deal more directly with the new phonology, just as Orm later adapted his orthography. In the next section, I outline precisely such a system.

3. Accents and a budding phonologist While we have seen in the previous section how a changing prosodic system could contribute to errors in the application of the acute, it is also natural to expect the development of new systems that could contribute to the "profusion" of accents (i.e. Hulme's first puzzle) in the sense that different types of diacritics could be used in an attempt to reflect the new phonological contrasts. In fact, an innovative accent system is applied to various homilies found in Bodley manuscripts 340 and 342. Sisam (1953: 186) suggests that the accents were added to the manuscript "presumably not earlier than 1050", and probably "round about the end of the eleventh century". In other words, we are dealing

104

Robert W. Murray

with approximately the same time frame as the Junius manuscript discussed above. As is commonly the case, the acute also appears in the two Bodley manuscripts in a variety of environments. 24 However, this individual (let's call him B342) was not content with the acute only, but also used three additional accents; the curl, the circumflex, and a 'very long macron which is usually extended over the following consonant and a vowel' (Sisam 1953: 188). Ignoring the circumflex, which appears to have been used for the same purpose as the curl, the nascent system can be summarized as follows: (12) Accent Acute Curl

Function Long Vowel

Short Vowel in Closed Syllable Long macron Short Vowel in Open Syllable

Example



'feet, dat.' 'word' 'man' 'hills, dat.' 'god's'

The use of the curl to indicate a short vowel is not unique. For example, Napier (1889: 221) notes the curl in the Cotton MS. Cleopatra B.13, which "is like a small c written over the vowel, or like the " used in Latin MSS. ... turned over on its side".25 However, B342 seems to have been innovative in introducing the long macron. In attempting to address its purpose, Sisam (1953: 188) discusses two possible factors, the first of which is purely speculative: It is at least curious to find that, quite apart from their importance in metre, the Latin short vowels in open syllables had been recently the subject of controversy, ^lfric in the Preface to his Grammar speaks of the many who pronounce pater, malus, etc., in prose after what he calls the "Welsh manner" (Brittonice), i.e. with a short vowel, and defends the lengthened pronunciation, which had become established in late Vulgar Latin, by arguing that to address the Heavenly Father with a short vowel in Pater is to subject God to the rules of grammar [...]. Such a dispute might lead to the practice of distinguishing the short vowels of open syllables in Latin, whence it might be transferred to English.

Sisam provides no evidence of the practice being applied to Latin and, perhaps more importantly, it is worth noting that the acute and curl alone would be adequate for the task of distinguishing long and short vowels in open syllable. The macron is superfluous in this regard: 26

Accents and medieval English phonologists

(13a) Possible:

curl

macron < godes >

105

acute vs. vs. vs. acute vs.

At the same time, of course, there is certainly no law enforcing logic and simplicity in the case of orthographic systems. Sisam's (1953: 186) second proposed motivation for the accent system lies in his entirely reasonable contention that B342's marks were intended as a 'guide ... in reading aloud'. 27 However, even this sensible claim is difficult to sustain from the perspective of the traditional reconstruction, since it is not at all apparent how the three-way contrast conveyed by the accents in (12) would serve to facilitate the reading of a text. First, as already indicated in (13), the long macron is superfluous. Second, the use of the curl before geminate graphs (e.g. 'men, dat.') and in forms such as 'hills, dat.' is completely redundant, since there is no chance of ambiguity in these environments where long vowels do not occur. Third, there is no apparent orthographically-based rationale for distinguishing a short vowel in closed syllable (the curl) versus a short vowel in open syllable (the macron). In fact, this practice seems to work against Sisam's proposal, since different diacritics are used for monosyllables and their inflected counterparts: (14) Short vowel in closed syllable: Short vowel in open syllable:



[god] [god-]

'god' 'god's'

Such a system does not appear to provide any apparent advantage to the reader, so the introduction of the macron is at best unnecessary and at worst serves to obscure morphologically related forms. The system is even more mysterious given that we are dealing with prose, so we cannot assume the imposition of some artificial metrical template to assist in the reading or intoning of a poem. 28 At the same time, it is reasonable to conclude that we are not dealing with an ad-hoc or decorative application of accents, but rather with a nascent system built on phonological underpinnings. Once again, I suggest that the inadequacy of previous approaches is directly traceable to the traditional reconstruction of Middle English phonology. Once the role of syllable cut is determined, the interpretation of the accent system is straightforward. In closed syllables the

106

Robert W. Murray

acute and the curl reflect smoothly versus abruptly cut syllables, and in open syllables the acute and the macron serve to distinguish long and short nuclei. In other words, B342 is making precisely the same phonological distinctions Orm makes about a century later (cf. Sisam 1953: 188-195). In the closed syllable environment, B342's use of the acute versus the curl parallels Orm's simple versus double graphs: (15) Smooth cut: Abrupt cut:

B342





Orm





'good' 'word' 'god' 'man' 'men, dat.'/'men's' 'thrown, scattered' 'hills, dat.'/'hills'

Similarly, in (smoothly cut) open syllable both systems distinguish long and short vowels, B342 with the acute and macron, Orm with the acute and breve: (16)

B342 Smooth cut Long nucleus Short nucleus < godes >

Orm

'feet, dat.' 'god's'

The relevant contrasts are portrayed in (17):29 (17a) Closed syllable Smooth ('good') B342

Orm

Abrupt ('god') B342 Orm

Accents and medieval English phonologists

(17b) Open syllable Long nucleus ('good') B342

Orm





*eompse (fem. *eampse) as a means of motivating the introduction of the ρ into the paradigm, which would then have been analogized throughout from the accusative case. While the phonology of this proposal is quite ordinary, a serious problem arises with motivating a reflexive form behind the -se. Despite the fact that ipse is often found with the true reflexive pronoun se (cf. Cie., Rose., 142, 9: se ipsum probe novit), and the fact that ipse often functions with reflexive force (cf. Cie., Fin., 5, 9, 24: nec quid ipsius natura sit intellegit), there is in fact no morphological or syntactic evidence for the juxtaposition of oblique forms of is, ea, id with reflexive se, especially in a relationship strong enough to promote a fusion and complete morphosyntactic reanalysis of the form. In other words, there is no construction such as *vidit eum se anywhere in Latin, thus depriving this proposal of a syntactic source, and leaving it without merit. An alternative version of this theory suggests the possibility that the se of ipse continues some particle or particle-like element, with possible origins somewhere in the pronominal system. The strongest case to be made (mentioned cautiously but favorably by Ernout-Meillet s.v.) is for a connection with the emphasizing particle {nota augens) so (sa, se) found in Old Irish (cf. Lewis and Pedersen 1961 [1937]: 221), which probably has its origins in the demonstratives. This version is compati-

Latin ipse, Continental Celtic -xsi: a tentative proposal

143

ble with what Leumann seems to have in mind in deriving ipse from */p(e)-so, with *so- comparable to Gk. ό (1977: 471). While compositionally justifiable and surely a more likely source than reflexive se, this form of the proposal depends again on the epenthesis of ρ (easily enough justified), and then an analogical spread of the accusative form with the newly installed -p- throughout the paradigm. Through this process the ρ is present in all cases, genders and numbers, a development which in turn is followed by the complete remaking of the paradigm on the basis of the nominative ipse. In other words, in this scenario the accusative first is established as the founding form, then, after the paradigm has been reshaped with the originally epenthetic p, the nominative takes over, serving as the head of the paradigm ipse, ipsa, ipsum. The phonological approach to the -p- is all quite ad hoc, regardless of the source of the -se element, depending as it does on the strength of the accusative to serve as a founding form for a complete morphological restructuring. In a non-scientific statistical exercise, I checked the number of accusative and nominative forms of ipse in all the plays of Plautus. The results are not surprising: there are thirty-two instances of accusative singular forms of ipse (ipsum, ipsam) in the Plautine corpus, while there are one hundred twenty-nine instances of the nominative singular1 forms ipse, ipsa, ipsum. These frequencies suggest that the nominative was the more salient form of this pronoun, thus weakening any claim that the accusative guided a complete analogical reshaping of the form with a phonologically-induced p. These objections apply to the solution utilizing reflexive se as well as one using particle -se. This brings us now to the second major alternative, a morphological one, which combines some element se with some further particle element -ρ-, resulting in a unified form -pse. It is possible that the -p- of ipse reflects the same element found in the Latin form -pte, a strengthening particle which shows up in suopte (cited in Paulus ex Festo 409, 1 as suopte pro suo ipsius, ut meopte pro meo ipsius...), and which may have a Greek equivalent in -πτε (τίπτε < τίποτε 'what then?'), though this is less secure. Ernout-Meillet (s.v. -pte) mention the possibility that -pte represents the accusative form of a particle which appears in the nominative as -pse (possibly represented by Gk. (Dor.) ψε 'they, them'), thereby unifying the two forms. Alternatively, the -p- of a putative form *i-p-se might contain the same element -p- that is seen in nem-pe, quippe and quis-p-iam. Given the inconclusive nature of the evidence for the

144

Philip Baldi

origin of the -ρ-, it seems quite unlikely that we will be able to arrive at an irrefutable conclusion. But in fact, either of the possible sources for the particle -p- will serve the needs of a credible derivation, since both served as strengthening or focusing particles in the cliticized constructions in which they are found (perhaps they should be unified). Alternatively, but not crucially for our current purposes, we might argue that in fact the elements underlying Latin -pse represent an unanalysable whole, and that, given the lack of secure comparanda for either -p- or se, we might do better to treat it as such and abandon the search for a compositional source *-p-se. But this seems not only unnecessary, but unwise. The -p- of Latin ipse is best thought of as continuing a ProtoIndo-European focusing particle -p-. But can we offer anything on the origin of the -sei In visiting this issue, we take advantage of the recent discovery of a Transalpine Celtic text at Chäteaubleau (Seine-et-Marne) in France, about 50 km. east of Paris. Uncovered in August 1997 in a rubbish pit in the Grands Jardins of Chäteaubleau, it is probably to be dated to the second or third century CE, perhaps somewhat later. The inscription is engraved on a tile containing eleven complete lines of text (Eska 2001).2 Since the inscription is written in a difficult Roman cursive script and has many uncertain readings, it is only now beginning to be published in the general literature, and many details remain uncertain. The relevant lines are 6 and 7, but only two words from them concern us directly.4 We are specifically interested in the forms ateri at the end of line 6 and xsi at the beginning of line 7. Lambert reads these two words as voc. ater plus a particle ixsi which he translates as 'a moi meme'. Lambert's interpretation of the two words together suggests that Λ

ixsi is functioning as a kind of dativus

sympatheticus

to voc.

ater,

equivalent to a Latin ο pater mihi, 'ο my father'. But as Eska has pointed out to me (p.c.), this interpretation depends ultimately on a most unlikely word division, with only a single character of putative ixsi before the line break. And although Lambert does suggest a connection between ixsi and Latin (dat.) ipsi, he does not elaborate the interpretation of ixsi as a dativus sympatheticus with 1 st person reference, viz. a moi. Finally, Lambert's apparent correlation of -xsi with Latin ipsi conceals a problem in the correspondence of the final vowel, long in Latin, but apparently short in Transalpine Celtic (e is expected).

Latin ipse, Continental Celtic -xsi: a tentative proposal

145

I propose here an alternative reading in which, following Eska (2001) ateri is interpreted as a dative 'to/for the father' (< loc. sg. PIE *ph2teri, syncretized with the dative), with completely regular loss of initial ρ and other phonological developments. Furthermore, I propose here to treat xsi as a pronominal clitic which reflects the same origins as the -pse of Latin i-pse, with the translation of ateri \ xsi as 'to/for the father himself. To justify this proposal, several criteria must be met. The first one is orthographic. Is it legitimate to treat xsi as a bound clitic form when it is in fact separated from its host word by a line break? While this treatment of clitics is common enough in Celtic and particularly Gaulish inscriptional materials, we can in fact find a supporting example right in the current inscription, in which line 7 appears to end in a 2nd pi. verb ending in -ete (bejiassusete), while line 8 (not reproduced here) begins with a clitic nota augens sue, showing that the separation brought about by the line break between ateri and xsi is not an impediment to our interpretation. The second part of our proposal requires a morphological justification. That is, is there a form which can be independently verified in the Celtic and Italic pronominal systems, and which can be reasonably postulated in a clitic-like function attached to a nominal form in combination with the focusing particle -ρ-Ί The answer again appears to be favorable to our interpretation, and supports the spirit of Ernout-Meillet's and Leumann's conjectures. Celtic is full of emphasizing pronominal particles such as se, sa, so and indeed si. While it is true that these particles typically attach themselves to verbs in Continental Celtic, attachment of these historically demonstrative elements to nouns is common in Old Irish, which at least lays the possibility of a general Celtic pattern which is typologically compatible with some form of a putative preLatin form *i-p(e)-so, as Leumann suggested. Since we are dealing here with what we are claiming to be a historically archaic form, the fact that there is no attested pattern of productive noun-particle suffixation in Continental Celtic would not appear to damage the current hypothesis. We propose that this form was grammaticalized as a particle of emphasis in both Italic and Celtic. Third, there is the phonological aspect of the proposal. Can we justify a protoform *{p)si as a legitimate ancestor to both -xsi and -psel The answer is securely positive. A proto sound *p normally and regu-

146

Philip Baldi

larly becomes χ before s in Gaulish, as in Latin crispus, Gaul. Crixus, Crixsus, with metathesis of -sp- to -ps- (cf. Lewis and Pedersen 1961 [1937]: 19), thereby yielding -xsi. As for Latin -pse, the phonology follows regularly in the history of Latin, with ps preserved, and final -e arising from final */, as in Latin mare < *mari (cf. pi. marid). Thus we can safely postulate a form *-psi as a justifiable predecessor to both -pse and -xsi. Whatever the philological merits (or demerits) of this proposal might be, the larger implications will certainly not have escaped attention. That is, given the lack of previously identified secure comparanda for the -pse of Latin ipse, if the current proposal to link it with the Continental Celtic -xsi is correct, then we have perhaps identified an ItaloCeltic isogloss. This isogloss is of the type which has long been identified as maximally diagnostic for the postulation of genetic connections between languages, which is to say that it is most unlikely that such a common formation of emphasizing particle plus pronominal reflects anything other than a shared morphological innovation. And the fact that Latin and Continental Celtic demonstratives have previously been identified in a special relationship, as Eska (1989) has shown for Hispano-Celtic IsTe and Latin iste, only serves to strengthen the current suggestion that Latin i-pse and Continental Celtic -xsi continue a common Italo-Celtic formation.

Notes 1. The proportions in the plural are even more unbalanced in favor of the nominative. 2. I became aware of this inscription from a presentation by Joseph Eska at the XIII UCLA Indo-European Conference, November, 2001.1 have profited from several exchanges with Prof. Eska, but he is blameless for any errors of fact or interpretation which may find their way into my conjectures. 3. See the recent papers by Lambert (1998-2000a, 1998-2000b), Schrijver (1998-2000). 4. The lines (nos. 7 and 8), based on Eska's text, are reproduced at the end of this note. Ligatures are in italic type; sequences with variant readings are in bold type; an underdot indicates an uncertain reading (e.g. e), the /'-longa is transcribed with ); and ss represents "tau Gallicum", which is a kind of dental fricative (see Eska 1998 for a review and narrowing of the possibilities). A precise translation of the text, including these lines, has yet to be offered,

Latin ipse, Continental Celtic -xsi: a tentative proposal

147

though the readings of Lambert (1998-2000a, b) can serve as a starting point (it should be noted that Lambert's transcription of these lines differs significantly from that of Eska, though not on the two critical forms under discussion here, namely ateri and xsi): siaxsiou bejiassuneftiti motun jummi ateri xsi jad ore core mwanajegumi sine bejiaseusete.

References Ernout, Alfred and Antoine Meillet 19674 Dictionnaire etymologique de la langue latine. Paris: Klincksieck. Eska, Joseph 1989 Italic e/iste, Hispano-Celtic IsTe. Emirita 57:317-323. 1998 Tau gallicum. Studio Celtica 32: 115-127. 2001 Remarks on linguistic structures in the Transalpine Celtic inscription of Chäteaubleau. Paper presented at the XIII Indo-European Conference, UCLA, November, 2001. Lambert, Pierre-Yves 1998-2000a La tuile gauloise de Chäteaubleau (Seine-et-Marne). Etudes celtiques 34: 57-115. 1998-2000b Les autres tuiles inscrites de Chäteaubleau (Seine-et-Marne). Etudes celtiques 34: 117-133. Leumann, Manu 1977 Lateinische Laut- und Formenlehre. Vol. 2, Part 2, No. 1 of Manu Leumann, J. Hofmann, and A. Szantyr. Lateinische Grammatik. München: Beck. Lewis, Henry and Holger Pedersen 1961 A Concise Comparative Celtic Grammar, [corrected reprint of 1937 edition], Göttingen: Vandenhoeck and Ruprecht. Schrijver, Peter 1998-2000 The Chäteaubleau tile as a link between Latin and French and between Gaulish and Brittonic. Etudes celtiques 34: 135-146. Walde, Alois and Johann Baptist Hofmann 1938-19543 Lateinisches etymologisches Wörterbuch. Heidelberg: Winter.

Der Name al-Andalus: ·· neue Überlegungen zu einem alten Problem Georg Bossong

Abstract Only a few years after the Islamic conquest of Spain, al-Andalus appears in coin inscriptions as the Arabic equivalent of Hispania. The traditionally held view that the etymology of this name has to do with the Vandals is shown to have no serious foundation. The phonetic, morphosyntactic, and also historical problems connected with this etymology are too numerous. Moreover, the existence of this name in various parts of central and northern Spain proves that al-Andalus cannot be derived from this Germanic tribe. It was the original name of the Punta Marroqui near Tarifa; very soon, it became generalized to designate the whole Peninsula. Undoubtedly, the name is of Pre-Indo-European origin. The parts of this compound (anda and luz) are frequent in the indigenous toponymy of the Iberian Peninsula. A semantic explanation is difficult, but Basque (or Vennemann's "Vasconic") may provide some clues.

Al-Andalus ist der arabische Name für Hispanien. Diese Bezeichnung ist seit den allerersten Anfangen belegt, und zwar nicht in schriftlichen Dokumenten aus späterer Zeit, sondern in einer Münzaufschrift; auf zwei zweisprachigen DTnärstücken, aufbewahrt im Museo Arqueologico Nacional in Madrid bzw. der Bibliotheque Nationale Paris, lesen wir auf der Vorderseite in lateinischer Sprache: ferit(us) soli(dus) in Span(ia) ann(o) XCVII\ und auf der Rückseite auf Arabisch: duriba hädä ldinäru bil-Andalus sanata tamänin wa-tiscTna „geprägt wurde dieser Dinar in Al-Andalus im Jahre 98" (zitiert nach Halm 1989: 254). Ob nun 97 oder 98 Higra (solche Diskrepanzen sind häufig), geprägt wurde die Münze zwischen dem 5. September 715 und dem 13. August 717 christlicher Zeitrechnung (Ocana Jimenez 1946: 46), nur wenige Jahre nachdem die ersten Muslime ihren Fuß auf hispanischen Boden gesetzt hatten. Der Name diente im Arabischen immer als Bezeichnung für den

150

Georg Bossong

islamisch beherrschten Teil der Iberischen Halbinsel, also ein Gebiet, das in den Jahrzehnten nach der Eroberung fast das gesamte Territorium außer einem schmalen Saum im Norden umfasste, während es in den Zeiten der 1492 entthronten Dynastie der Nasriden nur noch aus dem kleinen Königreich von Granada bestand - eine „peau de chagrin" für fast acht Jahrhunderte. Im heutigen Arabisch wird das moderne Isbäniyä vom historischen al-Andalus differenziert; das Adjektiv andalusi hat auch aktuelle Bedeutungen: in Marokko wird die klassische Musiktradition bis heute als müsJqä andalusiyya bezeichnet. Im Spanischen ist aus der arabischen Bezeichnung der Name der Region Andalucia entstanden, wobei man heute die Adjektive andaluz ,auf Andalusien bezogen' und andalusi ,auf al-Andalus bezogen' voneinander differenziert. Der älteste spanisch-arabische Historiker, Ahmad ibn Muhammad alRäzi aus Cordoba (gest. 955) fuhrt den Ursprung des Namens auf ein mythisches post-diluviales Volk zurück, das Andalus geheißen habe (nach Halm 1989: 252). Bei al-Himyari (gest. 1327) erscheint der Name in der Form Andalis; der Herausgeber Levi-Provengal erklärt ihn mit ,Wandalen', ohne dass der arabische Text dafür einen Anhaltspunkt liefert: summiyat bacda dälika bil-'andalus min JasmäJi l-1 andalis alladina sakanühä „Puis on l'appela al-Andalus (Andalousie), du nom d'al-AndalTs (les Vandales), qui vinrent s'y fixer." (1938: 2/ 4). Es ist klar, dass diese Etymologie ebenso märchenhaft ist wie die unmittelbar zuvor gegebene Ableitung von Isbäniyä von einem legendären Herrscher namens Isbän\ Schon der berühmte Arabist Dozy zweifelte daran, dass der Name alAndalus etwas mit den Vandalen zu tun habe, allein wegen der Kürze ihres Aufenthalts in Hispanien. Er war vielmehr der Auffassung, dass sich der Name ursprünglich nicht auf die ganze Halbinsel bezogen habe, sondern auf den Hafen, von dem aus die Vandalen sich nach Nordafrika eingeschifft haben; nach der „Frankenchronik" des Gregor von Tours war dieser Hafen das antike (Julia) Traducta. Wo dieses Traducta lag, ist unbekannt. Dozy setzte es mit dem späteren Tarifa gleich und vermutet einen zwischenzeitlichen Namen Vandalos, der von den „Vandalen" stamme; er schreibt: „il n'est pas surprenant que les ignorants Berberes de Tarif, debarques ä Vandalos, aient applique ce nom ä toute la contree qu'ils pillerent, et que plus tard les soldats de Täric l'aient donne, d'abord ä toute la Betique, ensuite ä toute l'Espagne." (Dozy 1881: 303). In dem Artikel „Al-Andalus" in der ersten Auflage der Encyclo-

Der Name al-Andalus: neue Überlegungen zu einem alten Problem

151

pedie de Γ Islam, der 1913 erschienen ist, nahm Streck diesen Gedanken wieder auf und modifizierte ihn dahingehend, dass er Traducta mit Algeciras gleichsetzte und statt Dozys Vandalos nunmehr Vandalicia schrieb. Wir können diese beiden leicht divergierenden Auffassungen wie folgt schematisieren. Laut Dozy ist die Abfolge so: Traducta [Abfahrtshafen der Vandalen, daher] i * Vandalos [Landeplatz des Berberführers Tarif, daher]

verallgemeinert zu: ,Baetica' (Andalusien), dann , Spanien' Laut Streck hingegen so: Traducta [Abfahrtshafen der Vandalen, daher] i * Vandalicia [Stützpunkt des Heerführers Täriq, ar. al-gazira, daher] A

lgeciras\^^

verallgemeinert zu: ,Baetica' (Andalusien), dann , Spanien' Diese Auffassung, wird - in summarischer, nicht nach einer der beiden Varianten differenzierten Form - auch noch in der neuesten CDRom Version der Encyclopedia of Islam vertreten. Letztlich stützt sich die Annahme einer Bedeutungsverallgemeinerung vor allem auf einen Hinweis in der anonymen Chronik JAhbär magmüca „Gesammelte Nachrichten" aus dem 11. Jahrhundert; in der spanische Übersetzung von Lafuente (1867: 6) lesen wir dort folgendes: Mu^a envio a uno de sus libertos llamado Tarif Abo Zorä, con 400 hombres, entre ellos 100 de caballeria, el cual paso en cuatro barcos y arribo a una isla llamada Isla de Andalus, que era arsenal de los cristianos y punto desde el cual zarpaban sus embarcaciönes. Por haber desembarcado alli, tomo el nombre de Isla de Tarif.

152

Georg Bossong

Ich denke, die Variante von Streck können wir getrost vergessen; es ist bekannt, dass Algeciras von Täriq gegründet worden ist. Aufgrund der Aussage aus °Ai}bär magmü'a ist davon auszugehen, dass die Filiation der Namen nach Dozy die richtige ist: Andalus ist ein alter Name, der sich darauf bezieht, was heute Tarifa heißt. Die arabische Chronik präzisiert, dass es sich um eine Insel handele. Die Frage ist nur, ob der von Dozy angenommene Name * Vandalos, sich wirklich auf die Vandalen bezogen hat. Ein Unbehagen darüber, diesen Namen von den Vandalen abzuleiten, hat es seit jeher gegeben. Keine der als Zwischenglieder angenommenen Namensformen ist belegt. Der Aufenthalt der Vandalen auf der Halbinsel sei, so Dozy, zu kurz gewesen, um dem ganzen Land den Namen zu geben, woraus er schließt, er könne sich ursprünglich nicht auf den gesamten Landstrich oder gar die ganze Halbinsel bezogen haben, sondern eben auf den Hafen, von dem aus dieses Volk sich nach Nordafrika eingeschifft hat. Damit bleibt aber die Frage, warum dieser Hafen über mehrere Jahrhunderte hinweg (die Vandalen sind 429 nach Afrika übergesetzt, Ankunft der Muslime war 711) untrennbar mit dem Namen der Vandalen verbunden geblieben sein soll. Dieser traditionelle Ansatz wurde 1997 von Volker Noll mit neuen Vorschlägen verteidigt; seine Konstruktionen nicht-belegter Zwischenglieder können insgesamt aber nicht überzeugen. Heinz Halm hat 1989 einen radikalen Gegenvorschlag gemacht. Nach dem Zeugnis des Orosius und später des Isidor von Sevilla hätten die germanischen Stämme der Alanen, Sueven und Vandalen bei ihrem Einfall in Spanien das Land per Los unter sich aufgeteilt. Später wird der Begriff gothica sors dann bezüglich der Landverteilung unter den Goten und sogar für das gesamte ,den Goten auf dem Boden des Imperiums zugefallene Landlos' verwendet. Halm schließt daraus auf das Vorhandensein eines - in den literarischen Quellen nicht belegten Kompositums *landa-hlauts für ,Landlos' —> ,Machtbereich der Goten' —•,westgotisches Spanien'. Schematisch: gotisch *landa-hlauts ,Landlos'

lateinisch (gotica) sors ,(gotisches) Spanien'

arabisch al-Andalus ,(maurisches) Spanien'

Der Name al-Andalus: neue Überlegungen zu einem alten Problem

153

Diese - vom Autor ausdrücklich als solche bezeichnete - Hypothese ist zweifellos geistreich und in mehr als einer Hinsicht bestechend. Ihr größter Vorteil fallt sofort ins Auge: sie vermeidet den in jeder Hinsicht unglücklichen Bezug auf die „Vandalen" als Etymon. Ein weiterer Vorteil liegt darin, dass sie die Konstanz des Artikels in der arabischen Bezeichnung al-Andalus erklärt. In der Tat haben fremde Toponyme im Arabischen meist keinen Artikel; durch die Re-Interpretation von *landa-hlauts —> *landalos als (a)l-andalos kommt hier ein arabischer Artikel an eine Stelle zu stehen, wo normalerweise kein solcher vorhanden ist. Allerdings bleibt bei Halm ein Einwand unberücksichtigt: nach allem was wir wissen, haben die Goten schon längst kein Germanisch mehr gesprochen, als sie in Spanien eingedrungen sind, sie waren längst zum Lateinischen übergegangen. Dass sich Germanisches noch bis ins 8. Jahrhundert hinein hätte halten sollen, scheint schwer vorstellbar. Hätte sich das gotische Wort schon vor dem Sprachwechsel als Appellativum fest etabliert haben sollen, das dann als germanisches Einsprengsel im Lateinischen weiterlebte? Warum finden wir dann davon keine Spur in den überlieferten Dokumenten? Mit dieser Überlegung kann die faszinierende und innovative These von Halm zwar noch nicht als widerlegt angesehen werden, aber es ist auch klar, dass man ihr mit einer gewissen Skepsis begegnen muss. Ich komme später noch einmal darauf zurück. Seit jeher weiß man um die Bedeutung der Ortsnamen als ältester Sprachschicht, in der Substrate auch dann noch fortleben, wenn die betreffenden Sprachen längst verstummt sind. Theo Vennemann hat die Toponomastik in den Dienst einer Forschung gestellt, welche die prähistorischen Verhältnisse in Europa in einem ganz neuen Licht erscheinen lässt. Seine eigenen Arbeiten beziehen sich zwar primär auf die Germania, auf Mittel- und Nordeuropa; aber die Iberische Halbinsel nimmt in den von ihm seit Jahren mit großer Intensität betriebenen Forschungen eine Schlüsselrolle ein. Als wir uns 1994 in Alicante trafen, um vor einem marokkanisch-spanischen Publikum über neue Entwicklungen in der präindogermanischen Toponomastik zu referieren, ahnten wir noch nicht, zu welchen Reisen im wörtlichen wie im übertragenen Sinne unser gemeinsamer Freund Juan Luis Roman del Cerro uns mitnehmen, wozu er uns anregen würde. Unsere Streifzüge durch die iberische To-

154

Georg Bossong

ponomastik in den entlegensten Winkeln der Provinz Alicante werden uns unvergesslich bleiben. Eine methodologische Lektion, die uns Theo Vennemann gelehrt hat, kann man so formulieren: gegenüber Etymologien aus historischer Zeit ist bei Ortsnamen grundsätzlich Skepsis angebracht; Re-Interpretation, „Volksetymologie" ist um so eher anzunehmen, je mehr Schichten das ursprüngliche Substrat überlagert haben. Jeder Sprachwechsel fuhrt auch zu immer neuen Deutungsversuchen längst opak gewordener uralter Bezeichnungen fur Flüsse, Berge und prähistorische Siedlungskerne. Und meist greifen auch die Thesen der modernen Wissenschaft zu kurz: wir sind allzu sehr geneigt, Römer, Kelten und Germanen als den letzten zeitlichen Horizont anzunehmen, den wir erreichen können. Woher wollen wir eigentlich wissen, dass Zaragoza von lateinisch Caesarea Augusta kommt, trotz der lautlichen Ungereimtheiten? Vielleicht haben ja auch schon die Römer einen präexistenten Namen umgedeutet. Ja, auch Cädiz, die älteste Stadt Europas, hat ihren Namen vielleicht nicht aus dem phönizischen Wort fur ,Mauern' (vgl. Porlan 1998: 30ff, 63ff). Ein anderes, näher liegendes Beispiel: Theo Vennemann wird wissen, worauf ich anspiele, wenn ich sage, dass ich kurz nach unserer Entdeckung des Berg-,Rückens' Bizcoy im Hinterland von Alicante auf einen prachtvollen Berg-,Rücken' namens Bischenberg zwischen dem Flughafen von Strassburg und dem Vogesenkamm gestoßen bin - und darauf liegt ein Dorf namens Bischofsheim, in dem es natürlich nie einen ,Bischof gegeben hat! Solche Schlüsselerlebnisse haben die Beweiskraft unmittelbar anschaulicher Evidenz. Kommen wir zurück zu al-Andalus. Dass dieser Name nicht von den Vandalen kommen kann, ergibt sich allein schon aus seiner Form: der Bestandteil -us ist absolut konstant, seit den ältesten Belegen. Woher sollte ein solches Element kommen? Etwa von einem lateinischen Nominativ Singular Vandalusl Wie könnte dies sein bei der Bezeichnung nicht eines Individuums, sondern eines Volkes? Wie könnte dies sein in einem Zusammenhang, wo man, wenn es schon eine lateinische Form sein soll, viel eher einen Genitiv Plural erwarten würde (*insula Vandalorum als Muster für gazlrat al-Andalus)? Wie könnte dies sein, wo sich ja auch im Spanischen ein solches Genitiv-s ausschließlich in dem ekklesiastisch beeinflussten Dios erhalten hat? Wie könnte dies sein angesichts der klaren arabisch-lateinischen Äquivalenz al-Andalus =

Der Name

al-Andalus: neue Überlegungen zu einem alten Problem

155

Spania auf den ältesten Münzen? Sollte nicht wenigstens damals noch eine Erinnerung an die Vandalen vorhanden gewesen sein, wenn sie denn tatsächlich die Namensgeber gewesen wären? Auch die Deutung des Anlauts ohne w- mithilfe des Berberischen ist unbefriedigend: zwar gibt es im Kabylischen ein Präfix, das des Status constructus im Maskulin Singular markiert, aber auch wenn man hier eine Re-Analyse des Typus lateinisch Vandalus —> kab. w-/0-andalus annähme, bleibt das Problem des unerklärten Singular: ein berberisches Singular-Präfix vor einem lateinischen Nominativ Singular für den Begriff ,Halbinsel der Vandalen'? Das passt weder grammatisch noch semantisch. Nein, ich glaube, jeden Rückgriff auf die Vandalen können wir getrost ad acta legen, gegen die heute noch herrschende communis opinio. Auch Noll (1997) hat gezeigt, dass diese lautlich-morphologischen Einwände gegen die Ableitung des Namens von den Vandalen letztlich unwiderleglich sind. Wo können wir also suchen, wenn wir gegenüber der interessanten These von Halm aus den erwähnten Gründen trotz allem skeptisch bleiben müssen? Was liegt, im Sinne der Forschungsperspektiven von Theo Vennemann, näher als die Vermutung, es handle sich um einen prähistorischen Ortsnamen? Dieser kann sich dann allerdings nicht auf die Iberische Halbinsel insgesamt bezogen haben, denn Toponyma aus Substratsprachen sind kleinräumig, sie beziehen sich auf konkrete Gegebenheiten in einem überschaubaren geographischen Rahmen. Wenn es sich tatsächlich um einen solchen Substratnamen handelt, muss er eine Referenz in dem Bereich haben, mit dem die Muslime bei ihrer Eroberung erstmals in Berührung kamen. Ich denke, dass an der These von Dozy soviel richtig ist, dass das Etymon von al-Andalus ursprünglich der Name einer geographischen Gegebenheit gewesen sein muss, die mit der später so genannten - Stadt Tarifa etwas zu tun hatte und von dort verallgemeinert worden ist. Das erste Vorauskommando der Muslime, eine Truppe von 400 Mann Infanterie und 100 Berittenen unter der Führung des Berberführers Tarif ibn Mallük al-MacäfirT, kunya Abü Zura, landete mit vier Schiffen im Ramadan des Jahres 91/ 710 auf „der Insel al-Andalus" (siehe das oben angeführte Zitat aus 'Afabär magmü'a; zu den historischen Details vgl. beispielsweise Chejne 1993: 19ff; Makki 1992: 6ff). Sofort danach sei dieser Ort nach dem berberischen Heerführer in Tarif umbenannt worden. Es ist anzunehmen, dass sich der Name auf die

15 6

Georg Bossong

kleine Insel bezieht, welche der Stadt vorgelagert ist (die Stadt gab es damals wohl noch gar nicht, die Quellen sprechen lediglich von einem „Arsenal" der Christen). Diese Insel ist heute über eine Dammbrücke mit dem Festland verbunden; ihre Südspitze trägt den Namen „Punta Marroqui", sie bildet den südlichsten Punkt des europäischen Festlandes, genau auf dem 36. Breitengrad (nur Kreta liegt noch südlicher). Die Muslime hatten damals noch keine genaue Vorstellung von den Ausmaßen der Iberischen Halbinsel, von der man ihnen allenfalls in Ceuta mündlich berichtet hatte; sie konnten im Moment der ersten Erkundungen noch nicht wissen, dass das ganze Land eine große Halbinsel ist; die Bezeichnung gazira konnte sich zu diesem Zeitpunkt also noch nicht auf die Iberische Halbinsel insgesamt, sondern nur auf die winzige Insel ganz in ihrem Süden beziehen. (Gazira bedeutet im Arabischen sowohl ,Insel' als auch ,Halbinsel'; gazirat al-cArab war schon zu Zeiten des Propheten Muhammad die Bezeichnung der Arabischen Halbinsel.) Wie unsicher die Kenntnisse zu Beginn noch waren, geht auch daraus hervor, dass der Befehlshaber der muslimischen Truppen, Müsä ibn Nusayr, bei dem Kalifen al-Walld um Erlaubnis zur Eroberung des Landes jenseits des Meeres nachfragte und dieser antwortete, er solle die Muslime nicht unnötig den Gefahren und Stürmen des Meeres aussetzen; darauf schrieb Müsä ibn Nusayr zurück, es sei nur ein Meeresarm, man könne von der marokkanischen Küste aus gut die andere Seite sehen. Trotzdem hielt er es für nötig, erst einmal das Vorauskommando unter Tarif hinüber zu schicken, ehe er seinen Stellvertreter Täriq ibn Ziyäd im darauf folgenden Jahr mit 7000 Mann in der Bucht von Algeciras landen ließ. Täriq ließ ebendort alle Schiffe verbrennen, um seine Mannen zum Durchhalten gegen einen scheinbar übermächtigen Feind zu zwingen. Die Muslime nahmen die Halbinsel dann aber im Sturm ein, in einem Blitzkrieg ohnegleichen, anfangs mit kriegerischen, später dann fast nur noch mit diplomatischen Mitteln. Die Übertragung des Namens von der kleinen Insel auf die große Halbinsel muss unmittelbar nach der Errichtung des ersten Brückenkopfes in Tarifa erfolgt sein; der ursprüngliche Bezug geriet nach der Umbenennung der kleinen Insel Gazirat al-Andalus in Gazirat Tarif sogleich in Vergessenheit. Auf der Münze von 716 deutet die Übersetzungsäquivalenz al-Andalus = Spania darauf hin, dass die Bedeutungsübertragung damals, als die Halbinsel zum allergrößten Teil unter muslimischer Herrschaft stand, bereits vollzogen war.

Der Name al-Andalus: neue Überlegungen zu einem alten Problem

157

Bis dahin hat die hier vertretene Hypothese eine gewisse sachliche Plausibilität; sie passt zu dem Zeugnis der Quellen und leuchtet im Hinblick auf die Psychologie der vorwärtsstürmenden berberischen und arabischen Eroberer der Halbinsel ein: binnen kürzester Zeit erschloss sich ihnen eine ganz neue Welt, so dass man verstehen kann, wie sie den Namen ihres allerersten Stützpunktes sogleich auf die ganze riesige Halbinsel übertragen konnten. Der Vorgang ist auf einer anderen Ebene damit vergleichbar, dass Asien, der Name einer Provinz in Westanatolien, schließlich auf den größten Kontinent der Erde übertragen wurde; oder dass der Subkontinent Indien seinen Namen von dem Strom in seinem äußersten Nordwesten erhalten hat, auf den man als ersten stößt, sobald man den Khyber-Pass überwunden hat. Nachvollziehen kann man einen solchen Vorgang also sehr wohl, aber gibt es weitere, womöglich noch zwingendere Argumente? Die präromanische Toponymie der Iberischen Halbinsel liefert uns zwei solche Argumente. Das erste liegt in unmittelbarer Nachbarschaft. Wie soeben gesagt, stand das Expeditionsheer, das sich 711 auf den Weg machte, um Spanien zu erobern, unter der Führung von Täriq ibn Ziyäd. Der Berg, der nach ihm Gabal Täriq —> Gibraltar genannt ist, hat einen prähistorischen Namen, der sich bis heute erhalten hat: Calpe. Es gibt einen weiteren steil aus dem Mittelmeer ragenden Berg, der allgemein bekannt ist: der Pehön de Ifach zwischen Benidorm und Denia. Zu seinen Füssen liegt ein Dorf mit Namen Calpe. Die Wurzel ^kal ,Felsen' gehört zu den weit verbreiteten und in der Forschung generell anerkannten Bestandteilen der prähistorischen Toponomastik Europas; wenn man die Interpretation von -pe mit Hilfe des Baskischen als Postposition mit dem Sinn ,unter(halb)' annimmt, lässt er sich deuten als ,unterhalb des Felsens' - eine perfekte Beschreibung der frühen Siedlungen sowohl in Gibraltar als auch am Pefion de Ifach. Wir finden hier die Konstellation, dass ein prähistorischer Name Konkurrenz erhielt durch eine arabische Bildung, in der sich der neue Eroberer verewigt hat, also ein Vorgang, der genau parallel zu der hier postulierten Entwicklung von al-Andalus verlaufen ist - nur mit dem Unterschied, dass al-Andalus seinen ursprünglichen Bezug infolge von Bedeutungserweiterung verloren hat, während Calpe neben Gibraltar bis heute fortbesteht. Schematisch:

158

Georg Bossong Tarifa [nach Tarif al-Mullük, 710]

al-Andalus , Spanien'

Kal-pe


an und abe > ab. Wie schwierig es ist, aus heutiger Sicht die einzelnen Reime dieses Typs zu bewerten, soll an ausgewählten Beispielen illustriert werden. Ich greife zur Veranschaulichung dieses Problems zunächst das literaturgeschichtlich und auch bei Heine überaus häufige Reimpaar ab bzw. herab - Grab heraus. Im Mhd. war der Vokal in diesem Reimpaar kurz. Die Partikeln weisen im Neuhochdeutschen einen scharf geschnittenen Vokal auf, Grab hat einen sanft geschnittenen Vokal. Der Vokal in Grab wurde in der Standardsprache des Neuhochdeutschen nicht aufgrund seiner Silbenstruktur, sondern in Analogie zu den flekierten Formen Grabe und Grabes ,gedehnt'. Dieser Ausgleich unterblieb in niederdeutschen Dialekten, wo Grab, Tag, Weg, Glas u. Ä. einen Kurzvokal aufweisen (vgl. Schirmunski 1962: 187), so dass eine Reimauflösung mit , Kurz'vokal möglich erscheint. Dasselbe geschah in einigen Dialekten mit ab, das aufgrund der mhd. Nebenform mit offener Tonsilbe abe ,gedehnt' wurde. In einigen Dialekten (ζ. B. Schlesischen, vgl. Neumann 1920: 193) wäre also die Reimauflösung dieses Reimpaares mit sanftem Schnitt möglich. Ähnliches gilt für die Partikel an, die an recht vielen verdächtigen Reimen beteiligt ist. Mit ,Lang'vokal wurde sie ζ. B. in Schlesien , Thüringen und Obersachsen ausgesprochen (vgl. Gottsched 1762: 82; Neumann 1920: 141, 194). Große Variation herrscht bis in die Gegenwart hinein bei den Suffixen -sam, -bar, -tum

286

Beatrice

Primus

sowie bei Bräutigam. Man bedenke auch Fälle wie betreten - Städten (Goethe, WD), die bis ins heutige Deutsche variabel gelesen werden können. Stadt hat nur scharfen Schnitt, aber für Städten erlaubt die heutige orthoepische Norm scharfen und sanften Schnitt (Duden 1990). In der heutigen Umgangssprache gibt es eine viel größere Variation im Bereich der Vokalopposition, so dass wir davon ausgehen können, dass dies auch für die Umgangssprache der ostmitteldeutschen Kulturzentren der Goethezeit gilt. Einen guten Überblick über die Variation im Bereich der Vokalopposition auf älteren Sprachstufen bietet Neumann (1920). Nun können nicht alle Reime dieses Typs als rein wegerklärt werden. Ein verlässlicher Indikator der Vokalopposition ist die Orthographie. Orthographisch gekennzeichnete ,Länge' bzw. ,Kürze' galt als gesicherte Norm schon in der Goethezeit (vgl. Gottsched 1762: 81-83). Unrein erweisen sich somit Reime wie Paradies - gewiß, alle - Strahle (Schiller, ET), rief- Felsenriff (Schiller T), wohl - ahnungsvoll (Busch, RO), vermutlich auch ziehn - dahin (Goethe, "Anklang"), glühn - Königin (Schiller, A). Der Vergleich der folgenden Reime von Heine wundersam - Eichenstamm, wunnesam - Bräutigam, nahm - Bräutigam (JL) ergibt, dass nicht alle rein gelesen werden können.

2.4. Frikativierung velarer Verschlusslaute Wenden wir uns nun den konsonantisch unreinen Reimen zu. Da die wichtigsten Probleme bei der Bewertung vokalisch unreiner Reime ausfuhrlich erläutert wurden, sollen in den Abschnitten 2.4-2.5 nur die wichtigsten einschlägigen Annahmen hervorgehoben werden. Es geht um den Nachweis einer phonologischen Beschränkung mit überregionaler Verbreitung und einer darauf aufbauenden literarischen Reimnorm. Da Neumann (1920: 289-295) die Zeit vor Goethe sehr gut abdeckt, wird im Folgenden nur die spätere Reimdichtung mit Belegen illustriert. Ein erster konsonantischer Reimtyp betrifft die Frikativierung velarer Konsonanten: (28) Ist 's euch nicht genug, Daß ins Leichentuch, Daß Ihr früh mich ins Grab gebracht?

(Goethe, BK)

Unreine Reime undphonologische

Theorie

287

Dass wir es auch hier mit literarisch zugelassenen Reimen zu tun haben, belegen fur die Zeit vor Goethe Neumann (1920) und die folgenden Reimstellen für die Zeit danach: (29) Goethe: genug - Leichentuch (BK); Sprache - Tage (ZX) Schiller: zugleich - Zweig (T) Heine: steigt - erbleicht, hergebracht - Liebesjagd, Taschentuch - trug (JL); Pracht - Smaragd (GD) Busch: Geschichte - kriegte (MM) Folgende phonologische Beschränkung lizensiert diesen Reimtyp: (30) Frikativierung: Es gibt keine velaren Plosive. D.h. den velaren Plosiven bestimmter Sprachvarietäten entsprechen velare und palatale Frikative. Das Ergebnis der Frikativierung hängt von der Lautumgebung bzw. dem Lautbestand der Frikative ab. Was das heutige Standarddeutsche betrifft, so kommt nach einem hinteren Vokal nur der velare Frikativ vor (vgl. Tuch, Sprache), sonst der palatale (vgl. zugleich, erbleicht, Geschichte). Der Stimmtonverlust ist im Silben- und Wortauslaut durch die Auslautverhärtung bedingt. Aber auch sonst, ζ. B. in neige - reiche, ergibt er sich von selbst dadurch, dass im nativen Wortschatz des Standarddeutschen kein stimmhaftes Gegenstück zu [9] existiert. Dialektal kommen auch andere Frikative in Frage (vgl. Schirmunski 1962: 301— 318). Wie alle hier diskutierten phonologischen Beschränkungen, die unreine Reime lizensieren, erfahrt auch die Frikativierung eine weite überregionale Verbreitung und betrifft nicht nur die velaren, sondern auch andere Plosive (besonders den bilabialen). Sie ist im Niederdeutschen, Mitteldeutschen und einem beträchtlichen Teil des Oberdeutschen, besonders in intervokalischer Stellung und im Auslaut attestiert (Schirmunski 1962: 302-318; Simmler 1983). Wichtig fur die laufende Argumentation ist die Feststellung, dass der literarische Reimkanon selektiv verfahrt und nur die Variation der velaren Plosive, die tatsächlich beträchtlicher ist als die der anderen Plosive, zulässt. Adelung bemerkt dazu (1782,1, 152-3): "g wird unter allen Buchstaben in den Deutschen Mundarten am schwankendsten und unbestimmtesten ausgesprochen." Aber schon bei Adelung wird die Frikativierung als normwidrig abgekanzelt. Dennoch hat sie auch im heutigen Standarddeutschen Spuren

288

Beatrice Primus

hinterlassen. So wird auslautendes -ig (vgl. König, gütig) frikativiert ausgesprochen. Im Lichte dieser Überlegungen darf die Abwertung des in literarischen Metriken vielfach erwähnten Goetheschen Reims neige - reiche in Frage gestellt werden. Vgl. (31)-(32): (31) Ach neige, Du Schmerzensreiche, Dein Antlitz gnädig meiner Not! (Faust, Verse 3588-3590) (32) Neige, neige, Du Ohnegleiche, Du Strahlenreiche, Dein Antlitz gnädig meinem Glück! (Faust, Verse 12069-12072) Nicht nur die nachgewiesene phonologische und literarische Lizensierung dieses Reimtyps, sondern auch die wichtigen Textstellen, in denen Goethe, der sehr sorgfältig reimt, diese Reime unterbringt, lassen ihre Geringschätzung fragwürdig erscheinen.

2.5.

Stimmtonneutralisation

Die letzte einschlägige phonologische Beschränkung, die für unreine Reime in Frage kommt, betrifft die Stimmtonneutralisation. Vgl. (33): (33) Recht aber soll vorzüglich heißen, Was ich und meine Gevattern preisen

(Goethe, ZX)

Den literarischen Reimkanon vor Goethe behandelt Neumann (1920: 286-287, 298-299.) ausführlich, die Reimtradition danach wird in (34) belegt: (34) Goethe: Freude - Seite (WA); Wiesen - umfließen (Ζ); heißen preisen, Abstruse - Buße, Propheten - Reden, reden töten (ZX) Schiller: Helden - Welten, baden - Saaten, Frühlingsrosen hingegossen, Rose - schlosse (A); Sarge - Marke, Mutter - Bruder, Werken - Tatenbergen, vergolden - rollten (ET); gelten - melden (F); Getose - Schöße (2 x) (T); Zeitenschoße - Lose (LG)

Unreine Reime undphonologische

Heine: Busch:

Theorie

289

Gruße - Muse (JL); Kameraden - waten, Kaiser - heiß er (GD) dieses - hieß es (MM)

Auch dieser Reimtyp ist durch eine phonologische Beschränkung mit ausgeprägter überregionaler Verbreitung lizensiert: (35) Stimmtonbeschränkung: Es gibt keinen Stimmtonkontrast. Die Neutralisierung von stimmhaften und stimmlosen Konsonanten herrschte mehrere Jahrhunderte lang in der hochdeutschen Aussprache in nahezu allen Sprachregionen (vgl. Schirmunski 1962: 330-346; Esawa 1972; Simmler 1983) und erfasste auch das Obersächsische (vgl. Adelung 1782, II: 684). Sie blieb als Auslautverhärtung im Silbenauslaut auch im heutigen Standarddeutschen erhalten. Intervokalisch, wie in den hier diskutierten Reimen, wurde der Stimmtonkontrast in vielen Dialekten zugunsten einer stimmhaften bzw. lenisierten Aussprache aufgehoben (zur /s/-/z/-Opposition im Besonderen vgl. Adelung 1782,1: 166-172; Neumann 1920: 298-299, Schirmunski 1962: 357f.). Die Ergebnisse dieses Abschnitts können nun zusammengefasst werden. Die unreinen Reime werden durch die Anwendung einer geringen Zahl von überregional verbreiteten phonologischen Beschränkungen lizensiert. Im Bereich der Vokale haben wir die Entrundung vorderer Vokale, die Aufhebung des Höhenkontrasts bei e-Lauten und den Zusammenbruch der Quantität. Bei den Konsonanten geht es um die Frikativierung velarer Verschlusslaute und die Stimmtonneutralisation. Wichtig ist dabei die Feststellung, dass alle diese Beschränkungen auch in der obersächsischen (,meißnischen') Variante des Hochdeutschen, die als nationales Vorbild Ende des 18 Jhs. immer mehr an Bedeutung gewinnt, ihre Wirkung zeigten. Die Übertragung dieser phonologischen Erscheinungen auf Endreime hat eine literaturhistorische Tradition, die nicht alle überregional operierenden phonologischen Beschränkungen heranzieht, sondern wie jede Normierung selektiv vorgeht und sich genau auf die hier besprochenen beschränkt. Diese Ergebnisse befähigen uns, den Stellenwert unreiner Reime in der Literatur- und Sprachwissenschaft, besser zu verstehen. Dieser Frage widmen sich die folgenden Abschnitte.

290

Beatrice Primus

3. Die Relevanz der phonologischen Reimanalyse für die Literaturwissenschaft Die sprachwissenschaftliche Reimanalyse ergab, dass die behandelten Dichter, die selbst zur Bildung bzw. Festigung des behandelten Reimkanons beitrugen, eine sehr sichere Sprach- und Reimkompetenz unter Beweis stellen. Sie verwenden fast ausschließlich nur solche Unreinheiten, die phonologisch und literarisch lizensiert sind. Andere Reimtypen, wie ζ. B. Odem - Boden (Heine, JL), ihm - hin (Goethe, F), begreifen weichen (Goethe, ZX) werden nur isoliert verwendet, obwohl sie lautlich recht unauffällig sind. Die lizensierten Reimtypen werden im Gegensatz dazu recht häufig (ca. 10% der Reime von Heine und Goethe in meinem Korpus) und über die verschiedenen Schaffensperioden recht konstant verwendet. Ausgenommen davon sind sehr frühe Reimdichtungen Goethes (z.B. das Versdrama "Die Laune der Verliebten") und frühe Werke Schillers mit einem höheren Anteil an unreinen Reimen. Bei den untersuchten Dichtern zeigt sich außerdem keine Tendenz, die lizensierten unreinen Reime unauffällig unterzubringen, sie in benachbarten Paarreimen, an exponierten Textstellen (Zueignung, Motto, Schlussverse) oder in streng reglementierten Formtypen (z.B. Sonetten) zu vermeiden. Die kanonstiftende Wirkung Goethes darf dabei nicht übersehen werden. Sie zeigt sich für Neumann (1920) darin, dass die Vielfalt der unreinen Reime in der Folge von Goethe stark zurückgeht. Dies bestätigt auch meine Stichprobe aus Schillers Reimdichtung, die vor 1797, dem Jahr der engsten Zusammenarbeit mit Goethe, 27% betrug und danach auf 17% zurückgeht. Wichtiger als diese vage quantitative Statistik ist die qualitative Entwicklung Schillers. Während in frühen Werken die schwäbische e/z-Neutralisation vor Nasal noch durchschlägt {Gesängen - schwingen, geringe - Gesänge, beschimmert - (nieder)gedämmert (A)), beschränkt sich Schiller in der späteren Schaffensperiode - bis auf isolierte Ausnahmen - auf die lizensierten Reimtypen. Am feinen Gespür dieser Dichter für die Sprachentwicklung und Reimtradition ihrer Zeit dürfte aufgrund dieser Ergebnisse nicht gezweifelt werden. Dennoch findet man in einschlägigen literaturwissenschaftlichen Standardwerken gegenteilige Aussagen. Nicht nur dass Karl Martin Schiller (1963: 8) Goethes Reim neige - reiche als falsch erklärt, für ihn

Unreine Reime undphonologische

Theorie

291

stellt jede "Nichtübereinstimmung der (vermeintlichen) Reimworte [...] einen Kunstfehler dar." Wagenknecht wiederum begeht den Fehler, vor dem Neumann (1920: 305) warnt, nämlich Unreinheiten als dialektal reine Reime wegzuerklären. Bei Wagenknecht heißt es (1989: 36): "Im übrigen bemißt sich die Übereinstimmung am jeweils vorausgesetzten Dialekt - weshalb "Ereignis" : "Gleichnis" bei Goethe und "Tuch" : "genug" bei Brecht für reine Reime anzusehen sind." Weder bei diesem noch bei den anderen lizensierten Reimtypen handelt es sich um dialektale Ausrutscher, was im Falle der Entrundung in dieser Arbeit näher begründet wurde. Im Rückblick muss außerdem hervorgehoben werden, dass alle lizensierten unreinen Reime, zu denen auch die von Wagenknecht genannten gehören, auf phonologischen Erscheinungen basieren, die im Hochdeutschen des obersächsischen Raums ihre Wirkung zeigten und meistens unter niederdeutschem Einfluss zurückgedrängt wurden. Natürlich gibt es auch dialektale Reime im untersuchten Korpus. Diese wurden jedoch - wie oben fur Schiller geltend gemacht - in reiferen Werken beseitigt oder - wie im Falle Heines (vgl. Abschnitt 2.1 weiter oben) - orthographisch explizit gemacht. Wir kommen den hier diskutierten unreinen Reimen eher auf den Grund, wenn wir bedenken, dass es eine streng normierte und kodifizierte Nationalsprache, wie sie seit dem Humanismus über Opitz, Gottsched und Adelung zwar gefordert und gefordert wird, aber erst im späten 19. Jahrhundert z.B. von Siebs und Duden kodifiziert und garantiert wird, in der Goethezeit noch nicht gab. Wir finden in der hier besprochenen Reimtradition überregionale Züge der deutschen Sprache, die im Ringen um die Normierung der deutschen Standardsprache erst allmählich untergegangen bzw. nicht zum Zuge gekommen sind. Die komplexen Vorgänge, die zur Bildung unserer nationalen Standardsprache führten, werden in dieser Reimtradition partiell widergespiegelt.

4. Die Relevanz literarischer Reimdaten für die phonologische Theoriebildung Literarische Reime spielen auch für die phonologische Theoriebildung eine wichtige Rolle. Für ältere Sprachstufen dienen sie der sprachlichen Rekonstruktion. Bei fehlender lautsprachlicher Evidenz muss man sich

292

Beatrice Primus

mit solchen Daten zufrieden geben. Anders verhält es sich mit den in Gebrauch befindlichen Sprachen, obwohl Reime auch hierfür sehr gerne (z.B. Ohala 1986, Berg 1990) als linguistische Daten zur Untermauerung phonologischer Hypothesen herangezogen werden. Im Folgenden soll gezeigt werden, dass ihr Wert überschätzt worden ist. Es handelt sich um primär literarische Daten, die - wie oben demonstriert - einem impliziten oder expliziten literarischen Reimkanon unterliegen, der verhindert, dass sie die gesamte Sprachwirklichkeit abbilden. Außerdem sind Endreime im Besonderen defizitär, weil sie keinen Einblick in phonologische Prozesse gewähren, die den fußinitialen Silbenonset betreffen. Fangen wir mit dem Reimbegriff selbst an. Der literarische Begriff wird gerne als Evidenz für die Annahme einer phonologischen silbischen Reimkonstituente herangezogen (vgl. Ramers 1998: 101), für die sich auch primär sprachliche Evidenz finden lässt (vgl. Neef 1996). Besonders wichtig ist die Reimkonstituente für die Berechnung des Silbengewichts. Aber hier zeigen sich schon die ersten Probleme beim Vergleich mit dem literarischen Reimbegriff. Denn zum einen gibt es Sprachen, in denen auch der Silbenonset zur Berechnung des Gewichts zählt (vgl. Auer 1991). Zum anderen deckt sich der literarische Reimbegriff - zumindest partiell - mit dem linguistischen Fußbegriff. Dies erkennt man schon daran, dass es sowohl einsilbige (Bsp. Tag - wach) wie mehrsilbige Endreimstellen (Bsp. neige - reiche) gibt. Entscheidend ist, dass Endreimstelle wie Fuß nur einen Akzent aufweisen dürfen. Das Problem ist, dass man für die Berechnung des Silbengewichts nicht einen gesamten zweisilbigen Fuß benötigt. Es dürfte somit klar sein, dass für die Berechnung des Silbengewichts eine linguistische Einheit in Frage kommt, die sich nicht genau mit dem literarischen Reimbegriff deckt. Wie defizitär literarische Reimdaten für den Nachweis phonologischer Regeln bzw. Beschränkungen sind, soll anhand der Arbeit von Berg (1990) veranschaulicht werden. Ich greife zur Illustration Bergs Hypothese heraus, dass der Stimmton im heutigen Standarddeutschen eine prosodisch-suprasegmentale Erscheinung sei. Untermauert wird diese Hypothese in erster Linie mit den unreinen Reimen Wilhelm Büschs. Zum ersten geht Berg methodisch ungenau vor, weil er nur eine Auswahl der lizensierten Reime behandelt, die Busch verwendet. Unreinheiten bezüglich der Höhe der e-Laute und Frikativierungen werden nicht beachtet.

Unreine Reime undphonologische

Theorie

293

Als Evidenz für die suprasegmentale Natur des Stimmtons beansprucht Berg die Tatsache, dass er sich besonders leicht vom vokalischen ,Kern', also von den übrigen phonologischen Merkmalen, abtrennen lässt. Dies ist zweifellos richtig, belegt aber zunächst nur die Tatsache, dass der Stimmton ein eigenständiges phonologisches Merkmal ist. Alle phonologischen Regeln heben nämlich genau ein phonologisches Merkmal auf, was in dieser Arbeit weiter oben an unreinen Reimen, deren Merkmal nicht suprasegmental interpretiert werden kann (Rundung, Höhenkontrast, Frikativ/Plosiv), hinreichend demonstriert wurde. Berg geht in seiner Argumentation weiter und weitet seine Hypothese hinsichtlich des laryngalen Knotens der Merkmalsgeometrie aus (Clements 1985). Hier kommt die Problematik seines Datenmaterials sehr deutlich zum Vorschein. An laryngalen Kontrasten sind nämlich nicht nur stimmhafte und stimmlose Konsonanten, sondern auch Hauchlaut und glottaler Plosiv beteiligt. Das Problem ist, dass diese Laute fur Endreime irrelevant sind. Sie kommen nämlich nur im Silbenonset vor. Erst wenn man alle laryngalen Merkmale heranzieht, fällt auf, dass sie allesamt prosodischen Beschränkungen unterliegen (vgl. Neugebauer 2001). Durch die Vernachlässigung primär linguistischer Daten und die Überbewertung literarischer Reimdaten hatte Berg keine Chance, dieses einheitliche Verhalten laryngaler Merkmale aufzudecken. Das ist um so bedauerlicher, als eine so auffallig einheitliche prosodische Lizensierung die Hypothese Bergs, dass Laryngalität suprasegmentaler Natur sei, viel besser untermauert hätte als die Unreinheit von Reimen. Fazit ist, dass Reime nicht genuin linguistische, sondern primär literarische Daten sind. Sie sind bei umsichtigem Gebrauch fur linguistische Hypothesen dennoch interessant, wie dieser Beitrag zu zeigen versucht hat.

Anmerkungen 1. Mein Datenmaterial umfasst nahezu die gesamte Reimdichtung Heines und umfangreichere Stichproben aus der Reimdichtung Goethes und Schillers. Einbezogen wurden die empirischen Befunde aus der Reimuntersuchung von Friedrich Neumann (1920), der den Zeitraum von 1620 bis ca. 1750, also von Opitz bis Wieland, abdeckt, sowie die Befunde von Thomas Berg (1990), der die gesamte Reimdichtung Wilhelm Büschs untersucht hat.

294

Beatrice Primus

2. Der Diphthong [oy], der der Entrundung unterliegt, entsteht dadurch, dass die Aussprache der Diphthongbestandteile das vom Sprecher intendierte artikulatorische Ziel nicht erreicht (vgl. Vennemann 1982: 275; Wiese 2000: 159). Es ist plausibel anzunehmen (vgl. Wiese 2000: 159), dass dieser Diphthong phonologisch als /ay/ zu repräsentieren ist. Durch Entrundung des zweiten vorderen Bestandteils entsteht /ai/. Vgl. jedoch Berg (1990: 10), der die Diphthonge als /oi/ und /ai/ analysiert und den Unterschied im ersten Bestandteil vermutet. Das Problem dieser Analyse ist, dass die Entrundung den hinteren runden Vokal des ersten Bestandteils nicht erfassen kann.

Literatur

Literarische Autoren und Textnachweise mit den verwendeten Siglen (chronologisch geordnet) Martin Opitz (Schlesien, 1597-1639): Über des Hochgelehrten und weltberühmten Danielis Heinsij Niderländische Poemata (DH) Liedt Andreas Gryphius (Schlesien, 1616-1664): Cardenio und Celinde (CC) Christian Hofmann von Hofmannswaldau (Schlesien, 1617-1679): Beschreibung vollkommener Schönheit (BS) Daniel Casper von Lohenstein (Schlesien, 1635-1683): Cleopatra (C) Barthold Heinrich Brockes (Hamburg, 1680-1747): Kirschblüte bei der Nacht (KN) Die kleine Fliege (KF) Christoph Martin Wieland (Württemberg, 1733-1813): Musarion oder die Philosophie der Grazien (M) Johann Wolfgang von Goethe (Frankfurt a. M., 1749-1832): An Luna (L) Willkommen und Abschied (WA) Der König in Thüle (KT) Jägers Abendlied (JA) Wanderers Nachtlied (WN Der du von dem Himmel) Mit einer Hyazinthe (MH) Der Fischer (F) An den Mond (M) Erlkönig (E) Harfenspieler (H) Der Schatzgräber (SCH)

Unreine Reime undphonologische

Theorie

295

Die Braut von Korinth (BK) Zueignung (Z) Sonette (SO) Zahme Xenien II (ZX) Westöstlicher Diwan (WD die ersten 103 Reimpaare) Friedrich Schiller (Württemberg, 1759-1805): Der Abend (A) Elegie auf den Tod eines Jünglings (ET) An die Freude (F) Sängers Abschied (SA) Der Taucher (T) Die Bürgschaft (B) Das Lied von der Glocke (LG) Heinrich Heine (Düsseldorf, 1797-1856): Junge Leiden (JL) Die Heimkehr (HK) Neuer Frühling (NF) Romanzen (RO) Deutschland Ein Wintermärchen (DW) Romanzero - Historien (HS) Gedichte 1853-1854 (GD) Wilhelm Busch (Niedersachsen, 1832-1908): Das Glöcklein im Walde (GW) Romanze von einem Maler und einem Mägdelein (RO) Liebestreu und Grausamkeit (LG) Max und Moritz (MM)

Zitierte wissenschaftliche

Literatur

Adelung, Johann Christoph 1782 Umständliches Lehrgebäude der Deutschen Sprache zur Erläuterung der Deutschen Sprachlehre für Schulen. 2 Bde. Leipzig: Breitkopf. 1790 Vollständige Anweisung zur Deutschen Orthographie. 2. Aufl. Leipzig: Weygandsche Buchhandlung. 1. Aufl. [1788]. Auer, Peter 1991 Zur More in der Phonologie. In: Karl Heinz Ramers und Richard Wiese (eds.), Prosodische Phonologie, 3-36. Göttingen: Vandenhoeck und Ruprecht. Berg, Thomas 1990 Unreine Reime als Evidenz für die Organisation phonologischer Merkmale. Zeitschrift für Sprachwissenschaft 9: 3-27.

296

Beatrice Primus

Besch, Werner et al. (eds.) 1983 Dialektologie. Ein Handbuch zur deutschen und allgemeinen Dialektforschung. 2. Bd. Berlin: de Gruyter. Chomsky, Noam und Morris Halle 1968 The sound pattern of English. New York: Harper und Row. Clements, George N. 1985 The geometry of phonological features. Phonology Yearbook 2: 225252. Duden 1962 Aussprachewörterbuch. Wörterbuch der deutschen Standardaussprache. Bearbeitet von Max Mangold. 1. Auflage. Mannheim/Wien/Zürich: Dudenverlag. 1990 Aussprachewörterbuch. Wörterbuch der deutschen Standardaussprache. Bearbeitet von Max Mangold in Zusammenarbeit mit der Dudenredaktion. 3. Aufl. Mannheim/Wien/Zürich: Dudenverlag. Ezawa, Kennosuke 1972 Die Opposition stimmhafter und stimmloser Verschlußlaute im Deutschen. Tübingen: Narr. Gottsched, Johann Christoph 1762 Deutsche Sprachkunst. 5. verb. Auflage. Leipzig: Breitkopf. Grubmüller, Klaus 1998 Sprache und ihre Verschriftung in der Geschichte des Deutschen. In: Werner Besch, Anne Betten, Oskar Reichmann und Stefan Sonderegger (eds.), Sprachgeschichte. Ein Handbuch zur Geschichte der deutschen Sprache und ihrer Erforschung. 1. Teilband. 2. neu bearb. Auflage, 300-310. Berlin/New York: Mouton de Gruyter. Kloeke, Wus van Lessen 1982 Deutsche Phonologie und Morphologie: Merkmale und Markiertheit. Tübingen: Niemeyer. Moulton, William G. 1961 Zur Geschichte des deutschen Vokalsystems. Beiträge zur Geschichte der deutschen Sprache und Literatur 83: 1-35. Neef, Martin 1996 Wortdesign. Eine deklarative Analyse der deutschen Verbßexion. Tübingen: Stauffenburg. 1997 Die Alternationsbedingung: Eine deklarative Neubetrachtung. In: Christa Dürscheid, Karl-Heinz Ramers und Monika Schwarz (eds.), Sprache im Fokus. Festschrift für Heinz Vater zum 65. Geburtstag, 17-31. Tübingen: Niemeyer. Neugebauer, Moritz 2001 Laryngale Merkmale im Deutschen. Merkmalsgeometrie und prosodische Lizensierung in einem constraint-basierten Grammatikmodell. Magisterarbeit, Institut fur Deutsche Sprache und Literatur, Universität zu Köln.

Unreine Reime undphonologische

Theorie

297

Neumann, Friedrich 1920 Geschichte des neuhochdeutschen Reims von Opitz bis Wieland. Studien zur Lautgeschichte der nhd. Gemeinsprache. Berlin: Weidmannsche Buchhandlung. Opitz, Martin 1624 Buch von der deutschen Poeterey. Hrsg. von W. Braune. Tübingen 1955. Ohala, John J. 1986 Consumer's guide to evidence in phonology. Phonology Yearbook 3: 3-26. Ramers, Karl-Heinz 1988 Vokalquantität und -qualität im Deutschen. Tübingen: Niemeyer. 1998 Einführung in die Phonologie. Fink: München. Restle, David 1998 Silbenschnitt - Quantität - Kopplung. Zur Geschichte, typologischen Einordnung und Repräsentation der Anschlußprosodie unter dem Blickwinkel einer Oszillationssilbentheorie. Dissertation, Universität München. Ryder, Frank G. 1963 Off-rhymes and consonantal confusion groups. Lingua 12: 190-198. Schiller, Karl Martin 1963 Vorwort. In: Willy Steputat, Reimlexikon. Stuttgart: Reclam. Schirmunski, Viktor M. 1962 Deutsche Mundartenkunde. Berlin: Akademie-Verlag. Siebs, Theodor 1898 Deutsche Bühnenaussprache. Berlin/Köln/Leipzig: Ahn. Simmler, Franz 1983 Konsonatenschwächung in den deutschen Dialekten. In: Besch et al. (eds.): 1121-1129. Spiekermann, Helmut 2000 Silbenschnitt in deutschen Dialekten. Tübingen: Niemeyer. Ternes, Elmar 1987 Einführung in die Phonologie. Darmstadt: Wissenschaftliche Buchgesellschaft. Vennemann, Theo 1982 Zur Silbenstruktur der deutschen Standardsprache. In: Theo Vennemann (ed.), Silben, Segmente, Akzente, 261-305. Tübingen: Niemeyer. 1991 Syllable structure and syllable cut prosodies in Modern Standard German. In: Pier Marco Bertinetto et al. (eds.), Certamen Phonologicum II: Papers from the Cortona Phonology Meeting 1990, 211243. Turin: Rosenberg und Sellien

298 1995

Beatrice Primus

Der Zusammenbruch der Quantität im Spätmittelalter und sein Einfluß auf die Metrik. Amsterdamer Beiträge zur älteren Germanistik 42: 185-225. Wagenknecht, Christian 1989 Deutsche Metrik. Eine historische Einfiihrung. München: Beck. Wiese, Richard 20002 The phonology of German. Oxford: Oxford University Press. Wiesinger, Peter 1970 Phonetisch-phonologische Untersuchungen zur Vokalentwicklung in den deutschen Dialekten. 1-2. Berlin: Walter de Gruyter. 1983a Phonologische Vokalsysteme deutscher Dialekte. Ein synchronischer und diachronischer Überblick. In: Besch et al. (eds.): 1042-1076. 1983b Dehnung und Kürzung in den deutschen Dialekten. In: Besch et al. (eds.): 1088-1101. 1983c Rundung und Entrundung, Palatalisierung und Entpalatalisierung, Velarisierung und Entvelarisierung in den deutschen Dialekten. In: Besch et al. (eds.): 1101-1105.

Choctaw intensives and syllable theory David Restle

1. Introduction In Choctaw, a Muskogean language of North America, a complex type of ablaut is used to form intensives. Starting in 1991, several analyses were brought into discussion, which describe the intensives in the framework of a theory known under the term Prosodic Morphology (cf. McCarthy and Prince 1995, 1998).1 The so-called >-grade" of Choctaw and the broken plural of Arabic have been presented as classic examples suited to illustrate the advantages of this theory which aims to give elegant and insightful descriptions of instances of morphological operations where phonology impinges on morphology, especially reduplication, ablaut and infixation. This article does not provide new data on the topic but merely presents a reanalysis of the intensives of Choctaw on the basis of the syllable theory developed in Restle (1998). It shows, from the viewpoint of theoretical phonology, that the use of this different syllable model is suited to reduce the complexity of previous analyses of the _y-grade and similar phenomena. After a brief presentation of the most important data (2.), the previous analyses are outlined in section 3. Based on the theory of Syllabic Oscillation, whose main tenets are discussed in 4.1, a reanalysis is proposed in 4.2, while section 5 compares the result with the other "battlefield" of Prosodic Morphology, the analysis of the broken plural in Arabic.

2. Choctaw intensives Choctaw, as other Muskogean languages, makes use of several verb grades derived from verbal bases by various types of modifications, as

300

David Restle

Munro (1985: 583f) illustrates with examples from Chickasaw, a language closely related to Choctaw: (1) Chickasaw verb grades2

///V-grade TV-grade F-grade G-grade H- grade

kosoma fatabli 'to be smelly' 'to have dandruff / kosohöma fatahambli fatämbli kosöma fatay:abli kosoy:o?ma fattabli kos:o?ma fatdhiabli kosoh:o?ma

oklhili 'to be dark, night' oklhihili /

s

/

oklhili oklihiy:i?li oklhi?li oklhih:i?li

Γ-grade and g-grade are collectively referred to as intensives by Nicklas (1974, 1975) and Ulrich (1986, 1994). The term > g r a d e " is used by Nicklas (1975), Lombardi and McCarthy (1991), and Hammond (1993) as a cover-term for intensives (including those called "g-grade" by Munro 1985 and Ulrich 1994). In Chickasaw, y-grade and g-grade differ in meaning (cf. (2)), while such a distinction is neutralized in Choctaw (cf. (3)), examples from Ulrich 1994: 326). (2)

wasaca wäs:a?ca3 wasay:a?ca

'it's sour' 'it's kind of sour' (g-grade) 'it's really sour' (y-grade)

(3)

talowa-h polak tal:orwa-h poläk taloy:owa-h

'she sang' 'she finally sang' (g-grade) 'she finally sang' (y-grade)

Nicklas (1974, 1975) (and, following him, both Lombardi and McCarthy 1991 as well as Hammond 1994) treats these grades as variants of the same form. Lombardi and McCarthy and Hammond seem to presuppose that j-grade and g-grade are in something like a complementary distribution (cf. (4)). Contrary to that position, Ulrich (1994) shows that both forms are attested at least for some bases (cf. (5)).

Choctaw intensives and syllable theory

(4)

Only one grade (y-grade), Lombardi and McCarthy (1991: 47f) Base intensive form LL ona 'to arrive there' oy:o:na täyiakci HL takci 'to tie' kobiaifa LLL kobafa 'to break' LHL täliakci talakci 'to be tied' toksäy:a:li HLL toksali 'to work' HHL oktäyiabli oktabli 'to dam up'

(5)

G-grade and y-grade Base LL ona HL takci LLL kobafa LHL talakci HLL toksali HHL oktabli

301

differentiated, Ulrich (1994: 326) y-grade g-grade 4 oyioina 'to arrive there' ό :na täyiakci 'to tie' täikci kobäyiaifa kobiaifa 'to break' täliakci taläyiakci 'to be tied' 'to work' toksaili toksäyiaili oktäibli 'to dam up' oktäyiabli

There is good reason to follow Ulrich in distinguishing the two grades even in Choctaw. Firstly, this differentiation allows for better compatibility of analysis with the data from Chickasaw. Secondly, and decisively, we encounter problems if we try to derive the g-grade forms from their corresponding y-grade forms (Vy/V > V by coalescence), as Lombardi and McCarthy (1991: 68 Fn. 8) propose for bases with a heavy or no antepenult. 5

3. Previous analyses The first formal analysis of the intensives was carried out by from Lombardi and McCarthy and is founded upon the following observations regarding the form of data presented in (4), which lacks a distinction between y-grade and g-grade: (6) a. All intensives have a heavy penult whose onset consonant is geminated (i.e., all antepenults are heavy too). Nuclei of light penults are lengthened to engender heaviness.

302

David Restle

b. Forms with infixed (V)y/V (only) arise if the antepenult of the base is heavy or missing. c. The position of the infix is constant; it appears after the nucleus of the penult of the base. Lombardi and McCarthy make heavy use of the theory of prosodic circumscription, which formalizes the intuition "that a morphological operation can apply to a prosodically delimited constituent within a morphological base rather than to the morphologically delimited base as a whole" (1991: 38). The authors derive the intensives of Choctaw by the sequence of rules in (7) (Lombardi and McCarthy 1991: 50), which are illustrated in (8). The derivation starts by fading-out the last syllable of the base qua extrametricality and describing an iambic foot (light-heavy) at the right edge of the remaining portion of the base. Sequences of the type σ σ are stripped-down to σ, while σ σ passes unchanged. If the input to rule 2 (cf. (7)) only consists of one syllable, a degenerate foot is built. The following "edge-in" association of the result of rule 2 stipulates that the iambic template used in 3 is filled by aligning the edges of template and input. Then, a default onset is provided for the second syllable of the template in order to avoid a naked syllable. The last derivational rule infixes a mora, which is lexically linked to a high tone, after the first mora of the output of rule 4. Strictly speaking, this medial gemination has to be seen as a combination of making the first syllable of the iambic template extra-prosodic and then prefixing a mora to the remaining base. (7)

Lombardi and McCarthy's (1991: 328) derivation of ^-grades 1. Final Syllable Extraprosodicity

σ-(σ)/_] 2. Prosodic Base Circumscription Φ (Iamb, Right) 3. Template Mapping 'Edge-in' association to iambic template 4. Default Onset Rule 0 —• y, when required by syllabic well-formedness 5. First Mora Extraprosodicity6 6. Mora Pre fixation

Choctaw intensives and syllable theory

303

(8) ona

takci

kobafa

talakci

toksali

oktabli

1.

ο (na)

tak(ci)

koba (fa)

talak (ct)

toksa(7z)

oktab(7/j

2.

ο (na)

tak(ci)

koba (fa)

talak (ci)

(tok)sa(li)

(ok)tab(li)

F

F

F σ

σ

I

κ

μ

μμ

ο

σ

σ

σ

AJT

t

a

k

ί

σ

Φ

koba

σ

F σ

σ

F σ

σ

σ

/F/FR ΑΑ AA

talak

s

a

t

a

b

F σ

σ

ψο F σ

σ

W ο oy:o.na

täy:akci

t al ak tähakci

oktäy:abli

The rather simple observations made at the beginning of this section (cf. (6)) do not support Lombardi and McCarthy's (1991: 48) intuition that the "rich array of regularities turns out to have a fairly straightforward interpretation within prosodic theory, once the requisite parameters of the analysis are recognized." On the contrary, for those not familiar with the complexities that are introduced by prosodic circumscription, the question may arise: What is achieved by the formalism in (7) in relation to the relatively easy description given in (6)? Yet, no competitive framework seems to be available, and both Hammond (1993) and Ulrich (1994) use the same method of prosodic circumscription. Hammond's analysis (1993) leaves the data basis and rules 1, 2, and 4 of the derivation untouched, but uses a heavy trochee σ σ instead of an iamb for template mapping. This change avoids problems resulting from Lombardi and McCarthy's mora prefixation but introduces a foot type not well-attested elsewhere. 7

304

David

Restle

As mentioned, Ulrich (1994) distinguishes the two intensive forms and therefore has to provide for two sets of rules. (9)

Ulrich's (1994: 328) derivation ofy-grades 1. Final Syllable Extraprosodicity

σ-(σ)/_] 2. Prosodic Base Circumscription Φ (σ, right) 3. Template Mapping 'Edge-in' association to heavy trochaic template 4. Default Onset Rule 0 —• y, when required by syllabic well-formedness 5. >>-Spreading Spread y to the mora to the left 6. Vowel Spreading Spread vowels to unfilled moras

(10) Ulrich's (1994: 331) derivation ofg-grades 1. Final Syllable Extraprosodicity

σ^(σ)/_] 2. Prosodic Base Circumscription Φ (iamb, right) 3. Template Mapping 'Edge-in' association to heavy trochaic template 4. Onset Filling Spread from the left to fill the empty onset 6. Contraction Contract feed with no medial onset

The analysis proposed by Ulrich has the advantage that both forms of intensives are derived with equal complexity and that the analysis can also be easily applied to the Chickasaw data. All three analyses depend on the mechanism of prosodic circumscription, a device whose enormous complexity is justified solely by a lack of simpler theories that would be able to deal with the data of ablaut systems and reduplication. Its main potential stems from its microsurgical character, namely the capability to cascade negative and/or positive descriptions of portions of a base in combination with rule ordering. Another problem of all three analyses is the type of default onset creation rules used. If we create an intermediate iambic foot without any onset consonants (cf. (8), step 3 first column), we wonder why an onset

Choctaw intensives and syllable theory

305

consonant is provided only for the second foot (i.e. o.o > o.yoo) and not for the first one (i.e. o.o: > yo.yo:, which would create an intensive form *yoy:o:na instead of oy:o:na). Both Lombardi and McCarthy (1991: 47) and Ulrich (1994: 329) treat this lack of default onset creation by stipulation. The main objection against the proposed analyses arises from the fact that the basically iconic character of the prosodic extension of the base to signal intensifying meaning becomes totally opaque. The j-grade evidences infixation of (V)y/V, but no such infix is found in their rules. Instead, a process of complex remapping (which is called "edge-in association") from the iambic material factored-out to some type of foot (iambus or heavy trochee) is called for. This move leads technically to correct output forms, but not to intuitively fathomable derivations. Therefore, observation (6) b) is not adequately represented in the proposed rule sets. Observation (6) a), namely that both antepenult and penult is heavy in the output form, has only been adequately represented in the analyses of Hammond's and Ulrich's (by template mapping to a heavy trochee). In Lombardi and McCarthy's analysis, however, the lengthening of a penultimate vowel is addressed by mapping to an iambic template, whose last syllable has to be heavy obligatorily, while the lengthening of an input consonant, or a default consonant (y), is addressed by prefixation (sic) of a mora at a different stage of derivation. Finally, the fixed position of the infix (V)yyV (cf. observation (6) c)) is an accidental outcome of rule ordering and the type of template mapping ("edge-in"). All in all, while the proposed analyses are able to generate the correct output, they are not really suited to depict obvious facts about the data adequately and elegantly.

4. Reanalysis The reanalysis proposed in this article is based upon the use of the syllable theory developed in Restle (1998), which is briefly sketched below.

306

David Restle

4.1. The theory of Syllabic Oscillation The model rests on two primitive prosodic categories, the opening and closing movement, which correspond phonetically to opening and closing of the vocal tract. Neither the single segment nor the classical syllable receives a comparable primitive status. • > * · ο (11) Primitives (incomplete) opening

closing

The concept of the phonological segment plays a role only as a qualitative property of primitives, namely as vocal tract configurations at the start, the transition, and the end of opening and closing. They are represented as diacritics on opening and closing, as can be seen for ona 'to arrive there': (12) ona ηη More complex openings and closings, which are extended by transitional configurations, are represented analogously: (13) German Kamm m

German Qualm m

These primitives are combined by basic relations, whose number is language-specific. For Standard German, three relations are necessary (loose coupling, tight coupling and amalgamating coupling).9 These are given along with their phonetic correlates:

Choctaw intensives and syllable theory

307

(14) Basic relations in German loose coupling

tight coupling

amalgamating coupling

V unlengthenable V shorter V more centralized

V unlengthenable V short or absent V central

/w /w

Between opening and closing Phonetic V lengthenable correlates V longer V peripheral 10 Representation Between closing and opening Phonetic C lengthenable correlates (C longer) 11 C strong articulation Representation At word edges Phonetic correlates

C unlengthenable (C shorter) C weak articulation

12

V lengthenable V longer V peripheral

V unlengthenable V shorter V more centralized

C lengthenable (C longer) C peripheral

C unlengthenable (C shorter) C weak articulation

V unlengthenable V short or absent V central

Representation Phonetic correlates Representation

Oscillatory syllables emerge out of sequences of opening plus closing. Word and phrase-final closings and word and phrase-initial openings render defective oscillatory syllables. In the theory of Syllabic Oscillation, the traditional distinction made between open and closed syllables is reanalyzed. The structures CV.CV, CVC.CV, and CV.CCV all have a closing phase after the articulation of the first vowel, cf. (15) for German examples.

308

David Restle

(15) Miete

Mitte

Mitra

A contrast between tauto- and heterosyllabicity of consonants is interpreted as a contrast of consonantal coupling. In order to accomplish this, we have to leave the implicit null hypothesis aside that the end configuration of the closing must be the same as the start configuration of the opening. This step is borne-out by the observation in languages like German, which have a maximal binary length contrast, that the final consonant of a closed syllable can be lengthened in the same way as ambisyllabic consonants. (16) fallen

falten

This parallel analysis of syllables closed by ambisyllabicity and those closed by the first consonant of a heterosyllabic cluster is further supported by the parallel behavior of both types of syllables in syllable weight assignment: both types are treated as heavy in stress assignment rules. For a detailed discussion about the theory of Syllabic Oscillation, see Restle (1998).

4.2. The new picture If we utilize of the representation provided by the syllable model presented in the previous section, a new picture arises. We follow Ulrich (1994) in separating j-grades from g-grades.13 The relevant data that has to be covered ((5) above) is repeated in (17).

Choctaw intensives and syllable theory

309

(17) Base LL HL LLL LHL HLL HHL

'to 'to 'to 'to 'to 'to

ona takci kobafa talakci toksali oktabli

arrive there' tie' break' be tied' work' dam up'

g- grade

>>-grade

oma ta:kci kob;a:fa taliakci toksä: li okta\ bli

oy:o:na täy.akci kobäy:a:fa talay/akci toksäy\ a: li oktäy\abli

4.2.1. G-grade formation We start the analysis with the description of the processes involved in creating g-grade forms. There are two types of output forms, those with a lengthened consonant and those without such a lengthening, which in turn have a falling tone in the resulting penultimate syllable. Representations of the first group of forms are given in (18).

(18) kobafa

kob.a:fa k

f k ' bb a. a : ο ο

ά

b

1 1

asa

a ,/a

f 4

ä

talakci V ι

k a, ä

b

O .0

talakci t

/ w

c

t

,v

ι

1 ΛΛ.

.

k a

a

w

v

L·

In these cases, we find a single consonant configuration in the domain of the closing of the antepenult and of the opening of the penult (b and / respectively). Since the coupling between opening and closing is tight, it can be changed to loose coupling, with this modification being transparent. Obviously, the aim of intensive formation is to lengthen the base without inserting distinctive segmental material. Therefore, the means to be used in accomplishing this task are likely to be restricted to prosodic elongation and infixation of non-distinctive material. On the other hand, this purely prosodic prolongation should be transparent in order to fulfill

310

David Restle

its function. When the coupling between antepenultimate closing and penultimate opening is already loose in the base, the loose coupling alone is not capable of signaling a g-grade. But, as the penultimate intravocalic coupling (between opening and closing) is affected by g-grade formation only in those cases where the penult would otherwise be light and as the coupling between the closing of the penult and the opening of the ultima is never affected, there must be other means to unambiguously mark g-grade-formation. It is exactly in these cases, where an onset to the penult can not be lengthened distinctively (due to a preexisting loose coupling realized by consonant gemination or heterosyllabicity of a cluster), that a distinctive falling tone is introduced and the penultimate vowel is lengthened. In contrast to the forms of g-grade with geminated consonants, where only the vowels of open penults are lengthened, all forms with falling tone have long penultimate vowel, thus possibly creating super-heavy syllables; cf. (19)-(20). (19)

toksali

toksä'Ai

HL

Disyllabic bases like tahli and ona do not have a suitable consonant for gemination because word-initial geminates are prohibited, and ona does not show any penultimate opening that could be coupled loosely with a closing on its left side.

Choctaw intensives and syllable theory

311

(20) tahli

ta:hli

HL

ona

b:na

HL

In summary, g-grades can be derived by a single rule: (21) G-grade formation: Change the coupling of a penultimate opening and its left neighbor to loose and then separate it from a following opening by a further loose coupling. If the left coupling of the penultimate opening is already loose in the input, mark the penult with a loose intra-vocalic coupling and a falling tone. The rule first tries to geminate the onset of the penult. If successful, the loose penult is further separated from the following material by loosening the coupling to the last syllable. This condition may already be fulfilled in the input (if the penult is heavy qua coda consonant), but if it is not, the first coupling after the penultimate opening is made loose. If unsuccessful, the alternative strategy is used in order to change the left coupling of the penult opening.

4.2.2. F-grade formation Taking the analysis proposed here into consideration, the strategy of building g-grades rests on the modification of existing prosodic categories, namely changes from tight to loose coupling. It is iconic in prolonging pauses between the proposed phonological primitives, and this corresponds phonetically to a slower succession of articulatory movements.

312

David Restle

The other logically possible iconic means of slowing down the articulation of phonologically distinctive primitives is to insert non-distinctive material, i.e. infixation. This seems to be so with the ^-grade. In all cases, a complete oscillatory movement, i.e. a closing of the vocal tract immediately followed by a reopening, is infixed at the nucleus of the penult, providing for the configuration at which the infixed oscillation begins and ends (the consonantal configuration is characterized by the default^). Parallel to the g-grade, the infix is separated from the last opening by a loose coupling, the effect of which is to obligatorily make the output penult heavy; cf. (23). (22)

7-grade formation: Infix an additional syllabic oscillation (closing and opening) at the position of the nucleus of the base penult and make sure that it is separated from the following opening by a loose coupling.

(23)

kobafa k

bb

kobay.a:fa f

a a

0 0

mr \

a,

t talakci

talay.akci 1

«

11

t toksali t a a

t oktabli

t

y

toksay:a:la 71

ο ο

1 St

a a

t

k ό

oktay:abli

Choctaw intensives and syllable theory

tahli

313

tayiahli

In this section, we have tried to show that the syllable model proposed in Restle (1998) provides for representations of the Choctaw forms which lead to a straight-forward analysis of the verb grades collectively known as intensives. We will now compare the analysis proposed with the well-known problem of broken plurals in Arabic.

5. The broken-plural problem in Arabic One may wonder if parallels for the infixation of an additional unspecific oscillation, as proposed in the previous section for deriving ygrades, are found in other languages as well. In Restle (1999), a similar method is proposed for the derivation of broken plurals in Arabic. In Arabic, there are two ways of forming plurals, namely a) affixation (the so-called pluralis sanus or external plural), and b) internal modification of the singular form {internal or broken plural). McCarthy and Prince (1990) have demonstrated that the vast majority of nouns build internal plurals, i.e. broken plurals, this part of the lexicon itself being dominated by one special type of broken plural, namely the iambic broken plural. The term stems from the initial iambic sequence CVCVV shared by this subgroup. The generalizations made so far concentrate on this subgroup of broken plurals, which is exemplified in (24)-(28) (examples taken from McCarthy and Prince 1990: 217).

314

David Restle

(24) CVCV:C jazi:r + at kari:m + at

jaza:?ir kara:?im

'island' 'noble'

(25) CV:CVC xa:tam ja:mu:s

xawaitim jawa:mi:s

'signet-ring' 'buffalo'

(26) CVCCV(:)C jundub sulfam

janaidib sala:t?im

'locust' 'sultan'

(27) CVCC nafs qidh

nufu:s qida:h

'soul' 'arrow'

?usu;d /?ana:b/

'lion' 'grape'

(28)

cvcvc ?asad ?inab

[?a?naab]

These plurals have the following properties in common: 1. In most cases the vocalism of the last three syllables is /a-a-i/. The exceptions, e.g. the /i-a/ and /u-u/ in (27), appear to be unpredictable and therefore lexical. 2. The vowel of the first syllable of the plural form is always short, that of the second syllable always long. 3. The vowel length of the third syllable of the plural form depends upon the situation in the singular. This may be seen in (25), where short /a/ in xaatam appears as a short /i/ in xawaatim. The long /u/ in jaamuus corresponds to a long /i/ in jawaamiis. 4. Both the 111 in (24) and the /w/ in (25) are default consonants. The position of these consonants seems to depend upon the position of the long vowel in the singular form. In his earlier work, McCarthy's (1979, 1983) commences with the idea of infixation of some structure after the first syllable in the singular form. In the work of 1990, collaborated with Alan Prince, this idea is no longer pursued. Rather, they use the device of Prosodic Circumscription: the first two moras are factored-out and mapped to an iambic template. Restle (1999) proposes a different means which parallels that of deriving y-grades.

Choctaw intensives and syllable theory

315

(29) Mechanism of deriving broken plurals from singular forms: 1. Insertion: Insert an additional oscillation (opening + closing or closing + opening) into the position of the first loose coupling (if none is available, do nothing). 2. Specification: a. Make the coupling inside the second syllable (i.e. the coupling between the second opening and closing phase) loose. b. Change the target position of the first three opening phases to /a-a-i/. The main difference between the broken plural and the y-grade is the position at which the infix appears relative to the input: It is fixed in Choctaw (before the closing of the penult) and variable in Arabic (at the first loose coupling), which leads to a greater diversity of the various output forms in the latter case. This major difference remains neglected in the derivations proposed by McCarthy and Lombardi and by McCarthy. (30) ja:mu:s j

jawa:mi:s

mm a

t

a janaidib

jundub J

d

η

b

&

t / r a a

?

Jm

§v Λ Τ ?inab ηη

Φ

Φ

k

/ r a

l ^ - u , ?ina:b

b

ίγί

k

/b I

kara:?im

kari:m(at) ik\

d

Y\ A is

n

η\

7 b «ΧΛΛ a

m

316

David Restle

6. Conclusion From the perspective of theoretical phonology, I have intended to provide a view of Choctaw intensives that seems to be rather different from the known analyses. Nevertheless, the use of the new syllable model allows for descriptions of ablaut systems and infixation close to those intuitions or observations obtainable by approaching these data from a theory-independent view. Both the intensives of Choctaw and the broken plural of Arabic can be seen as in part iconic because the phonological mechanisms used to lengthen forms seem to mirror their respective semantic functions (increased intensity, increased number).14 It is very inspiring that the other main area of current research in the domain of prosodic phonology, namely reduplication, shows a similar preference for iconicity15 and that it can be handled by the model introduced here in the same way (cf. Restle 1999).

Notes 1. I wish to thank Gaby Hollmann and Thomas Becker for helpful comments concerning this paper. 2. The semantic function of the above verb grades is given in Munro (1985: 588f., fn. 9). 3. Munro (1985: 581): "In Choctaw studies, the high pitch accent is referred to simply as the 'accent' (Nicklas 1972: 12); there is at most one accent per word, and many words have no accent. In Choctaw verbal forms, the accent is placed morphologically, on the penultimate vowel of deverbal nouns and nominal modifiers, and on the penultimate or (in specifiable cases) antepenultimate vowel of certain aspectual formations or 'grades'." 4. V refers to a falling tone distinctive from the high tone (V). 5. We would need two rules for such a derivation, one for light antepenults and one for heavy or missing antepenults. Furthermore, as Ulrich (1994: 333) points out, "... g e m i n a t e ^ optionally has an assimilatory effect on a preceding back vowel: ayya is often pronounced as eyya or iyya. [...] G-grades of bases with heavy or no antepenults, on the other hand, never exhibit fronting or raising of the accented vowel." 6. Rules 5 and 6 are presented in combination by Lombardi and McCarthy (1991: 50) as "Medial Gemination/Accentuation" and explained on page 54: "The rule of medial gemination applies to the canonical iamb obtained by template mapping (indicated below as Base 1). The gemination rule prefixes a mora to a

Choctaw intensives and syllable theory

7. 8. 9.

10.

base (indicated by Base 2 below) created by making the first mora of Base 1 extraprosodic." Hammond (1993: 325) tries to refute this objection. This set of primitives is extended in Restle (1998) for appendices und hiatus configurations. For languages of the Estonian type, three grades of coupling are needed, cf. sada 'hundred', sa:da 'send!', sa::da 'become', kat:us 'roof, and katr.us 'covered himself. Additionally, in order to elucidate the degrees of coupling, loose couplings are represented by horizontal separation where possible. Loose coupling: tight coupling:

11.

12.

13. 14. 15.

317

-

γ

or

,

or

,

amalgamating coupling: Δ or . This is, however, only possible, if both primitives to be coupled appear in the representation. Not all of the possible phonetic correlates of a coupling degree have to be realized at the same time. In German, for example, consonant length after accented short vowels is no longer realized. Nevertheless, the contrast in readiness to lengthen is still active (consonants after short vowels can be lengthened without semantic effect; those after long vowels can not). A short accented vowel must be followed by a consonantal element in the same syllable, i.e. by a default consonant (glottal stop) or by the onset consonant of a following word that becomes ambisyllabic. The combination of both grades ä la Lombardi and McCarthy (1991), cf. (4), would not complicate the proposed analysis. I am very grateful to Theo Vennemann for emphasizing this point in a previous presentation. Cf. Moravcsik (1978: 330): "A tendency has been noted for languages to use reduplicative patterns - i.e. quantitative form differentiation - for the expression of meanings that have something to do with the quantity of referents. This kind of onomatopoetic use of a form device also appears to have parallels outside the domain of reduplicative constructions."

References Haas, Mary 1940 Ablaut and its function in Muskogee. Language 16: 141-150. Hammond, Michael 1993 Heavy trochees in Choctaw morphology. Phonology 10: 325-336.

318

David Restle

Lombardi, Linda and John McCarthy 1991 Prosodic circumscription in Choctaw morphology. Phonology 8: 3771. McCarthy, John J. 1979 Formal problems in Semitic phonology and morphology. Ph.D. Dissertation, MIT. 1983 A prosodic analysis of Arabic broken plurals. In: Ivan R. Dihoff (ed.): Current approaches to African Linguistics. Volume I, 289-320. Dordrecht: Foris. McCarthy, John J. and Alan S. Prince 1990 Foot and word in prosodic morphology: The Arabic broken plurals. Natural Language and Linguistic Theory 8: 209-283. 1995 Prosodic Morphology. In: John A. Goldsmith (ed.), The Handbook of Phonological Theory, 318-366. Oxford: Blackwell. 1998 Prosodic Morphology. In: Andrew Spencer and Arnold M. Zwicky (eds.), The Handbook of Morphology, 283-305. Oxford: Blackwell. Moravcsik, Edith A. 1978 Reduplicative constructions. In: Joseph H. Greenberg (ed.), Universals of human language. Volume 3: Word structure, 297-334. Stanford: Stanford University Press. Munro, Pamela 1985 Chicasaw accent and verb grades. In: Ursula Pieper and Gerhard Stickel (eds.): Studia linguistica diachronica et synchronica. Werner Winter sexagenario, 581-591. Berlin/New York: Mouton de Gruyter. Nicklas, T. Dale 1974 The elements of Choctaw. Ph.D. Dissertation, University of Michigan. 1975 Choctaw morphophonemics. In: James M. Crawford (ed.), Studies in Southeastern Indian languages, 287-250. Athens: University of Georgia Press. Ratcliffe, Robert R. 1998 The "broken" plural problem in Arabic and comparative Semitic. Allomorphy and analogy in non-concatenative morphology. Amsterdam/Philadelphia: John Benjamins. Restle, David 1998 Silbenschnitt - Quantität - Kopplung. Zur Geschichte, Charakterisierung und Repräsentation der Anschlußprosodie unter dem Blickwinkel einer Oszillationssilbentheorie. Ph.D. Dissertation, Universität München. 1999 Reduplication as pure constituent copying. In defence of the syllable against the phonemic melody. In: John Rennison and Klaus Kühnhammer (eds.): Phonologica 1996. Proceedings of the 8th International Phonology Meeting, 195-217. The Hague: Thesus.

Choctaw intensives and syllable theory

319

Ulrich, Charles H. 1986 Choctaw morphophonology. PhD. Dissertation, University of California, Los Angeles. 1994 A unified account of Choctaw intensives. Phonology 11: 325-339.

Imperatives: the relation between meaning and form Renate Raffelsiefen

To evaluate claims of non-arbitrary relationships between meaning and form in morphology it is necessary to understand how complex expressions are formed. Specifically, to determine whether a certain category is encoded by an affix, "zero morph", stem modification, or subtraction it is necessary to compare the expression of a derived form with that of the base. The much neglected but non-trivial prerequisite is proper identification of the base. Consider the rule for Latin in (1) (cf. Matthews 19912: 194): (1)

Present Infinitive Active [X]

-

First Singular Imperfect Subjunctive [Xm]

According to the rule in (1) the first singular imperfect subjunctive is formed by adding -m to the surface form of the present infinitive active. That is, the word form in the left-hand column in (1) is identified as the phonological, but not the morphosyntactic or semantic, base of the first singular imperfect subjunctive. The rule is illustrated in (2): (2)

vo'care mo'nere

vo'carem mo'nerem

Ί would call' Ί would admonish'

'esse

'mem

Ί were'

Matthews (1991: 195), who cites the Latin grammarian Priscian as an inspiration for positing the type of rule (1), comments as follows:1 "On the semantic plane, the transformation cuts across the paradigm. But it is justified because the rule is formally both simple and absolute."

322

Renate Raffelsiefen

While most linguists would presumably be willing to grant that the rule in (1) has didactic value as a mnemonic device, few would be willing to posit such rules in morphological descriptions.2 However, although it is conceivable that the effect described by the rule in (1) is accidental for words derived by regular morphological rules, it is implausible to attribute to pure chance the applicability of the rule to irregular verbs like esse. This is because there is strong evidence that such forms are learned by rote memorization and stored as unanalyzed units (cf. Bybee 1985). Based on material from some middle and northern European languages, it will be shown below that once Priscian rules are allowed, the relation between meaning and form in imperative formation is more homogeneous than is commonly assumed.

1. The typology of imperative formation So-called imperative paradigms are typically characterized by both formal and semantic non-uniformity. In Finnish there are distinct case markings for the direct object in second and third person as is shown in (3): (3)

Indicative: Syön omenan]AkkiImperative 2: 3:

Syö omena\ Akk π = nom ! (*Syö omenan) Syököön (vain) omenan]A\±

Ί eat the (whole) apple.' ' E a t the a p p l e ! '

i. 'Let him/her eat the (whole) apple.'

In Latvian, the alleged third person imperative also differs from the second person imperative in that object pronouns precede the verb. Semantically only second person imperatives express direct commands whereas the mood of the first or third person forms is often referred to as "optative" or "hortative". For these reasons, imperatives are here understood narrowly as semifinite categories which exhibit neither tense nor person contrasts (cf. Eisenberg 1998: 195).3 The restriction of imperatives to direct commands, known as "imperative proper", is essential for a study of the relation between meaning and form, as can be inferred from Greenberg (1966: 47): "Hortatives then, whether confined

Imperatives: the relation between meaning andform

323

to the first and/or third persons or including also a second person distinct from the imperative, show the characteristics of marked categories. On the other hand, imperatives proper often have zero expression, particularly in the singular." Somewhat mysteriously, Greenberg continues as follows: "In such cases, however, there is sometimes a difference of stress pattern or in other suprasegmentals so that the form is not in fact a 'pure stem' form." As will be shown, the effect described by Greenberg indicates the following common pattern of imperative formation, which describes the clipping of final segments from a surface word form. The rule is intended as a redundancy rule describing a counter-iconic relation: (4)

Word form [XS]

-

Imperative Singular [X]

Bybee (1985) does not mention clipping but instead states that imperatives are often formed by zero markers or suffixes, thereby arguing for non-iconic or iconic markings. Below I will argue that several cases, where zero markers or suffixes are commonly posited, turn out to represent the pattern in (4) when inspected more carefully. The assumption that imperatives tend to correspond directly to bare stems is sometimes explained as a consequence of their primary function (cf. Korhonen 1967: 164, 167).4 Evidence for clipped imperatives is significant then in that it shows that imperatives are derivative.

2. Norwegian, Danish In grammars of Norwegian Bokmäl, it is usually claimed that imperatives are based directly on stems (cf. Faarlund, Lie and Vannebo 1997). To evaluate this claim, consider the different phonological effects of stem derivation versus clipping illustrated with examples from Icelandic. Words derived from stems necessarily exhibit "repair" to match the regular phonological patterns in the language. For example, the bare stem /akr/ undergoes epenthesis to satisfy the constraint SON in (6), which refers to the sonority hierarchy in (7). Epenthesis does not apply if a vowel-initial suffix follows because the cluster Ar violates SON only in coda position.

324

Renate Raffelsiefen

(5)

/akr/ σ σ I a: k y r

(6)

SON Sonority increases in the syllable head and decreases in the coda.

(7)

Sonority hierarchy low

stop

•

fricative

nasal

1

r

high

vowel

Compare the relation between the abstract stem /akr/ and the surface form [a:kyr] with the examples of clipping in (8):5 (8)

[pYikra] pukra [klifira] klifra

'to whisper' 'to climb'

[pYikr] pukr [klif:r] klifr

'whisper' 'climbing'

Clippings are by definition based on surface forms. The violation of regular prosody exhibited by the clippings in (8) indicates the effect of segmental deletion without any concomitant restructuring of syllabic associations. That is, in the clipped forms, open syllable lengthening applies but epenthesis does not, as if the final consonant cluster were invariably associated with onset position. Crucially, there is no adjustment or repair.

(9)

pukra

pukr

Consider now the imperatives in (10) in southern and western Norwegian dialects (cf. Kristofferson 1991: 247): (10) [syk:b] sykle [o:png] opne [vo:kns] vokne

'to go by bicycle' 'to open' 'to wake up'

[syk:l]! [o:pn]! [vo:kn]!

Imperatives: the relation between meaning andform

325

The monosyllabic imperative [syk:l] violates regular phonotactics, thereby differing from the related noun [sykisl], which exhibits regular prosody. The irregular application of open syllable lengthening in [o:pn] or [vo:kn] also indicates the origin of the imperatives as clipped forms. Interestingly, in Oslo such imperatives are accepted only if followed by a vowel-initial word, but are judged ungrammatical and avoided otherwise (cf. Kristofferson 1991: 247) (11)

^syklut!

'go out by bicycle!'

*sykl noI

'go now by bicycle!'

The pattern in (11) makes sense because the combination with a vowel-initial word is the only type of "repair" available to a clipped form with final sonority-increase. Bokmäl imperatives accordingly fit the pattern in (4). Kristofferson (1991) also concludes that Bokmäl imperatives are formed by clipping based on surface word forms, but this finding goes unmentioned in the description in Faarlund, Lie and Vannebo (1997), which suggests non-iconic imperative formations. For Danish, Mikkelsen (1894: 209) claims that imperatives are based on stems, but he also observes that imperatives ending in a cluster with increasing sonority are combined with vowel-initial words by poets and paraphrased by the ordinary speaker.6 This type of prosodic sensitivity reeks of clipping. Eighty years later, the analysis of Danish imperatives as clipped surface forms has been firmly established by the evidence from stßd. Andersson (1975), who refers to the description in Basboll (1969), shows that the occurrence of sted in Danish imperatives cannot be explained on the basis of stem phonology but requires reference to longer surface forms. It is unfortunate that the phonological evidence has been ignored in Danish grammars. For example, in the comprehensive grammar of Danish by Allan, Holmes, and Lundskser-Nielsen (1995), we are again informed that Danish imperatives are based directly on stems and involve Zero marking. Norwegian Bokmäl and Danish are thus two languages which are generally claimed to have zero marking but for which careful phonological analysis indicates formation by clipping. Moreover, the special phonology resulting from clipping fits Greenberg's notion of "impure" stems.

326

Renate Raffelsiefen

3. Imperatives in German In Old High German, singular imperatives of strong verbs are generally monosyllabic whereas imperatives of weak verbs end in an unstressed vowel -e, -o or -i.1 For Middle High German, Paul, Wiehl, and Grosse (1989: 254) note that the /-e/ (i.e. the reduced vowel reflecting Old High German -e, -o, -i) fails to appear after short Irl and l\l preceded by a short stressed vowel even in weak verbs. There is evidence that this particular gap relates to the overall distribution of schwa in the respective paradigms as is illustrated in (13): Imp. SG 1 2 3 PL 1 2 3

a.

mach[s]! 'make!' mach[s] mach[3]st mach[3]t mach[^]n mach[s]t mach\$\nt

still[3]! 'nurse!' still[s] still[s]st still[d]t still[s]n still[3]t still[^]nt

b.

hoi! 'call!' hoi hoist holt holn holt holnt

ner! 'cure!' ner nerst nert nern nert nernt

Verbal paradigms in Middle High German are characterized by a uniform distribution of schwa. That is, schwa occurs either in all members of a paradigm or in none. The distinguishing trait of schwaparadigms is the inclusion of at least one member which requires schwa to satisfy constraints on syllabic wellformedness. Specifically, given that the most sonorant suffix in the verbal paradigm is -n, it follows that all stem-final consonants other than -/ and -r need the schwa to satisfy the constraint SON defined in (6) (cf. machen instead of *machn). The preference for uniform paradigms explains the occurrence of the schwa in the remaining members. Similarly, the relevance of both vocalic and consonantal length for the occurrence of schwa indicates that both types of segments still exhibit moraic contrast such that the presence of schwa is needed to satisfy a limit on maximal weight in stressed syllables. In Middle High German, the weight of a syllable with either a long vowel or a long consonant exceeds the weight of a syllable with a short vowel followed by a short consonant. Schwa is accordingly needed to break up heavy syllables such as * [mailt] (cf. the bimoraic syllable in the actual form [ma:, tat] 'paints') or *[stil:t] (cf. the bimoraic syllable in the actual form [stil.tat] 'nurses'), whereas lighter syllables

Imperatives: the relation between meaning andform

327

such as [holt] 'calls' or [nert] 'cures' do not exceed maximal syllable weight and are therefore schwaless. Assuming that the occurrence of schwa in (13) is indeed determined by constraints on syllabic wellformedness which inherently affect specific suffixed forms in the paradigm, the question arises of how to analyse the (secondary) effect on the imperative forms. One possibility is to include the imperative forms in the paradigm, with the result that they are affected directly by leveling. There are two reasons for rejecting such an analysis. First, it can be shown that Middle High German leveling in general affects narrowly defined paradigms where members must not differ in tense. Semifinite tenseless forms like imperatives, accordingly, do not belong to present tense paradigms. Second, the inclusion of imperative forms in paradigms would entail many exceptions for strong verbs to otherwise perfect leveling, as is illustrated in (14): (14) Imperative SG 1 2 3 PL 1 2 3

nim! 'take' nim[s] nim[5]st nim[s]t nem[d]n nem [a]/ nem[s]nt

For these reasons I suggest that Middle High German imperative forms indicate the type of rule shown in (15). I leave open here the question of which exact form is chosen as the base. (15) Singular Present Indicative Weak XY

Imperative Singular

X

Paul, Wiehl, and Grosse (1989: 254) note that the imperative /-e/ occasionally extends to strong verbs in analogy to weak verbs (cf. nimel, 'take!' gibe! 'give!' tribe! 'drive!'). However, this observation does not necessarily warrant the conclusion that there is a spreading suffix but rather indicates that the rule in (15) generalizes to the effect that the feature "weak" loses relevance. The latter interpretation is supported by the subsequent development. That is, the strict ungrammaticality of "e-

328

Renate Raffelsiefen

suffixation" to the New High German reflex of Middle High German nime!, i.e. nimm!, is unexpected, given the notion of a spreading suffix. By contrast, the ungrammaticality is explained by the type of rule shown in (15), as can be inferred from the members of the current paradigm: (16) Imperative SG 1 2 3 PL 1 2 3

mm/w/'take!' nehm[s] nimmst nimmt nehm\pi\n nehmt nehm[a]n

The claim is then that the strict ungrammaticality of New High German *nimme! is a direct consequence of the non-existence of forms like *nimmest, *nimmet. There simply is no form in the paradigm where a stem with "vowel raising" co-ocurrs with schwa. Similarly, the marginal acceptability of the imperative komme! of the strong verb kommen 'to come' is a direct consequence of the existence of the word forms in the paradigm in (17) in which the stem komm- does co-occur with schwa. (17) Imperative Komm! ?Komm[d]! SG 1 komm[s] (komm) 2 kommst 3 kommt PL 1 komm[d]n 2 kommt 3 komm[d)n The imperative komme! is comparable to the alternative comparative of nehmen, i.e. nehme!, which also has an archaic ring.8 This observation suggests the following rules: (18) a. Standard New Second Person Present Indicative [Xst]

High German: b. Archaic German: Imperative First Person Imperative Singular Present Singular Indicative Χ [X] X

Imperatives: the relation between meaning andform

329

According to the rules in (18), the imperative forms are inferred directly from finite forms in the paradigm. In (19), the forms marked with are inferred by rule (18a), those marked with ' ? ' are inferred by rule (18b), and those marked with '*' cannot be inferred. According to the rules in (18), the distinction between weak and strong verbs is no longer relevant for imperative formation in New High German: geben 'to give' and heben 'to lift' are strong verbs, whereas leben 'to live' is a weak verb. (19) a. SG 1 2 3 b. SG 1 2 3 c. SG 1 2 3

?

gebe gibst gibt hebe hebst hebt lebe lebst lebt

gebe! gib! *gibe!

7

hebe! ^heb! *hib! *hibe!

?

lebe! 07 *lib! Hibe!

The choice of the second rather than the third person as a base for imperative formation is called into question by a certain type of neutralization occurring in the second person singular indicative. The examples in (20) show that a rule based on the third person forms (i.e. Xt —• X), which are not affected by neutralization, would yield the correct imperative forms: (20) a. SG 2 [fli:st] fließt 3 [fli:st] fließt

*[fli:]! [fli:s]! fließ 'flow!' V b. SG 2 [fli:st] fliehst [fli:]! 3 [fli:t] flieht V[fli:]! flieh 'flee!' V

(cf. (18a)) (based on X t ^ X ) (cf. (18a)) (based on Χ / ^ X)

However, additional data indicate both the relevance of the second person as a base for imperative formation and the cause of ungrammaticality in (20a). Consider the judgments in (21):

330

Renate Raffelsiefen

(21) SG 1 werde 2 wirst 3 wird ^werde! *wird! 'become!'

trete trittst, tritt ^ tritt! "tretet 'kick!'

gelte gil(t)st, gilt

fechte fich(t)st9 ficht

igelte! ^fechte! -gilt! -ficht! 'bevalid!' 'fence!'

flechte flich(t)st flicht ^flechte! "flicht! 'braid!'

The interpretation of the variation in (21) is complicated by the fact that the verbs differ considerably in frequency. Yet, a striking phonological generalization emerges from those data. The ungrammaticality of the imperative *wird! correlates with the obligatory absence of the stem-final stop in the second person singular (i.e. wirst rather than *wirdsf). In other words, the imperative *wird! is ungrammatical because there is no precedent for this form in the relevant slot of the paradigm (i.e. the second person singular indicative). The preference for the imperative tritt! correlates with the obligatory presence of the stem-final stop in the second person (i.e. trittst rather than *trist). The variation in the remaining cases correlates with the phonologically conditioned variable pronunciation of the stem-final stop (e.g .fichtst ~ fichst). The generalization suggested by these patterns is that imperatives are based on the second person singular unless that form has a deficient stem. Stem deficiency is also the cause of ungrammaticality in (20a). Assuming that this generalization is correct, the data in (21) not only indicate the relevance of the second person singular for determining imperatives but also reveal the inadequacy of a simple Priscian rule which operates in a strictly syntagmatic fashion. Rather, the recognition that the stem wir- in wirst, but not tritt- in trittst, is incomplete requires reference to other forms in the respective paradigms. 10 Additional evidence for the need to consult multiple forms in the paradigm to form imperatives concerns umlaut alternations. The data in (22) show that both second and third person forms are rejected as a base for imperative formation when they exhibit umlaut. The right-most example in (22) shows that the problem is not caused by umlaut per se; the problem is again solved by basing imperatives on the first person in such cases (the rule will be stated in (26b) below).

Imperatives: the relation between meaning andform

(22) SG 1 2 3

fahre fährst fährt

wasche wäschst wäscht

^ fahr! 1 fahre! **fähr! W

^ wasch! ? wasche! ** wäsch! 'wash!'

331

stärke stärkst stärkt v

stärk! stärke! 'strengthen!'

What exactly is the property of umlaut (as opposed to vowel raising exemplified above) which causes the remarkably strict unacceptability? Significantly, the same judgments are found in German dialects where vowel alternations differ from those in the standard language. The examples below are from Linkeier Piatt, a dialect spoken east of Cologne:

(23) SG 1 2 3

va.-rsn vi:rs yiirt A

vi:r! 'go!'

ko:fon kö:fs kö:ft V

ko:f.

**kö:f!

'buy!'

A possible cause relates to markedness: umlaut alternations always involve an increase in segmental markedness (e.g. ö vs. o), whereas the other alternations do not exhibit this property (e.g. i vs. e). Consider next the imperative forms of trochaic second person forms. (24) SG 1 2 3

wackele, wackle, wackel wackelst wackelt wackle! wackel! wackele! 'wobble!'

Here the choice of the first person as a base is both necessary and sufficient for describing the variation in imperative forms. The generalization is that final schwa is dropped unless it is necessary for pronunciation. For instance, in wackle, final schwa is necessary to satisfy SON (* wackf). In predige, another verb with a trochaic second person form, final schwa is necessary to satisfy a constraint which prohibits voiced obstruents in coda position (*predi[g]).11 Imperatives for which final schwa is not phonologically conditioned are associated with a different register and are described by rule (18b).

332

Renate Raffelsiefen

(25) SG 1 2 3

predige predi[$]st predi[q]t predige! *predi[ g]/ 'preach'

'predigst' 'predigt'

To ensure that the formation of the imperatives in (24, 25) is based on the first rather than the second person (cf. *predi{9]!), the rule for forming imperatives on the second person can be made sensitive to stress. As a result of specifying stem-final stress in rule (26a) that rule takes precedence by the Elsewhere Condition. For all rules it holds that imperatives are not based on word forms with incomplete stems or (alternating) umlaut. (26) a.

b.

Second Person Present Indicative Xlc(e)st First Person Present Indicative Xe

Imperative Singular

Χιφ) Imperative Singular X (if oronounceable) Xe (otherwise)

The reason for specifying stem-final rather than word-final stress relates to inflected forms with epenthetic schwa, such as 'arbeitest '(you) work', 'rettest '(you) save'.12 Here rule (26a) should apply to make sure that the schwa appears in the imperative forms, even though it is not needed on phonological grounds. In all cases, imperative forms are substrings of inflected present tense surface forms: schwa is sometimes obligatory, sometimes impossible, and sometimes marked (archaic), depending on the distribution of schwa in the forms serving as precedents (as well as on pronounceability). The refusal to recognize Priscian rules entails considerable loss of generalization. Consider the description of imperative formation in Drosdowski (1984: 174): "Some 'ablauting' verbs form the imperative singular by changing the e (ä, ö) of the present tense stem to i (ie); -e is

Imperatives: the relation between meaning andform

not attached." A list of the relevant formations follows {lies!, birg!,

333 wirf.,

stirb!...).

The formulation "some ablauting verbs" misses the true generalization determining the applicability of "vowel raising" in imperatives: the rule affects precisely those verbs which have a raised vowel in the second or third person. That is, both geben 'give' and heben 'lift' are characterized by ablaut (i.e. geben, gab, gegeben; heben, hob, gehoben), but the paradigm of only geben, and not of heben, includes raised stems and, consequently, raised imperatives. In addition, rule (26) accounts for historical changes in imperative forms. As a result of the regularization undergone by verbs like löschen (cf. f lischst > löschst) or melken (cf. + milkst > melkst), the corresponding imperatives disappear as well (cf. 'lisch! > lösch(e)!, ^milk! > melk(e)!). The second major problem with the description in Drosdowski concerns the seemingly disconnected statement "-e is not attached." As has been emphasized above, on the assumption that there is no imperative suffix -e, the ungrammaticality of liese! can be explained as a direct consequence of the nonexistence of finite forms like * liesest, *lieset.]3 Consider three additional rules for e-suffixation stated in Drosdowski (1984: 174): (27) a. The imperative -e is generally attached to verbs whose stem ends in -dor -t. b. The imperative -e is generally attached to verbs whose stem ends in Iml or Inl preceded by a consonant. c. The imperative -e is not attached to verbs whose stem ends in Iml or Inl preceded by Iml, Inl, 111, Irl or single Ihl. In all cases, it can be shown that the generalizations, including occurring variations, follow straightforwardly from the distribution of schwa in the second person singular of the relevant paradigms: (28) a. arbeitest (*arbeitst) ^ arbeite! *arbeit! 'work!' b. atmest (*atmst) ^ atme! *atm! 'breathe!'

leidest (*leidst) ^ leide! Heid! 'suffer!'

redest (n redst) ^ rede!??red! 'talk!'

334

Renate Raffelsiefen

c. kämmst (*kämmest) ^ kämm! 'comb!'

lernst (*lernest) ^ lern! 'learn!'

The description in Drosdowski (1984) wrongly suggests the existence of an imperative suffix in German.14 In reality, German mostly fits the pattern in (4) (cf. rules (26a, b)) and partially exemplifies "zero derivation" (cf. rule (18b, 26b)). German differs from Norwegian and Danish in that the evidence for Priscian imperatives is based not on phonological "irregularities" resulting from clipping but on distributional evidence (e.g. the distribution of raised vowels, schwa).

4. Imperatives in Icelandic In Icelandic, imperatives are always followed by a subject clitic (cf. colloquial German hören Sie! —> [hornza] 'hear!'). The imperative stems typically end in a consonant, except for class IV verbs, which end in -a. The formations are thus generally described as shown in (29) (the exclamation mark is not used in Icelandic): (29) heyr+pii heyröu! kalla+pü=>kallaöu!

'hear!' 'call!'

For some verbs, we find both a form described by the rule in (29) and a seemingly idiosyncratic form. Examples are given in (30): (30) a. kaup+ρύ yrk+pu scek+pu

b. kauptu! yrktu! scektu!

c. keyptu! ortu! sottu!

'buy!' 'build!' 'fetch!'

Assuming the inputs in (30a) the forms in (30c) have the appearance of relics doomed to become obsolete. However, Oresnik (1981) informs us that the forms in (30c) are the innovative colloquial variants, whereas the seemingly "regular" formations in (30b) are archaic. Helgason (1970) identifies the source of the innovative forms: they are based on the third person plural preterite forms. The relevant paradigms are given in (31):

Imperatives: the relation between meaning andform

SG 1 2 3 PL 1 2 3

Present kaupi kaupir kaupir kaupum kaupi ö kaupa

Preterite keypti keyptir keypti keyptum keyptuö keyptu

SG 1 2 3 PL 1 2 3

yrki yrkir yrkir yrkjum yrkiö yrkja

orti ortir orti ortum ortuö ortu

SG 1 2 3 PL 1 2 3

sceki scekir scekir scekjum sceki ö scekja

sotti sottir sotti sottum sottuö sottu

(31)

335

The innovative imperatives in (30) thus indicate the prime example of a Priscian rule in (32): (32) Third Person Plural Preterite Indicative X

Imperative Singular

It seems clear that rule (32) originates from the coincidental homophony between cliticized imperative forms such as heyr+pu —• heyrdu and the third person plural preterite of weak verbs where the dental suffix is followed by u. (33) SG 1 2 3

heyri heyrir heyrir

heyrdi heyröir heyrdi

336

Renate Raffelsiefen

PL 1 2

heyrurn heyriö heyra

heyröum heyröup heyröu = heyröu < heyr+pü

One factor which may have prompted speakers to construct rule (32) relates to the fact that the subject clitic is on its way to becoming further grammaticalized into an imperative suffix. On this hypothesis, speakers are reluctant to form imperatives proper iconically (i.e. by suffixation) and opt for zero derivation instead. Another factor concerns the (even for native speakers) formidable allomorphy found in imperative singular cliticization. Specifically, the junctures involving obstruents give rise to speech errors and uncertainties. Some examples for such junctures are listed below (cf. Petursson 1992: 121): (34)

ö+p f+P d+p g+P k+p p+P t+p

> > > > > > >

dd fö t gd kt pt tt

[t:] [vö] [t] [kt] [xt] [ft] [ht]

It seems natural that speakers prefer to fall back on already familiar forms in the paradigm instead of tediously constructing forms according to the rules in (34). However, there is one case where speakers systematically shun the convenient third person plural preterite: that form is never used when it exhibits (alternating) umlaut. In such cases, speakers choose a stem which supplies the relevant allomorphy without umlaut and attach the vowel -u. As a result, the imperative of the verb etja 'to egg on' is neither ettu, nor öttu, but attu: 1 2 3 1 2 3

et etur etur etjum etjiö etja

att / att ir att/ öttum öttuö öttu

Oresnik (1981) considers the output attu (rather than öttu) evidence for the incorrectness of Helgason's generalization stated in (32) and

Imperatives: the relation between meaning andform

337

concludes that imperative singulars are based on second person singular preterite in (better) accordance with semantics. However, his conclusion lacks cogency since we know independently from German that umlaut forms are systematically avoided as bases for imperative formation and that imperatives need not be based on a unique word form (cf. rule (26)). Additional evidence against "synthetic" formations of imperatives can be gleaned from imperatives with emphatic subjects. The relevant forms are not those shown in (36a) but rather the clipped forms shown in (36b) (cf. Oresnik 1980): (36) a

99

kauppu yrk pu. 99 seek pu 99

b. keypt pu! ort pit! sott pü!

The type of emphatic imperatives shown in (36a) occurs regularly for class IV verbs (e.g. kalla pit!). This observation correlates with the fact that rule (32) never applies to class IV verbs, which necessarily have umlaut in the third person plural preterite (e.g. kölluöu).15

5. Finnish Historically, Finnish imperatives singular were formed by Ä>suffixation, as shown in (37a). Such forms still exist in some dialects, in Southwestern dialects, the -k has vanished, as is shown in (37b): (37) maksaa antaa vetää hakata

'to 'to 'to 'to

pay' a. maksak! b. maksal give' annak! anna! pull' vedäk/ vedä! beat' hakkaak! hakaa!

In Standard Finnish, however, the -k is claimed to have left a reflex. This reflex is manifested in gemination, when a consonant-initial word follows, and in the occurrence of a glottal stop, when a vowel-initial word follows. Examples are given in (38):16 (38) maksa[pp]ois! maksa[?o]mena!

'Pay, will you!' 'Pay for the apple!'

338

Renate Raffelsiefen

A possible representation of the imperative reflex in Finnish is the empty coda node in (39), which remained after the loss of the segment: (39) σ

σ

σ

O N C O N C I I I I I I m a k s a k

>

σ

O N C O N C I I I II I m a k s a 0

The empty coda is then associated with a following consonant, giving rise to ambisyllabicity (reflected in segmental length) or, if there is no following consonant, it is filled by a glottal stop:17 (40) a. σ

σ

σ

O N C O N C O N C m a k s a pois b. σ

σ σ σ I Γ\ Γ\ O N C O N C N O N O N m a k

σ

s

a [ ? ] o m e n a

While the phonological effects described above are generally analysed in terms of reflexes of the historical imperative suffix, there is evidence for a historical re-analyis of the reflex. Consider first the relation between imperative forms and other forms in the paradigm illustrated below (cf. Karlsson 2000: 181) (41) Infinitive

l.Pers. SG Present Tense

anta/a vetä/ä sulke/a

'to give' 'to pull' 'to close'

annan vedän suljen

luke/a

'to learn'

luen

Imperative Singular αηηα\Ί\ vedaf[?] sulje\7]

/ae[?]

Imperatives: the relation between meaning andform

avat/a maat/a tavat/a määrät/ä hakat/a tarjot/a teh/dä juos/ta ajatel/la harkit/a paet/a

'to open' 'to lie' 'to find' 'to determine' 'to beat' 'to offer' 'to make' 'to run' 'to imagine' 'to consider' 'to flee'

avaan makaan tapaan määrään hakkaan tarioan teen juoksen aiattelen harkitsen pakenen

339

αναα[Ί] makaayi ] tapaayi ] määrää\1] hakkaa\l] tarjoa\7} tee[1] iuokse\?] aiattele\T\ harkitse[?] pakene\1]

Comparing infinitive stems and imperatives, we find a range of alternations many of which exemplify the so called "Consonant Gradation", a historical rule which affected stops in closed syllables. However, the exceptionless generalization emerging from the data in (41) is that imperatives - apart from the final "reflex" - are identical to the first 1δ person singular without the -n. In fact, the generalization works for any finite verb form, where the stem ends in a closed syllable, as is illustrated in (42):19 (42) SG 1 an. nan 2 an.nat 3 an.taa PL 1 an.nam.me 2 an.nat.te 3 an.ta.vat Historically, the identity effect shown in (41) is of course caused by the syllable structures in the individual forms: both the imperative -k and the first person -n (as well as -m and the second person -t) close the stem-final syllable, thereby triggering Consonant Gradation. However, although the glottal stop or the geminate also close the syllable in imperative forms and could therefore serve as the synchronic trigger of Consonant Gradation, it seems implausible that Finnish speakers fail to exploit the identity patterns illustrated in (41) and tediously derive imperative forms "from scratch" (cf. the related discussion of the Icelandic data). Rather, it is likely that the historical reflex of the imperative suffix described in (39) (i.e. gemination or the glottal stop) has been re-

340

Renate Raffelsiefen

analyzed as a synchronic reflex of syllable closure in present tense paradigms described in (43): (43) Standard Finnish Present Tense Indicative Singular

Imperative Singular

Coda

Coda X

X

Some evidence for reanalysis relates to colloquial reductions of high frequency verbs such as tule- 'come', mene- 'go' and ole- 'be', illustrated in (44): (44) SG 1 2 3 PL 1 2 3

tu.len tu. let tu.lee tu.lem.me tu.let.te tu.le.vat

tuun tuut tu.lee tuum.me tuut.te tu.le.vat

Like Consonant Gradation, "High Frequency Reduction" is sensitive to syllable closure. The rule entails the loss of a short unstressed vowel and the preceding sonorant (if there is one), such that only the mora remains: (45)

σ

σ

Ο Ν Ο Ν C μ μ t

ύ 1 e

η

The fact that High Frequency Reduction also applies in the related imperative forms in Standard Finnish (e.g. tuu\7J) could of course be attributed to the final glottal stop, which closes the syllable. However, reduced imperatives such as tuu! 'come' also occur in Southwestern

Imperatives: the relation between meaning andform

341

dialects, where the crucial prerequisite of syllable closure is not given. Another argument concerns the imperative of the verb nähdä 'to see'. This verb also undergoes High Frequency Reduction in the first and second person finite forms as is shown in (46): (46) SG 1 2 3

näen > nään näet > näät näkee näkee

Here the rule should not apply to the imperative because the imperative of this verb is not used.20 However, despite the lack of frequency, the imperative wää[?]! (as opposed to näeft]}.) is easily elicited from native speakers. This fact indicates that imperatives are synchronically derived by rule (43), which expresses a counter-iconic relation.

6. Conclusion The non-arbitrary relation between meaning and sound observed in imperatives supports a central tenet of functionalist language theories: that there exists a unity of form and function (cf. Bybee 1985). In this paper, I have presented evidence from five languages that imperatives are formed by clipping final segments from independently existing, inflected verb forms. The imperatives in these languages are usually analysed as based on stems (rather than surface word forms) to which zero marking or suffixation is applied. I conclude that the mode of marking in these languages is more homogeneous than is commonly assumed. Evidence for word formation based on inflected word forms supports the notion of the paradigm consisting of related surface word forms (cf. Vennemann 1974). The relevance of paradigms is also supported by the observation that base selection can require systematic comparison of at least two distinct members (cf. German, where imperatives are based on second person forms only if that form includes (a) correspondents to all consonants occurring in first person forms and (b) a stem vowel which corresponds to that of the first person with respect to the feature [±front]). Since the frequency with which imperatives are formed by clipping can be established only on the basis of a far larger and more varied

342

Renate Raffelsiefen

sample of languages, the contribution of this study lies mainly in alerting researchers to focus their attention on proper identification of base 01 forms. Specifically, the possibility should be taken into account that a word can serve as a phonological base, without contributing morphological features to the derived form (cf. Icelandic, where the features third person, plural, preterite are clearly not part of the imperative meanings). The identification of phonological bases should be guided by phonological criteria alone, specifically, the occurrence of systematic identity patterns which cannot be explained as the result of rule application to forms in isolation, including violations of canonical prosody.22 The claim that a phonological base must not necessarily form a semantic base - anticipated by Priscian and other grammarians - in one respect contradicts the tenet of the unity of form and function in functionalist theories (supported by the evidence for counter-iconicity). An important question here is whether "Priscian rules" arise only under certain phonological conditions.23 In two cases (Icelandic, Finnish), the choice of the base forms appears to have been determined by fortuitous identity patterns in surface forms, which spares the learner the effort of applying difficult morphophonological rules. When constructing a rule based on an identity relation between two slots in the paradigm, there is of course the question of which form is chosen as a base and which is derived. The very limited sample suggests that imperatives are based on indicatives rather than vice versa (cf. the innovative Icelandic imperatives in (30c)). As for imperatives, a preference for counter-iconic formation may also cause the selection of semantically unmotivated base forms. Confirmation of the tentative conclusion that imperatives are preferably formed by clipping would be remarkable in that imperatives are possibly the only inflectional category exhibiting this property.24 Formation by clipping would clarify Greenberg's (1966) characterization of imperatives as 'impure stems'. Notes * I would like to thank Ulrich Groenke, Lutz Gunkel, Tuija Hämäläinen, Martin Haspelmath, Marja Järventausta, Magnüs Petursson, David Restle, Roger

Imperatives: the relation between meaning and form

1.

2.

3.

4. 5. 6.

7.

8. 9.

343

Schwarzschild, Dietmar Zaefferer, and an anonymous reviewer for help and discussion. Assuming that the rule in (1) is intended as a redundancy rule (rather than a transformation), it fits into Vennemann's (1974) concept of a mental lexicon consisting of only surface word forms. The question of whether a word can be the phonological base of another word, without also being the semantic or morphosyntactic base, is not explicitly discussed there, but part of the discussion of the example from Granadense (cf. 1974: 353) suggests that such divergences are unexpected. Not surprisingly, linguists with a strong interest in phonology are exceptions here, cf. the 'rules of referral' discussed by Zwicky (1991), Stump (1993). Steriade (1999) introduces the notion 'split base effect', which says that the phonological, morphosyntactic, and semantic notion of a base need not coincide. Needless to say, to 'morpheme-morphologists' the rule in (1) is an absurdity. I dispute the claim that the main syntactic characteristic of imperatives is the optionality of the subject (cf. Donhauser 1986, Eisenberg 1998). Clearly, polite imperatives in German such as Hören Siel with an obligatory subject fall in the same category as subjectless imperatives like Hör\. In Icelandic, all imperatives have an obligatory subject clitic. Korhonen cites Revesz (1946: 136), who argues that imperatives are prelinguistic. See Benua (1995) for a discussion of the correspondence in surface structure within Optimality Theory. Mikkelsen writes: „Digterne lade undertiden den sidste Medlyd i Udtalen gaa over paa en fülgende med en selvlyd begynende Stavelse....Slige Former omskrives dog i Regien" (1894: 209). For Old High German, reference to the first person singular present indicative accounts for the imperative stem vowel in all strong verbs. Assuming that the base is identified correctly, the formation involves clipping of the final vowel and is accordingly counter-iconic (e.g. nimu —• niml, rato —*• rät!). Significantly, vowel raising was at no point in time phonologically conditioned in imperative forms (as opposed to the second and third person singular). Class I weak verbs are based on second or third person present indicative with clipping of the final consonant (suochit—> suochi!). Class II and Class III weak verbs require vowel shortening in addition ( s a l b o t s a l b o ! , habet —*• habe!). According to Drosdowski (1984: 174), such forms were used frequently by Goethe and exclusively by Heine. Needless to say, the argument applies only to those speakers who have regular vowel raising in the finite forms but not in the imperative form. For the not particularly frequent verbs fechten and Hechten there are colloquial variants without raised vowels, which indicate that these verbs are about to regularize. Perhaps this regularization is partially triggered by the imperative formation,

344

10.

11.

12.

13.

14.

15.

16.

17.

Renate Raffelsiefen which for the deficient stems fichst, fliehst is based on the first person singular. According to Bielenstein (1883: 48, 128), clipping of stem consonants in imperative forms, which are based on the second person present tense indicative, occur occasionally in Latvian. German differs thus from Norwegian Bokmäl or Danish in that pronounceability is always ensured in imperatives (instead of avoiding unpronounceable forms). The occurrence of the schwa is caused by a constraint against adjacent stops with identical place features, which applies to third person singular and second person plural forms (e.g. arbeitet, rather than * arbeitt, red[s]t rather than *redt). The occurrence of the schwa in the Second Person singular forms is due to analogy. The form siehe as in siehe oben 'see above' is not an exception to rule (26) (the relevant finite forms of the verb sehen are sehe, siehst, sieh ή; rather siehe is not an imperative. A phrase like "siehe oben" does not express a command to the reader but rather informs that additional relevant information appeared earlier in the text. Very similar rules are stated in Zifonun, Hoffmann, and Strecker (1997: 1725). Eisenberg (1998: 194, 195) attempts to capture the Priscian generalization without positing a Priscian rule when he writes that imperatives singular are formed by raising the vowel in those strong verbs which also undergo vowel raising in the present tense and that they are without ending then. This description raises the question of why these dependencies exist. It is of course also true that the allomorphy rules for cliticization are simple for class IV verbs, because the dental fricative occurs intervocalically. As a result, speakers have little to gain from resorting to preterite forms as bases for imperative formation. Cf. the minimal pair maksa\pp\ois! 'Pay, will you!' versus maksa\$\ois 'Liver away' from maksa 'liver' and pois 'away'. The context generally given is an operating room where the surgeon discovers that the liver cannot be saved [Marja Järventausta, p.c.]. The contrast between (40a) and the expression maksa pois 'liver away' is represented as follows (cf. the preceding footnote): (i)

maksa[p]ois σ

σ

σ

O N C O N

ONC

m a k s a

p o i s

Imperatives: the relation between meaning andform

345

18. This is the rule given in Karlsson (2000) and, less explicitly, in Terttu (1993). By contrast, Schmeidler (1989) and Abondolo (1998) are so averse to Priscian rules that they posit rules for deriving imperatives from the infinitive stem. 19. That is, the stems in the first and second person singular and plural are always identical. 20. The form katso^i] 'look!' is used instead. 21. On strictly phonological grounds, Prince (1975) identifies perfect forms as the base for clipped imperatives in Tiberian Hebrew. 22. Sometimes the crucial effects will be observable in only a tiny fraction of all imperative forms and are therefore easily overlooked. For example, in Norwegian Bokmäl only verbs with a stem-final cluster with increasing sonority exhibit proof of clipping (rather than stem-based derivation). 23. For example, the choice of present infinitive active forms as bases for msuffixation in Latin may be partially due to the fact that such forms consistently end in a vowel (cf. (1)). This is because a non-coronal consonant like m, which may not occur as an appendix, is most likely to satisfy constraints on coda structures if a vowel precedes. 24. Perhaps counter-iconicity is also characteristic for vocatives which would mean that clipping is associated with direct address.

References Abondolo, Daniel 1998 Colloquial Finnish. The complete course for beginners. London/New York: Routledge. Allan, Robin, Philip Holmes, and Tom Lundskaer-Nielsen 1995 Danish. A comprehensive grammar. London/New York: Routledge. Anderson, Stephen R. 1975 On the interaction of phonological rules of various types. Journal of Linguistics 11: 39-62. Basboll, Hans 1969 Notes on the phonology of Danish imperatives with a digression on vowel quantity. ARIPUC 4: 15. Benua, Laura 1995 Identity effects in morphological truncation. In: Jill Beckmann, Laura Walsh, and Susann Urbanczyk (eds.), Papers in Optimality Theory, 77-136. (University of Massachusetts Occasional Papers in Linguistics 18.) Amherst: GLSA. Bielenstein, August 1883 Lettische Grammatik. (Ausgabe: 1972 Zentralantiquariat der Deutschen Demokratischen Republik: Leipzig.)

346

Renate

Raffelsiefen

Bybee, Joan L. 1985 Morphology: α study of the relation between meaning and form. Amsterdam/Philadelphia: John Benjamins. Drosdowski, Günther (ed.) 1984 Duden. Grammatik der deutschen Gegenwartssprache. Mannheim/Wien/Zürich: Bibliographisches Institut. Eisenberg, Peter 1998 Grundriss der deutschen Grammatik. Das Wort. Stuttgart/Weimar: Metzler. Faarlund, JanTerje, Svein Lie, and Kjell Ivar Vannebo 1997 Norsk referanse-grammatikk. Oslo: Universitetsforlaget. Fennel Trevor G. and Henry Gelsen 1980 A grammar of Modern Latvian. The Hague: Mouton. Frege, Gottlob 1923 Das Gedankengefuge. In: Günther Patzig (ed.): Frege, Logische Untersuchungen, 72 91. Göttingen 1976: Vandenhoeck and Ruprecht. Greenberg,Joseph 1966 Language universals. The Hague: Mouton. Hakulinen, Lauri 1957 Handbuch der finnischen Sprache. Wiesbaden: Otto Harrassowitz. Helgason, Jon 1970 Islenzkfornkvasöi. Vol. VII. (Editiones arnamagnaeanae, series Β, vol. 16.) Kobenhavn: Munksgaard. Karlsson, Fred 2000 Finnische Grammatik. Hamburg: Helmut Buske. Korhonen, Mikko 1967 Die Konjugation im Lappischen. Suomalaisen Kiijallisuuden Kirjapaino Oy. Helsinki: Helsinki University Press. Kristofferson, Gjert 1991 Aspects of Norwegian Syllable structure. Ph.D. Dissertation. University of Tromso. Matthews, Peter H. 1991 Morphology. Cambridge: Cambridge University Press. Mikkelsen, K. 1894 Dansk sproglcere. Kobenhavn: Lehmann and Stages. Oresnik, Janez 1980 On the modern Icelandic clipped imperative. In: Even Hovdhaugen (ed.) The Nordic languages and modern linguistics, 305-314. Oslo: Universitetsforlaget. 1981 On some Icelandic irregular imperative singular forms. Afmseliskvedja till Halldors Halldorssonar 13. jüli 1981, 211-218. Reykjavik: Islenska Mälfraeöi felagiö.

Imperatives: the relation between meaning andform

347

Paul, Hermann 198923 Mittelhochdeutsche Grammatik. Neu bearb. von Peter Wiehl u. Siegfried Grosse. Tübingen: Niemeyer. Petursson, Magnüs 1992 Lehrbuch der isländischen Sprache. Hamburg: Helmut Buske. Prince, Alan 1975 The phonology and morphology of Tiberian Hebrew, Doctoral Dissertation. MIT. Revesz, Geza 1946 Ursprung und Vorgeschichte der Sprache. Bern: Francke. Schmeidler, Marie-Elisabeth 1989 Der Formenbau des finnischen Verbs. Hamburg: Helmut Buske. Steriade, Donca 1999 Lexical conservatism in French adjectival liaison. In: J. Marc Authier, Barbara E. Bullock, and Lisa A. Reed (eds.), Formal Perspectives on Romance Linguistics. Amsterdam/Philadelphia: John Benjamins. Stump, Gregory T. 1993 On rules of referral. Language 69: 449-479. Terttu, Leney 1993 Finnish. A complete course for beginners. NTC/Contemporary Publishing Company. Vennemann, Theo 1974 Words and syllables in natural generative grammar. In: Anthony Bruck, Robert A. Fox, and Michael W. La Galy (eds.), Papers from the parasession on natural phonology. Chicago: Chicago Linguistic Society. Xrakovskij, Victor S. (ed.) 2001 Typology of imperative constructions. München: Lincom Europa. Zifonun, Gisela, Ludger Hoffmann, Bruno Strecker, et al. 1997 Grammatik der deutschen Sprache. Berlin: Walter de Gruyter. Zinkevicius, Zigmas 1996 The History of the Lithuanian Language. Vilnius: Moksklo ir enciklopediju leidybos institutas. Zwicky, Arnold M. 1977 Systematic versus accidental phonological identity. In: Frans Plank (ed.), Paradigms: The economy of inflection, 113-131. Berlin/New York: Mouton de Gruyter.

Struktur und Akzent komplexer Komposita Peter Eisenberg

Abstract More than ten years ago Theo Vennemann (1991) considered the possibility of describing German words such as Attentat and Abenteuer as morphologically simple but prosodically complex. More precisely he proposed to assign to them the prosodic structure of compounds such as e.g. Wagenburg or Haufenwolke respectively. The mismatch based on the fact that two phonological words are distributed over one morphological word marks them as borderline cases. In some respects they are simplexes, in some others they show properties of compounds. In the present paper a fairly common type of composition in German is treated in an anlogical way to what Vennemann proposed. The words in question are complex compounds (i.e. compounds with more than two immediate constituents) such as Waldschadensbericht, Verkehrsbeeinflussungsanlage or Stadtplanungsbüro. The second component is a deverbal nominalization. Their analysis poses problems because the prototype can bear the main accent on the first constituent as well as on the second one (e.g. Wdldschadensbericht vs. Waldschddensbericht). In the literature they are normally considered as having two different constituent groupings. With a left branching structure the first constituent should bear the main accent ([['a b] c]), with a right branching structure it should be placed on the second constituent ([a ['b c]]]. Yet most authors are very reluctant with repect to the specification of semantic affects based on this ambiguity. At best one speaks of "slightly different compositional readings" depending on the two structures. Similar differences arise when compounds are analyzed syntactically, no matter whether they are related indirectly via argument inheritance to, or whether they are derived directly from underlying syntactic structures. We therefore propose to treat such words again as borderline cases. As in the examples discussed by Vennemann it is the prosodic structure which shows properties of the higher level. This is now the level of syntax. Inside morphological domains the accent should be fixed. But instead of being fixed it can be shifted as it is typically possible within syntactic domains. It is argued that accent placement is controlled by the information structure of the wider context of the word. The accent can be shifted between the first and the second constituent of the compound by changing the focus structure. Roughly speaking the placement of the accent is to be understood as indicating a focus exponent located at the respective constituent of the word.

350

Peter Eisenberg

1. Simplexakzent In seiner Skizze der deutschen Wortprosodie fasst Theo Vennemann die in mehreren früheren Arbeiten entwickelten Regularitäten zur Akzentuierung morphologisch einfacher Wörter des Deutschen in wenigen Regeln zusammen. Nur Vollsilben können akzentuiert werden, und außerdem gilt: (1) eine bedeckte reduzierte Ultima arretiert den Akzent auf der letzten Vollsilbe, (2) nur die drei letzten Vollsilben können akzentuiert werden und (3) der Akzent geht nicht über eine schwere Pänultima hinaus. Schwer ist jede Silbe, die nicht sowohl offen als auch monophthongisch wie sanft geschnitten ist (Vennemann 1991: 97ff., s. auch Vennemann 1990, 1991a). Zu jeder der Regeln gibt es Ausnahmen, aber zur Pänultimaregel (3) wahrscheinlich besonders viele und von ganz unterschiedlichem Typ. Bestätigt wird die Regel durch Wörter wie Veranda, Petrarca, Balalaika, Eleusis, Canaletto und Kentucky. Infrage gestellt wird sie einmal durch Eigennamen wie Turandot und Valentin, Heribert und Hildegard. „Die zweite Gruppe enthält Wörter, die sich wie Komposita „anfühlen": A.mei.se (vgl. scherzhaft Be.mei.se), Ta.lis.man (wie Flach.mann, Bai. ler. mann) und insbesondere solche, die durch interne Reduktionssilben das prototypische Kompositum zu imitieren scheinen: At.ten.tat, A.ben.teu.er, Ta.mer.lan." (1991: 100). Vennemann kommt dann darauf zu sprechen, dass in den mehrsilbigen Stämmen bei bestimmten Silben durch realisationsphonologische Transformation Vokalschwächung eintreten kann. Das geschehe bevorzugt zwischen rhythmisch prominenten Silben, etwa zwischen der haupt- und nebenakzentuierten Silbe wie der zweiten in Känguruh oder Dromedar, dann artikuliert als Kän.g[^].ruh und Dro.m[^].dar. Es gibt ganz unterschiedliche Fußstrukturen, die derartigen Prozessen unterworfen sind. Die beschriebene ist hier von besonderem Interesse, weil realisationsphonologisch erneut das prosodische Muster eines Kompositums entsteht. Mit der Deutung eines phonologisch irregulären Akzents als komposital nimmt Vennemann einmal Stellung in der bei Erscheinen der Skizze noch nicht beendeten Auseinandersetzung darüber, ob das gegenwärtige Deutsch Initial- oder Finalakzent habe. Kann ein Akzent mit den Regeln 1 bis 3 nicht als Finalakzent ausgewiesen werden, dann ist er aus speziellen Gründen an bestimmte Wortklassen (etwa Eigennamen) ge-

Struktur und Akzent komplexer Komposita

351

bunden oder er ist ein morphologischer Akzent, etwa der eines Kompositums. Entscheidend für die Gangbarkeit dieses Weges ist, was als Analogiebasis zugelassen wird. Für die Gruppe Ameise, Talisman, Nachtigall, Bräutigam spricht man üblicherweise vom Kompositionsakzent, weil wenigstens ein lexikalischer Bestandteil in einer ftir Komposition möglichen Stellung isoliert werden kann. Für Attentat und Abenteuer ist das zweifelhaft, für Tamerlan, Känguruh, Dromedar ausgeschlossen. Wenn man ihnen einen Kompositionsakzent zuweist - was Vennemann nota bene für die beiden letztgenannten Wörter nicht explizit tut - dann wäre die prosodische Struktur allein dafür verantwortlich. Das Wort hätte zwei "phonologische Wörter" innerhalb eines morphologischen (Raffelsiefen 2000). Die prosodische Struktur könnte demnach auch unabhängig vom lexikalischen Aufbau einen Reanalyseprozeß Ingangsetzen. Es bestünde damit die Möglichkeit, einen Ebenenwechsel primär an den Akzenteigenschaften eines Wortes festzumachen. Das Wort hätte die Eigenschaften eines morphologisch einfachen, bis auf seine Prosodie. Der folgende Beitrag möchte dafür argumentieren, dass man eine analoge Möglichkeit auch im anderen Übergangsbereich kompositaler Einheiten ins Auge fassen sollte, nämlich am Übergang zur Syntax. Für einen Typ von Kompositum soll gezeigt werden, dass sein Akzentverhalten so erklärt werden kann. Es handelt sich um dreigliedrige Komposita vom Typ Waldschadensbericht, die regelmäßig sowohl auf dem ersten wie dem zweiten Bestandteil akzentuierbar sind. Die angestrebte Analyse hat sich vor allem mit Folgerungen aus der üblichen Annahme auseinanderzusetzen, morphologische Konstituentenstrukturen seien binär verzweigend. Abschnitt 2 resümiert die Hauptlinien der Behandlung des Kompositaakzents in der Literatur unter diesem Gesichtspunkt. In Abschnitt 3 wird ein Analyseweg skizziert, der dem Akzentverhalten komplexer Komposita Rechnung tragen kann.

2. Wortstruktur und Mehrdeutigkeit 2.1. Binarität Anders als für phonologische und syntaktische Strukturen ergibt sich die vergleichsweise einheitliche Auffassung, morphologische Strukturen

352

Peter Eisenberg

seien binär verzweigend, aus einem speziellen Verhältnis von Kombinatorik und ihrer Basis im Lexikon. Kaum ein grammatiktheoretischer Ansatz sieht für die Phonologie überhaupt die Notwendigkeit einer generellen Beschränkung auf Binarität. Multiple Verzweigungen wie in Vennemeanns (1991) Morenraster, auf der Skelettschicht von Silben, bei den und innerhalb der Silbenkonstituenten, bei Silben als Fußkonstituenten usw. sind an der Tagesordnung. In der Syntax ist das viel weniger selbstverständlich, nicht selten wird aber gerade vom Binaritätspostulat abgesehen. Sind morphologische Strukturen wie syntaktische formatiert, dann ist das Binaritätspostulat in der Regel vorausgesetzt, wie beispielsweise in Ansätzen, die mit einem X-bar-Format arbeiten. Morphologische Strukturen gelten dann als binär, weil syntaktische als binär gelten. Das ist hier nicht gemeint. Herausgestellt werden soll vielmehr, dass in der Morphologie ein Binaritätsprinzip vorherrscht, das gar nichts mit einem bestimmten Beschreibungsformat oder sprachtheoretischen Ansatz zu tun hat. Beispielsweise ist unerheblich, ob man morphologische Regularitäten kombinatorischsyntaktisch oder wortbezogen-paradigmatisch versteht und darstellt. Eine Suffigierung wie in (la) sieht im paradigmatischen Ansatz von Thomas Becker wie in (lb) aus, bleibt also dem Binaritätsprinzip verpflichtet (Becker 1993). Ähnliches gilt fur wortbezogene Regularitäten wie Rückbildung {Bausparer - bausparen) oder Cross-formation (psychologistisch - Psychologismus). (1) a.

deceivableA deceivev

ableA

b.

X V

liable A

Erinnert wird auch an die Bemühungen zur Darstellung morphologischer Gehalte bei nichtsegmentaler Kodierung, vor allem Vokalwechseln, im klassischen Strukturalismus. Für die englische Verbform took erörtert Hockett entsprechend der Denkweise des Item and Arrangement mehrere Möglichkeiten. Bei allen besteht die Form aus genau zweierlei, z.B. aus einem diskontinuierlichen Allomorph It...kl mit einem Infix oder aber der Präsensform take mit einem Ersetzungsmorph /u/ —» /ey/ usw. (Hockett 1954, ausfuhrlich dazu Eisenberg 1993). Im Allgemeinen verstehen wir morphologische Regularitäten als lexikonbasiert. Im Prinzip wird ein Stamm, ein Wort oder eine Wortform auf der Basis eines Stammes, eines Wortes oder einer Wortform gebil-

Struktur und Akzent komplexer Komposita

353

det. Liegt nicht kategoriale Umsetzung in dieselbe Form vor, dann ist die morphologische Regularität an Veränderungen oder Unterschiede zwischen den Formen gebunden, die als eine Einheit aufgefasst werden können und aufgefasst werden. Und schon vom zweiten Wirksamwerden einer Regularität (segmental gesprochen: vom dritten morphologischen Bestandteil) an stehen bei binärer Gliederung Darstellungsalternativen zur Verfügung, die nicht an kategoriale, sondern ausschließlich an strukturelle Merkmale gebunden sind, etwa an hierarchische: „Hierarchical tree structures are capable of showing semantic relations in a salient way", z.B. [un[do able]] ,which cannot be done' und [[un do]able] ,which can be undone' (Haspelmath 2001: 83f.). Ein Strukturierungsprinzip macht Hall daraus, wenn er schreibt, aus „semantischen Gründen müssen manche Komposita mit drei Bestandteilen" als linksverzweigend, andere als rechtsverzweigend analysiert werden (Hall 2000: 287). Es wird nun gezeigt, was das für die Plazierung des Hauptakzents bei zwei- und dreigliedrigen Komposita bedeutet.

2.2. Einfache und komplexe Komposita Ausgangspunkt ist die Annahme, dass die Normalplazierung des Hauptakzents auf dem ersten Bestandteil einfacher Komposita als morphologisierter Kontrastakzent verstanden werden kann. Wortstrukturell wird das damit begründet, dass der Akzent auf dem Bestandteil liegt, der weder morphologischer Kopf noch semantischer Kern der Gesamteinheit ist. Funktional entspricht dem, dass wie beim syntaktischen Kontrastfokus paradigmatische Reihen als mögliche Kookkurrenzklassen zum zweiten Bestandteil induziert sind. Der Akzent verweist auf semantische Alternativen in dem Sinne, wie Fokusstrukturen in der sog. Alternativensemantik behandelt werden. Als Folge dieser Akzentuierung kann gelten, dass im Normalfall der zweite Bestandteil zur Grammatikalisierung in Richtung Suffix oder Halbsuffix neigt. Grammatikalisierungen des ersten Bestandteils sind dagegen selbst paradigmatisch gebunden, sei das Grammatikalisierungsziel nun ein echtes Präfix oder sei es eine trennbare Partikel. Den Prototypen dreigliedriger Komposita mit festem Hauptakzent werden meist die Konstituentenstrukturen in (2a) oder (3a) zugewiesen.

354

Peter Eisenberg

Illustriert werden sie zuerst mit zwei Gruppen von Beispielen, die den Akzent auf dem linksperipheren Bestandteil haben. (2)

(3)

a. [['ab] c] b. Haustürschlüssel, Blumenkohlsuppe, Lebenshaltungskosten

Schadenersatzforderung,

a. ['a [b c]] b. Männerwohnheim, Fingerhandschuh, Altersteilzeit, sonnenwende

Sommer-

Beide Fälle können als Fortschreibung der Grundregel für einfache Komposita angesehen werden: Akzentuiert wird der erste Hauptbestandteil, der erste vom ersten usw. Was die Konstituentenstruktur betrifft, dürfte (2) für das Deutsche präferiert sein. Linksverzweigende Komposita werden leichter gebildet als rechtsverzweigende. Sie haben verarbeitungstechnische Vorteile und enthalten außerdem den morphologischen Kopf als unmittelbare Konstituente der Gesamteinheit. In zahlreichen Fällen kann nun der Hauptakzent auch auf dem zweiten Bestandteil liegen, wobei wieder nach der Verzweigungsrichtung zu unterscheiden ist. Linksverzweigung kommt bei einem vergleichsweise seltenen Typ von erstem Hauptbestandteil mit markiertem Akzent vor, zum Beispiel Jahrtdusend. In unserem Zusammenhang geht es dabei allein um die Akzentplazierung und nicht um die allgemeinere Frage, ob es im Deutschen Komposita mit „umgekehrtem Determinationsverhältnis" gibt. In der Literatur wird diese Frage meist nicht formbezogen, sondern in Hinblick auf das semantische Verhältnis zwischen den Bestandteilen erörtert. Man fragt, ob mit Kakaopulver nicht Kakao mit bestimmten Eigenschaften oder mit Härtegrad nicht Härte in bestimmter Ausprägung gemeint sei (zusammenfassend Donalies 2000). Eine Klassifizierung der auf dem zweiten Bestandteil akzentuierten einfachen Substantivkomposita des Deutschen lässt sich möglicherweise in Analogie zu dem vornehmen, was Olsen (2000b) fiir das Englische vorschlägt. Olsen bindet den auch im Englischen markierten Zweitakzent an wenige semantische Grundrelationen: Die Zuschreibung von gleichgewichtigen Eigenschaften bei bestimmten Kopulativkomposita, lokale und temporale Modifikation sowie einige Typen von Prädikation wie ,besteht aus' bei Artefakten können zu markierter Akzentplazierung führen. Zumindest die drei zuerst genannten Gruppen scheint es auch im Deutschen zu geben (4).

Struktur und Akzent komplexer Komposita

(4)

355

a. Whiskysoda, Gintonic, Bitterlemon, Coxorange b. Nordwesten, Mittelamerika, Hessen-Nassau, Frankfurt-Oder c. Wochenanfang, Jahrhundert, Monatsmitte

Becker (1992) spricht bei allen und weiteren Gruppen wie zusammengesetzten Eigennamen (Meier-Leibniz) von appositiven Strukturen. Klar zu sein scheint, dass der Typus im Deutschen wesentlich restringierter ist als im Englischen, dass der Akzent längst nicht immer auf dem zweiten Bestandteil fixiert ist und dass pragmatisch begründete Lexikalisierung eine Rolle spielt. Anders werden sich Unterschiede wie in Südfrankreich vs. Norddeutschland oder Wochenanfang vs. Wochenende kaum erklären lassen. Eine weiterreichende Erfassung hätte einen größeren Bestand und verwandte Konstruktionen wie Whisky pur, Forelle blau, Berlin-Ost usw. zu berücksichtigen. Als erste Bestandteile komplexer Komposita führen die Stämme aus (4) zu Linksverzeigung mit Akzentuierung des zweiten Bestandteils wie bei den Beispielen in (5). (5)

a. [[a'b]c] b. Nordwestpassage, Norditalienkatalog, Jahrhundertwerk, Jahresendfigur

Wochenanfangspreise,

Sehr viel häufiger und für unsere Fragestellung wichtiger ist der rechtsverzweigende Typ gemäß (6). (6)

a. [ a ['b c]] b. Schienenersatzverkehr, Müttergenesungswerk, lungsplan, Stadtplanungsbüro

Schulentwick-

Zu vergleichen ist (6) zunächst mit (3). Eine Mindermeinung vertritt (3) als unmarkierten Fall bei Rechtsverzweigung (z.B. Benware 1987), während die Mehrheit, meist im Anschluss an Giegerich (1983 und 1985), (6) als unmarkiert ansieht (z.B. Löhken 1997, Jessen 1999, Hall 2000). Die Akzentplazierung gemäß (3) kommt nach dieser Auffassung nur zum Zug, wenn die komplexe Konstituente idiomatisiert oder wenigstens eindeutig lexikalisiert ist. (3) hat eigentlich den Akzent einfacher Komposita, d.h. die Akzentuierungsregel sieht nicht drei, sondern nur zwei Bestandteile. Für (6) als Normalfall wird die Verzweigungshypothese in Anschlag gebracht: Akzentträger ist die verzweigende Konstituente, (6) steht (2) gegenüber.

356

Peter Eisenberg

Für die Verzweigungshypothese sprechen gute Gründe. Wenn Linksverzweigung präferiert ist, muss Rechtsverzweigung besonders markiert werden, zumal ja der morphologische Kopf hier eingebettet ist. Der Akzent zeigt unter diesen Bedingungen die morphologische Hauptgrenze an. Liegt diese wie in (3) durch Lexikalisierung fest, kann der Akzent woanders liegen. Offenbar kommt (3) nur infrage, wenn [a b] nicht einen wesentlich höheren Lexikalisierungsgrad hat als [b c], und umgekehrt kommt (6) nur infrage, wenn [b c] nicht einen wesentlich höheren Lexikalisierungsgrad hat als [a b]. Dass die Hauptkonstituenten eines komplexen Substantivkompositums besonders markiert werden, ist auch aus der Untersuchung von Fugenelementen bekannt. Fuhrhop (2000: 209) bringt es so auf den Punkt: „Gegenüber dem Frühneuhochdeutschen hat sich die Struktur der Komposita geändert: damals waren komplexe Substantive als Erstglieder ungewöhnlich, heute sind sie typisch für Fugenelemente". Es wird sogar die These vertreten, dass an sich (d.h. bei einfachen Komposita) ungrammatische Fugen bei komplexen akzeptabel werden (z.B. * Autosverkäufer aber Spielzeugautosverkäufer, Heide Wegener, pers. Mitteilung). Auch insofern fügt sich die Deutung des Akzents in (6) als Mittel zur Sicherung einer weniger präferierten Struktur gut in die Gesamtgrammatik der Komposita ein. Bei Linksverzweigung wird die Hauptgrenze eher durch ein Fugenelement angezeigt, bei Rechtsverzweigung eher durch den Akzent.

2.3. Morphologische Ambiguität bei nur einer Bedeutung? Wie gezeigt wurde, bindet zumindest ein Teil der Literatur strukturelle Ambiguität an prosodische und beide an Mehrdeutigkeit. Zwei Akzentuierungsmöglichkeiten sind Anzeichen für strukturelle Ambiguität, beide haben ihre Funktionalität in der Signalisierung von Mehrdeutigkeit. Das sollte bedeuten, dass zwei Wortbedeutungen vorliegen. Sehen wir uns an, wie diese Zusammenhänge in der Literatur dargestellt werden. Wiese (1996: 298) schreibt: „... identical strings of words may display different stress patterns depending on their structure. In a compound such as Stadtplanungsbüro, the main stress on Stadt or Planung dissolves the structural ambiguity of these words ...". Zwei der drei zu

Struktur und Akzent komplexer Komposita

357

korrelierenden Größen werden genannt. Mit doppelter Akzentuierbarkeit sind zwei Strukturen verbunden, von Bedeutungen ist nicht die Rede. Für das englische Beispiel baggage sizing device heißt es bei Olsen (2000: 904; ähnlich 2000a: 57f.): „In complex compounds of three and more elements, different constituent structures are often possible, each reflecting a slightly different compositional reading". Es handele sich einmal um ,a device for baggage sizing', zum anderen um ,a sizing device for baggage'. Hier stimmt alles zusammen, nur wird in Hinsicht auf Bedeutung sehr vorsichtig formuliert. Olsen spricht nicht einfach von zwei Bedeutungen, und in der Tat hat man es schwer, sie sich vorzustellen. Einen Schritt weiter geht Mötsch, indem er einen Hinweis auf den Bildungstyp gibt, der hier offenbar die Hauptrolle spielt (1999: 386): „Sehr häufig sind Bildungen, die eine Ergänzung zum Verb enthalten: Betonmischmaschine, Abfallsortieranlage, Schönschreibübung. Hier ist eine Doppelanalyse möglich: Abfallsortieranlage ,Anlage, die Abfall sortiert' ,Sortieranlage, die Abfall bearbeitet'". Die Beispiele sind nicht reine Substantivkomposita, sondern haben einen verbalen zweiten Bestandteil. Auch sind sie nicht durchweg auf zwei Weisen akzentuierbar. Trotzdem gehören sie in den gegebenen Zusammenhang. Die gerade aus der Literatur zitierten wie die Beispiele in (6) haben in zweiter Position den Stamm eines Verbalabstraktums oder jedenfalls eines Substantivs, bei dem man von einer spezifischen Argumentstruktur der Art sprechen kann, wie sie bei Verbalabstrakta durch Argumentvererbung entsteht. Was die Bedeutung betrifft, sind wir bei Mötsch in ähnlicher Lage wie bei Olsen. Auch bei Mötsch ist nicht von zwei Bedeutungen, sondern nur von einer „Doppelanalyse" die Rede. Machen wir einen weiteren Versuch, zu einer systematischen Doppeldeutigkeit dreigliedriger Komposita zu gelangen. Es sind ja leicht Fälle konstruierbar, in denen jeweils zwei der drei Bestandteile hochgradig lexikalisiert sind, z.B. Haustür und Türklinke. In aus ihnen kontaminierten Wörtern gibt es aber offenbar nur eine Möglichkeit der Akzentuierung und folglich Strukturierung, nämlich die von (2). Haustürklinke, Sachbuchrücken, Hosentaschenmesser sind linksverzweigend, so wie nach den oben gemachten Präferenzaussagen erwartbar. Analysierbarkeit allein reicht für die Möglichkeit von Rechtsverzweigung nicht aus. Vielmehr tritt sie bei bestimmten Klassen von zweiten Be-

358

Peter Eisenberg

standteilen auf, bei denen sie dann aber nicht zu einer nur dieser Struktur entsprechenden Bedeutung führt.

3.

Morphologie und Syntax

3.1. Interface Wenn es um das Verhältnis von Morphologie und Syntax geht, kann sehr Unterschiedliches thematisiert sein, insbesondere kann entweder ein grammatiktheoretischer oder ein sprachtheoretischer Aspekt des Verhältnisses im Mittelpunkt stehen. Man vermeidet falsche Frontstellungen, wenn der Unterschied präsent bleibt. Als grammatiktheoretisch verstehe ich die Frage, ob Wortstrukturen insgesamt oder fur bestimmte Wortklassen, etwa die nach produktiven Mustern gebildeten, denselben Prinzipien unterworfen sind wie syntaktische Strukturen. Ein Ansatz dieser Art kann etwa postulieren, dass Wortstrukturen im selben oder in einem vergleichbaren Sinn wie Phrasenstrukturen mit X-bar-Formaten zu beschreiben sind. Begründet und durchgeführt ist das für die Rektionskomposita des Deutschen in Rivet (1999). Weintrinker, Wetterbeobachter oder Umweltverschmutzer werden aus einer X-bar-Struktur abgeleitet, die den ersten Bestandteil als direktes Objekt zum Verbstamm und das Nominalisierungssuffix als Kopf der übergeordneten NP enthält (7). (7)

[ \er]NO [[wetter]NoNP [beobacht]Vo ]VP]Np

Durch zweimalige Kopfbewegung und Inkorporation entsteht eine zugehörige Oberflächenstruktur. Ein grammatiktheoretisches Argument für die Vorgehensweise ist, dass die charakteristische Eigenschaft von Rektionskomposita „auf die regulären Mechanismen der Θ-Rollen Zuweisung zurückgeführt werden" kann, was etwa auch bedeutet, dass für Argumentvererbung „nicht ein eigener grammatischer Mechanismus postuliert wird". (Rivet 1999: 312). Was man als syntaktische Repräsentation ansetzt, ist nun am einfachsten in zweifelsfrei morphologischen Einheiten zu realisieren. Die Ableitung deverbaler Substantive vom Typ Trinker, Beobachter, Verschmutzer ist problemlos. Die Ableitung von Rektionskomposita ist

Struktur und Akzent komplexer Komposita

359

weniger problemlos. Ein Grund dafür ist die Frage, wie die komplexe Einheit zu klammern wäre. Enthält sie einen Verbstamm wetterbeobacht, der dem Suffix er inkorporiert wird, oder einen Substantivstamm Beobachter, dem Wetter inkorporiert wird? Eine primär grammatiktheoretische Argumentation kann in einem solchen Fall auf ihre Festlegungen zurückgreifen, zum Beispiel auf die Verzweigungshypothese. Einen Verbstamm wetterbeobacht gibt es zwar nicht, aber es gibt linksakzentuierte Komposita wie Phrasendreschmaschine. Postuliert man Linksverzweigung, lässt sich ihre Akzentuierung „korrekt voraussagen" (Rivet 1999: 324). Der Akzent folgt nicht mehr der Verzweigung, sondern diese dem Akzent. Das Argumentationsdilemma ist doch gravierend, wenn prototypisch morphologische Einheiten wie die er-Substantive von einer syntaktischen Basisstruktur leichter ableitbar sind als die am Rande zur Syntax liegenden Rektionskomposita. Schon deshalb kann grammatiktheoretisch auch ganz anders argumentiert werden. Es gibt sogar gute Gründe, Inkorporation als Mechanismus zur Ableitung morphologischer Einheiten selbst am Übergangsbereich von Morphologie und Syntax abzulehnen (z.B. Olsen 1999). Ein sprachtheoretischer Zugriff auf das Verhältnis von Morphologie und Syntax fragt zuerst nach Übergangserscheinungen. Er setzt den Unterschied zwischen morphologischer und syntaktischer Strukturiertheit voraus, kommt aber nicht überall mit ihm zurecht. Ganz so, wie bestimmte morphologisch einfache Wörter nicht einen phonologisch determinierten, sondern einen kompositalen Akzent erhalten, gibt es Wörter, deren Bestandteile sich in mancher Hinsicht nicht wie morphologische, sondern wie syntaktische Einheiten verhalten. Der vielleicht bekannteste Fall dieser Art ist die teilweise schwierige Unterscheidung von syntaktischen Attributen und morphologischen Modifikatoren bei Komposita. Und weiter fungiert das Determinans eines Kompositums nicht nur selbst wie ein syntaktisches Attribut, sondern es kann auch Kern für syntaktische Attribute sein. Bei klarer Trennung von Morphologie und Syntax ergeben sich dadurch „falsche Bezüge" wie in verregnete Feriengefahr. Die Vermischung von Morphologie und Syntax gilt als grammatische Entgleisung, Normaussagen und grammatiktheoretische Festlegungen treffen sich. Genauere Untersuchungen ergeben in diesem wie in vielen vergleichbaren Fällen, dass auch ein Übergangsbereich seine Prototypen

360

Peter Eisenberg

haben kann. Unter bestimmten Bedingungen ist das Determinans in Kernfunktion weitgehend unauffällig, die Gesamtkonstruktion akzeptabel. In Ausdrücken wie die Absturzursache des Flugzeugs, der Befreiungsversuch von zwanghaften Strukturen oder ein erhöhtes Spannungsverhältnis muss man zweimal hinsehen, um sie als grammatisch falsch einzustufen. Offenbar legen sie eine Interpretation mit „umgekehrtem Determinationsverhältnis" nahe, was sogar zur Suspendierung ihrer morphologischen Strukturiertheit fuhren kann. Die Frage ist, ob es außer semantischen auch morphologische Eigenschaften der Wörter gibt, die ihre Strukturiertheit unsicher machen (Fabricius-Hansen 1993). Das Morphologie-Syntax-Übergangsfeld zu untersuchen ist etwas anderes als Syntax in prototypisch morphologischen Domänen zu verankern. Im folgenden wird in groben Umrissen skizziert, warum und wie die doppelte Akzentuierbarkeit von Komposita als Übergangserscheinung gedeutet werden kann.

3.2. Fokusexponent Im Berliner Tagesspiegel (19. September 2001, S. 14) findet sich unter der Überschrift Zeichensprache auf der Stadtautobahn ein Bericht, der mit den Sätzen in (8a) beginnt. (8)

a. Vor Baustellen, Staus, Unfällen, Nässe oder anderen Problemen auf der Stadtautobahn A 100 warnen jetzt 19 neue elektronische Verkehrsschilder. Auf speziellen Stahlbrücken gingen die zusätzlichen , Verkehrsbeeinflussungsanlagen' - so der amtliche Begriff - gestern zwischen Hohenzollerndamm und Oberlandstraße in Betrieb. b. Zur Verbesserung der Lichtverhältnisse auf der Stadtautobahn A 100 wurden zwischen Hohenzollerndamm und Oberlandstraße 250 Peitschenleuchten mit einer Leistung von je 25 Watt installiert. Auf speziellen Stahlbrücken gingen zusätzlich 19 Verkehrsbeeinflussungsanlagen in Betrieb. c. Beim Stadtautobahnbau hat man zur Bewältigung des ständig steigenden Verkehrs jahrelang auf einen weiteren Ausbau durch Vermehrung der Fahrstreifen gesetzt. Neuerdings geht

Struktur und Akzent komplexer Komposita

361

man einen anderen Weg und versucht, mit Verkehrsbeeinflussungsanlagen weiterzukommen. Die Texte (8b) und (8c) wurden mit dem Ziel konstruiert, eine je spezifische Akzentuierung von Verkehrsbeeinflussungsanlage zu erreichen. Und in der Tat akzentuiert eine deutliche Mehrheit der befragten Personen in (8b) auf dem ersten und in (8c) auf dem zweiten Bestandteil. Viel unsicherer ist die Akzentuierung in (8a), jedenfalls im Rahmen der bisher geringen Anzahl von Befragungen. An einer systematischen Fundierung der unterschiedlichen Akzentplazierung in den drei Beispielen besteht u.E. kein Zweifel. Als Deutung ausgeschlossen werden darf wohl die Annahme, in (8b) liege Linksverzweigung, in (8c) Rechtsverzweigung und in (8a) morphologische Ambiguität vor. Ein theoretisches Postulat hätte sich verselbständigt, schon weil eine Korrelierbarkeit von struktureller und lexikalischsemantischer Mehrdeutigkeit in keiner Weise gegeben ist. Wie die Beispiele präsentiert sind, legen sie eine Korrelation von Akzentplazierung und Informationsstruktur nahe. Durch den Wortakzent wird ein Fokusexponent ausgezeichnet. Funktional ist ein Kontrastfokus realisiert, der innerhalb einer morphologischen Domäne gebunden ist. Von den zur Kodierung informationsstruktureller Verhältnisse verfügbaren Mitteln Intonation und Reihenfolge steht innerhalb morphologischer Domänen nur ersteres zur Verfugung. Generelle Bedingung fur die Realisierbarkeit eines Kontrastfokus ist, dass das Kompositum in einer kontrastfähigen Konstituente steht. Im Rahmen der übergeordneten Topik-, Fokus- oder Themastruktur kann sie ganz unterschiedliche Funktion haben. Die vorgeschlagene Deutung der Akzentverhältnisse hat schon auf den ersten Blick eine Reihe von Vorteilen. Einmal passt sie zu der Sicht, der Akzent einfacher Komposita sei als grammatikalisierter Kontrastakzent anzusehen. Zweitens scheint sie verträglich zu sein mit weiteren Eigenschaften von Rektionskomposita, die auf Analogien zur Verbalphrase abheben. Zu untersuchen ist beispielsweise, wie weit Analogien in Hinsicht auf das Vorhandensein von Fokusprojektionen reichen. Ein Nachteil der Lösung besteht darin, dass die üblichen Annahmen über den Bau von Komposita für einen Teil von ihnen aufzugeben wären. Vielleicht ist es möglich, eine der Akzentplazierungen als die unmarkierte zu erweisen. In der Literatur erscheinen solche Wörter meist

362

Peter Eisenberg

mit Akzent auf dem zweiten Bestandteil, auch in (6) wurden sie so eingeführt. Man könnte dann erwägen, bei Rechtsverzweigung zu bleiben und die Akzentuierung des ersten Bestandteils aus anderen Eigenschaften abzuleiten, etwa aus der Rektionsbindung vom zweiten zum ersten Bestandteil. Wahrscheinlich ist es aber angemessener, so nicht zu verfahren. Komplexe Rektionskomposita scheinen gerade nicht auf binären Bildungsmustern zu beruhen. Dies und manches andere bleibt zu klären. Das betrifft auch die empirische Basis, auf der man operiert. Haben wir es mit einem mühsam zusammengesuchten Haufen von Einzelfällen zu tun? Wie identifizieren wir den Typ, bei dem zweifache Akzentuierbarkeit systematisch gegeben ist oder gibt es mehrere Typen dieser Art?

3.3. Daten Zur Bearbeitung solcher Fragen werden wir es mit Daten aus dem Projekt TIGER versuchen, das Potsdamer Germanisten gemeinsam mit Projektgruppen an den Universitäten Saarbrücken und Stuttgart durchführen. Im Projekt werden Zeitungstexte fortlaufend und reproduzierbar morphologisch und syntaktisch einschließlich eines Zugriffs auf das ΘRaster von Verben annotiert, in einer Baumdatenbank niedergelegt und verschiedenen Abfragemöglichkeiten zugänglich gemacht. Das Korpus umfasst derzeit etwa 50.000 Sätze. Aus ungefähr 9.000 davon wurden die dreigliedrigen Substantivkomposita extrahiert, wobei wir die mit Abkürzungen, Bindestrich, Klammern u.ä. als Sonderfälle ausgeschlossen haben. Es bleibt eine Liste mit etwa 600 Einträgen, aus der (9) den Anteil des Buchstaben Α wiedergibt. (9) A dven ts kränz b inden Allgemeinkrankenhäusern Altersübergangsgeld Altfallregelung Altweibersommer A rbeitgeberverband A rbeitgebervereinigung A rbeitnehmerinteresse

Arbeitszeitgestaltung A rbeitszeitregel Α rbeitszeitverkürzung A rchitekturzeitschrift Armeehubschrauber A rmeerundfunk Α rtenschutzkonferenz Arzneimittelagentur

Struktur und Akzent komplexer Komposita

Arbeitnehmerseite Arbeitnehmervertreter Arbeitnehmervertretung A rbeitsbeschaffungsprogramm Α rbeitskräftewanderung Arbeitslosenanteil Arbeitslosengeld Arbeitslosenhilfe Arbeitslosenproblem A rbeits losenrate Arbeitslosenversicherung Arbeitslosenzahl Arbeitsmarktforschung A rbeitsmarktpolitik Arbeitsplatzabbau A rbeitsplatzangebot Arbeitsplatzdynamik

363

Arzneimittelbehörde Arzneimittelkomitee Arzneimittelkommission Α temweginfektion A tombombenversuch Atomkraftwerk Atomtestfrage Atomtestgegner Atomversuchsreihe A ufsichtsratschef Aufsichtsratsentscheidung Aufsichtsratsvorsitzende Augenblickskontrolle Α ußenamtschef Autobahnbau Α utofirmengipfel Automobilbau

Den wirklich repräsentativen Buchstaben scheint es hier wie sonst nicht zu geben, aber wir nehmen das schiefe Bild, das durch einige häufig wiederkehrende erste Bestandteile zustande kommt, inkauf. Dadurch erscheint der Anteil von etwa 20% Kandidaten fur doppelte Akzentuierbarkeit viel zu hoch. Er liegt jedoch überall bei einigen Prozent, d.h. man hat durchaus die Möglichkeit, zu großen Zahlen von Komposita zu kommen. Andererseits erkennt man schon an den wenigen Beispielen, dass Wörter mit denselben oder analogen Bestandteilen hinsichtlich des Akzentverhaltens nicht einheitlich sind. So scheint der Akzent bei Arbeitnehmer- weniger beweglich zu sein als bei Arbeitgeber-, bei Arbeitnehmerseite (und bei Arbeitgeberseite) dürfte er trotzdem fest bzw. nur als echter Kontrastakzent verschiebbar sein. Wo der Akzent überhaupt verschiebbar ist, liegt auch eine Form von Rektionskompositum vor, vielleicht mit der einen Ausnahme Autofirmengipfel. Weitere Vermutungen und vorläufige Schlüsse sollen an dieser Stelle nicht zum besten gegeben werden, wohl aber die Gewissheit, dass man nur mit sehr viel größeren Datenmengen und einer methodisch (nicht grammatiktheoretisch!) gut begründeten Hypothese über „doppelt akzentuierbar" weiterkommen wird.

364

Peter Eisenberg

Literaturangaben Becker, Thomas 1992 Compounding in German. Rivista di Linguistica 4: 5-36. 1993 Back-formation, cross-formation, and „bracketing paradoxes" in paradigmatic morphology. In: Geert Booij and Jaap van Marie (eds.), Yearbook of morphology 1993, 1-25. Dordrecht/Boston/London: Kluwer Academic Publishers. Benware, Wilbur A. 1987 Accent variation in German nominal compounds of the type (A(BC)). Linguistische Berichte 108: 102-127. Donalies, Elke 1999 Das Kakaopulver im Moralkorsett des Ministerfreundes. Gibt es Substantivkomposita mit umgekehrtem Determinationsverhältnis? Zeitschrift für germanistische Linguistik 27: 322-343. Eisenberg, Peter 1995 Grammatik der geschriebenen Sprache als Symbolgrammatik. Ein Versuch an ausgewählten Themen der Morphologie. In: Vilmos Agel und Rita Brdar-Szabo (eds.), Grammatik und deutsche Grammatiken, 23-38. (Linguistische Arbeiten 330). Tübingen: Niemeyer. Fabricius-Hansen, Cathrine 1993 Nominalphrasen mit Kompositum als Kern. Beiträge zur Geschichte der deutschen Sprache und Literatur 115: 193-243. Fuhrhop, Nanna 2000 Zeigen Fugenelemente die Morphologisierung von Komposita an? In: Rolf Thieroff, Matthias Tamrat, Nanna Fuhrhop und Oliver Teuber (eds.), Deutsche Grammatik in Theorie und Praxis, 201-213. Tübingen: Niemeyer. Giegerich, Heinz J. 1983 Metrische Phonologie und Kompositionsakzent im Deutschen. Papiere zur Linguistik 28: 3-25. 1985 Metrical phonology and phonological Structure. German and Englisch. Cambridge: Cambridge University Press. Hall, Tracy Alan 2000 Phonologie. Eine Einfuhrung. Berlin/New York: Walter de Gruyter Haspelmath, Martin 2001 Understanding morphology. Leipzig: Max-Planck-Institut für evolutionäre Anthropologie. Hockett, Charles F. 1954 Two models of grammatical description. Word 10:210-231. Jessen, Michael 1999 German. In: Harry van der Hülst (ed.), Wordprosodic systems in the languages of Europe Berlin/New York: Mouton de Gruyter.

Struktur und Akzent komplexer Komposita

365

Löhken, Sylvia 1997 Deutsche Wortprosodie. Abschwächungsund Tilgungsvorgänge (Studien zur deutschen Grammatik 17). Tübingen: Stauffenburg. Mötsch, Wolfgang 1999 Deutsche Wortbildung in Grundzügen. Berlin/New York: Walter de Gruyter. Olsen, Susan 1999 Gegen Inkorporation. Papiere zur Linguistik 61: 75-91. 2000 Composition. In: Gert Booij, Christian Lehmann et al. (eds.), Morphology. Ein internationales Handbuch zur Flexion und Wortbildung, 897-916. (Handbücher zur Sprach- und Kommunkkationstheorie 17.1). Berlin/New York: Walter de Gruyter. 2000a Compounding and stress in English. A closer look at the boundary between morphology and syntax. Linguistische Berichte 181: 55-69. Raffelsiefen, Renate 2000 Evidence for word-internal phonological words in German. In: Rolf Thieroff, Matthias Tamrat, Nanna Fuhrhop und Oliver Teuber (eds.), Deutsche Grammatik in Theorie und Praxis, 43-56. Tübingen: Niemeyer. Rivet, Anne 1999 Rektionskomposita und Inkorporationstheorie. Linguistische Berichte 179:307-342. Vennemann, Theo 1990 Syllable structure and simplex accent in Modern Standard German. Papers from the Annual Regional Meeting of the Chicago Linguistic Society 26: 399^112. 1991 Skizze der deutschen Wortprosodie. Zeitschrift für Sprachwissenschaft 10: 86-111. 1991a Syllable structure and syllable cut prosodies in Modern Standard German. In: Pier M. Bertinetto, Michael Kenstowicz, and Michele Lopocaro (eds.), Certamen Phonologicum II, 211-243. Turin: Rosenberg and Sellier. Wiese, Richard 1996 The phonology of German. Oxford: Clarendon Press.

Warum wir zusammenschreiben nicht immer zusammenschreiben Präferenzgesetze im Schriftsystem Joachim Jacobs

Abstract I argue that word separation by spaces (WS) is governed by preference laws in the sense of Vennemann (1986, 1988). These laws are based on different universally valid criteria for word boundaries that may be in conflict with each other. Using methods of Optimality theory, I try to show that a specific ordering of three laws of this kind explains how such conflicts are resolved in Modern German and thereby accounts for a wide variety of German WS data, including hitherto unexplained variations in the WS of German particle verbs.

1. Einleitung Dieser Beitrag soll zeigen, dass die Spatiensetzung im Deutschen, die sog. Getrennt-ZZusammenschreibung (GZS), unter dem Einfluss von Präferenzgesetzen im Sinne Vennemanns (1986, 1988) steht. Ich werde deutlich machen, dass bisher noch nicht verstandene GZS-Variationen bei Verben mit trennbaren Bestandteilen (kurz: T-Verben) aus einer unabhängig begründbaren Ordnung von drei in Präferenzgesetzen verankerten Bedingungen folgen. Die Beispiele (1) und (2) illustrieren das Phänomen: (1)

(2)

a. b. c. a. b. c.

368

Joachim Jacobs

In Fällen wie (1) wird das Erstglied eines T-Verbs mit dem Zweitglied zusammengeschrieben, während in Fällen wie (2) mindestens ein Spatium interveniert.1 Diese variierende GZS von T-Verben wurde durch die Rechtschreibreform von 1998 nicht verändert. Die Reformer hatten offenbar nicht das Gefühl, dass hier eines der Phänomene unseres Schriftsystems vorliegt, denen es an „innerer Logik" 2 gebricht und die deshalb neu geregelt werden müssten. Das heißt aber nicht, dass sie die hier waltende Logik explizieren und das Phänomen erklären konnten. Der einschlägige Paragraph 34 der amtlichen Neuregelung begnügt sich vielmehr mit einer Aufzählung der Fälle, in denen zusammengeschrieben wird: „Partikeln, Adjektive oder Substantive können mit Verben trennbare Zusammensetzungen bilden. Man schreibt sie nur im Infinitiv, im Partizip I und im Partizip II sowie im Gliedsatz bei Endstellung des Verbs zusammen". Diese Darstellung stellt gegenüber der Behandlung des Phänomens in den GZS-Richtlinien der letzten Duden-Ausgabe vor der Reform (Duden 1991, R205) keinen Fortschritt dar, in der immerhin der Versuch einer Erklärung erkennbar ist: „Ein zusammengesetztes Verb wird getrennt geschrieben, wenn das erste Glied am Anfang des Satzes steht und dadurch besonderes Eigengewicht erhält [...] Dies gilt nicht für Infinitiv und Partizip". (Der erste Satz bezieht sich auf Fälle wie (2c), der zweite soll wohl verhindern, dass auch in Fällen wie (lc) getrennt geschrieben wird.) Mit der Rede vom „ersten Glied am Anfang des Satzes" ist offensichtlich die syntaktische Konstellation gemeint, bei der das Erstglied eines T-Verbs allein das Vorfeld von Sätzen mit Verbzweitstellung füllt (eine Option, die allerdings nicht bei allen T-Verben besteht).3 In diesem Fall muß das Erstglied tatsächlich ein gewisses „Eigengewicht" haben, nämlich mindestens einen Sekundärakzent tragen, der oft mit einem expressiv wirkenden, zum Finitum hin ansteigenden Ton gesprochen wird. Die Regelformulierung des Dudens legt nahe, dass die Vorfeldstellung bzw. das mit ihr verbundene „Eigengewicht" die Getrenntschreibung erklärt. Diese Idee findet sich auch bei Schaeder (1997: 169f), der sich dabei ähnlichen Überlegungen von Maas (1992: 181f) anschließt: „Verantwortlich für die getrennte Schreibung ist die syntaktische Regel, dass im Aussagesatz das finite Verb die zweite Stelle einnimmt". Schaeder und der Duden gingen anscheinend davon aus, dass man die Getrenntschreibung in den fraglichen Fällen durch die folgende Generalisierung erklären kann:

Präferenzgesetze

im Schriftsystem

369

Gl Wenn ein Ausdruck 4 allein das Vorfeld eines Verbzweit-Satzes füllt, wird er vom nachfolgenden finiten Verb getrennt geschrieben. Tatsächlich ist Gl eine ausnahmslos zutreffende Aussage über das heutige Deutsch.5 Aber Gl erklärt nichts. Vor allem trägt Gl nichts zum Verständnis der unterschiedlichen Schreibung in Fällen wie (2c) und solchen wie (la-c) bei. Warum werden Ausdrücke gerade dann, wenn sie allein das Vorfeld füllen, notwendigerweise getrennt geschrieben, vgl. (lc) vs. (2c)? Warum üben nicht auch Positionen im Mittelfeld einen entsprechenden Druck in Richtung Getrenntschreibung aus, vgl. (la-b)? Ebenso läßt Gl offen, wieso auch in (2a - b) zwischen Partikel und Verb mindestens ein Spatium steht. Eine Antwort auf die letzte Frage gibt allerdings die folgende nahe liegende Verallgemeinerung von G1:

G2 Die topologischen Felder deutscher Sätze werden in der Schrift durch Spatien voneinander abgegrenzt. Auch diese Generalisierung trifft ausnahmslos zu: An den Grenzen der traditionell für deutsche Sätze angenommenen topologischen Felder (Vorfeld, linke Klammer, Mittelfeld, rechte Klammer, Nachfeld) 6 stehen stets Spatien. Also muss auch in (2a-b) zwischen Verb und Partikel ein Spatium liegen, da zwischen beiden ja die Grenze zwischen linker Klammer und Mittelfeld liegt. Aber wie Gl gibt auch G2 keine Auskunft darüber, warum das Erst- und das Zweitglied eines T-Verbs nicht auch in anderen Fällen getrennt geschrieben wird, etwa wenn sie zusammen im Vorfeld oder im Mittelfeld stehen, vgl. (lc) bzw. (lb). Dieses Problem kann man offensichtlich nur durch eine generellere Behandlung der GZS im Deutschen lösen, die die Gründe für die Anoder Abwesenheit von Spatien auch in jenen Fällen identifiziert, in denen keine Feldergrenzen intervenieren. Tatsächlich wird sich im folgenden zeigen, dass eine adäquate Analyse der GZS innerhalb der einzelnen Felder (etwa innerhalb des Mittelfelds) auch die Daten in (l)-(2) erklärt. Die Feldergliederung spielt also für die deutsche GZS keine wesentliche Rolle, und die auf diese Gliederung Bezug nehmenden Generalisierungen Gl und G2, auf denen bisherige Erklärungsversuche basierten, beschreiben Epiphänomene anderer, allgemeinerer Zusammenhänge.

370

Joachim Jacobs

2. Theoretische Grundlagen Diese Zusammenhänge sind Präferenzgesetze im Sinne von Vennemann (1986, 1988). Nach ihm (vgl. 1986: 32f.) ist ein Präferenzgesetz eine universelle Aussage über natürliche Sprachen, in der festgehalten wird, welche sprachlichen Strukturen generell bevorzugt werden, indem sie bezüglich eines angegebenen Parameters besser sind als andere. Präferenzgesetze sind also Bewertungsregeln fur sprachliche Strukturen.7 Ihre allgemeine Form ist: PG Eine sprachliche Struktur S ist universell umso mehr bevorzugt, je mehr sie die Bedingung Β erfüllt. Im Kontext dieses Aufsatzes geht es um Präferenzgesetze für die Spatiensetzung, also um Gesetze der folgenden Art: Spat-PG Eine graphische Struktur ist universell umso mehr bevorzugt, je mehr sie die Bedingung Spat-B erfüllt. Mit graphischer Struktur ist die Graphemfolge gemeint, die einen natürlichsprachlichen Ausdruck repräsentiert, inklusive Zeichen- und Spatiensetzung;8 Spat-B ist eine Bedingung, die sich auf die Spatiensetzung in der graphischen Struktur bezieht. (Das wird in 3. präzisiert).9 Ein weiterer theoretischer Baustein, den ich zur Erklärung der in 1. aufgeführten Daten einsetzen werde, ist die in der Optimalitätstheorie10 (OT) entwickelte, in der Präferenztheorie Vennemanns und in Ansätzen der Natürlichkeitstheorie bereits angelegte11 Idee, dass sich einzelne Sprachsysteme durch das Gewicht unterscheiden, das sie den in den verschiedenen PGen enthaltenen Bedingungen bei der Bewertung sprachlicher Strukturen beimessen. Nach dieser Idee gelten PGe zwar grundsätzlich für alle natürlichen Sprachen (sie tun es ja per definitionem), sie wirken sich aber nicht in allen Sprachen in gleichem Maße aus. Vielmehr unterscheiden sich Sprachen durch verschiedene Rangordnungen der in den PGen enthaltenen Bedingungen. Eine solche Ordnung Bi » B2 » ... Bn legt das für die jeweilige Sprache anzunehmende relative Gewicht der Bedingungen von PGen fest und schreibt damit auch vor, wie die häufig auftretenden Konflikte zwischen den Bedingungen zu lösen sind. Auf der Grundlage einer solchen für eine Sprache

Präferenzgesetze im Schriftsystem

371

L anzunehmenden Ordnung der Bedingungen von PGen können die in L zulässigen - oder optimalen - von den nicht-zulässigen Strukturen nach einem bestimmten Berechnungsmodus abgegrenzt werden. Damit weist die OT einen Weg, wie aus einer auf PGen beruhenden universellen Sprachsystemtheorie Theorien einzelner Sprachsysteme - also Grammatiken - gewonnen werden können.12 Entsprechend werde ich zunächst drei Spat-PGe identifizieren, und zwar anhand von Überlegungen, die unabhängig von der GZS-Variation bei deutschen T-Verben sind. Danach werde ich zeigen, dass eine bestimmte Ordnung der Bedingungen dieser Spat-PGe die Daten in (l)-(2) ohne zusätzliche Annahmen und ohne Bezug auf die Feldergliederung (vgl. 1.) erklärt sowie auch die sonstigen im zentralen Bereich der heutigen deutschen GZS bestehenden Verhältnisse erfasst. Durch die Konzentration auf den zentralen Bereich soll das Problem vermieden werden, dass seit der Rechtschreibreform in Deutschland zwei GZS-Systeme in Gebrauch sind, die Alt- und die Neu-GZS. In der großen Mehrzahl der Fälle besteht jedoch kein Unterschied zwischen den in der Alt- und in der Neu-GZS vorgesehenen Schreibungen, denn die Neuregelung hat die Spatiensetzung nur punktuell verändert. Es gibt also einen großen von der Reform unberührten Bereich der deutschen GZS, und dieser bildet die Kern-GZS, auf die ich mich hier konzentrieren will. In ihr manifestieren sich die Gesetze, die im Lauf der Entwicklung des deutschen Schriftsystems zum selbstverständlichen Maßstab richtiger Spatiensetzung geworden sind. Wie schon in 1. angemerkt, ist auch das hier zu untersuchende Phänomen Bestandteil dieser Kern-GZS, also sowohl der Alt- als auch der Neu-GZS. Im übrigen beschränken sich die Abweichungen der Alt-GZS von der Kern-GZS, wie ich in Jacobs (in Vorb.) zeige, auf eine Anzahl von Irregularitäten, d. h. von Schreibungen, die nicht den einschlägigen Gesetzen (expliziert als eine Ordnung von in Spat-PGen verankerten Bedingungen, s.o.) folgen und deshalb einzeln im Lexikon festgehalten werden müssen. Für die NeuGZS muß dagegen ein anderes, quantitativ komplexeres System angenommen werden, das zudem eine nicht in einem Präferenzgesetz verankerte (also unnatürliche) Bedingung enthält, vgl. ebd. Diese Systemveränderung ist es, die bei vielen Kritikern den Eindruck erweckt hat, die Neu-GZS sei kontraintuitiv und verstoße gegen „wesentliche Grundsätze der Zusammenschreibung". 13 Für die Fragestellung dieses Aufsatzes spielt das jedoch keine Rolle.

372

Joachim Jacobs

3. Wortgrenzen und Spatiensetzung aus präferenztheoretischer Sicht Die Annahme, dass die GZS durch mehrere in Spat-PGen verankerte Faktoren beeinflusst wird, die im Konflikt stehen können, mag zunächst überraschen, geht man doch für das Deutsche und viele andere Sprachen davon aus, dass die Spatiensetzung einem einzigen, einfachen Gesetz gehorcht: Grundgesetz der Spatiensetzung Spatien markieren die Grenze zwischen im Text benachbarten Wortvorkommen.14 Doch die multifaktorielle Natur der GZS ist hierin bereits angelegt. Die Lage von Wortgrenzen kann nach verschiedenen grammatischen Kriterien bestimmt werden, und diese Kriterien führen nicht immer zum selben Ergebnis.15 So können phonologische Kriterien für Wortgrenzen - in jüngerer Zeit im Zusammenhang mit dem sog. phonologischen Wort diskutiert16 - im Widerspruch zu (morpho-)syntaktischen Kriterien stehen, und beide können mit semantischen Kriterien in Konflikt geraten. Darüber hinaus konvergieren auch Kriterien derselben Ebene nicht immer. Einige Beispiele hierzu. Phonologisch koinzidieren Wortgrenzen u.a. mit Grenzen für den Einflussbereich bestimmter Silbenstrukturgesetze, etwa für den von Vennemanns (1988) Silbenkontaktgesetz, nach dem eine Silbenfolge um so weniger präferiert ist, je geringer der Sonoritätsunterschied zwischen den Lauten an der Kontaktstelle ist. Dieses Gesetz wird ζ. B. in der Fuge deutscher Komposita systematisch ignoriert, vgl. [blu:t.Ro:t] blutrot statt [blu:.tRo:t], was ein Indiz dafür ist, dass diese Komposita in phonologischer Hinsicht aus mehreren Wörtern bestehen. In dieselbe Richtung weisen auch verschiedene semantische Kriterien, etwa die Tatsache, dass die Komposita in der Regel mehrere Autosemantika enthalten. Nach vielen morphologischen und syntaktischen Kriterien bilden deutsche Komposita dagegen nur je ein Wort. Dass sich auch Kriterien derselben Ebene widersprechen können, exemplifizieren die syntaktischen Kriterien Beweglichkeit und Wohlgeformtheit. Damit einer von zwei adjazenten Ausdrücken X und Y bewegt werden kann, muss zwischen X und Y eine syntaktische Konstituentengrenze liegen, also auf der terminalen Stufe der syntaktischen

Präferenzgesetze

im Schriftsystem

373

Struktur eine Wortgrenze.17 Außerdem unterliegen X und Y, wenn sie durch eine syntaktische Grenze getrennt sind, Bedingungen, die ihren inneren Aufbau beschränken. So müssen im Deutschen syntaktisch selbständige Nominalausdrücke, deren Kern ein zählbares Singular-Appellativum ist, einen Determinator enthalten, vgl. dass er den Weihnachtsbaum schmückt vs. *dass er Weihnachtsbaum schmückt. - Diese Kriterien können jedoch in Konflikt geraten: So wäre in dass er nicht Partei ergreift nach dem Beweglichkeitskriterium zwischen Partei und ergreift eine Wortgrenze anzunehmen, vgl. Ergreift\ er nicht Partei ex. Nach dem Wohlgeformtheitskriterium wäre aber wegen des fehlenden Determinators keine syntaktische Konstituentengrenze zwischen Partei und ergreift und damit auch keine Wortgrenze zu vermuten. Diese multifaktorielle und in sich widersprüchliche Determination von Wortgrenzen überträgt sich durch das Grundgesetz der Spatiensetzung auf die Spat-PGe. Diese beruhen nämlich auf solchen potentiell konfligierenden Kriterien für Wortgrenzen, wie sie gerade erwähnt wurden.18 Genauer gesagt ist die Bedingung Spat-B eines Spat-PGes (s. 2.) eine Aussage, die abhängig vom Zutreffen eines universell gültigen Wortgrenzenkriteriums WK auf Teilausdrücke Χ, Y ein Spatium zwischen und , d. h. zwischen den X bzw. Y repräsentierenden Teilen der graphischen Struktur, fordert oder verbietet. Spat-PGen haben also die folgende präzisierte allgemeine Form: Spat-PG Eine graphische Struktur ist universell umso mehr bevorzugt, je mehr sie die folgende Bedingung erfüllt: Wenn WK auf Χ, Y zutrifft, liegt zwischen und (k)ein Spatium. Das Maß, in dem eine graphische Struktur die jeweilige Bedingung erfüllt, ergibt sich aus der Zahl der in der Struktur enthaltenen Verletzungen der Bedingung. - Ein Beispiel für ein entsprechendes Spat-PG: Getrenntschreibung beweglicher Ausdrücke (Getr-Bew) Eine graphische Struktur ist universell umso mehr bevorzugt, je mehr sie die folgende Bedingung erfüllt: Wenn X oder Y bewegt werden können, liegt zwischen und ein Spatium. Dass damit eine tatsächlich geltende universelle Präferenz erfasst wird, müsste anhand einer großen Zahl von Schriftsystemen belegt werden, was in diesem Aufsatz nicht möglich ist. Ich bin mir aber sicher,

374

Joachim Jacobs

dass eine entsprechende Untersuchung tatsächlich ergeben würde, dass graphische Strukturen, die durch Getr-Bew als schlecht bewertet werden, sprachübergreifend die Ausnahme darstellen. (In vielen Schriftsystemen, wie dem des Englischen, fehlen sie völlig.) Andererseits würde sich zeigen, dass solche Strukturen gelegentlich eben doch vorkommen, etwa in deutschen Beispielen wie (la). Die Teilausdrücke finiter TVerben in Sätzen mit Verbletztstellung sind ja beweglich (vgl. etwa (2a-c)), werden aber im Deutschen, zumindest in der Kern-GZS (s. o.), stets zusammengeschrieben, wodurch graphische Strukturen entstehen, die nach Getr-Bew schlecht sind. Das traditionelle deutsche Schriftsystem scheint also dem Gesetz Getr-Bew ein geringeres Gewicht als bestimmten auf anderen Kriterien beruhenden Spat-PGen beizumessen und deshalb auch Konflikte zuungunsten von Getr-Bew aufzulösen. Aus der OT-Perspektive (s. 2.) heißt das, dass für die deutsche Kern-GZS eine Ordnung von Bedingungen Bi :» B2 ... » Bn anzunehmen ist, in der die Bedingung von GetrBew nicht den obersten Rang einnimmt und deshalb im Interesse der Einhaltung mindestens einer anderen Bedingung verletzt werden kann, so dass sich in Fällen wie (la) die Zusammenschreibung als optimal erweist. Später werden wir sogar sehen, dass die Bedingung, dass zwischen beweglichen Ausdrücken ein Spatium steht, in der für die KernGZS geltenden Ordnung einen so niedrigen Rang hat, dass sie gar nicht operativ ist, d. h. dass sie sich in keinem konkreten Fall auf die Spatiensetzung auswirkt.19

4. Im Deutschen operative Präferenzen der Spatiensetzung 4.1. Wortbildende Prozesse Das traditionelle deutsche Schriftsystem gewichtet also mindestens ein Spat-PG, das Zusammenschreibung fordert, höher als Getr-Bew. Dieses nun zu ermittelnde Spat-PG muß in Fällen wie (la) greifen, in denen die Teilglieder von T-Verben trotz Beweglichkeit nicht durch ein Spatium getrennt werden. Aber natürlich wäre es günstig, wenn sich aus diesem Gesetz die Zusammenschreibung auch in anderen Fällen ergäbe. Beides leistet das folgende Spat-PG:

Präferenzgesetze im Schriftsystem

375

Zusammenschreibung der Resultate wortbildender Prozesse (Zus-Wortbild) Eine graphische Struktur ist universell umso mehr bevorzugt, je mehr sie die folgende Bedingung erfüllt: Wenn X und Y durch einen wortbildenden Prozess miteinander verbunden wurden, liegt zwischen und kein Spatium. Zus-Wortbild expliziert die in der deutschen GZS-Forschung verbreitete Ansicht, fehlende Spatien seien ein Signal für „Zusammensetzungen", vgl. etwa die Vorbemerkung zur GZS-Neuregelung (Duden 1997: 35). Dabei ist „Zusammensetzung" offensichtlich nicht, wie in der Wortbildungslehre, als Synonym von „Kompositum" gemeint, sondern in einem viel weiteren Sinn. Nach Zus-Wortbild ist es generell das Wirken wortbildender Prozesse, das durch Zusammenschreibung signalisiert wird. „Wortbildender Prozess" ist dabei ein Terminus technicus, der a) die Anwendung morphologischer Bildungsmuster und b) die syntaktische Univerbierung umfasst, aber ζ. B. nicht Klitisierung (wie in Nehmen 'S doch Platz) oder rein semantische Inkorporation (wie Partei ergreifen).20 - Die ausfuhrliche Begründung von Zus-Wortbild in Jacobs (in Vorb.) kann ich hier auch nicht ansatzweise wiedergeben. Der Bezug dieses Spat-PGes zum Grundgesetz der Spatiensetzung sollte aber dennoch leicht erkennbar sein: Die Verbindung zweier Ausdrücke durch einen wortbildenden Prozess im erläuterten Sinn ist ein Kriterium par excellence für die Abwesenheit einer Wortgrenze - also nach dem Grundgesetz: für die Abwesenheit eines Spatiums. Entsprechend gibt es zweifellos eine universelle Präferenz zur Zusammenschreibung der Resultate wortbildender Prozesse. Hier konzentriere ich mich auf Möglichkeit a), also auf die Anwendung morphologischer Bildungsmuster. Eine geeignete theoretische Grundlage ist die paradigmatisch-relationale Morphologie.21 Nach ihr ist ein morphologisches Bildungsmuster die Charakterisierung einer rekurrenten Beziehung zwischen lautsprachlichen Zeichen, die in typischen Fällen Wörter sind. Solche morphologischen Muster können kreativ angewandt werden, also zur Bildung neuer Zeichen, oder analytisch, zur Ermittlung der Struktur oder Herkunft von Zeichen. Hier ist der kreative Modus relevant, der vorwärts bildend oder rückwärtsbildend sein kann. Im ersten Fall wird ein neues Zeichen im Nachbereich des Musters auf der Basis eines vorhandenen Zeichens im Vorbereich, im

376

Joachim Jacobs

zweiten ein neues Zeichen im Vorbereich auf der Basis eines vorhandenen im Nachbereich gebildet. Die meisten morphologisch komplexen Zeichen des Deutschen gehen auf eine vorwärtsbildende Anwendung morphologischer Muster zurück, etwa von Mustern der expliziten Derivation, der Komposition oder der Konversion. Vorwärtsgebildet sind etwa machbar aus machen nach dem Muster für bar-Adjektive, Reiseantrag aus Reise und Antrag nach dem Muster für Determinativkomposita und (das) Handys- Verleihen aus Handys verleihen nach dem Muster der Konversion infinitivischer Verb(alphras)en in Nomina. Die Rückwärtsbildung erweitert im heutigen Deutsch besonders den Bestand an komplexen Verben; so ist maschineschreiben auf der Basis von Maschineschreiben nach dem Muster der Infinitivnominalisierung rückwärtsgebildet worden, d. h. man hat in der Annahme, Maschineschreiben, Maschineschreiberi> sei in der diesem Muster entsprechenden Relation, aus dem im Lexikon vorhandenen Nachbereichselement das neue Vorbereichselement gebildet (und dabei ignoriert, dass Maschineschreiben tatsächlich keine Nominalisierung des komplexen Infinitivs ist, sondern ein Kompositum aus Maschine und dem einfachen nominalisierten Infinitiv Schreiben). - Nach Zus-Wortbild sollten so entstandene Zeichen zusammengeschrieben werden,22 und genau das sieht das deutsche Schriftsystem vor, vgl. vs. *, vs. *, 23 vs. *, vs. *e(V)-(sub)species are not known to me. The second small group of classificatory nouns refers to insects. Again, a large class should be expected, but so far only three nouns ar attested: (15) apa as au

'butterfly' 'grasshopper' 'flying insect; plane (of any size)'

Here the generic element of the nouns would be a-, which synchronically has the status of a prefix. Subspecies of butterflies - e.g., as manifested by color - do not show separate names, it is only the color adjective that is added: apa sei 'white butterfly', apayauso 'yellow butterfly' etc. Note that in contemporary Kilmeri au has gained a remarkable metaphorical extension of meaning in that it also means 'airplane'. The tertium comparationis between the two meanings 'flying insect' and 'airplane' is the type (not intensity!) of noise the two things produce as my teacher explained to me.12 Strategies of classification can also be observed in the field of natural groupings of plants. The following nouns refer to different (sub)species of bananas:

400

Claudia

(16) a. pewo

Gerstner-Link

'banana'

b. pailili, paipines, paipopo, paisol, panap, panoi, panpal, pasok, pasou, pumapup, penuwan, pesa, pewes, pewis, feles, femike c. buka, klapua, yawa, sukuru Pewo is the general term for banana. The common denominator of all nouns in (16) a. and b. is the morpheme pe-/pa~. The exact morphological status of this generic element is unclear; it can be the relic of a phonetically reduced noun, or it can be a former noun that does not occur any more as a free noun in contemporary Kilmeri. However, four of the terms for bananas do not fit into the pe-/pa- schema. Two of them are evidently Tok Pisin words: buka and klapua < karapua (cf. Mihalic 1971: 107). Buka and the Kilmeri word panoi may refer to the same widespread and tasty species of cooking bananas, but speaker opinions differ. Many speakers know and use buka only whereas panoi has disappeared from their vocabulary. Therefore buka is often understood as a Kilmeri word. The same holds for klapua; some people regard it as Kilmeri word, others as Tok Pisin. Yawa refers to an Indonesian banana species from the island of Java. The language status of sukuru is still unclear. The very existence of these four foreign terms for banana species, however, is strong indirect evidence for the classificatory device that underlies all of the other banana words. They constitute the 'endemic words' for the regionally endemic species of bananas. Edible mushrooms build another group of plants that are seen as a separate class. Consider the following nouns: (17) peir, pelap, pelili, penukap, pepual, pepuep, pepues, peisopo The common generic element is obviously pe-\ it is very likely a reduced form of pele 'leaf and should be analyzed as a prefix synchronically. The classificatory meaning is 'leaf-like (edible) plant', that is, here it is primarily the shape of the floral object that gives rise to the formation of the class. In Tok Pisin the cap of the mushroom is called lip 'leaf: this is certainly an interference of Kilmeri, because Mihalic (1971), still the best available dictionary of Tok Pisin, does not account for this meaning of lip. (It might be mentioned that the Kilmeri word for mould/mildew is alro what is totally different from the words in (17)). The largest class of plant terms is formed by the generic noun ri and a specific noun. Ri is the general term for tree and wood, that is, it is

Noun classification

and composition

in Kilmeri

401

simultaneously a count noun and a mass noun. Its extension of meaning is very broad: it refers to almost any type of wooden material as long as the specific function of the item is not in focus. (18) ri ri ri ri ri ri ri ri ri ri ri ri

'tree, bush, shrub; wood, firewood; stick; trestle' maro ruwi sap pep sawa il suo seli pol ali lop

'ironwood'

'wild mango' 'Tahitian chestnut' 'coconut palm' 'betel pepper' 13

/ / / / / / / / / / /

kwila amberoi ton wel mango wel galip diwai kokonas daka -

This is only a small selection of plants that are classified as ri by the speakers of Kilmeri; the list could be continued for long. However, the selection already shows some essential points. The functional value of the species of the r/-class is very different. There are trees whose timber is used (by the people themselves or for export: ri maro, ri ruwi, ri sap); there are trees whose fruits are eaten (ri pep, ri suo, ri sawa, ri il); there are trees or shrubs whose leaves have healing effects (ri pol, ri seli); there are smaller trees whose wood is particularly suitable as firewood (ri lop). The shape of the plant does not influence the classification; ri refers to strong, high trees as well as to many kinds of shrubs. What about the status of the second, specific elements of the nouns? At least some of them occur as free nouns or in other compositional environments: (19) a. ko seli lipeli ko pep lipeli

muli

b. ko yur pol riye pol yep am ar poll

Ί am looking for seli leaves' ni Ί am looking for ~ like ~ eat ton fruits' Ί see a bird's nest containing a clutch' 'there are no menstrual huts any more'

As free nouns the specific elements refer to the leaves or fruits of a tree. The compounds yur pol and pol yep have rather idiomatic mean-

402

Claudia Gerstner-Link

ings: 'a bird's nest with a clutch' and 'menstrual hut', respectively. These meanings can be explained as metaphorical extensions of the meaning of pol with respect to the healing, styptic effect of the betel peper leaves. When referring to a tree the compound structure of the nouns with generic ri is obligatory. Since the specific elements readily occur as free nouns even the generic-specific-constructions are orthographically represented as two words. As we have seen the rz-class is rather heterogeneous from a European perspective. However, this is certainly less the case from a Kilmeri point of view, since here one of the most relevant criteria for classification is 'food value'. With respect to tree-like plants this means that the terms referring to traditional staple food constitute a class of their own: the class consisting of all species of sago as opposed to all other tree-like floral objects: (20) due

'sago: swamp, palm, and product throughout the stages of processing before cooking'

duku 'sago palm species' duam " dikana " numomo " disou 'wild sago palm' The classifying generic element is obviously d-, yet this d- can be of double origin (the n- of numomo results from prenasalization and wordinternal assimilation of the onset of the syllables). The generic element d- may stem from due, a term of much broader generality than the other only species referring nouns. Then there would have been a coalescence of -ue with the initial vowel of the following specific noun. On the other hand, d(V)- may be the same as ri, since the regular correspondence d ~ r is diagnostic for the determination of cognates between Kilmeri and its neighboring language Pagi; and, furthermore, this correspondence d ~ r is still present in contemporary Kilmeri, e.g., as articulatory variation of loans from Tok Pisin: redim 'make ready' > rerim. It is clear, however, that the fusion of the two combined nouns is by far greater than with all the r/'-compounds; synchronically, these nouns referring to sago are not compounds, it does not even seem justified to speak of a prefix structure, rather, these nouns should be seen as one morphological unit. Whatever the exact morphological development was there is no doubt

Noun classification and composition in Kilmeri

403

that the terms referring to sago species receive a treatment that is conceptually different from that of terms referring to other trees and shrubs. Concluding this section of floral noun classification it should be mentioned that other cultivated plants, especially leaf vegetables and tuberous vegetables, are not involved in class formation processes. Thus it seems likely that the four floral classes discussed above are not joined by any further classes. Beyond the fields of fauna and flora it is the natural environment in terms of landmarks that shows at least traces of classification in toponyms. If one looks at names of rivers or creeks or of hills and mountain sites it cannot be overlooked that these landmarks, or habitable places, or sites of hamlets are named according to a special pattern. All places connected with water, be it a river or an artificially dug well, have pu 'water' as their first syllable and morphological element:14 (21) Puwani (main river of the area) Puel (second biggest river of the area) Puep, Pupual, Pusemo, Puwalae (small creeks in the area) Pule, Puniau (small lakes in the area) Pupulboli (a well dug by half-legendary ancestors of some clans) pupaek (waterhole in the sago swamps or bush) Similarly, many hillside places have names starting with o- from ol 'hill, mountain, riverbank': (22) Onume, Owanuf15 Osa, Onane, Osawa Omoi, Omula, Oiru Ofilaua

(small hills with hamlets) (names of the three table mountains of the area which are important landmarks) (names of villages) (name of a resting place)

Nowadays some of the terms refer to mountains, some to hills, some to villages; however, in former times, houses or hamlets were only built upon hills, and all the villages now located on the riverbanks have their origin on a closeby hill. Thus it seems quite reasonable that in former times locations were distinguished according to their main function: locations connected with water received pu 'water' as distinguishing and generic element of their names, and locations connected with home-

404

Claudia Gerstner-Link

steads received ol 'hill' as distinguishing and generic element of their names.

5. Morphosyntactic, semantic, and typological evaluation of noun classification in Kilmeri By way of summarizing the above discussion we note the following: there are three semantic fields where classification of nouns takes place: fauna, flora, and natural environment. Within these fields we find 14 criteria of classification that manifest themselves in generic nouns or generic prefixes as relics of generic nouns. It seems unquestionable that all these complex nouns, either referring to natural species or to landmarks, have a detectable morphosyntactic history in that they are diachronically compounds consisting of a generic or classifying noun and a specific noun. Lexicalized compounds result from (more or less) frequent or frozen syntagmatic structures, where often at least one of the elements lost its independent character. With respect to the lexicalized compounds dealt with in this paper most of the specific nouns don't occur free any more, that is, they have lost their status of independence. The (in)dependence of the specific part does not entirely depend on the degree of fusion or phonetic reduction of the first, generic part; even if the generic element is not reduced, the specific part normally doen't occur free in the language. As for the generic nouns, it is remarkable that they behave quite differently in terms of boundness. Seven of them occur as free nouns with a general meaning: wal 'fish', pial 'snake', ri 'tree', pele 'leaf, pu 'water', ol 'hill', and bi 'pig' (the only generic term that refers to a particular species of mammals). And the same number of (former) nouns occurs as bound form only: /'- 'bird-like', u'lizard-like', a- 'flying insect', suko- 'caterpillar', be(r)- 'edible grub', pe-/pa- 'banana', d- 'sago'. But note that wal, pial, pele and ol are phonetically reduced when occurring as generic elements in a complex noun.16 What does that mean for the synchronic status of the morphologically complex kind referring nouns? If we have either phonetic reduction or boundness, one should rather speak of generic prefixes than of generic nouns and therefore the resulting complex nouns are no proper compounds; however, in the remaining cases of unreduced and free

Noun classification and composition in Kilmeri

405

occurrence of the generic elements it seems justified to speak of proper compounds, that is, in all cases of composition with bi and ri. We turn now to the semantic function of the generic noun. What does the addition of a generic noun mean for a kind referring term? The generic noun establishes the referential setting for the specific noun. If one agrees that reference to an object is a combination of quality, quantity, and deixis, then the generic noun belongs to the qualificational domain of reference. 17 It sets up the quality frame into which the specific noun is embedded. In Kilmeri this referential structure holds for a large number of natural kind terms and for toponyms. The generic noun is rigid, there is no choice possible according to actual situational circumstance of reference. This is to be expected for 'inherent nature' classification, since the inherent nature of objects normally does not change. However, classification criteria in Kilmeri are not entirely reducible to inherent nature qualities. Quite often a functional component is present as well. Recall the case of bi: it refers to 'true animals' whose natural habitat is the ground as opposed to water or air. Yet it does not include wor 'dog' since there is the additional constraint of being man's prey and food. This is clearly a functional constraint. If nevertheless there is rigidity this implies that the functional value of the referents does not change; here: dogs are never eaten. Otherwise there is no distinction between living animals and meat, that is, bi also refers to the meat of the respective animals. Therefore the classifying generic noun does not distinguish between individual and mass reference in Kilmeri. Such a distinction would already be a step towards the transition from quality to quantity, because mass nouns and count nouns behave differently under quantification, e.g., under pluralization. Thus we can say that in Kilmeri the generic noun belongs entirely to the qualificational domain of the establishment of reference. From a typological point of view - which should comprise morphosyntactic and semantic considerations - the status of Kilmeri noun classification patterns is not self-evident. Since the classification is lexicalized it does not seem appropriate to speak of "classifier nouns" or "classifier prefixes". On the other hand, the Kilmeri strategy of classification is an elaborated system that should be distinguished from the mere possibility of combining class-nouns with specific nouns that we find in some languages. Mostly the occurrence of class-nouns in combination with specific nouns is rather restricted to a few words like, e.g., in Ger-

406

Claudia Gerstner-Link

man the combinations with -beere '-berry'. Furthermore, the addition of -bäum in Apfelbaum, Orangenbaum, Kastanienbaum, etc. distinguishes the tree from the fruit that is referred to by the specific noun alone. But this is the only area a class noun is systematically employed in German. Thus it would be inadequate to speak of systematic noun classification with respect to German. Although the referential distinction between 'tree' and 'fruit' is functionally parallel to the findings in Kilmeri, the Kilmeri strategy of classifying nouns cannot be reduced to this type of referential distinction. By contrast, we do have systematic classification in a wide section of terms for natural species. In order to obtain a suitable evaluation of the Kilmeri noun classification system we try to compare it with the Yidiny system of noun classification.18 Morphosyntactically, the classifying generic nouns in Yidiny are free nouns like the specific nouns, and both noun types can occur independently of each other, although particular semantic classes of nouns normally do occur together with a generic noun. These semantic classes are partly parallel to those of Kilmeri in that they encompass some subdomains of natural entities, but less than in Kilmeri: we have a 'bird'class, a 'frog'-class, an 'ant'-class, a 'tree'-class, and a 'vine'-class. On the other hand, Yidiny classification extends to selected classes of artefacts and natural objects; this holds for both, inherent nature and function/use generics. Kilmeri shows classification of objects of the natural environment in a very restricted sense only, namely with proper names. Note, however, that Yidiny bulmba 'habitable' - which would correspond to Kilmeri ol 'hill, mountain', because hills are the habitable places - is also often used with proper names (1977: 482). What about the Yidiny distinction between inherent nature and functional classifiers in Kilmeri? Kilmeri does not have two different series of classifiers or classification devices as Yidiny does, but nevertheless the conceptual difference is present: edible mushrooms form a separate class, edible grubs form a separate class, and all mammals that are potential prey or meat source form a separate class, which does not include dogs. Note that dogs are treated as a separate class of animals in Yidiny, too, since they are not eaten, and the word for dog can never be combined with the classifier minya 'edible flesh food' (1977: 490). As for the three classifiers of human beings that are found in Yidiny there are no lexical parallels in Kilmeri; yet there is one construction type that might be a remnant of a (formerly more widespread) classifying construction: yako

Noun classification and composition in Kilmeri

407

Samaria ako 'the Samaritan woman' where the first occurrence of yako could be seen as a classifier. The shorter version Samaria ako, however, is much more frequently used as translation of the Samaritan woman. Summarizing this comparison between the structure of classification in Kilmeri and Yidiny we can say the following: prima facie there is little similarity, since out of 20 Yidiny and 14 Kilmeri classificatory elements there are only two referring to the same classes, that is, only the 'bird'-class and the 'tree'-class we find in both languages. However, when looking at the classificatory strategy, analogies in the behavior of the systems become apparent. There is a correspondence in the lexical fields of classification; the only field that doesn't show any trace of classification in Kilmeri is the field of artefacts. But even with respect to Yidiny one should keep in mind "that these [the three classifiers for artefacts] only cover a fraction of the traditional artefacts known to the Yidinydi" (1977: 487). Furthermore, the functional domain of classification both languages treat alike by employing the criteria 'edible', 'meat source', 'habitable', and 'drinkable'. 19 Now given the fact that the details of classification strategies and of systems of languages with classifier nouns differ considerably,20 the analogies between Kilmeri and Yidiny suggest to view Kilmeri noun classification in the light of Yidiny noun classification. The idea is that the lexicalized classification system of contemporary Kilmeri represents a later stage of a Yidiny-like system of classification working by a combination of classifier or generic nouns with specific nouns that is still productive in Yidiny. Thus there would be external evidence for one particular aspect of the diachronic development in the morphosyntax of complex nominal expressions in Kilmeri.

Notes 1. Noun classification occurs in several Papuan languages. The systems so far known, however, differ from the structure of classification in Kilmeri in that they are systems of noun classes combined with certain types of agreement. Cf. Yimas (Foley 1991), Abu' (Nekitel 1986), and Manambu (Aikhenvald 1998). 2. The species of animals and plants named in the following section are not biologically identified; they are given by their Kilmeri words and, if possible, by an English translation. Very often, however, an English translation is not

408

3.

4.

5.

6.

7.

8.

9.

Claudia Gerstner-Link available and neither is a Tok Pisin translation. Thus, for the speakers of Kilmeri the transition from Kilmeri to Tok Pisin involves a loss of specificity in naming their natural environment. The term natural kind is a technical term in the semantic literature that is particularly concerned with the referential range of expressions referring to species. Here it seems reasonable to use this term because the domain of classification in Kilmeri is exactly the domain of natural kinds. The following data of the Kilmeri language are drawn from fieldwork that the author conducted over a period of three years from 2000-2002. An English translation of the Kilmeri terms is not generally available (cf. examples (2), (3), (5), (7), (12), (13), (14), (16), and (17)). If at all it is only the generic level names that can be given an English equivalent; the level of subspecification of endemic species is not accessible in English. Therefore throughout this article it is only the most general terms that receive a translation. Unfortunately, at the point of writing the biological terminology and systematization were not available to me. The form biras is not attested. Because of the synchronic accent pattern the syncope should have taken place before the process of lexicalization of the generic-specific construction was finished. Generally the cassowary is not seen as a bird in folk taxonomy; for instance, in Yidiny, an Australian language where nouns are classified by noun classifiers, it does not belong to the bird class (Dixon 1977: 485). Apparently it can be neglected that the young animals develop in an egg instead of the uterus; but note that for the speakers of Yidiny eggs count as flesh food (Dixon 1977: 489) and for Kilmeri speakers eggs are objects with the feature [+ANIM]: eggs are grammatically treated in the same way as animals and human beings with respect to agreement properties. See also the legend of the hornbill and the cassowary - a well-known traditional story throughout New Guinea - that 'explains' how the cassowary became a ground-roaming bird: the hornbill together with other birds broke the branch of a tree on which the cassowary sat, and so the cassowary fell down and broke his wings. Since then the smaller birds lived high up in the trees and the cassowary had to stay on the ground. That dogs are essential for hunting is shown by a traditional story, where the hunters, two brothers, go into the bush without their dogs. The missing dog gives rise to a transformation of the younger brother into a dog in order to ensure a successful hunt. (Stori Tumbuna, told by Andrew Wapi, September 2000 in Ossima.) The two exceptions warpas in (7) and walpop in (9) cannot be not explained along these lines; however, (8) states a phonotactic preference and not a strict phonotactic rule. My teacher uses the Tok Pisin phrase bikpela nem ('big name') to refer to generic nouns, and there is no doubt that she is well aware of the distinction between generic and specific nouns referring to natural species of different

Noun classification and composition in Kilmeri

10.

11.

12.

13.

14.

15.

16.

17.

18.

409

taxonomic levels and extension of meaning. So her proposal might be more than pure folk etymology. Note that the consonant cluster sk arises from the syncope of u in the first syllable su of sukolap\ it is the only case of a possible syllable-initial consonant cluster besides plosives + liquides. The same phonetic change holds for piaskil and piasko in (12). According to Mihalic 1971 spirits called "masalai" are good or indifferent spirits, but this definitely does not hold for the use of the word masalai by the people of the Vanimo area: for them masalai are evil and to be feared. Since frogs are also collected as food they should be expected to make up a further class of special animals as in some Australian languages. However, so far the data are not clear enough to draw any conclusions. The most frequent word for frog is kwer, which is evidently an onomatopoetic expression also referring to the late afternoon hours before dark when these frogs start making noise. Note that betel pepper and betel nut are two different species. Betel nuts are the fruits of a palm species (areca catechu, Tok Pisin buai), whereas betel pepper is a subspecies of pepper (piper betle, Tok Pisin daka). Usually the betel nuts are chewed together with the leaves of the betel pepper plant. This structure of hydronyms, namely a combination of the generic element 'water' with something else, is widespread and seems to be the semantic standard device of the naming of 'water-places'. In Europe it is often found as the element is related to Basque iz 'water': "Ich halte eine Herleitung der isNamen aus einer paläo-baskischen Wurzel mit der Bedeutung 'Wasser' für wesentlich besser als eine solche aus einer idg. Wurzel mit der Bedeutung 'sich heftig bewegen', und zwar sowohl aus semantischen Gründen als auch aus Gründen der Verbreitung." (Vennemann 1993: 472) In the light of a possible typology of hydronyms Vennemann's position concerning European isNames is certainly more reasonable than the older Indoeuropean one. When an /a/ is present in the second syllable of the name it can trigger assimilation of the initial o- to a-, such that these names occur with a twofold pronunciation as Osa ~ Asa, Owanuf ~ Awanuf etc. The presence of both free and bound classifying elements in one system or language is also attested in some Australian languages; Ngan'gityemerri, e.g., has nine class prefixes and ten classifiers, some of them available for the same classes (Sands 1995: 282). See also the case of Olgolo for the development of (optional) classifier prefixes from nouns (Dixon 1982: 207-210). This view of the semantic contribution of classifier nouns to the whole NP is also supported by Grinevald: "... noun classifiers and their predominantly material/essence semantics are operators of quality ..." (Grinevald 2000: 77). The Yidiny system is chosen because it is the most detailed description of classifier nouns available.

410

Claudia Gerstner-L ink

19. The classifier minya 'edible flesh food' as given in table 6.1. (1977: 483) seems to convey mass meaning rather than count meaning, yet from texts there is some evidence that minya can also be used independently as a count noun with the general meaning 'animal(s)' (see, for instance, the phrase minya maniiny jarrjii, line 62 in Dixon 1991: 37, or the phrase wurrgay/wurrgay janajanaany/minyaagu, line 9 in Dixon 1991: 55), that is, it is conceptually very close to Kilmeri hi 'pig; mammal; meat of an animal'. 20. Cf. Sands 1995; note that the reconstruction of classificatory elements for Proto-Australian concerns very general features and classifiers only (1995: 317).

References Aikhenvald, Alexandra Y. 1998 Physical properties in a gender system: an example from Manambu, a Ndu language of New Guinea. Language and Linguistics in Melanesia 27:175-187. 2000 Classifiers. A typology of noun categorization devices. Oxford: University Press. Dixon, R. M. W. 1977 A grammar ofYidiny. Cambridge: Cambridge University Press. 1982 Where have all the adjectives gone? and other essays in semantics and syntax. Berlin/New York: Mouton de Gruyter. 1991 Words of our country. Stories, place names and vocabulary in Yidiny, the aboriginal language of the Cairns-Yarrabah region. St. Lucia: University of Queensland Press. Foley, William A. 1991 The Yimas language of New Guinea. Stanford: University Press. Gerstner-Link, Claudia 2000-2002 Kilmeri: Fieldnotes. 2001 Review of Gunter Senft (2000): Nominal classification. LINGUIST List: Vol-12-645 (electronic publication) Grinevald, Colette 2000 A morphosyntactic typology of classifiers. In: Gunter Senft (ed.), 50-92. Mihalic, Francis S.V.D. 1971 The Jacaranda dictionary and grammar of Melanesian Pidgin. The Jacaranda Press: Queensland. Nekitel, Otto 1986 The sketch of Abu' Arapesh noun classes. (Papers in New Guinea Linguistics 24.) Pacific Linguistics A-70: 177-205.

Noun classification and composition in Kilmeri

411

Sands, Kristina 1995 Nominal classification in Australia. Anthropological Linguistics 37: 247-346. Senft, Gunter (ed.) 2000 Systems of nominal classification. Cambridge: Cambridge University Press. Vennemann, Theo 1993 Zur Erklärung bayrischer Gewässer- und Siedlungsnamen. Sprachwissenschaft 18: 425-483. Wilkins, David 2000 Ants, ancestors and medicine: a semantic and pragmatic account of classifier constructions in Arrernte (Central Australia). In: Gunter Senft (ed.), 147-216. Zavala, Roberto 2000 Multiple classifier systems in Akatek (Mayan). In: Gunter Senft (ed.), 114-146.

The puzzle of the autoantonymous argument role. Unraveling the polysemy of risk/riskieren Dietmar Zaefferer

1. Introduction It is a truism in lexical semantics that most lexemes have not just one meaning but several related ones. In other words, polysemy is rather the rule than the exception in the lexicon: Monosemy is the marked case, polysemy the unmarked one. Going up from word level to sentence level, we are faced with an interesting markedness reversal: Here monosemy is the unmarked case and polysemy, or ambiguity, as it is rather called in this context, the marked one. The reason for this is obvious: The co-occurrence of words co-constrains their sensible reading choices rather than multiplies them. So languages can have their cake and eat it too: They have both flexible, multi-purpose word meanings stored in economical lexica and sufficiently stable and unambiguous sentence meanings. Thus in order to understand the functioning of natural human languages (artificial ones tend to have different architectures), a proper understanding of the precise nature of polysemy and reading choices is of prime importance. The present case study is a rather modest contribution to the current debate and competition between different approaches in lexicology (cf. e.g. Behrens and Zaefferer 2002, Ravin and Leacock 2000) in that it aims less at sweeping generalizations than at a detailed description of an especially interesting case: The paradoxical polysemy of English risk and its German counterpart riskieren. Here are some pertinent data: (1)

a. John's death is a top priority for Mary. b. John's life is a top priority for Mary.

(2)

a. Mary tries to get rid of her illness. b. Mary tries to get rid of her health.

414

Dietmar Zaefferer

(3)

a. John makes every effort to avoid bankruptcy. b. John makes every effort to avoid solvency.

The sentence pairs under (1) through (3) have incompatible meanings, and since they differ just in one noun, it is likely that this incompatibility comes from these pairs of nouns. And indeed, death/life, illness/health, and bankruptcy/solvency are nice examples of antonyms in the sense of incompatible opposites,1 and this incompatibility of the word meanings seems to be inherited by the sentence meanings. The sentence pairs (4) through (6) are very similar, but here the incompatibilty of the same pairs of nouns is not only not inherited by the sentence, it is even neutralized to the extent that these sentence pairs can be considered synonymous or paraphrases. How come? The simplest answer is that the verb risk must be polysemous with one reading being in a way the opposite of the other. (4)

a. Mary risked death to save John. b. Mary risked her life to save John.

(5)

a. Mary risked illness. b. Mary risked her health.

(6)

a. In doing that, John risked bankruptcy. b. In doing that, John risked his own solvency.

This curious case of polysemy has been the subject of detailed investigations in the frameworks both of Frame Semantics (Fillmore and Atkins 1992) and of the Generative Lexicon (Pustejovsky 1995, 2000), but neither of these analyses can be considered satisfactory in the light of the data to be discussed below. It will turn out that the interest of this case lies less in its paradoxical behavior than in the precise nature of the corresponding semantic role. Before considering different solutions for this puzzle, a few words about terminology will be helpful. Antonymy, as mentioned above, is taken in this context in the sense of incompatibility of opposites, and autoantonymy means that different readings of the same lexical item are antonyms of one another.2 The notion of polysemy is not uncontroversial. In his highly influential book Lexical Semantics, Alan Cruse qualifies terms like polysemy and polysemous as "not entirely ideal" because they tend to imply a view of the lexeme as the primary semantic unit and of its different readings or lexical units as secondary, whereas he

The puzzle of the autoantonymous argument role

415

favors the opposite view with the (primary) lexical units forming a family, which is the (secondary) lexeme (Cruse 1986: 80). The following investigations do not take an a priori stand on this matter, but it will turn out that their outcome gives more support, at least in this case, for the view that Cruse disfavors.

2. Some data on risk and riskieren In their corpus-based study of the semantics of risk, Fillmore and Atkins arrive at eleven categories that "prove useful in a valence description of RISK and its semantic companions" (Fillmore and Atkins 1992: 81). Later, in the FrameNet project,3 categories of this kind are called Frame Elements, but since these elements belong to a decompositional analysis of lexical meanings, they will be called meaning factors or simply factors in the following. Here are the eleven factors that define the lexical field RISK according to Fillmore and Atkins:4 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

Chance: The uncertainty about the future, possibility Harm: A potential unwelcome outcome Victim: The individual affected by the harm Valued object: The endangered possession of the victim (Risky) situation: The state of affairs within which someone is at risk Deed: The act that brings about a risky situation Actor: The person who performs the deed (Intended) gain: The actor's hoped-for gain in taking a risk Purpose: What the actor intends to accomplish Beneficiary: The person for whose benefit something is done Motivation: The source of the actor's behavior

The data analyzed in the Fillmore and Atkins study are taken from a 25,000,000-word corpus and consist of 1,770 examples. The resulting observations comprise, among other things, the obvious noun-verb polycategoriality with the corresponding polysemy and two verbal uses, a Stative {run a risk) and a dynamic one (take a risk). The latter is by far the most interesting one and therefore the only one which will be further discussed. The dynamic verb risk occurs in eighteen valence patterns that form three groups. Four of the eleven factors mentioned above turn out to be

416

Dietmar Zaefferer

privileged: The Actor, invariably coded as the subject, the Harm and the Valued Object, invariably coded as the direct object, and the Deed, which may be coded as direct object but also as an instrumental phrase. None of the other factors can be coded by one of these major grammatical roles. Thus, the valence patterns form thus three groups, one with the Valued Object as direct object with eight subpatterns, one with the Harm in this role and six subpatterns, and one with the Deed as direct object and four subpatterns. This third group has not yet been mentioned in the introduction above and complicates of course the picture, but since it cannot be done away with as a marginal phenomenon it has to be considered as well. The following examples illustrate the three major groups just defined. (7)

a. Why did he risk his life for a man he did not know? (SU: Actor; DO: Valued Object;/or-PP: Beneficiary) b. Better be blamed for weak leadership than risk the cause by attempting too much. (SU: Actor; DO: Valued Object; by-PP: Deed)

(8)

a. The board was risking a liquidity crisis. (SU: Actor; DO: Harm) b. Rather than risk waking Peggy in searching for my pajamas, I crept... (SU: Actor; DO: Harm; in-PP: Deed)

(9)

No skipper or fisherman would risk the crossing. (SU: Actor; DO: Deed)

Turning now to the corresponding expressions in German, two major differences can be expected: the lack of a noun-verb polycategoriality (the German noun differs from the verb, the former being Risiko, the latter riskieren) and the absence of gerunds.5 Furthermore, the German verb does not seem to have a stative reading. Apart from that, the corresponding data for the verb are remarkably similar: They also clearly show three major groups of valence patterns. The German data are taken from the 'Projekt Deutscher Wortschatz' 6 (German Vocabulary Project) of the University of Leipzig. Its database includes 1336 examples with riskieren, 256 of which have been analyzed for the present investigation. The following examples have a slightly different annotation in that

The puzzle

of the autoantonymous

argument

role

417

the terms Valued Object, Harm, and Deed are replaced by the terms POSSESSION, DAMAGE, and TRIGGER, respectively: (10) a. Wir dürfen das bislang Erreichte nicht riskieren, ... 'We should not risk what we have obtained so far' ( S U : ACTOR; D O : POSSESSION)

b. Und einige - wie Soros - riskieren auch eigenes Geld. 'And some - like Soros - risk also some of their money' ( S U : ACTOR; D O : POSSESSION)

(11) a. Warum soll ich riskieren, dass sie sich verletzt, wenn ...? 'Why should I risk that she injures herself, i f . . . ? ' ( S U : ACTOR; D O : DAMAGE)

b. Um kein Defizit in diesem Jahr zu riskieren, ... 'In order not to risk a deficit this year, ...' ( S U : ACTOR; D O : DAMAGE)

(12)

..., ob es lohnt, ein Experiment zu riskieren. '..., whether it pays to risk an experiment' ( S U : ACTOR; D O : TRIGGER)

A numerical comparison of the readings of the direct object role gives the following figures: In 170 sentences the direct object codes the DAMAGE, i n 1 1 5 t h e POSSESSION, a n d i n 6 7 t h e TRIGGER. 7

3. First attempt: Three readings of risk A first and obvious way of dealing with the data outlined above is to assume three readings or variants of risk/riskieren and to characterize them by different meaning postulates, e.g. in the following way: MP 1 ( D A M A G E ) MP 2 ( P O S S E S S I O N ) MP 3 ( T R I G G E R )

Vy . ] [ (riski'(JC, y) Λ things_go_wrong') —» occur'(Y) ] Vx Vy • [ (risk2'0, y) Λ things_go_wrong') get_lost'(y)] Vx Vy • [ risk3'(x, y) -» perform'(x, y) ] VX

This shows that the first two readings are much more closely related to one another than to the third reading: If someone risks a liquidity crisis, this implies that a liquidity crisis occurs only under the additional condition that things go wrong. Similarly, when someone risks the cause by attempting too much, this implies that the cause gets lost only if

418

Dietmar Zaefferer

things go wrong. But when someone risks the crossing, he performs the crossing. So risk?, is unconditionally implicative with respect to the realization of the denotation of its direct object, risk\ is conditionally implicative in this respect, and riskj is conditionally implicative with respect to the realization of a loss of the denotation of its direct object. But the way the above meaning postulates are formulated, they overstate the differences between the three readings. Risk, for instance, is not only unconditionally but also conditionally implicative, namely with respect to an unexpressed DAMAGE. And the other two readings are not only conditionally but also unconditionally implicative, namely with respect to an unexpressed TRIGGER. SO a more complete formulation of the meaning postulates should show more clearly not only the differences, but also the commonalities of the three readings: MP 1' MP 2' MP 3'

Vx Vy 3z • [risk/(Λ:,y) —» [ perform'(x, ζ) Λ [ thingsgowrong' —> occur'(y) λ bad'(y)]]] Vx Vy 3z • [ risk,'(x, y) —» [ perform'(x, ζ) Λ [ things go wrong' —> get_lost'(y)]]] Vx Vy 3z • [ risk3'(x, y) —> [ perform'(x, y) Λ [ things go wrong' —> occur'(z) Λ bad'(z)]]]

This looks much better insofar as the common structure in all three cases is clearly visible: To risk is to do something with a conditional or potential bad outcome. The qualification of the potential outcome as bad became necessary because MP 3' would be too weak without that requirement: To risk the crossing is not just to do it with some event potentially occurring but rather with some damage potentially occurring. For the sake of schema uniformity, this qualification has been included in MP Γ as well, although one might consider it redundant there: A liquidity crisis is inherently bad and does not need the qualification. But this objection doesn't carry over to cases like (8)b. above: waking Peggy cannot be considered by any means inherently bad; our assumption that in this case it is bad obviously derives from our knowledge of the meaning of risk} But now we have a problem with MP 2'. Here we cannot add the same qualification since it is not the possessed object that is bad, but its loss. So we need an event variable for the qualification to hook up to. And this holds, of course, for the other cases as well: They in MP l1 and both the y and the ζ in MP 3' are event variables and not ordinary indi-

The puzzle of the autoantonymous argument role

419

vidual variables. So the second set of meaning postulates understates the ontological differences between the readings: The DAMAGE and the TRIGGER readings are relations between an individual and an event, and only the POSSESSION reading is a relation between an individual and an object. So it is about time for an ontologically richer representation, in other words, a Davidsonian recasting of the postulates: MP 1" Ve0 Vx Ve; 3 e2 •

[ risk|'(E0, x, ei) —» [ perform'(x, e2) A [ thingsgowrong' —» occur'(e;) A bad'(e/)]]] MP 2" Ve0 Vx Vy 3 e; 3 e20 [ risk2,(e(>, x, y) —>• [ perform'(x, e2) A [ things go wrong' —» get_lost'(e/,_y) A bad'(e/)]]] MP 3" Ve0 VJC \/e2 3 ei • [ risk3'(e0, x, e2) —> [ perform'(x, e2) A [ things go wrong' —» occur'(E/) A bad'(E7)]]] This set of postulates seems to give a pretty fair picture of the polysemy of risk. The overall structure reflects the common denominator, the opposition of DAMAGE and TRIGGER reading is reflected in the opposition of the direct object denoting either the potential bad outcome or the unconditional input action, and the closeness of the POSSESSION and the DAMAGE reading is spelled out as an analytical inference: Instances of the former are special cases of instances of the latter in that the bad outcome consists in the loss of the possession. There is, however, one problem left with this kind of representation and analysis, which could be considered a case of a fallacy of misplaced information:9 On the one hand it is obvious that the ambiguity lies in the semantic interpretation of the DO argument and not in the event itself, but on the other this semantics of the object is not explicitly represented in the formulas. So it seems worthwhile to make a second attempt not simply with three readings of risk but more precisely with three readings of the object role of risk.

4. Second attempt: Three readings of the object role of risk If the source of the polysemy of risk is not just anywhere in its meaning, but precisely in the semantic role associated with the direct object, a decompositional semantic representation is called for that allows for an explicit representation of semantic roles, in other words, a neo-David-

420

Dietmar Zaefferer

sonian representation. Here is a neo-Davidsonian restatement of the meaning postulates: M P 1"' \fe0 Vx \/e, 3 e2 • [ risk'(E 0 ) Λ ACTOR(e0, x) A DAMAGE(E0, ei)

-»

[ ACTOR(E?, x) A TRIGGER^, e2) A [ things_go_wrong' —> occur'(E/) ]]] M P 2"' V e 0 Vx Vy 3 et 3 e 2 • [ risk'(e 0 ) A ACTOR(e0, χ) A POSSESSION^, y) ->• [ ACTOR(E2, x) A TRIGGER(e0, e2) A DAMAGE(e0, e/) A UNDERGOER(e/, 7) A get_lost'(e/) A [ things_go_wrong' —> occur'(e/) ]]] MP 3"' Ve„ Vx Ve 2 3 e/ • [ risk'(e 0 ) A ACTOR(e0, x) A TRlGGER(e0, e2) [ ACTOR(e2, x) A DAMAGE(e0, et) A [ things g o wrong' —> occur'(ey) ]]]

In all three cases, the event type and the semantic role of the subject argument are the same, but the semantic role of the object changes between the three possibilities discussed above. Again, in a way, the first reading with its specialization, the second one, is shown to be in some kind of complementary opposition to the third: Either the damage is expressed and the triggering event must be inferred or the other way around. The negative qualification in the last conjunct of the formulas is not needed anymore since it is inherent in the DAMAGE role. Since it is not the event type itself that makes the difference between the readings but the semantic interpretation of the direct object role, the polysemy cannot be represented by predicate (i.e. event type) indices anymore, it must rather show up in the linking pattern, the matching of event structure and syntactic structure. The English verb risk and its German counterpart riskieren enter the following three linking patterns (LP stands for linking pattern, GR for grammatical relation, SC for semantic constituent): (LP 1) (LP 2) (LP 3)

GR SC

1. subject

2. direct object

3. predicate

1.ACTOR

2 . DAMAGE

3 . RISKING

GR SC

1. subject

2. direct object

3. predicate

1.ACTOR

2 . POSSESSION

3 . RISKING

GR SC

1. subject

2. direct object

3. predicate

1. ACTOR

2 . TRIGGER

3 . RISKING

There are, however, at least three problems with this analysis: (a)

The semantic roles associated with the direct object are far more specific than usual; especially the relation of the three roles with

The puzzle of the autoantonymous argument role

(b)

(c)

421

the Undergoer role standardly found in this position remains unclear; there are cases where the denotation of the direct object fits all three roles at the same time and where the distinction seems unnecessary; there are cases where the denotation of the direct object fits neither of the three roles.

The easiest solution for the first problem would be to say that risk and riskieren are simply exceptions to the general preference of direct objects for an Undergoer role interpretation. This preference, as is well known, is characteristic of accusatively organized languages and subsystems thereof, whereas in ergatively organized systems the secondranking grammatical role tends to be linked with an Actor role interpretation instead. However, the latter option would offer no way out either since, first, English and German are not good candidates for ergative languages, and second, at most the TRIGGER bears some kind of similarity to the Actor role. Furthermore, there are no other indications that the verbs in question are dramatically different from normal transitive verbs, and finally, the three postulated semantic roles seem to be rather idiosyncratic and there are few other places where they can be observed. The last statement, however, needs some qualification and does so again with respect to the TRIGGER or Deed role. A closer look at this role reveals a well-known semantic relation, namely nothing else than causation: The TRIGGER event is some activity of the actor that causes the state of danger, the possibility of a bad outcome. But causation is a characteristic rather of the Actor role than of the Undergoer role, as David Dowty has convincingly shown in his ground-breaking article on the preferred linking patterns of grammatical with semantic roles (Dowty 1991). Does this mean that we have found an ergative pattern in two predominantly accusative languages? The TRIGGER reading of risk/riskieren seems be worth closer scrutiny. So the next section will be devoted to this interesting reading.

422

Dietmar Zaefferer

5. Risk and riskieren as modality verbs First, it will be helpful to look at some more data. Examples (9) and (12) above are repeated here for convenience, (13)—(15), also corpus data, are added to indicate the bandwidth of the event types that can trigger a danger (in each example the object NP is given in roman). (9)

No skipper or fisherman would risk the crossing.

(12)

ob es lohnt, ein Experiment zu riskieren. '..., whether it pays to risk an experiment'

(13) Den Frauen wird an der Heidelberger Universitätsklinik empfohlen, bei mehr als drei herangereiften Follikel [sie!]10 eine Befruchtung nicht zu riskieren. 'The women get the recommendation from the Heidelberg University Hospital not to risk a fertilization if more than three follicles are ripe.' (14) I didn 't dare risk a pause to let that sink in. (15) Man muss als Schriftsteller das Recht haben, eine Sprache zu riskieren, die nicht im Schwange ist. 'One must have the right as a writer to risk a language that is not in vogue.' Example (13) is especially interesting insofar as the concept of fertilization associated with its direct object is one of the notions that could in principle play all three roles with respect to risk we have considered so far: In a context where a woman tries to avoid a pregnancy, to risk a fertilization will have the DAMAGE reading, where she wants to have a baby, it will have the (hoped for) POSSESSION reading, and where, as in (13), the context suggests that the fertilization would be the beginning of a risky pregnancy, there it has the TRIGGER reading. Example (14) is interesting for another reason, its strengthening effect: I didn't make a pause would be neutral, I didn't risk a pause would presuppose that making a pause would have been dangerous, and the way (14) is formulated, it emphasizes the danger and gives the impression of a considerable threat. This observation throws some light on risk itself: If the modal verb dare can be used to reinforce the effect of risk, then the question arises whether the latter could be some kind of modal verb it-

The puzzle of the autoantonymous argument role

423

self. And there is indeed, as I have shown elsewhere (Zaefferer 2001), a modal category that specifies the circumstances under which actions take place and which is sometimes expressed by a modal verb like English dare. French oser, Middle High German turren, and Mandarin gan are other cases in point. The same concept is sometimes expressed by nonmodal verbs like German wagen and sich trauen or by an adverb like bravelyI mutigerweise }x In order to determine the semantic role linked to the object complement, it is helpful to compare different realizations of the latter: Whether one dares a jump, or dares to jump, or dares jump, in all cases one performs a jump, one produces an instance of this kind of motion. Whether it is called product, or incremental theme, or effect, the semantic role in question is that of a rather prototypical Undergoer. Transferring this result to the TRIGGER role of risk, it turns out that although it was correct to consider its instantiations as causes, namely of some danger, this observation was incomplete because they are effects as well, namely products of their actors' intention to bring them about.12 So at least one third of the problem with the semantic role of the object of risk seems to be solved. Example (15) finally demonstrates that the assumption of an underlying predicate perform' in the first three sets of meaning postulates was a little too restrictive: One cannot literally perform a language that is not in vogue, but one can produce instances or tokens of such a language. And as this fits well with the analysis just presented, the latter seems to be confirmed. Still it is always safer to double-check every result. And indeed a second look reveals that although it is true that the TRIGGER is not only a cause but also an effect, it is rather hard to consider it as the effect of the risking. Consequently, the problem remains to be solved.

6. Further problems Two of the three problems with the three readings of the object analysis mentioned in section four above have not been discussed yet. The second one stated that

424

Dietmar Zaefferer

(b)

there are cases where the denotation of the direct object fits all three roles at the same time and where the distinction seems unnecessary.

What is meant here are cases like the following: (16)

... ein Stück über Leidenschaften und Wünsche und was man manchmal riskieren muß, um diese Wünsche zu erßillen. '... a play about passions and wishes and what one has to risk sometimes in order to fulfill these wishes' (SU: Actor; DO: ?)

(17) Die Ärzte wollen nichts riskieren. 'The doctors do not want to risk anything.' (SU: Actor; DO: ?) What one has to risk sometimes in order to fulfill some wishes may equally well be risky actions, or possible bad outcomes, or possible losses of possessions, and analogously for what the doctors are avoiding in (17). Of course, one could say that in cases like these the reading choice is irrelevant,13 but it would be more appealing to have one global meaning that covers in a way the other three. But this would mean that a fourth semantic role has to be looked for, on top of the three already problematic ones. The third problem stated in section 4 is, in a way, the opposite of the second one. It consists in the fact that (c)

there are cases where the denotation of the direct object fits neither of the three roles. Cases in point are the following:

(18) Captain Prien risked treacherous tides and currents and a narrow passage to penetrate British defenses and torpedo the battleship Royal Oak. (SU: Actor; DO: ?) (19) She risks the hazards of the highway to catch a glimpse of him as he addresses wayside crowds. (SU: Actor; DO: ?) Neither treacherous tides and currents nor a narrow passage are actions that could trigger a danger nor could they possibly be bad out-

The puzzle of the autoantonymous argument role

425

comes or possessions to be lost, and the same holds for the hazards of the highway. Is it necessary therefore to look for a fourth and a fifth reading of the object role? Instead of further pursuing this little attractive option, let us have a closer look at (19). Here the head noun of the object NP denotes a concept that is closely related to the concept of risk itself, the concept of a hazard. Webster's Ninth Collegiate Dictionary defines the fitting reading of hazard as 'source of danger', confirming thus the impression that (19) represents a reading with a cognate object of risk. It would be easy to dismiss (19) as marginal, but a quick glance at the internet shows that cognate object uses of risk are far from exceptional, witness for instance (20): (20) My father risked the danger of being labeled tionary ,..14

counter-revolu-

Before considering what moral can be drawn from this somewhat surprising observation, it will be useful to recall the findings of an often cited and highly influential contribution to lexical typology.

7. Interlude: Talmy's Law and participant co-encoding expressions It is not an exaggeration to claim that the history of modern decompositional lexical typology goes back to 1985 when Leonard Talmy's groundbreaking article Lexicalization patterns: Semantic structure in lexical forms (Talmy 1985) appeared. Comparing the lexical coding of motion events in a sample of typologically quite heterogeneous languages, Talmy found that, alongside with the superconcept of motion itself, four additional factors are of prime importance: The entity that is in motion (its exponent, often also called theme), the background against which motion takes place, the path along which motion takes place, and the manner or cause of motion. Here are the factors together with their labels and corresponding abbreviations: 0. The 1. The 2. The 3. The 4. The

motion exponent of motion background of motion path of motion manner or cause of motion

superconcept exponent ground path manner/cause

S Ε G Ρ Μ

426

Dietmar Zaefferer

What Talmy discovered, among other things, is a very remarkable preference law of lexical typology that could be called, and it seems only fair to do so although he never formulated it in these words, Talmy's Law: (TL) In natural human languages, monomorphemic expressions for motion events tend to co-encode with the concept of motion itself exactly one additional factor of motion. (TL) entails, among other things, that languages where the shortest translation of the English verbs walk or enter would correspond literally to something like 'move on one's feet' or 'move into' are dispreferred, since no additional factor is co-encoded with motion, and so are languages with monomorphemic verbs that translate as 'roll downwards over grass as or like a rock', since more than one additional factor is co-encoded with the superconcept. Walk, on the other hand, is well-behaved with respect to (TL), since it co-encodes with the superconcept only manner of motion. Schematically and using the abbreviations introduced above, with '+' joining intramorphemically co-encoded factors, this could be represented as follows: [S + M]. The coding of path information or of the factors exponent and background of motion, on the other hand, is left to other components of the clause, e.g. a pronoun and a preposition as in he walked in. Here is a schematic representation of this case with '&' joining factors encoded in different morphemes: [E & [S + Μ] & P]. Enter conforms with (TL) in a different way: It co-encodes with the motion its path ([S + P]), leaving to other elements the coding of exponent and manner (e.g. he entered walking easily [E & [S + Ρ] & M]) or other factors. Talmy's typological distinction between what he calls verb-framed and satellite-framed languages, between languages like Spanish on the one hand and English on the other, the former preferring path verbs in the coding of motion events whereas the latter prefer manner verbs, has spurred a large discussion (cf. e. g. Slobin 2000 for a modified Whorfian interpretation of this opposition and Papafragou et al. 2001 for an opposing view), and it is possibly due to this discussion that other interesting aspects of Talmy's findings have been neglected. One of them are motion verbs of the type [S + E], where the exponent of the motion is co-encoded in the motion verb. Cases in point are German kugeln 'to

The puzzle of the autoantonymous argument role

427

move as or like a sphere over a plane, to roll' or English rain 'to move as or like rain, to fall as a distributed collection of small items'. Both verbs have causative readings as well and the concepts they express could be called exponent co-encoding motion concepts. Rain and its counterparts in other languages show interesting patterns of polymorphy, i.e. polyvalency and polysemy, as well as linking mismatches, which I have studied elsewhere (Zaefferer in print), but they also have, and that is why they are mentioned here, cognate arguments, as a logical consequence of their decompositional set-up: If the exponent of the motion is already encoded in the root of the motion verbs, then any overt expression of it must be a kind of cognate argument. Examples (21)-(24) demonstrate that in English the exponent of the rain motion can surface both as cognate subject and as cognate object and that it can consist of the prototypical rain drops or of other matter. Note that confetti in (23) and (24) denotes a cognate concept insofar as it cannot characterize confetti in any kind of state, say glued together, but only as a distributed collection of small items. (21) (22) (23) (24)

It is raining a hard rain. The rain is raining all around. (Robert Louis Stevenson)15 It rained confetti as the Big Apple preparedfor the big night. ... where confetti rained at the end of the game ...

Now the question arises if this kind of decompositional analysis of verbs with cognate arguments as being of the type [S + E] (or more generally [S + X] with X being any participant role) is restricted to motion verbs, and the answer is obviously negative. Fight in (25), live in (26), die in (27), dream in (28), play in (29) are not verbs of motion and even dance in (30) does not behave in the relevant aspect as a verb of motion in that its object does not denote the exponent of motion. (25) (26) (27) (28) (29) (30)

to fight a battle to live a life to die a death to dream a dream to play a game to dance a dance

The next question that comes up is the following: Is it plausible that Talmy's Law is restricted to monomorphemic expressions of motion or

428

Dietmar Zaefferer

could it be valid for other concepts as well? Here the answer is much less obvious, but it seems a viable hypothesis to assume the following generalization of Talmy's Law: 16 (TL + )In natural human languages, monomorphemic expressions for events tend to co-encode with the most general event type itself exactly one additional factor of that event type. Based on this hypothesis, we can venture a factorial analysis of verbs of performance and production like the ones in the examples above; let us call them verbs of tokenization: 0. The 1. The 2. The 3. The

tokenization actor of the tokenization event type tokenized instrument of the tokenization

superconcept actor event type instrument

S A Τ I

All six verbs in (25)-(30) are well-behaved with respect to (TL+), since they co-encode with the superconcept only the tokenized event type, schematically [S + T], Shovel in shovel a ditch and author in author a book would be other well-behaved English verbs of tokenization, being of type [S + I] and [S + A], respectively.

8. Third attempt: One reading of the object role of risk Equipped with the lessons to be learned from the preceding section on preferred codings and participant co-encoding expressions, we can now return to the cognate object reading of risk and ask for the relation between this reading and the other readings discussed so far. The first observation in this respect is that what has been illustrated by examples (16) and (17) above and characterized as having a global meaning coincides precisely with the cognate object reading: For instance, not to risk anything means simply not to produce a dangerous situation or a token of risk. And the second observation is that all the other readings considered so far, the three main readings as well as marginal readings like the one exemplified in (18), can be considered partial characterizations of the event type central to the cognate object reading. So the somewhat surprising result of these lengthy investigations into the precise nature of

The puzzle of the autoantonymous argument role

429

the polysemy of risk!riskieren is that on some level of analysis there is no polysemy at all! Risk and riskieren are monosemous tokenization verbs with an Actor subject and an Undergoer cognate object. Having said this, the obvious diversity of interpretations does not have to be explained away, but it is treated on another level, the place where the most obvious problem with cognate arguments is handled: the level of redundancy reduction. By definition, participant co-encoding event expressions do not require overt expression of the participant in question as long as it is not further specified. Therefore, in this case, zero or expletive realization of the argument is optimal and other realizations are marked (cf. Zaefferer in print and the poetic example (22) above). Pustejovsky (1995, 2000) speaks here of shadowing, which he defines as "the relation between an argument and the underlying semantic expression, which blocks its syntactic projection in the syntax" (2000: 69). He goes on to claim that "the grammatical expression of a particular argument NP [...] is licensed only when the NP stands in a subtype relation with certain 'implicit semantic content' of the verb" (2000: 69). This is, of course, a slight exaggeration, as we know from poetic usage, but it is doubtlessly correct as a strong preference. And this explains why the most simple realization of the direct object only rarely surfaces: To risk a risk or, in German, ein Risiko riskieren are avoided, especially since there are alternatives with a much lower degree of redundancy, namely to take a risk and ein Risiko eingehen. So one way of redundancy reduction is weakening the content of the verb and the other one is strengthening the content of the argument: As soon as the risk is specified, its expression is required or, more precisely, the expression of that part of its expression that carries the specification. But sometimes more redundant formulations occur as well. The German corpus offers two cases in point: (31)

Um „ nicht das leiseste Risiko " beim noch laufenden Bieterverfahren für das Olympiastadion zu riskieren, ... 'In order to risk "not even the slightest risk" in the ongoing bidding process for the Olympic stadium, ...'

430

Dietmar

Zaefferer

(32) . . o h n e schwerwiegende gerichtliche Auseinandersetzungen und das Risiko hoher Schadensersatzforderungen zu riskieren. '..., without risking serious legal controversies and the risk of high compensation claims.' The second case is a little difficult in that it can also be interpreted as a case of second-order risk (the possible bad outcome consists of controversies and of a second risk whose possible bad outcome are compensation claims), but the first one is an unambiguous illustration of the fact that sometimes a formulation escapes the redundancy reduction razor: the author could have chosen the option of weakening the verb meaning by writing einzugehen 'to take' instead. Note, by the way, that the alternative option of reducing the object expression to the specification of the risk is not available in this case, since the degree of the risk can only be expressed as an attribute. Maybe this is the reason why it is not included in the list of Frame Elements in Fillmore and Atkins (1992) although it is a prime element of non-linguistic discussions about risks. As mentioned in note 4 above, that list of Frame Elements is incomplete in one important respect, and this has to do with another shortcoming of the analysis: It does not define its core notions. If one defines a risk for some participant as a state with a non-negligible probability of a DAMAGE to this participant, then it becomes clear that there are at least four scales involved in the logic of risk: To risk a lot can mean (a) to do many risky things (TRIGGERS), (b) to expose highly valued POSSESSIONS to the danger of getting lost, or (c) to face a considerable DAMAGE. By contrast, a high risk is one with (d) a high probability that the DAMAGE will occur. Now, what exactly is the claim of the one-reading-of-the-object analysis of risk1 It consists of two subclaims: 1. The lexical syntax and semantics of risk!riskieren is characterized by the following linking pattern, where COGNATE TOKEN is the specialization of the UNDERGOER role for cognate object verbs: (LP 4)

GR

1. subject

2. direct object

3. predicate

SC

1. ACTOR

2 . COGNATE TOKEN

3 . RISKING

2. The instantiations of the direct object role of risk!riskieren are characterized by the following principles: 2.1. Quantified forms of the direct object may denote general cognate tokens.

The puzzle of the autoantonymous argument role

2.2.

431

Other forms of the direct object metonymically characterize a special cognate token; • in the unmarked case they do so by naming one of its three main factors: (a) the DAMAGE, i.e. the possible bad outcome of the dangerous situation; (b) the POSSESSION, i.e. the possessed or hoped for good that is threatened and whose loss constitutes a damage; (c) the TRIGGER, i.e. the action that causes the dangerous situation; • in the marked case, i.e. if none of the above applies, they do so by naming (d) another factor of the dangerous situation.

9. A special problem: Heterogeneous lists as objects of risk In his article 'Lexical shadowing and argument closure', James Pustejovsky (2000) treats the relation between the HARM and the POSSESSION role as a case of complementary shadowing in that the expression of one precludes the expression of the other. In this section, some data will be discussed that seem to contradict this analysis, but first a few words are in order about the relation between Pustejovsky's analysis and the one advocated here. At first glance they seem rather unrelated in that the present approach sees shadowing effects between the verb and its object, whereas Pustejovsky talks about shadowing between different roles associated with the object. But the connection is closer than it seems. With respect to the HARM and the POSSESSION role, Pustejovsky argues that "if these roles stand in complementary distribution then they are probably just different expressions of the same underlying role" (2000: 78). So, in a way, the view presented here could just be regarded as a generalization of Pustejovsky's idea: Not only the HARM and the POSSESSION role are conceived of as different expressions of the same underlying role, but also the TRIGGER and, in fact, all other roles. This means, of course, that the idea of the complementary distribution has to be given up since now the complement of say the POSSESSION role would be the union of all other non-general roles. But is the obser-

432

Dietmar Zaefferer

vation of the complementarity of roles correct? There are three examples in the German data that seem to be at variance with it. (33) ..., dabei Stürze im Zielspurt bei Tempo 70 km/h, die Gesundheit, ja das Leben riskieren, ... thereby risking crashes during the final sprint at 70 km/h, their health, even their life, ...' (34) Sie riskieren Drohungen, Berufsverbote, oft sogar ihr Leben. 'They risk threats, job bans, often even their lives.' (35) Wollte man wirklich noch mehr Pleiten, noch mehr Arbeitsplätze riskieren? 'Would one really want to risk even more bankruptcies, even more jobs?' The first example, (33), is about cycling races and contains a list of three object noun phrases, the first one coding a DAMAGE, the other two coding POSSESSIONS. (34) lists two DAMAGES and one POSSESSION, (35) one DAMAGE and one POSSESSION. How are these examples to be understood? A closer look reveals that they are not necessarily about different aspects of the same risk, but about different related risks. In (33), falling off the bike in the final phase of a race is a DAMAGE in any case since it costs time; at the given speed it also involves at least painful injuries, but it can also threaten the health or even the life of the racer, hence the sequence of noun phrases codes three risks with three DAMAGES in increasing order from the minimum to the maximum. The structure of (34) is remarkably similar, except that the switch from DAMAGE to POSSESSION occurs later, between the second and the third noun phrase. (35) is different in that an increase can be assumed only if jobs are more important than avoiding bankruptcies, which is a plausible but not a cogent assumption. But apart from that, the two risks mentioned, one partially characterized by its DAMAGE, the other one by its POSSESSION, are interrelated but not identical: A bankruptcy may involve a loss of jobs, but only if the bankrupt entity involves jobs, and a loss of jobs may result from a bankruptcy, but also, for instance, from efforts to avoid one. This means that examples (33)—(35) are not counterexamples to Pustejovsky's claim that the DAMAGE and the POSSESSION role are in a relation of complementary shadowing. A counterexample would involve

The puzzle of the autoantonymous argument role

433

something like to risk death and one's life, which would go against the redundancy reduction razor and which cannot be found because according to our analysis to risk one's life is a metonymical short hand for to risk the loss of one's life, the object of which in turn is synonymous with the relevant reading of death. So Pustejovsky is right in this respect, but he is wrong in his explanation for the phenomenon. When he writes (Pustejovsky 2000: 77): "... the HARM role always indicates the privation of a possible POSSESSION role ...", then this is not quite correct. When the HARM consists, e.g., as in example (8) above, in a liquidity crisis, it is hard to restate this in terms of the privation of a possible POSSESSION (possession of the absence of a liquidity crisis?). Rather, it seems that the inverse statement makes more sense: The loss of a POSSESSION is always a DAMAGE - remember that POSSESSION means possessed or hoped for good - just as expressed in the meaning postulates MP 1"' and MP 2'" above. At least as interesting as the fact that examples (33)-(35) exist seems to be the observation that a similar combination of the TRIGGER with either the DAMAGE or the POSSESSION role occurs nowhere in the corpus. The explanation for this absence, if it is not simply due to the sample size, could be the availability of alternative more transparent ways of coding. In principle, sentence (8)b. above, repeated here for convenience, could be reformulated as (36), but the gain in brevity is more than compensated for by the loss in comprehensibility. (8) b. Rather than risk waking Peggy in searching for my pajamas, I crept ... (36) Rather than risk searching for my pajamas and waking Peggy, I crept ...

10.

Conclusion

The puzzle of the autoantonymous argument role of risk has been solved. It results from the combination of two facts: First, the cognate object of the verb in question is often only metonymically characterized by its possible bad outcome, and second, the latter consists often in the loss of some possessed (or at least hoped for) good, and this loss can be again metonymically characterized by this good alone. It is a perfect

434

Dietmar Zaefferer

analogon to the translation paradox encountered, for instance, in the French/German pair danger de mort!Lebensgefahr, where French characterizes the danger directly by its possible bad outcome and where German indirectly qualifies the danger by what gets lost in such a case, in other words, by the endangered good. In a way, it is the difference between an unergative construction that features the threatening entity and an unaccusative one that highlights what is threatened. But looking back at the course of this case study of polysemy it seems fair to say that the discoveries made on the way to this solution seem to be more interesting than the finding itself. It first turned out that what looked like a polysemous verb is a verb with a monosemous event type and a polysemous object role. And then it turned out that the systematic relatedness of the readings of this argument role is due to the fact that at one level of abstraction it is a monosemous argument role with different ways of characterization at a lower level of abstraction, some of them holistic, some of them metonymical, and one of them even second-order metonymical. And at the higher level of abstraction, the standard assumptions about prototypical Undergoers or Patients could be preserved. The detailed exploration of the uses of English risk and German riskieren has shown that both the Frame Semantics and the Generative Lexicon approach to the analysis of the former give only a partially correct picture: The Frame Semantics account is more complete, but it presents a mere enumeration of senses and overlooks the factor of degree; the Generative Lexicon analysis is more explanatory in its use of the conceptual apparatus of what it calls shadowing, but it gets the kind of shadowing wrong and overlooks another, higher shadowing and also the factor of degree. The new approach owes a lot to both of its predecessors, in fact it would have been impossible to develop it in such a short time without them. If all of its findings are correct, then an interesting case of verbal polysemy has been discovered, and it seems worthwhile to cast out the net and look for analogous or similar cases. Especially the aspect of the hidden cognate argument offers an interesting starting point for future investigations. In trying to combine the corpus-based degree of completeness of FrameNet with the detail of analysis and explanatory ambition of the Generative Lexicon, the present paper has given an example of what could develop into a new paradigm of polysemy research in

The puzzle of the autoantonymous argument role

435

special and lexical semantics in general. It will have served its purpose if it has spurred the reader's curiosity for things to come.

Notes * I am grateful to the members of my 2001/2002 Oberseminar, especially Andrea Schalley, for stimulating discussions on the subject, but very special thanks go to my most influential teacher, the honoree of this volume, who taught by example that in order to do interesting theoretical linguistics, it is sometimes necessary to take certain risks. 1. To be more precise, they are privative opposites in that they are asymmetrical insofar as the first member of each pair is conceived of as the result of losing the second, but not vice versa. The category of privative opposition goes back, of course, to Aristotle; cf. Horn (1989: 7). 2. This term seems to go back to Dennis Baron, who used it on page 73 of his 1989 book Declining grammar and other essays on the English vocabulary. 3. Fillmore 2000-2003. 4. Although this impressive list of factors looks rather over than undersized (most of the time Actor, Victim, and Beneficiary coincide, and the Purpose is to realize the Intended gain), it is incomplete in one important respect that will be discussed in section 8 below. 5. On the typologically 'exotic' features of English (and Celtic) cf. Vennemann (2001), where the gerund is discussed especially in the section 'The rise of the verbal noun' (pp. 353-356). 6. http://wortschatz.uni-leipzig.de/html/sitemap.html. 7. The attentive reader will have noticed that these three figures do not add up to 256 but to 352 examples. This puzzle will be resolved below. 8. The attentive reader will ask for the other two meanings: How do we know that waking Peggy is the DAMAGE? By exclusion: It cannot easily be the POSSESSION, and the TRIGGER is already coded in the PP in searching for my pajamas. 9. The nice term is borrowed and adapted from Barwise and Perry (1983: 38), who use it in a different sense, namely for the misled assumption that all the information in an utterance must come from its interpretation. 10. The quoted noun form is either a typo or it demonstrates the loss of inflectional endings in contemporary German; the correct form would be Follikeln. 11. Interestingly, the dual counterpart of this concept, expressible in adverbs like cowardlyIfeigerweise, does not seem to have a verbal coding, at least not in English or German. 12. Philosophically, it is a big problem whether intentions can cause actions, but that need not concern linguists more than the fact that physically the moon

436

13. 14. 15. 16.

Dietmar Zaefferer does not change in size, although people say that the moon is growing or waning. This is the solution of the puzzle mentioned in note 6: These cases were counted in all three categories. From: Sichos in English: Night Following Vov Tishrei, 5750 (http:// www.sichosinenglish.org/books/sichos-in-english/43/05.htm). From his collection Ά Child's Garden of Verses and Underwoods' (1913). The weak point in this generalization of Talmy's Law is, of course, the wording 'most general event type'. It is clear what is meant, namely something comparable to motion, but it will be hard to pin it down more precisely.

References Barwise, Jon and John Perry 1983 Situations and attitudes. Cambridge, MA: MIT Press. Behrens, Leila and Dietmar Zaefferer (eds.) 2002 The lexicon in focus. Competition and convergence in current lexicology. Frankfurt am Main: Peter Lang. Cruse, D. Alan 1986 Lexical semantics. Cambridge: Cambridge University Press. Fillmore, Charles J. (PI and Director) 2000-2003 FrameNet++: An On-Line Lexical Semantic Resource and its Application to Speech and Language Technology. http://www.icsi.berkeley.edu/~framenet/mainpage.html. Fillmore, Charles J. and Beryl T. Atkins 1992 Towards a frame-based lexicon: The semantics of RISK and its neighbors. In: Adrienne Lehrer and Eva F. Kittay (eds.), Frames, fields and contrasts: New essays in semantic and lexical organization, 75-102. Hillsdale, NJ: Lawrence Erlbaum Associates. Horn, Laurence R. 1989 A natural history of negation. Chicago/London: The University of Chicago Press. Papafragou, Anna, Christine Massey, and Lila Gleitman 2001 Motion events in language and cognition. In: Anna H.-J. Do, Laura Dominguez, and Aimee Johansen (eds.), BUCLD 25: Proceedings of the 25th annual Boston University Conference on Language Development, 566-574. Somerville, MA: Cascadilla Press. Pustejovsky, James 1995 The generative lexicon. Cambridge, MA: MIT Press. 2000 Lexical shadowing and argument closure. In: Yael Ravin and Claudia Leacock (eds.), 68-90.

The puzzle of the autoantonymous argument role

437

Ravin, Yael and Claudia Leacock (eds.) 2000 Polysemy: Theoretical and computational approaches. Oxford: Oxford University Press. Slobin, Dan I. 2000 Verbalized events: A dynamic approach to linguistic relativity and determinism. In: Susanne Niemeier and Rene Dirven (eds.), Evidence for linguistic relativity, 107-138. Amsterdam/Philadelphia: John Benjamins. Vennemann, Theo 2001 Atlantis Semitica. Structural contact features in Celtic and English. In: Laurel J. Brinton (ed.), Historical linguistics 1999. Selected papers from the 14th International Conference on Historical Linguistics, Vancouver, 9-13 August 1999, 351-369. Amsterdam/Philadelphia: John Benjamins. Zaefferer, Dietmar 2001 Modale Kategorien. In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher, and Wolfgang Raible (eds.), Sprachtypologie und sprachliche Universalien (HSK 20.1), 784-816. Berlin/New York: de Gruyter. in print Polysemy, polyvalency, and linking mismatches. The concept of 'rain' and its codings in English, German, Spanish, and Italian. In: DELTA - Documentagäo de Estudos em Lingüistica Teörica e Aplicada 18, Special volume on polysemy, edited by Heronides M. Moura.

Be brief and vague! And how Bidirectional Optimality Theory allows for verbosity and precision1 Manfred Krifka

1. The sad story of the metric system in America Given the beginnings of the United States of America, its sympathy with the French revolution and its rationalist attitude towards the institutions of society, one would have expected that it would have been one of the first nations to adopt the new metric system that was introduced in France in 1800. But the history of the attempts to do so is decidedly mixed. American Congress authorized the use of the metric system in 1866. In 1959, American measurements were defined in relation to the metric system. In 1968, the government ordered a study which was published three years later under the title "A Metric America: A Decision Whose Time Has Come". The year 1975 then saw the Metric Conversion Act, leading to the establishment of the US Metric Board. Amended in 1988, it resulted in the Metric Program, an organization founded to support the various federal agencies, which are required since 1991 to file an annual report on their efforts to change to the metric system. In spite of all these attempts, the United States of America is still the one major industrial nation that does not use the metric system as the predominant one. To this day, American schoolchildren have to count with miles that contain 1760 yards, yards that contain 3 feet, and feet that contain 12 inches. They have to memorize that an acre is 4840 square yards, and that a gallon contains 231 cubic inches. The costs of this are undoubtedly huge - they include, for example, the Mars mission of 1998 that failed because measurements were converted wrongly. Still, the general attitude of the American public towards the metric system is largely negative. There are websites with telling addresses like

440

Manfred Krifka

www.metricsucks.com, and it is not uncommon that common people suspect a secret communist, catholic, or Jewish plot behind the attempts to go metric. Why did the metric system not catch on? There are many reasons.2 But one that cannot be taken lightly is that certain well-intended public relation attempts intended to familiarize the American people with the metric system just did not work. Since the Metric Conversion Act, road distances in National Parks are often given in miles and kilometers. And since then, travelers encounter signs like the following one:

(1)

^ Eagle Pass 7 miles 11.265 km

It is not hard to see why road signs like (1) suggest that the metric system is something for intellectuals, or "rocket scientists", far too unwieldy for everyday purposes.

2. Problems with precision The problem with (1) leads us to the question: How much precision is enough? When can we stop being precise, relax and be a little vague? I won't have much to say about this in general, but I will have to say something about the relation of precision level and linguistic form. Assume that the distance between Amsterdam and Vienna, measured as usual from city border to city border along the shortest connecting road path, is 965 kilometers. Now consider the following examples: (2)

A: The distance between Amsterdam and Vienna is one thousand kilometers. B: #No, you 're wrong, it's nine hundred sixty-five kilometers.

B's reaction strikes us as inadequate; what he says is true but pedantic. But not so in the following exchange, where A's utterance is actually closer to the truth.

Be brief and vague!

(3)

441

A: The distance between Amsterdam and Vienna is nine hundred seventy-two kilometers. B: No, you 're wrong, it's nine hundred sixty-five kilometers.

The road sign example (1) showed that the phenomena to be talked about here constitute a problem for translation. And indeed, if we translate (2) into the American measurement system, the oddness of the exchange vanishes (one has to know that 600 miles are 965 kilometers, and 621 miles are 1000 kilometers). (4)

A: The distance between Amsterdam and Vienna is six hundred and twenty-one miles. B: No, it's six hundred miles.

It is quite obvious that the oddness of (2) cannot simply be stated in terms of truth conditions. Otherwise, the following exchange should be odd as well, but it isn't (if we disregard the fact that it is odd to render the distance of two cities with a precision of ±50 meters in the first place!). (5)

A: The distance between Amsterdam and Vienna is one thousand point zero kilometers. B: No, you 're wrong, it's nine hundred sixty-five kilometers. Examples like (2) can be multiplied at will. Consider the following:

(6) (7)

A: The number π is 3.14159. B: Mo, it's 3.1415926535. A: The number π is 3.1415926534. B: No, it's 3.1415926535.

Clearly, the reaction of Β in (6) is pedantic, but it is not pedantic in (7)· The first generalization that we can draw from examples (2) to (7) is that one should not increase the level of precision that was set by the first speaker. To say that the distance between Amsterdam and Vienna is one thousand kilometers sets the level of precision to +50 km. The speaker indicates with the choice of words that the only distance values that should be mentioned are 800 km, 900 km, 1000 km, 1100 km, and so on. Changing this level in the reaction is pedantic. Similarly, to say that the value of π is 3.14159 sets the level of precision to 5 decimal

442

Manfred Krifka

points; again changing this level is pedantic. If the first speaker starts out with a higher level of precision, as in (3), (5), or (7), the reactions of the second speaker at the same level are not considered pedantic. The translation problems we encountered with (1) and (5) show that the level of precision does not translate under a "precise" translation of terms. A better translation of 7 miles into the metric system would have been 11 kilometers, and there is no conceivable translation of (2) into mile measurements that would preserve the oddity of the exchange. The precision level of an expression can be marked explicitly by modifiers like roughly or exactly. We virtually have to employ exactly or precisely if we want to have a round number understood in a precise way: (8)

The distance between Amsterdam and Vienna is exactly one thousand kilometers.

We can also use these modifiers to support an interpretation that an expression would have had anyway: (9)

a. The distance between Amsterdam and Vienna is exactly nine hundred sixty five kilometers. b. The distance between Amsterdam and Vienna is roughly one thousand kilometers.

But it is ludicrously pedantic to utter (10), because it suggests that one could be even more precise: (10) The distance between Amsterdam and Vienna is roughly nine hundred sixty five kilometers. The notion of precision level is applicable only to the values of measurement terms, and not in cases in which numbers are used in an ordinal way. There is nothing pedantic in B's reaction in the following example: (11) A: Her phone number is sixty-five one thousand. B: No, her phone number is sixty-five one-thousand and one. I will refer to the phenomenon we are after by the term precision level choice. The principle of proper precision level choice can be formulated as follows:

Be brief and vague!

443

(12) Precision level choice: When expressing a measurement of an entity, choose a level of precision that is adequate for the purpose at hand. In (1), the adequate level of precision for motorists is on the mile or kilometer level; it is simply irrelevant for their concerns, like driving times or the amount of gas in one's tank, to give distances to a precision down to meters. In (2), the questioner A indicates a particular precision level as adequate for the purpose at hand (roughly, ±50 km), and B's reaction is appropriate only at another precision level (roughly, ±0.5 km). Notice that B's reaction is fine as a rejection of the precision level selected by A. For example, if Β is the manager of a Dutch transportation company that wants to pay as little as possible to their truck drivers, the exchange sounds quite idiomatic. How do we recognize precision levels? The empirical generalization is easy enough: Round Numbers suggest Round Interpretations in measure expressions. I will call this the RN/RIprinciple: (13) RN/RI principle: a. Short, simple numbers suggest low precision levels. b. Long, complex numbers suggest high precision levels. This is not necessarily so. In the mathematical language of science we can combine any measurement with any error margin. But it is an evident pragmatic principle in the interpretation of natural language. The RN/RI principle is, as I said, an empirical generalization only. We should be able to derive it from more general principles of language use, that is, of linguistic pragmatics, which in turn should be motivated as general principles of human behavior. This is the main topic of this article. After a short review of the few previous attempts to deal with the phenomenon, I will try to identify simpler principles that are independently motivated and lead to the RN/RI principle. We will see that two such principles can be identified that give us the correct result, provided they interact in a way that is suggested by Bidirectional Optimality Theory, a theory of the interaction of interpretative principles developed in Blutner (1998, 2000) and subsequent work.

444

Manfred Kriflia

3. Previous work The phenomenon we are after is, of course, well-known. It belongs to our everyday experience with language, and it has been noted in linguistic work at various places, for example in Sadock (1977), a response to Lakoff (1972), and in Pinkal (1995: 271), a comprehensive theory of the semantics of vagueness. A somewhat more detailed treatment can be found in Wachtel (1980), who develops a theory of rounded numbers for examples like the following: (14) a. Sam is approximately six feet tall. b. Odessa has approximately 1,000,000 inhabitants. Rounded numbers are marked by indicators of precision levels, like approximately or roughly. Wachtel also observes that we get the approximate interpretation without overt marking with "round" numbers like 1.000.000, but he does not explain why this is so. Curtin (1995), in an unpublished MA thesis, introduces the notion of granularity that can be used in the linguistic representation of reality. He develops the idea that fewer entities and distinctions exist on a coarse-grained level than on a fine-grained level, but that these levels are related to each other by homomorphic mappings. He observes that, with measuring terms, expressions on the coarse-grained level are generally shorter (e.g., one hundred, two hundred, three hundred) than on the fine-grained level (e.g., one hundred, one hundred and one, one hundred and two), but he does not give an explanation why this is so. Most recently, Lasersohn (1999) developed a theory of imprecise interpretations in terms of pragmatic "slack" of expressions. But the relation to the complexity of expressions is not addressed at all.

4. A preference for short expressions It is quite obvious that, everything else being equal, shorter and simpler expressions are preferred. This is a well-known law of the economics of language, and has found a well-known formulation in Zipf (1949). It is also behind a submaxim of manner in Grice's William James Lectures of 1967, published as Grice (1975), which states, briefly, "Be brief!". In the Neo-Gricean accounts of Atlas and Levinson (1981), Horn (1984),

Be brief and vague!

445

and Levinson (2000) it corresponds to the I-principle (or Horn's Rprinciple), which says: "Produce the minimal linguistic information sufficient to achieve your communicational ends!" (cf. Levinson 2000: 114). The principle is typically motivated as one that is rooted in the interest of the speaker; speakers are lazy, hence like to be brief. However, experience with chatterboxes will make one doubt that (i) this is true of speakers in general in all situations, and (ii) that it is solely based on the interest of speakers. In any case, one cannot deny that it is a powerful pragmatic principle. The principle plays an obvious role in precision level choice. It explains why, if the distance between Amsterdam and Vienna is 965 kilometers, one can quite truthfully utter (15.a) instead of (b). (15) a. The distance between Amsterdam and Vienna is one thousand kilometers. b. The distance between Amsterdam and Vienna is nine hundred sixty-five kilometers. Whatever (15.a) is lacking in precision is compensated by the gain in brevity, at least in typical situations. In general, in a decimal-based system, number words referring to the powers of ten, like 10, 100, or 1000, or multiples thereof, like 30, 400, or 7000, are shorter than number words referring to adjacent natural numbers. This suggests that principle (a) of the RN/RI (cf. (13)) can be explained by a preference of brevity. In many situations, brevity is a worthier goal to aim for than precision. But can preference for brevity also explain principle (b) of the RN/RI? Perhaps it can: There is no gain when we interpret (15.b) vaguely, hence it is interpreted in a precise way. I will argue that this line of reasoning is right, and I will develop it more precisely below. But before we do that, a somewhat longish detour concerning brevity is in order.

446

Manfred

Krißa

5. A closer look at brevity Consider the following, where we talk about different distances. (16) a. The distance A is one thousand one hundred kilometers. b. The distance Β is one thousand and one kilometers. Clearly, (16.a) can be interpreted in a less precise way, even though it is more complex in terms of numbers of syllables and perhaps even in its morphosyntactic structure. We can counter such examples by saying that in general, multiples of powers of ten are expressed in a simpler way, and therefore allow for a less precise interpretation, even though there might be exceptions of shorter expressions that are not multiples of powers of 10. The following examples also show an apparent exception to brevity: (17) a. The train will arrive in five / fifteen / twenty-five minutes. b. The train will arrive in four / sixteen / twenty-four minutes. We allow for more lax interpretations in the cases of (17.a) than in the corresponding cases of (17.b). The reason cannot be greater brevity of expression in (17.a), as the expressions are equally complex on all counts (morphosyntactic complexity, number of syllables). Obviously, the number or digit "5" is treated in special ways in our decimal-based system, just like the numbers that are powers often or multiples thereof. The reason is that they represent the half point between a multiple η of a power of 10 and the multiple (n+1) of the same power of 10, and that halves are the conceptually most prominent fractions. The prominence of half points can also be observed in cases like the following: (18) a. The train will arrive in half a minute. b. The train will arrive in thirty seconds. c. The train will arrive in forty seconds. Here, (18.a) allows for an imprecise interpretation. This also holds, interestingly, for (18.b), perhaps to a somewhat lesser extent, in contrast to examples like (18.c), which are interpreted at a higher precision level. It appears that thirty seconds is just another standardized way to refer to one half of a minute, just as sixty seconds is a standardized way to refer to a minute. We find similar phenomena for other measures of time that

Be brief and vague!

447

are not based on the decimal system, cf. e.g .fifteen seconds, fifteen minutes, forty-five minutes, twelve hours, and twenty-four hours. Another case in which brevity all by itself does not explain everything is the following. In English, there are traces of competing numbering systems, based on 12 {dozen) or 20 (score). I found that people share the intuitions that (19.a) is more easily interpreted in a lax way than (19.b): (19) a. A dozen / two dozens bandits were approaching. b. Twelve / twenty-four bandits were approaching. Notice that the complexity of two dozen bandits and twenty-four bandits is roughly the same: The first expression is syntactically more complex, but the latter one has fewer syllables. It seems that brevity of expression is related to another factor that is the critical one. The idea is the following: When we want to express certain meanings, we select a particular set of expressions, which I will call the expression-choice space, out of which we choose the expressions that identify the semantic objects. The expressions of the set that we do not choose are possible alternatives, and the fact that they were not chosen may lead to pragmatic implicatures. Expressions that are not in the set do not count in the semantic or pragmatic interpretation. The notion of expression-choice space is similar to Curtin's notion of granularity. The difference is that granularity concerns meanings, whereas expression-choice space concerns expressions. To illustrate this idea, consider again the following examples (we assume as above that the "real" distance is 965 kilometers): (20) a. The distance between Amsterdam and Vienna is one thousand kilometers. b. The distance between Amsterdam and Vienna is nine hundred seventy-two kilometers. In (20.a), the expression-choice space for the number word are number words based on the multiples of 100, that is, eight hundred, nine hundred, one thousand, one thousand one hundred, etc. In (b), the expression-choice space for the number word consists of number words based on multiples of 1. In the expression-choice space of (a) there doesn't even exist a number word nine hundred sixty-five', the number word one thousand is the closest one to the real distance. In the expres-

448

Manfred Krifka

sion-choice space of (b), on the other hand, this number word exists, and hence the choice of nine hundred seventy-two is not the best one. How do we know which items are contained in the expression-choice space? The expression that we are actually using tells us. If, in a measuring context, we use the number word thirty, then the expressionchoice space will probably contain the number words that are multiples of 10, such as twenty, thirty, forty. If we use a number word like three hundred, the most likely alternatives are two hundred, three hundred, four hundred, and so on. If we use a number word like one thousand, we can either assume an alternative set like one thousand, two thousand, three thousand, etc., or an alternative set of number words that are factors of 100, such as nine hundred, one thousand or one thousand one hundred. If we use a number word that refers to the mid number between two adjacent multiples of powers of 10, such as thirty-five, then the alternatives are the multiples of this power of 10 and their mid values, such as twenty-five, thirty, thirty-five, forty, forty-five, etc. If we evoke numbering systems with different bases, then similar principles apply to the powers of these bases; for example, the number word two dozens invokes alternatives like one dozen, two dozens, three dozens, etc. If we refer to dimensions that are measured by measure functions that are not based on the decimal system, such as time, then the expression-choice space may be determined by prominent halves and quarters of the basic measurement units of these dimensions. For example, choosing fifteen minutes will evoke alternatives such as thirty minutes, forty-five minutes, and one hour as alternatives. One important point to be mentioned is that a particular expression may allow for different expression-choice spaces; for example, one thousand is, in principle, compatible with number words that express multiples of 100 or multiples of 1000. In general, I will assume that expression-choice spaces of number words based on multiples of lower powers of 10 are possible. The number word one thousand then allows for the following expression-choice spaces: (21) a. b. c. d.

one thousand, two thousand, three thousand, ... ... nine hundred, one thousand, one thousand one hundred, ... ... nine hundred ninety, one thousand, one thousand and ten, ... ... nine hundred ninety-nine, one thousand, one thousand and one, ...

Be brief and vague!

449

There is a systematic relationship between these expression-choice spaces: The upper ones are more coarse-grained, the lower ones are more fine-grained. We can model this relationship by set inclusion of sets of expressions. An expression like one thousand presumably will prefer an expression-choice space like (21.a) or (b), and not (c) or (d). In general, we can state that the use of an expression α that comes with expressionchoice spaces introduces the most coarse-grained, or a more coarsegrained, expression-choice space in the actual pragmatic evaluation. What is the relation of brevity of expressions to expression-choice space? Put simply, the average complexity of expressions in more finegrained expression spaces is greater than the average complexity of expressions in less fine-grained expression spaces. Take the following example: (22) a. one, two, three, four, five, six, seven, ... one hundred b. five, ten, fifteen, twenty, twenty-five, thirty, ... one hundred c. ten, twenty, thirty, ... one-hundred If we measure complexity by number of syllables, the average complexity of (22.a) is 273/100 = 2.73; of (22.b) it is 46/20 = 2.3, and of (22.c) it is 21/10 = 2.1. Measurement by morphological complexity would yield a similar result. Hence, expressions in less fine-grained expression-choice spaces are, in general, shorter. We can then qualify the tendency for brevity as one for expressionchoice spaces: Speakers prefer expression-choice spaces with expressions of low average complexity. I will call this principle or constraint BRIEFEXPRESSION: (23)

BRIEFEXPRESSION:

Expression-choice spaces with shorter, less complex expressions are preferred over expression-choice spaces with longer, more complex ones. There actually may be systematic pressures that make expressions of coarse-grained expression-choice spaces shorter. Notice that in English, fifteen is a phonologically and orthographically reduced form (*fiveteen but fourteen, sixteen). The shortness of the form twelve obviously relates to the duodecimal system found in forms like dozen. Such phonological simplification may be mediated by frequency of use. Of course,

450

Manfred

Krijka

more comprehensive studies would be necessary to establish any such connection between expression-choice spaces and phonology.

6. A preference for precise interpretations? After this somewhat longish digression on brevity, let us return to the general theme of preference for short expressions. We can now phrase it as follows: The pragmatic principle BRIEFEXPRESSION is responsible for our interpretation of a measure term like one thousand kilometers in a vague way. Even if the speaker knows the distance more precisely, he may choose a term that is true only under a vague interpretation because of the resulting gain in brevity. It is unclear whether this also explains why we interpret a measure term like nine hundred sixty-five kilometers in a precise way. Intuitively, in this case nothing is to be gained by a vague interpretation, as the expression is long anyway, and therefore we settle on a precise interpretation. In order for this argumentation to go through, we have to assume that, as a matter of principle, precise interpretations of measure terms are preferred over imprecise, vague ones: (24)

PRECISEINTERPRETATION:

Precise interpretations of measure functions are preferred over vague ones. We can think of vague interpretations of measure functions as measure functions that allow for errors. For this we can introduce the following precise notation: (25) If μ is a measure function and ρ is a positive real number, then μ+ρ is the vague measure function defined as: For all χ, μ±ρ(χ) = [μ(χ)-ρμ(χ), μ(χ)+ρμ(χ)] For example, if km(x) = 1000, then km+o.osW = [950, 1050], Why should precise interpretations of measure terms be preferred over vague ones? The obvious reason is: Because they are more informative! This is a general pragmatic principle; witness Grice's first submaxim of Quantity, which states that a speaker should make a contribution as informative as required by the current purpose of information exchange.

Be brief and vague!

451

With BRIEFEXPRESSION and PRECISEINTERPRETATION as two independent constraints we can explain the preference patterns as in the following OT-tableau. We assume, as before, that the true, precise distance is 965 kilometers. (26) Expression

BRIEF-

PRECISE-

EXPRESSION INTERPRETATION

car (a) The distance between A and V is one thousand kilometers. es" (b) The distance between A and V is nine hundred sixty-five kilometers. (c) The distance between A and V is nine hundred seventy-two kilometers.

*

*

*

*

Here, sentence (a) is the shortest candidate, but it violates PRECISEINTERPRETATION. Sentence (b) violates BRIEFEXPRESSION, but it allows for a precise interpretation. If we do not rank the two constraints (as indicated by the dotted line), then it is predicted that both sentences are fine. In contrast, (c) violates both constraints, and hence is dispreferred over the other two candidates. The problem of this view, however, is that an addressee will preferably interpret an expression like The distance between Amsterdam and Vienna is one thousand kilometers as saying that the distance is precisely one thousand kilometers. Under this interpretation, the expression is both brief and precise. But this is clearly not what we find - see the RN/RI-principle, (13.a). If we indeed want a precise interpretation of round numbers, we have to mark this explicitly with expressions that indicate the precision level, such as exactly. (27) The distance between Amsterdam and Vienna is exactly one thousand kilometers. How can we avoid this problem? Perhaps we should change our line of reasoning by roughly 180 degrees and assume instead a preference for vague interpretations.

452

Manfred Krijka

7. A preference for vague interpretations The opposite of PRECISEINTERPRETATION is VAGUEINTERPRETATION: (28)

VAGUEINTERPRETATION:

Measurement terms are preferably interpreted in a vague way. How could such a tendency be motivated? It clearly seems to run against Grice's first submaxim of quantity, which states that information content should be maximized. But already Grice added a second submaxim, namely, that the speaker should not give more information than is necessary for the current purpose of information exchange. Preference for imprecise expressions has been noticed by linguists working in empirical pragmatics. Ochs Keenan (1976), in her seminal study of the pragmatics of rural speech communities in Madagascar, describes conversational maxims that prefer expressions with less informative content over those with more content. One of the reasons, Ochs Keenan claims, is that vague expressions help to save face: One is not as easily proven wrong if one stays vague. This preference is by no means only to be found in exotic rural communities. The philosopher of science Pierre Duhem, noted, famously, that there is a balance between precision and certainty: One cannot be increased except to the detriment of the other (cf. Duhem 1904: 178f., cited after Pinkal 1995: 262). So, it appears that a tendency for vague interpretation can be motivated. But does it give us what we want? Consider the following tableau: (29) Expression

BRIEF-

I

VAGUE-

E X P R E S S I O N ! INTERPRETATION 3

cs (a) The distance between A and V is one thousand kilometers. (b) The distance between A and V is nine hundred sixty-five kilometers.

I *

!

(*)

•

Now (a) is preferred, and we also predict that this sentence is preferably understood under a vague interpretation. But (b) is dispreferred; I put the star for a violation of VAGUEINTERPRETATION in parentheses to

Be brief and vague!

453

indicate that the sentence could be seen either under a vague or a precise interpretation. There are very obvious problems with this approach. First, (b) is not really dispreferred over (a); it just suggests a precise interpretation. And more importantly, if (b) is actually uttered, a way to reduce the violations on the hearer's side is to assume not a precise but a vague interpretation! This is because this would reduce the violation from two stars to one. But, of course, this is exactly the opposite of what we find. Something must be fundamentally wrong. Nevertheless, I will argue that BRIEFEXPRESSION and VAGUEINTERPRETATION are indeed the two constraints that explain the RN/RI principle. But we have to consider the way how these constraints interact. Put informally, we must find a way to express, and motivate, the idea that one constraint can be violated just in case the other is violated.

8. The interaction of brevity and vagueness The theoretical framework that gives us what we need is Bidirectional Optimality Theory, as developed by Blutner (1998) and Blutner (2000); cf. also Dekker and van Rooy (2000) for a game-theoretic interpretation working with the notion of Nash equilibrium. In classical Optimality Theory (OT), the input to a rule expressed in a tableau of ranked or unranked constraints is a set of expressions, and the output is the one expression, or the set of expressions, that violate the constraints the least. In Bidirectional OT, the input are pairs of objects, constraints are independently specified for the members of the pair, and the output are those pairs that violate the constraints the least. The constraints are formulated in a modular fashion, for the members of the pairs. But finding the optimal solution(s) requires optimization for both members. In semantic and pragmatic applications of Bidirectional OT, the pairs are pairs (Ε, I) of an Expression and its Interpretation. There are different possibilities for comparing pairs (Ε, I) of expressions and interpretations. A particularly interesting one has been suggested by Jaeger (2000); it is called bidirectional super-optimality, and it is defined as follows:

454

Manfred Krifka

(30) A pair (Ε, I) of a set of candidate expressions GEN is superoptimal iff: i. ii.

There is no superoptimal (Ε', I) e GEN such that