Perspectives on Variation: Sociolinguistic, Historical, Comparative [Reprint 2011 ed.] 9783110909579, 9783110182842

Citation preview

Perspectives on Variation

W DE G

Trends in Linguistics Studies and Monographs 163

Editors

Walter Bisang (main editor for this volume)

Hans Henrich Hock Werner Winter

Mouton de Gruyter Berlin · New York

Perspectives on Variation Sociolinguistic, Historical, Comparative

edited by

Nicole Delbecque Johan van der Auwera Dirk Geeraerts

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication

Data

Perspectives on variation : sociolinguistic, historical, comparative / edited by Nicole Delbecque, Johan van der Auwera, Dirk Geeraerts. p. cm - (Trends in linguistics. Studies and monographs ; 163) Includes bibliographical references and index. ISBN 3-11-018284-X (hardcover : alk. paper) 1. Language and languages — Variation. I. Delbecque, Nicole. II. Auwera, Johan van der. III. Geeraerts, Dirk, 1995IV. Series. P120.V37P47 2005 417'.7-dc22 2005001780

ISBN 3-11-018284-X Bibliographic information published by Die Deutsche Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at .

© Copyright 2005 by Walter de Gruyter GmbH & Co. KG, 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.

Table of contents

Introduction

1

Europe's sociolinguistic unity, or: A typology of European dialect/standard constellations Peter Auer

1

From phonetic similarity to dialect classification: A principled approach Paul Heggarty, April McMahon and Robert McMahon

43

Inflectional variation in Belgian and Netherlandic Dutch: A usage-based account of the adjectival inflection Jose Tummers, Dirk Speelman, and Dirk Geeraerts

93

Interdialectal convergence between West-Flemish urban dialects Reinhild Vandekerckhove

111

Substitutions in epistolary forms of address in the seventeenth century Dutch standard variety Arjan van Leuvensteijn

129

LOVE in words: Experience and conceptualization in the modern English lexicon of LOVE Heli Tissari

143

On the role of semasiological profiles in merger discontinuations Clara Molina

111

The ANGER IS HEAT question : Detecting cultural influence on the conceptualization of ANGER through diachronic corpus analysis Caroline Gevaert

195

Development and motivation of marked plural forms in German Heide Wegener

209

vi

Table of contents

Not arbitrary, not regular: the magic of gender assignment Marcin Kilarski and Grzegorz Krynicki

235

Future time reference - English and Dutch compared Griet Beheydt

251

Cleft constructions in French and Spanish Katleen Van den Steen

275

How to express indifference in Germanic: Towards a functional-typological research programme Torsten Leuschner

291

The lexicalization of speech act evaluations in German, English and Dutch Gisela Harras and Kristel Proost

319

Index

337

Introduction

The significant advances witnessed over the last years in the broad field of linguistic variation testify to a growing convergence between sociolinguistic approaches and the somewhat older historical and comparative research traditions. Particularly within cognitive and functional linguistics, the evolution towards a maximally dynamic approach to language goes hand in hand with a renewed interest in corpus research and quantitative methods of analysis. Only in this way, many researchers feel, can one do justice to the complex interaction of forces and factors involved in linguistic variability, both synchronically and diachronically. The contributions to the present volume illustrate the ongoing evolution of the field. By bringing together a series of analyses that rely on extensive corpuses to shed light on sociolinguistic, historical and comparative forms of variation, the volume tries to highlight the interaction between these subfields. Most of the contributions go back to talks presented at the meeting of the Societas Linguistica Europaea held in Leuven in 2001. The volume leads off with a global typological view on the sociolinguistic landscape of Europe offered by Peter Auer. It is followed by a methodological proposal for measuring phonetic similarity between dialects designed by Paul Heggarty, April McMahon and Robert McMahon. Various papers deal with specific phenomena of socially and conceptually driven variation within a single language. For Dutch, Jose Tummers, Dirk Speelman and Dirk Geeraerts analyze inflectional variation in Belgian and Netherlandic Dutch, Reinhild Vandekerckhove focusses on interdialectal convergence between WestFlemish urban dialects, and Aijan van Leuvensteijn studies competing forms of address in the seventeenth century Dutch standard variety. The cultural and conceptual dimension is also very much present in the diachronic lexicosemantic explorations presented by Heli Tissari, Clara Molina and Caroline Gevaert for English expressions referring to the experiential domains of love, sorrow and anger, respectively: the history of words is systematically linked up with the images they convey and the evolving conceptualizations they reveal. The paper by Heide Wegener and that by Marcin Kilarski and Grzegorz Krynicki constitute a plea against arbitrariness of alternations at the level of nominal morphology: dealing with marked plural forms in German, and with gender assignment to English loanwords in the Scandinavian languages, respectively, their distributional accounts bring into the picture a variety of motivating factors. The four cross-linguistic studies that close the volume focus on the differing ways in which even

2

Introduction

closely related languages exploit parallel morphosyntactic patterns. They share the same methodological concern for combining rigorous parametrization and quantification with conceptual and discourse-functional explanations. While Griet Beheydt and Katleen Van den Steen confront the use of formally defined competing constructions in two Germanic and two Romance languages, respectively, Torsten Leuschner and Gisela Harras and Kirsten Proost analyze how a particular speaker's attitude is differently expressed in various Germanic languages. Peter Auer proposes a comprehensive typology of the different ways in which standard varieties, standard-oriented regional varieties, traditional and modern dialect varieties relate to one another. Approaching the standard-dialect dimension from a chronological perspective, he distinguishes five main configurations, together with an additional set of transitional constellations. The continuum goes from (i) exoglossic diglossia, without an endoglossic standard, over (ii) medial diglossia, with an endoglossic standard, and (iii) spoken diglossia, where the capitals play a major role in the emergence of written and spoken standard varieties, to the actually most wide-spread types of standard/dialect relationships, viz., (iv) diaglossia, characterized by the presence of intermediate varieties between standard and base dialect, and (v) dialect loss. The latter can occur either via a scenario of further infiltration of the dialect/regiolect by standard features in diaglossic areas, or, else, following a direct path from spoken diglossia, via dialect avoidance among middle class speakers. Paul Heggarty, April McMahon and Robert McMahon present a method for measuring the phonetic distance between historically related languages or dialects, like the Romance languages and their national varieties. Such a comparison of historical dialects against their common ancestor language intends to shed new light on the notion of family tree. The authors pay specific attention to a comparison of their approach (which includes a weighting of features on the basis of general phonetic principles) with dialectometrical approaches using so-called Levenshtein distances to measure phonetic distance. Jose Tummers, Dirk Speelman and Dirk Geeraerts focus on the use of attributive adjectives as indicators of language variation in the two national varieties of Dutch, Belgian and Netherlandic Dutch. They analyze the alternation between -e and zero in the declension of adjectives modifying definite neuter nouns. Using a sociolectometrical methodology, they examine the question whether this adjectival variation reveals a different stratification of Belgian and Netherlandic Dutch, and whether it differs from lexical variables as indicators of geographical and stylistic differences.

Introduction

3

Reinhild Vandekerckhove reports on recently collected data revealing an ongoing change in the Dutch dialects of the western part of Flanders. Although these processes result from the complex interaction of intralinguistic and extralinguistic factors, the author argues that they can best be interpreted in terms of speaker contact and in the light of the socioeconomic context in which these contacts takes place. The convergence patterns attested for the personal pronouns of the plural in five WestFlemish cities can, therefore, best be characterized as horizontal convergence leading to structural leveling towards a koine within the WestFlemish dialect area. Arjan van Leuvensteijn focusses on the use of competing forms of address in seventeenth century Dutch standard, as it is symptomatic of ongoing changes in the emerging standard variety. The distribution of different forms of address (the plural ghi(j), the third person singular, titles, and the declining singular du), sheds light on the pragmatic functions of respectful address in a class-ridden society. Specifically, he focusses on the abbreviation UE (Uwe Edelheid 'Your Honour') in letters addressed to the famous writer P.C. Hooft and in the personal correspondence of a wellknown upper-class family. Heli Tissari combines thesaurus data and corpus data to study the evolution of the vocabulary of LOVE in the English language since 1500. The analysis of the lexicon related to distinct love domains, defined in terms of the participants involved, brings to the fore a number of domainspecific metaphors that are revealing of the ways in which the different kinds of love have been conceptualized throughout the centuries. In discussing their relative centrality, the author also draws attention to metaphors that occur across the various participant domains. Clara Molina explores the history of the English nouns SORE (Old English sdr) and SORROW (OE sorg) from the point of view of diachronic prototype semantics. Whereas the medieval data support the hypothesis of an ongoing merger, the author shows that this process came to an end from the fifteenth century onwards. She explains the replacement of sore by the French pain by a series of prototype effects in the sore/sorrow network that go back to the very semasiological profile of the terms. Caroline Gevaert reconstructs the lexical and conceptual field of anger in Old and Middle English integrating historical semantics, prototype semantics and cognitive semantics. Her corpus-based diachronic analysis reveals that the heat-conceptualization of anger in English is not as embodied and universal as sometimes claimed, but to a large extent culturally determined. She further shows that before the heat metaphor took

4

Introduction

over in Middle English, the physiological metaphor that prevailed in Old English was a swell metaphor. Heide Wegener addresses the issue of the motivation of the different plural forms in German. She is particularly concerned with studying the function of the -er plural, substituting for the -e plural subjected to apocope in Early New High German, and the recent formation of the additional plural class in -s, the emergence of which seems to run counter the assumptions of Markedness Theory. Via the study of their frequency distribution, their communicative-pragmatic function is shown to obey transparency requirements, in line with various naturalness principles. Marcin Kilarski and Grzegorz Krynicki analyze gender assignment of English loanwords borrowed into Danish, Swedish and Norwegian. Relying on a corpus of over the two thousand nouns for each language, they measure the relative impact of a number of formal and semantic factors. Gender assignment appears to be neither fully arbitrary neither fully predictable: some significant patterns emerge, but the factors that form relevant clusters somewhat differ from language to language and operate with variable stringency. Griet Beheydt examines future time reference in English and Dutch. Her corpus based analysis reveals that of Dutch zullen 'will' + infinitive, allative gaan 'go' + infinitive, and the present tense, the latter two have a wider distribution than their English counterparts. Dutch 'go' periphrasis, in particular, has developed uses that are remote from its original allative meaning. For the present tense strategy, Dutch allows a wider variety of licensing mechanisms. For Dutch, she also examines sociolinguistic and regional variation. Katleen Van den Steen analyzes cleft and pseudo-cleft constructions in French and Spanish in a parallel corpus. She shows that Spanish may achieve the effect of the French cleft by word order variation or by the socalled 'inverse' pseudo cleft, as in el resultado es lo que cuenta 'the results is what matters'. Among the morphosyntactic and discourse parameters that determine the choice between the various strategies, she highlights the form and syntactic function of the extracted constituent, the tense of the copula, the type of contrast that is expressed, and the way it informationally relates to the context. Torsten Leuschner proposes a cross-linguistic study of the phrases routinely used in the Germanic languages to declare the speaker's indifference. Based on corpora, grammars, dictionaries, he studies the underlying patterns of conceptualization and metaphor as well the processes of grammaticalization and idiomatization. He also recognizes

Introduction

5

them as cultural artefacts, sensitive to register, taboo breaking and crosslinguistic borrowing. Gisela Harras and Kristel Proost study patterns of explicit and implicit speech act evaluation in English, Dutch, and German. They compare the use of speech acts verbs and speech act idioms. Focusing on lying and boasting, they show that speech act idioms are particularly useful for intensification, attributing values like 'unrealistic' and 'direct'. The three languages differ in the way the degrees are lexicalized, the specification of the propositional content and the manner of the speech act performance. The editors thank Barbara De Cock for taking care of the technical editing.

Europe's sociolinguistic unity, or: A typology of European dialect/standard constellations1 Peter Auer

1. Preliminaries At first sight, the diversity of the dialect-standard constellations found in Europe is enormous and seems to defy any attempt to find a common denominator. Many believe that the sociolinguistic situation, say, in England, Germany and Italy is fundamentally different and that what is meant by 'dialect' and 'standard' in these countries cannot be compared at all. In this paper, I would like to advocate the opposite position and argue for a uniform description of the European sociolinguistic repertoires. More specifically, my claim is that on a sufficient level of generalisation there is a systematicity behind the superficial heterogeneity which unfolds from a historical perspective. I will suggest a typology of speech repertoires which distinguishes five sociolinguistic types that also represent a chronological order. The typology aims at the standard-dialect dimension, while multilingual constellations will only be commented on to the degree that exoglossic standard languages enter the repertoire. There have been some previous attempts at more comprehensive typologies valid for all sociolinguistic areas in the world.2 By comparison, the present attempt is a good deal less ambitious since it is restricted to Europe and since it leaves out many multilingual repertoire types; however, it is supplemented by a historical component which is usually absent. A number of preliminary remarks are necessary. I will use the term 'dialect' as a purely relational concept (following, for instance, Coseriu 1980), which means that by definition, without a standard there can be no dialect. (The opposite is a contingent fact and an important generalisation over the European sociolinguistic history.) Also, the term will be used exclusively in order to refer to areal variability within a language (i.e., there is no 'standard dialect' of a language according to my usage of the term, but only a standard variety). On the other hand, the term 'dialect' will not be restricted to the 'base dialects', i.e. the most ancient, rural, conservative

8

Peter Auer

dialects, but will be used such as to include regional and urban varieties with a larger geographical reach as well. More controversial is the term 'standard'. Here, it is employed in order to designate a variety of a language (which follows a 'norm' or 'codex', i.e. 'standard' does not designate the norm itself), which is characterised by the following three features: (a) it is orientated to by speakers of more than one vernacular variety (which does not necessarily imply that it is mastered by everybody), (b) is looked upon as an Η-variety and used for writing3, and (c) it is subject to at least some codification and elaboration (Haugen 1966) or Ausbau (Kloss 1967).4 A standard variety therefore is more than a 'common variety' (Gemeinsprache), which only satisfies the first criterion, and more than an H-variety (Hochsprache), which only satisfies the second criterion. The third requirement excludes from consideration, for instance, the various Η-varieties of Middle High German as used in the medieval literature. But taken alone, it would not be sufficient to define a standard variety either, since there are codifications of dialects as well (for instance, there is a Zürichdeutsche Grammatik in which people look up what Zürichdeutsch dialect should 'really' be like, although no Swiss person would think of Zürichdeutsch as a standard variety). My focus will be on the emergence of the endoglossic national standard varieties in Europe in the second millennium A.D. and their relationship to the dialects in their respective geographical area. These standard varieties are closely linked to the emergence of the European nation states; since the codification of a national language has often been looked upon as an important if not an essential step in nation building, the history of the standard variety and its status today usually reflect the way in which nation building has proceeded. In some areas, there has been a straightforward development from the late Middle Ages to the present day (such as in the prototypical European nation states, e.g. in England or France), in other cases, the process set in much later (as in Romania, Luxembourg, etc.), or was restarted after a period of destandardisation (Ireland, Greece, etc.). Note, however, that not all national standard languages (e.g. those whose emergence is due to a nationalistic ideology) are state languages in Europe. Some have failed to reach this status and are used as minority languages within one or more European states (such as Standard Basque/Euskara Batua). There are also some recent legally recognised standard languages which only came into being in the post-nationalist era and therefore have never played a role in nation-building (such as Reto-Romance). However, the recent developments in east and south-east Europe after 1990 provide ample evidence for the fact that the nationalist equation of one language = one nation = one state still survives. Finally, it should be mentioned that

Europe 's sociolinguistic unity

9

there is at least one national language for which it is at least doubtful whether it fulfils my criteria for a standard language, since it is not a Hvariety in Ferguson's sense; this is Letzebuergesch, which is the national language of Luxembourg (and together with French and German one of the official languages of the state), but is almost exclusively spoken (Gilles 1999:8-9). The opposite is of course also true: there are standard languages in Europe which were no national languages (while Letzebuergesch is a national language which is no standard). They fall outside the scope of this paper. Among them is Latin, which, at least from the times of the Carolinian renaissance onwards, was not only used as an H-code, particularly in writing, but was also subject to codification. Other candidates for standard languages which were no (or not everywhere) national standards are the Attic-Ionic ('Hellenistic') koine of the fifth century, and Old Church Slavonic, at least in some periods and in some of the east and south-east European countries in which it was used.5

2. Type Zero repertoires: No endoglossic standard (exoglossic diglossia) In the historically primordial situation in medieval Europe, if a nonvernacular variety was used at all, it was an exoglossic6 high (standard) variety: Latin, Old Church Slavonic7 or Arabic.8 This repertoire type can be represented as in Fig. 1:9

(exoglossic standard)

vernacular varieties

(spoken/written)

Figure I. Type Zero Repertoire

·

10

Peter Auer

Since the Standard variety was exoglossic, no levelling through standardisation (levelling out of non-standard features) could take place. This does not imply, however, that the exoglossic standards had no influence on the vernaculars; the opposite is true. But here, we are dealing with language contact, not with dialect/standard levelling. Also, it should be noted that the vernaculars sometimes underwent horizontal levelling (koineisation) before the onset of standardisation. Examples are the dialect koinai of southern, especially Andalusian Spanish which developed in the late Middle Ages (thirteenth to fifteenth century) before standardisation set in (Villena 1996). It is possible that a similar pre-standard koineisation took place in the course of the German settlements in the Slavic east, particularly in the Upper Saxonian region, in the twelfth and thirteenth century (Frings 1957).

(exoglossic standard)

·

vernacular varieties

(spoken/written)

Figure 2. Koineisation in a Type Zero Repertoire

Inter-vernacular levelling is related to the development of a Gemeinsprache^10 (common language), and it may be a helpful step on the way to a national standard variety. In Europe today, non-standardised varieties are rarely written (and if so, only in personal genres such as in e-mail, a conceptually half-oral, halfwritten text type). The same is not true for the history of the European languages. Examples range from the dialects of Old High German and the Proto-Romance dialects/languages to the Greek demotic literature under Ottoman rule. Over the centuries, when the European nation states came into being, many national standard languages also assumed the status of an exoglossic standard variety in Europe outside the nation from which they had originated. The addition of an exoglossic standard language to the repertoire took place in various guises, from the imposition of the language

Europe 's sociolinguistic

unity

11

of the conquerors as a standard on top of the structurally distant local vernaculars or even in competition with an endoglossic standard (examples among many: German in Czechia, Ottoman-Turkish on the Balkan and in Greece/Cyprus, English on Malta or in Gibraltar), the marginalisation or suppression of linguistic and cultural minorities within the territory of the newly developing nation states (examples: the Celtic languages in France or Great Britain, Swedish in Finland, German in South Tyrol) to the voluntary acceptance of a foreign standard language as the language of writing, ceremony and/or educated speech (examples: the Russian variant of Church Slavonic in the eighteenth century in Serbia, a variant of Church Slavonic as a literary and ritual language in Romania from the fourteenth to the eighteenth century, French in the era of enlightenment in most of Europe, French by the Flemish speaking bourgeoisie in Flanders in the nineteenth century, or French in Luxembourg today). In some areas, a combination of these factors was in play, or the motivation for the use of an exoglossic variety changed over the course of time (such as in the case of German in the Baltic). Type Zero repertoires continued to exist until well into the twentieth century for many of the smaller (minority) languages of Europe. They resorted to the standard varieties of the nations to which the respective minority belonged as a (pseudo-)'roof. 11 Most of these language minorities have recently (with more or less success) developed standard varieties of their own, in a conscious attempt to retain the vitality and competitiveness of the respective languages against the larger European languages, usually the languages of the states to which they belong. The new endoglossic standard varieties in this case compete with the well-established exoglossic one. Examples for recent standardisation are the Sami languages (with Finnish, Swedish, or one of the Norwegian standards as the overarching heteroglossic roof), Basque with Spanish or French as the exoglossic roof, Reto-Romance with German, the Frisian varieties with Dutch and German, Letzebuergesch with French and German, or the Bretonic varieties with French as the respective roofs. There are only very few European languages which have never developed an endoglossic standard; an example would seem to be the varieties of Romani (overarched by a dozen or more European languages, with very minor and unsuccessful attempts at standardisation in the Balkan states). Some varieties of European languages have lost their (endoglossic) standard variety because the speech community has become completely detached from its 'homeland' as a consequence of migration, and therefore from the larger part of the speech community. This holds, for instance, for Csängo-Hungarian spoken in Romanian Moldavia which has been isolated

12

Peter Auer

from Carpathian-Basin Hungarian since at least the seventeenth century (Sandor 2000), for the German language enclaves in northern Italy, i.e. the Highest Alemannic Walser colonies and the South Bavarian so-called Cimbric language islands (both medieval migrant communities), for the Albanian, Greek and Croatian language enclaves in Italy, or for the Turkish language enclaves in Bulgaria. (Some of these communities have recently re-established links to their place of origin.) One may also think of Irish English as a set of (migrant) varieties which lack an endoglossic standard (given the non-acceptance of the British standard and the general concern with and symbolic function of Irish in the Republic of Ireland, but the lack of concern with and symbolic function of English). 12 This shows that the development from Type Zero to Type A (etc.) is not irreversible.

3. Type A repertoires: Medial diglossia with an endoglossic standard The first type of linguistic repertoire which is not bilingual but contains a structurally related standard variety in addition to the vernacular varieties (dialects) is called diglossic here following Ferguson's initial (and restrictive) use of the term (Ferguson 1959). Diglossia with an endoglossic standard variety - Type A+B repertoires of my typology - resembles Ferguson's diglossia in that (a) the two varieties are clearly delimited from each other in the perception of the speakers, (b) the varieties are genetically closely related to each other, (c) the standard represents the Η-variety and is used for writing and (if spoken at all) for formal situations, whereas the dialect as the L-variety is not (usually) written, (d) the Η-variety is not the language of primary socialisation (first language acquisition). The transition from Type Zero to Type A can be represented as in Figure 3. (exoglossic standard) endoglossic standard (mainly written) Ο A

Figure 3. From Type Zero to Type A - written diglossia

Europe 's sociolinguistic unity

13

Of course the notion structural relatedness (or, for that matter, close genetic affiliation) and therefore the distinction between an exoglossic and an endoglossic standard leaves room for redefinition. Depending on the linguistic ideology of a society, two varieties may be called the standard and dialect of one language, or be looked upon as two different languages by the speakers (and sociolinguists). For instance, Bavarian is considered a dialect of German in German dialectology, while Sicilian, structurally roughly at a similar distance from standard Italian than Bavarian from standard German, is not called a dialect of Italian in most of Italian dialectology. Letzebuergesch was considered to be a dialect of German at least until 1900 while there is a consensus of opinion today that it is a language of its own. Macedonian is looked upon as forming a language apart by some (and is officially recognised as such nowadays), while others think it is a dialect of Bulgarian.13 Galician was taken to be a dialect of Spanish at some time, it is considered by many to be a language of its own today, and there are some who believe it is a dialect of Portuguese. Some think that Aragonese and Leonese are not dialects of Spanish (Castilian), but languages of their own. The southern Catalan varieties in Valencia (and on the Balearic Islands) were roofed until recently by standard Castilian only, not by standard Catalan.14 Castilian is genetically clearly less closely related to Valenciä than Catalan, but compared to the genetic distance between, say, Basque and Castillian, the distance is minor. Nevertheless, Castilian is considered to be an exoglossic standard variety by most of the speakers. Low German, on the other hand, is not roofed by Standard Low German (since there is no such thing), but by standard German ('High German'), which is not directly related to it, although the structural distance is once again relatively small. The point is that in such closely related pairs of varieties, no structural reasons {Abstand) can be adduced to argue for or against Binnendiglossie vs. Außendiglossie. Kloss (1967: 34-37) describes the consequent instability of the 'more' endoglossic minority standards which are in constant danger of being reduced to mere dialects under political or attitudinal pressure, and he cites the example of the Imperial Academy at St. Petersburg which, in the nineteenth century, simply decreed that Ukrainian be a dialect of Russian from then on - which would not have been possible if the structural distance between Russian and Ukrainian had been more pronounced.

14

Peter Auer

3.1. Historical development Apart from the extraordinary case of Old Church Slavonic (Old Bulgarian) in Bulgaria, perhaps apart from the case of Old Irish in Ireland and apart from the somewhat special case of Greek in Byzantium, we have no reason to believe that the formation of a high, codified, common language set in on the vernacular languages anywhere in Europe before 1300, and in many areas not until much later. Common languages before 1300 (local koinai) were not codified, not linked to a national discourse and usually accepted in small areas only. The transition from Type Zero to Type A goes through a (sometimes prolonged) phase in which the exoglossic standard competes with it. I call this first phase in the development of an endoglossic standard medial diglossia (following a terminology used in Swiss sociolinguists), because new standard varieties were used for the medium of writing only in the beginning. An important role for the spread of these written standards was played by the printing technology, and in the Protestant areas in particular, the bible written in an endoglossic standard was often the most successful print product in the post-reformation period which contributed to the standardisation of the written variety. In general, it can be said that the spread of the endoglossic written standard was to a large degree achieved through the written medium; cf. the spread of the High German Standard to the Low German areas; the spread of Castilian Spanish to the Aragonese and Leonese areas in the late Middle Ages; the spread of London English as the dominant printers' variety to other areas of England in the fifteenth century; the use of Edinburgh Scots English as the written variety in Scotland during the same period; the spread of standard Polish to the periphery under the influence of the Cracow printing offices. The generalisation that new standard varieties are first introduced in writing also holds for the nineteenth and twentieth century standardisation processes (such as in Norway, Finland, Greece re. the katharevousa, 15 Czechia / Czechoslovakia, Bulgaria, etc.), in which philologists/linguists took a central part. The difference is that these 'late' standard languages often were construed by conscious reference to some older, sometimes quite distant variety which these philologists considered to be more prestigious than the vernacular language actually spoken at their time (under the roof of an exoglossic language associated with some external power). It is therefore typical for these late standardisation processes that they had a strong archaising component. The resulting standard variety is bound to be restricted to written usage first since it is perceived by the population as being far removed from their ordinary language practices.

Europe 's sociolinguistic

unity

15

Of course, written standards were sometimes also spoken (for instance, read out, or used in oratorical style); they were then pronounced according to the written form. This Sprechen nach der Schrift often led to hyperarticulate forms, and even to the introduction of phonemic differences into the phonology of the language which had not been there before. Some of them had a lasting impact. Examples are the (nowadays phonemic) distinction between /e:/ and / basilectal regional std. basilectal regiolects

Figure 9. From Type C (diaglossia) to Type D/l (dialect loss) repertoires, second stage

Europe 's sociolinguistic unity

29

Areas in which the process of dialect loss after previous diaglossia may be on its way are parts of England as well as presumably some areas of Greece (areas of Athens, Thessaloniki) and Russia/Belorus/Ukraine. 48

6.2. Type D/2 areas: From diglossia to dialect loss There is also a direct path from Type Β (diglossia) to Type D in which case the old dialects are not handed on to the next generation, i.e. they are lost by shift. In social terms, it is the middle classes and the young women who avoid the (base) dialect first. This sets in motion a mechanism by which the dialect is associated with lower social class, manual (industrial) or farm work, lack of education, etc., first of all in the cities, later in the countryside as well. As a consequence, the base dialect loses prestige and domains of usage; most notably, parents avoid dialect with their children. Since the base dialect is seldomn used, speakers' competence in that variety also diminishes, which leads to insecurity and reluctance to speak dialect in more out-group contexts. In the final stage before loss, the attitudes towards the now almost extinct base dialect are usually positive again, and folkloristic attempts at rescuing the dialect may set in - usually without success. standard (written and spoken)

standard (written and spoken)

Figure 10. From Type Β (diglossia) to Type D/l (dialect loss) repertoires

This scenario of dialect loss captures the Low German dialect areas in northern Germany and the (northern and eastern) middle German dialects,

30

Peter Auer

particularly in most of Hesse (with the exception of the Frankfurt area which leans towards D/l) and Thuringia (whereas Saxony is D/l). 49 It can be assumed that it equally applies to Hungary, where the remaining dialects are highly stigmatised and only used in in-group situations,50 and perhaps to the remaining Spanish base dialects spoken in isolated areas of Aragonese and Asturian-Leonese parts (with decreasing vitality and tendencies of dialect loss). What remains in such a case is again the standard variety alone, with some variation due to the usual lack of stability of any spoken language. Interestingly, processes of divergence from the national standard often set in after the dialects and regional dialects have vanished, usually beginning in the large cities (cf. Reyneland 2001). In other words: the victory of the standard language does not imply a lack of linguistic heterogeneity. On the contrary, speakers seem to feel the need to sound different from the codified standard. The so-called francais avanci may be mentioned in this context as well as ethnolectal urban varieties (such as Rinkeby Swedish in Stockholm).

7. Some possible exceptions and consequences The four types of repertoires cover almost of all Europe, diachronically and synchronically. The historical starting points for the development of national standard varieties are, in Coseriu's sense, "primary dialects" (1980) which are older than the national standard varieties that have developed out of them. However, there are some few areas which raise questions since they are not easily (or not at all) accommodated by the model. A first counterexample might seem to be those large parts of the Slavic language area, in which geographical diversification set in late. For instance, diversification in and between the east Slavic varieties (Russian, Belorussian, Ukrainian) does not seem to be older than the fourteenth century, and dialect diversification in these languages continues to be low compared to most other European languages. At the same period when the present-day dialects of east Slavic emerged, there were already tendencies to develop a common high variety based on a power centre (the Moscow Rus'). Different as this development may be from the older Germanic and Romance varieties (where geographical diversification set in much earlier and the vernacular varieties are older), it is no counter-example to the model presented here since the early Russian common language certainly was no standard variety (in this function, Old Church Slavonic was used).

Europe 's sociolinguistic unity

31

Characteristic for these areas is that the emergence of an acrolectal common variety and dialect diversification on the basilectal level went hand in hand. However, there is at least one nation-state and one national standard language for which it is claimed that no dialects have ever existed, and that a national standard variety has been codified directly on the basis of a common language. This is Iceland, which presents a true counter-example to my typology.51 Since Iceland was colonised by settlers who came to the island around 900 from a well-circumscribed part of west Norway, the vernacular varieties brought along by them would seem to have been relatively uniform, and it may also be assumed that there was levelling induced by the migration process itself (as is well-attested for modern migrant communities as well). What is remarkable is the fact that little geographical variation developed over a thousand years after the colonialisation. Milroy and Milroy (1985) explain this exceptional situation by dense networks in a closely knit, but socially non-stratified society inhabiting a small territory; these networks would not only have kept language change at a minimum, but also prevented symbolic barriers between social or geographically defined groups from being erected.52 Other factors such as a high level of literacy already during the middle ages may have played a role as well. Finally, an interesting and also deviant case is that of (Old) Greek and the Hellenistic koine. With some justification, it can be claimed that Greek had already gone through the development from Type A to Type D oncewhen the other European languages had not even started to develop an endoglossic standard. The Old Greek dialects became roofed by a common standard variety based on Attic, which started out as some kind of levelled regional variety and spread to more and more Greek-speaking areas until it was made the standard variety under Philipp and Alexander the Great (the Hellenistic koine). The new prestige variety gradually led to the disappearance of the old dialects. The present-day dialects of Modern Greek are not the offsprings, of the Old Greek dialects, but the result of a secondary/tertiary dialectalisation on the basis of the Hellinistic koine. In sum, on a map of Europe with the four types of standard/dialect relationships drawn out, the most wide-spread types would surely be Type C, the diaglossic one, and Type D, in which the base dialects have disappeared. The very fine-grained dialect differences which have been documented on the European dialect atlas based on materials which are now 50-100 years old are vanishing. On the other hand, many diaglossic repertoires are quite stable, with the regiolects even gaining the upper hand over the standard varieties, and sometimes introducing new variation.

32

Peter Auer

This empirical generalisation has some impact on social dialectology (method and theory building) in a European context. If it is true that the European sociolinguistic situation has been characterised until very recently (Type D) by the co-existence of the old local dialects and the national standard varieties (with the latter emerging from, but also influencing the former), then models for the description of linguistic variation in the European languages are called for which sufficiently integrate these dynamics. Whatever such theories would look like in detail, they would need to start from the assumption of two or more linguistic systems „in contact" and therefore (by empirical necessity) differ considerably from the dominant variational paradigm in linguistics, the latter being conceived in order to meet the needs of a completely different sociolinguistic context (i.e., that of secondary and tertiary dialects). 53 It is only for the last stage of repertoire development, i.e. Type D (dialect loss), that the variational paradigm seems to be suited.

Notes 1. Many colleagues have been of help in the preparation of this paper. In particular, I would like to thank Angel Angelov (Sofia), Ludo Beheydt (Louvain-la-Neuve), Fran9oise Gadet (Paris), Peter Gilles (Freiburg), Frans Hinskens (Leipzig), Jeffrey Kallen (Dublin), Maria Kakridi (Athens), Marilena Karyolemou (Cyprus), Miklos Kontra (Budapest), Jiri Nekvapil (Prague), Bengt Nordberg (Uppsala), Inge Lise Pedersen (Copenhagen), Unn Rjayneland (Oslo), AnneCatherine Simon (Louvain-la-Neuve), Marja-Leena Sorjonen (Helsinki), Mats Thelander (Uppsala) and Juan Villena (Malaga) for their partly extensive comments. The paper continues to be marred by my own ignorance on many European sociolinguistic contexts, and all mistakes are of course made entirely on my account. 2. Relatively close in scope and intention to my own is Berruto's typology (1989a) whose "dilalia" seems to be similar to my "diaglossic" Type C. 3. Note that this does not imply that the non-standard varieties are void of (overt) prestige, although this happens to be the case quite regularly. 4. This last criterion is an attitudinal one; it is not the fact of codification (such as the existence of a grammar and a dictionary) which makes a standard variety, but the fact that its speakers think that such things should exist and that, where they exist, they should determine how members of that society ought to express themselves in situations in which the standard is required. 5. Old (Church) Slavonic was a 'commissioned language' for the missionary activities in the West Slavic area, 'constructed' by Cyril and Method in the nineth century on the basis of Old Bulgarian. Since it played a role as the

Europe 's sociolinguistic unity

33

national standard of Bulgaria, and in its Russian, Romanian etc. variety also as a national standard in these countries, the status of Old Church Slavonic is ambiguous. 6. The terms endoglossic vs. exoglossic correspond with Kloss' (1976) distinction between Binnendiglossie and Außendiglossie. 7. Old Church Slavonic (Old Bulgarian) started out as an exoglossic standard variety in Russia, Romania, etc. despite local adaptations, particularly in orthography and phonetics. However, the use of Old Bulgarian in the late period of the first Bulgarian kingdom (863-1018) is an early case of a European H-variety (Hochsprache). Since this language also underwent some Ausbau as the language of christianisation, it may also qualify as a standard. 8.1 am referring to the Arabic part of Spain, of course. 9. This and the following figures are inspired by Hugo Moser's representations of the German language repertoire through history (1950). One should be careful not to lose sight of the simplifications which are inherent in any model of this type; in our case, this caveat refers in particular to the distance between the base dialects (ground line) and the standard variety (top point) which is systematically ambiguous between an attitudinal and a structural interpretation. Also, the model is simplified since it presupposes that the distance between any variety in the lowest circle and the standard is the same. Obviously, the opposite is usually true: some dialects are structurally and attitudinally closer to the standard variety than others. See below, section 4. 10. This obecnä cestina ('Common Czech') has never fully submitted to the new, articifically created (written) Czech standard variety of the nineteenth century (the spisovny jazyk, 'literary language'); cf. Nekvapil 2000, Sgall et al. 1992:174 for details. 11. The term roof is used in a rather loose way here, deviating from Kloss' original version (Kloss 1978) were the notion Überdachung is restricted to dialects and 'their' standard language (cf. Berruto 2001 for a valuable discussion). 12. J. Kallen (pers.comm.). Kallen (1997) points out that the notion of (English) dialects is disputed in the Irish case, since it is not clear whether the base dialects were a uniform variety with a strong Irish substratum (or even a postcreole variety), or a rural dialect of English extended from British (plus Scottish?) English. 13. Cf. Angelov 1999 for a discussion. 14. Cf. Martin Zorraquino 1998. 15. Standard Czech was fabricated by linguists such as Dobrovsky (around 1800) and Jungmann on the basis of written Czech of the late sixteenth century. The emerging Bohemian standard of the time around 1800 was rejected since it was taken to be associated with one region (Bohemia, leaving out Moravia, Silesia and possibly Slovakia) and since it was taken to be 'degenerated' by dialect and German influence. In addition, it lacked the prestige of a literary language since

34

Peter Auer

most serious written publications were in German. For details cf. Sgall et al. 1992:171. 16. Pedersen (in press). 17. Cf. Paunonen 1994. 18. Of course the degree of codification of Standard Swiss German or of Nynorsk is considerably higher than that of High German or English in the fifteenth or sixteenth century when graphical variation was still a marker of distinction (cf. Mihm 2000). 19. The majority of Norwegians use Bokmal as their standard variety which is structurally very similar to the urban dialect of Oslo. 20. The present-day Norwegian and Swiss situation is the result of a process in which the dialects have been upgraded in attitudinal terms; compared to the period 100 years ago, they are on the one hand seen in more positive terms today than they used to be then, on the other hand they are, at least in the case of German-speaking Switzerland, more strictly separated from the standard variety. 21. For instance, Swiss standard German is used in university lectures, in the parliament, or in the TV news broadcasts, but also when talking to foreigners, service personnel in hotels and restaurants, etc. 22. Cf. Siebenhaar 2000 and Christen 1998 for empirical studies on dialect levelling in Switzerland, Sandoy (in press) and Rjayneland (in press) for a discussion of the Norwegian situation. 23. Cf. Cheshire/Milroy 1993:10. 24. English continued to be used even after independence, particularly in jurisdiction. 25. When an exoglossic standard variety was already part of the repertoire, these elites sometimes were in the paradoxical situation of having to learn the endoglossic standard without the support of a dialect since their everyday language was this exoglossic standard. A good example is Finland in the nineteenth century where the elites propagating the codification and Ausbau of a Finnish standard language usually spoke Swedish and were not competent in a Finnish dialect (cf. Paumonen 1996, 1997). 26. Here, the case is somewhat more complicated since Swedish-speaking immigrants played a role in the formation of kyökki Finnish as well. 27. This difference may be regionalised, leading to regional standard varieties, in a larger language area such as the German one, while regional traits may be absent from it in a smaller language area such as the Danish one. 28. Cf. Pedersen 1996, Kristensen (MS). It should be added that Low Copenhagen speech also has a direct impact on the vernacular speech forms in Denmark today. 29. The katharevousa was given up in 1976 as the official standard language and replaced by a demotic variety, which, however, has integrated many structural elements of the katharevousa. This suggests that, contrary to what the term

Europe's sociolinguistic unity

35

„diglossia" used in the literature on modern Greek for the relationship between demotiki and katharevousa implies, the archaising written standard (katharevousa) and the spoken (common and also written) (standard?) language used in everyday life (demotiki) formed a continuum rather than two strictly separated varieties. The term diglossia is misleading also because the 'lowest' stratum of the Greek repertoire, the dialects (and 'idioms'), are by no means to be identified with the so-called demotic variety, the latter having been used in (literary) writing a long time before it was promoted to the national standard (in the version of the Koini Nea Eliniki). The demotiki was also codified by a grammar as early as 1938. 30. If this description is accurate, we are dealing with diglossia in the best sense of the word, and the addition of a „pseudo" seems unnecessary. It should be mentioned though that other Italian dialectologists such as Berruto (1989a, b) postulate (my) Type C for the whole of Italy. Berruto (1989b) presents a picture of Italy as a whole in which the dialects are levelled {koineizzazione) and where there is convergence towards the standard (italianizzazione), while the standard variety destandardises (dialettalizazione with the result of the creation of the italiano popolare) and standard and dialect combine into a hybrid (ibridazione). Basically, the discussion in Italian social dialectology has centred around the issue of how vast an area is covered by the emerging koinai, and this issue also offers the possibility of defining areas of diaglossia (or dilalia in Berruto's terms). 31. Cf. Map 1 in Hägen 1989. 32. The Greek scholars seem to disagree here; cf. Delveroudi 1999:562 for the first view (Peloponnesian koine), Alexiou 1982:162 for the latter. 33. Cf. Villena (in press). 34. Pedersen, in press (a) tracks this development through Danish history, showing the emergence of regiolects under the influence of the standard variety during the first part of the twentieth century. 35. The process has been strengthened by the displacement of a large segment of the Greek speaking population from the northern Turkish part of the island in 1974 which led to the dissolution of traditional network structures that used to shelter the traditional dialects. 36. Sanday, in press. In the 70s of the last century, Thelander in his investigation of Burträsk in northern Sweden found a C-type repertoire (cf. Kristensen and Thelander 1984), and Nordberg/Sundgren (MS 1999) in a recent follow-up (real time) study on Eskilstuna also found a persistence of diaglossia. 37. Moosmüller/Vollmann 2001. 38. Berruto (in prep). 39. Bücherl 1982. 40. Villena (in press). 41. For this reason, the horizontal arrows in Fig. (5) reach beyond the outer limits of the pyramid linking the base dialects to the standard.

36

Peter Auer

42. Sandoy (in press). 43. For Poland, Mazur (1996:97) suggests that de-standardisation of this type may have been promoted by the purposeful recruitment of the new intelligetsia among the workers and farmers during communist times. For Belorussia, cf. Woolhiser (2001). 44. As is well known, the 'Serbocroatian language' had a fragile status and was itself an invention of the late nineteenth century. 45. See Pedersen, in press, for a detailed account of these developments in Denmark in the second half of the twentieth century. 46. In Brussels, the Dutch standard variety is dominant (and competes with French, the traditional Η-variety used by the bourgeoisie), in Antwerp, an urban regiolect has taken over which is looked upon as standard Dutch by the speakers; cf. Willemyns 1997:140. Willemyns dates the transition from „dialect dominance" to „standard dominance" to the period between World War I and II in the Netherlands, but to the period after the second World War in Belgium. 47. Gadet (in prep.). According to estimates, hardly a quarter of the population of France spoke (standard) French as their mother tongue around 1914. Today, the old dialects have more or less disappeared, while a kind of regional standard has survived. 48. Tzitzilis 2000:88. 49. Cf. Dingeldein 1996, particularly the map on p. 130. 50. Miklos Kontra (pers. comm.) draws my attention to the fact that the Hungarian dialects survive nevertheless in the Hungarian minorities in Slovakia, Romania, Ukraine etc. (cf. for details Lanstyak and Szabomihäly 1996, Csemicsko and Fenyvesi 2000). 51. There is some variation in Icelandic which, however, is restricted to phonetics and vocabulary (cf. jDrainsson and Arnason 1992, Arnson 1987). 52. In Old Irish, no dialectal variation is attested but since dialect differences do show up at later stages of Irish history, it is usually presumed that they were covered under 'standardised' Old Irish which was already powerful and accepted enough to represent the language in writing alone. 53. See Auer (1995) for an overview, as well as Thelander (1976), Moosmüller (1987) and Auer (1997) for further details.

References Alexiou, Margret 1982 Diglossia in Greece. In Standard Languages, W. Haas (ed.), 156-190. Manchester University Press. Alfonzetti, Giovanna 1998 The conversational dimension in code-switching between Italian and dialect in Sicily. In Code-Switching in Conversation, P. Auer (ed.), 180-214. London: Routledge.

Europe 's sociolinguistic unity

37

Angelov, Angel G. 1999 The political border as a factor for language divergence. Unpublished manuscript. Arnason, Kristjan 1987 Icelandic dialects forty years later: the (non)survival of some northern and south-eastern features. In The Nordic Languages and Modern Linguistics 6, P. Lilius and M. Saari (eds.), 79-92. Helsinki. Auer, Peter 1986 Konversationelle Standard/Dialekt-Kontinua (Code-Shifting). Deutsche Sprache 2: 97-124. 1995 Modelling phonological variation in German. In Verbale Kommunikation in der Stadt, I. Werlen (ed.), 9-37..Tübingen: Narr. 1996 Co-occurrence Restrictions between Variables: A Case for Social Dialectology, Phonological Theory and Variation Studies. In Variation, Change and Phonological Theory, F. Hinskens, R. van Hout and L. Wetzels (eds.), 71-102. Amsterdam: Benjamins. 1999 From code-switching via language mixing to fused lects: toward a dynamic typology of bilingual speech. International Journal of Bilingualism 3 (4): 309-332. Bellmann, Günter 1997 Between Base Dialect and Standard Language. In Folia Linguistica (Special Isssue: ed. P. Auer), XXXII (1-2): 23-34. Berruto, Gaetano 1989a On the typology of linguistic repertoires. In Status and Function of Languages and Language Varieties, U. Ammon (ed.), 552-569. Berlin: de Gruyter. 1989b Tra italiano e dialetto. In La dialettologia italiana oggi = Festschrift for Manlio Cortelazzo, G. Holtus, M. Metzeltin, Μ. Pfister (eds.), 107122. Tübingen: Narr. 2001 Dialetti, tetti, coperture. Alcune annotazioni in margine a una metafora sociolinguistica. In Die vielfältige Romania (FS H. Schmid), Vigo di Fassa, Ist. Cultural Ladin "Majon di Fascegn ", M. Iliescu et al. (eds), 23-40. in prep. Dialect convergence, mixing, and models of language contact: the case of Italy. To appear in: Dialect Change, P. Auer, F. Hinskens, P. Kerswill (eds). Cambridge: Cambridge University Press. Bücherl, Rainald 1982 Regularitäten bei Dialektveränderung und Dialektvariation. ZDL IL, 1: 1-27. Cheshire, Jenny and Milroy, Jim 1993 Syntactic variation in non-standard dialects: background issues. In Real English, J. Milroy and L. Milroy (eds.), 3-33. Harlow: Longman.

38

Peter Auer

Christen, Helen 1998 Dialekt im Alltag: eine empirische Untersuchung zur lokalen Komponente heutiger schweizerdeutscher Varietäten. Tübingen: Niemeyer. Coseriu, Eugenio 1980 'Historische Sprache' und 'Dialekt'. In Dialekt und Dialektologie, J. Göschel, P. Ivic, K. Kehr (eds.), 106-116. Wiesbaden: Steiner. Csernicsko Istvan and Anna Fenyvesi 1999 The sociolinguistic stratification of Hungarian in Subcarpathia. Multilingua 19: 95-122. Delveroudi, Rea 2001 The language question and the modern Greek dialects (1880-1910). In 'Strong' and 'weak' languages in the European Union, A.-F. Christidis (ed. Thessaloniki: Centre for the Greek Language, Vol II, 561-568. Dingeldein, Heinrich J. 1996 Sprachvarietäten in .Mitteldeutschland'. In Varietäten des Deutschen, G. Stickel (ed.), 109-141. Berlin: de Gruyter. Ferguson, Charles 1959 Diglossia. Word 15: 325-41. Frings, Theodor 1957 Grundlegung einer Geschichte der deutschen Sprache. Halle: VEB Niemeyer. (1948 3 ) Gadet, F. inprep. France. In Soziolinguistik - ein Handbuch, U. Ammon et al. (eds.). Berlin: de Gruyter (second edition). Geeraerts, Dirk, Grondelaers, Stefan and Speelman, Dirk 1999 Convergentie en divergentie in de Nederlandse woordenschat. Amsterdam: Meertens Instituut. Gilles, Peter 2000 Dialektausgleich im Letzebuergeschen. Tübingen: Niemeyer (-Phonai No 44). Hagen, Anton 1988 The Netherlands. Sociolinguistica 3: 61-74. Haugen, Elmar 1966 Dialect, Language, Nation 'Dialect, nation, language'. American Anthropologist 68. (Reprinted in: J. B. Pride/J. Holmes (eds.), 1972, Sociolinguistics. Penguin Books, 97-111.) Herrgen, Joachim 1986 Koronalisierung und Hyperkorrektion: das palatale Allophon des /ch/Phonems und seine Variation im Westmitteldeutschen. Stuttgart: Steiner. Ivars, Ann-Marie 1996 Stad och bugd. Finlandssvenska stadsmäl i ett rgionalt og socialt perspektiv = Folkmälsstudier 37, Helsinki: Föreningen for nordisk filologi.

Europe 's sociolinguistic unity

39

Kallen, Jeffrey L. 1997 Irish English, In Focus on Ireland idem (ed.), 1-33. Benjamins. 2001 Two languages, two borders, one island: some linguistic and political borders in Ireland. IJSL 145: 29-63 (Dialect convergence and divergence across European borders, J.L. Kallen, F. Hinskens and J. Taeldeman (eds.)). Karyolemou, Marilena 2000 La variete chypriote: dialecte ou idiome? In La langue grecque et ses dialectes .A.-Ph. Christidis (Ed.), 111-115. Athens: Greek Republic/Center of the Greek Language. 2001 From linguistic liberalism to legal regulation: the Greek language in Cyprus. Language Problems and Language Planning 25 (1): 25-50. Keevallik, Leelo and Pajusalu, Karl 1995 Linguistic diversity and standardization in Estonian: the history of the active past participle. DiG 3: 13—41. Kloss, Heinz 1967 Abstand Languages' and ,Ausbau Languages'. Anthropological Linguistics 9 (7): 29-41. 1976 Über 'Diglossie'. Deutsche Sprache 4: 313-323 1978 Die Entwicklung neuer germanischer Kultursprachen seit 1800. Düsseldorf: Schwann. (1952 2 ) Kontra, Miklos 1994 On current research into spoken Hungarian. IJSL 111: 9-20. Kristensen, Kjeld Standard Danish, Copenhagen Sociolinguistic, and Regional Varieties in the 1900s. Unpublished manuscript. Kristensen, Kjeld, and Thelander, Mats 1984 On dialect levelling in Denmark and Sweden. FoL XVIII/1-2, 223246. Lanstyak Istvan and Gizella Szabomihaly 1995 Contact varieties of Hungarian in Slovakia. A contribution to their description. IJSL 120, 111-130. Mackridge, Peter 1998 Le discours public en Grece apres 1'Abolition de la Diglossie. In 'Strong' and 'weak' languages in the European Union, A.-F. Christidis (ed), Vol II, 133-136. Thessaloniki: Centre for the Greek Language. Marie, Jaap van 1996 Dialect versus standard language: nature versus culture. In Taming the vernacular. From dialect to written standard language, J. Cheshire/D. Stein (eds.), 13-34. London: Longman. Martin Zorraquino Maria Antonia 1997 Sociolinguistic attitudes and beliefs towards dialectal and standard varieties. In La Franja de Aragon (Spain). FoL XXXII/1-2 (Special

40

Peter Auer

Issue on Dialect Levelling and the Standard Varieties in Europe, P. Auer (ed.)), 131-144. Martin Zorraquino, Maria Antonia, Fort, Maria Rosa and Arnal, Maria Luisa (et al.) 1995 Estudio sociolingüistico de la Franja Oriental de Aragon. Zaragoza. Mazur, Jan 1996 Konvergenzen und Divergenzen in den polnischen Sprachvarietäten. Sociolinguistica 10: 53-74. Michelena, Luis 1981 Normalizacion de la forma scrite de una lengua: el caso vasco. In Demonte V. / Varela S. (eds.), El bilingüismo: problemätica y realidad. Madrid (=Revista de Occidente 10-11), 55-75. Mihm, Arend 2000a Zur Deutung der graphematischen Variation in historischen Texten. In Vom Umgang mit sprachlicher Variation. (=FS H. Löffler),A. Häcki Buhofer (ed.), 369-390. Tübingen: Franck. 2000b Die Rolle der Umgangssprachen seit der Mitte des 20. Jahrhunderts. In Sprachgeschichte - ein Handbuch, Ammon U. et al. (eds.), Berlin (second ed., vol. II), 2107-2137. Miller, Jim 1993 The Grammar of Scots English. In Real English, Milroy and Milroy (eds.), 99-138. Longman. Milroy Jim and Milroy Leslie 1984 Linguistic change, social network and speaker innovation. Journal of Linguistics 21: 339-384. Moosmüller, Sylvia 1984 Soziophonologische Variation im gegenwärtigen Wiener Deutsch. Wiesbaden: Steiner. Moosmüller, Sylvia and Vollmer, Ralf 2002 'Natürliches' Driften' im Lautwandel: die Monophthongierung im österreichischen Deutsch. Zeitschrift für Sprachwissenschaft 20 (1): 42-65. Moser, Hugo 1950 Deutsche Sprachgeschichte. Stuttgart: Schwab. Nekvapil, Jiri 2000 Language management in a changing society. Sociolinguistic remarks from the Czech Republic. In Die sprachliche Situation in der Slavia zehn Jahre nach der Wende Β. Panzer (ed.), 165-177. Lang. Nekvapil, Jiri and Chloupek, Jan 1986 Introduction: On Czech Linguistics and Sociolinguistics. In Reader in Czech sociolinguistics, idem (eds.), 7-18. Benjamins. Nordberg, Β engt and Sundgren, Eva 1999 On the rate of change of some variables in present-day Swedish. Paper presented at the NewWAVE 28, Toronto. (Unpublished manuscript, University Uppsala).

Europe 's sociolinguistic unity

41

Paunonen, Heikki 1994 The Finnish language in Helsinki. In The sociolinguistics of urbanization: the case of the Nordic countries, B. Nordberg (ed.), 223-245. Berlin: de Gruyter. 1996/97 Finland. In Kontaktlinguistik - ein Handbuch, Η. Goebl et al. (eds.). 993-1007. Berlin: de Gruyter. Pedersen, Inge Lise 1996 Regionalism and linguistic change. Sociolinguistica 10: 75-89. in press a Traditional Dialects of Danish and the De-dialectalization 19002000. In International Journal of the Sociology of Language in press b Processes of standardization in Scandinavia. To appear in Dialect Change. P. Auer, F. Hinskens and P. Kerswill (eds), Cambridge University Press. Royneland, Unn 2001 in press Dialects in Norway - catching up with the rest of Europe? to appear in: International Journal of the Sociology of Language Sand0y, Helge in press Nordic language history and current trends in dialectology. In The Nordic Languages, O. Bandle et. al. (eds.), Ch. 35. Berlin: de Gruyter. Sandor, Klara 1999 National feeling or responsibility. The case of the Csängo language revitalization. Multilingua 19(1-2): 141-168. Sgall, Petr, Hronek, Jifi, Stich, Alexander and Horecky, Jan 1992 Variation in Language. Code Switching in Czech as a Challenge for Sociolinguistics. Amsterdam: Benjamins. Siebenhaar, Beat 2000 Sprachvariation, Sprachwandel und Einstellung: der Dialekt der Stadt Aarau in der Labilitätszone zwischen Zürcher und Berner Mundart. Stuttgart: Steiner. Sobrero, Alberto 1995 Italianization and variations in the repertoire: the koinai. Sociolinguistica 10: 104-111. P. Auer, F. Hinskens and K.J. Mattheier (eds.). Thelander, Mats 1976 Code-switching or code-mixing? Linguistics 183: 103-123. 1982 De-dialectalization in Sweden. In Linguistica e Antropologia = Atti del XIV congresso internazionale die studi, 175-195. Roma. Trumper, John and Maddalon, Marta 1987 Converging divergence and diverging convergence: The dialectlanguage conflict and contrasting evolutionary trends in modern Italy In Variation and Convergence, P. Auer and A. di Luzio (eds.), 217— 259. Berlin: de Gruyter.

42

Peter Auer

Tzitzilis, Christos 2001 Dialectes et dialectologie du Grec moderne. In Variation and Convergence, Republique Hellenique/Ministere de l'education nationale et des cultes (ed.), La language Grecque et ses dialectes, 83-90. Athens. {^rainsson, Höskuldur and Arnason, Kristjan 1992 Phonological variation in 20th century Icelandic. In Islenskt mal 14: 89-128. Van de Velde, Hans 1996 Variatie en verandering in het gesproken Standaard-Nederlands (1935-1993). Ph.D. thesis, Katholieke Universiteit Nijmegen. Videnov, Michail 1998 The Present-Day Bulgarian Language Situation: Trends and Prospects. International Journal of the Sociology of Language 135/137: 11-36. Villena Ponsada, and Juan Andres 1996 Convergence and divergence in a standard-dialect continuum: networks and individuals in Malaga. Sociolinguistica X: 112-137. P. Auer, F. Hinskens and K.J. Mattheier (eds.). 2001 Identitad y variation lingüistica: prestigio nacional y lealtad vernacular en el Espanol hablado en Andalucia. In Identidades lingüisticas en la Espana Autonomica, Bossong Georg/Bäez de Aguilar, Francisco (eds.), 107-150. Frankfurt. in prep. The Iberian Peninsula, in: Soziolinguistik - Ein Handbuch (U. Ammon et al., eds.). Berlin: de Gruyter (second edition). Widmark, Gün in print Standard and colloquial languages in Scandinavia in the 19th century. In The Nordic Languages, O. Bandle et. al. (eds.). Berlin: de Gruyter. Willemyns, Roland 1996 Dialektverlust im niederländischen Sprachraum. ZDL 64 (2): 129-154. Woolhiser, Curt 1999 Language ideology and language conflict in post-soviet Belarus. In Language, Ethnicity and the State, Camille C. O'Reilly (ed.), Vol. 2, 91-122. London: Palgrave.

From phonetic similarity to dialect classification: A principled approach Paul Heggarty, April McMahon and Robert McMahon1

1. Language versus dialect comparison In many ways, dialects are like related languages: in both cases, we are dealing with synchronic similarities, and an underlying history of divergence. One might therefore expect that any tools developed for classifying language families would generalize naturally to the dialect level. Yet in fact the preoccupations of classificatory historical linguistics particularly following the recent controversies over Greenberg (1987) and Dixon (1997), for instance - lie mainly with very large-scale comparison, and dialects tend to be ignored. Even if we did try to include dialects, the questions to be asked are not the same: we are not worrying about whether varieties are related, but how closely, and the usual models in historical linguistics simply do not work at the required level of resolution, or of detail. To illustrate this, consider what is perhaps the best known approximately quantificational method for calculating degrees of similarity between languages, namely lexicostatistics. This involves selecting some basic word meaning; identifying (or choosing - and this begs the significant question of just how we determine what the 'normal' translation is) its corresponding lexemes in a number of languages; and seeing if they are cognate. With the meaning 'castle', for instance, lexicostatistics would identify these corresponding forms in various Romance languages: Italian castello, Spanish Castillo, and French chäteau (fort); and all of these are derived from Latin castellum. It follows that a lexicostatistical quantification of the 'linguistic distance' between some Romance language varieties on the basis of this set of words would produce the outcome illustrated in Figure 1 below. Of course, Figure 1 is rather obvious, and might be thought a little unfair. However, it does make clear one most important point, which is that lexicostatistics is blind to what it scarcely takes a linguist to tell: namely that Italian castello [kas'tello] is far more like Spanish castillo [kas'tiXo] than either is to French chäteau [jato]. There is a great deal of comparative data in these words, awaiting quantification; but this cannot be achieved on

44

Paul Heggarty, April Mc Mahon and Robert McMahon

French: Poitou

French: Paris

Portuguese: Rio

Spanish: S. del Estero

Spanish: Buenos Air.

Spanish: Caracas

Spanish: Bogota

Spanish: Madrid

Italian

the level of meaning. And lexicostatistics passes by the level of sound completely. The difficulty in applying lexicostatistics at the level of dialects rather than distinct languages is that they generally share far 'too many' cognates, even though there may be fairly stark differences in phonetics. Yet it is only the limitations inherent in the method which make us see this as a problem; we can just as well see these cognates as an unusually rich data source. In the case of long-range, or even family-level comparison, we tend to find that cognates are hard to come by, so that a great deal of weight is carried by a restricted number of data points.

100 100 100 100 100 100 100 100 100 Classical Latin 100 100 100 100 100 100 100 100 Italian

*kastellum kastello

100 100 100 100 100 100 100 Spanish: Madrid

kastiXo

100 100 100 100 100 100 Spanish: Bogota

kastijo

100 100 100 100 100 Spanish: Caracas

kahtijo

100 100 100 100 Spanish: Buenos Aires kahtijo 100 100 100 Spanish: S. del Estero

kasti30

100 100 Portuguese: Rio

kajtelu

100 French: Paris French: Poitou

Jato Jato

Figure 1. Quantification of the similarity between a selection of Romance varieties as calculated by lexicostatistics for the meaning 'castle' 100 = identity

In dialect comparison, meanwhile, we do have a great deal of directly comparable and potentially quantifiable data at our disposal, which we could exploit to produce finer-grained measurements, if only the method of comparison were not limited to lexical meaning. Far from the castellum

From phonetic similarity to dialect classification

45

words being of no use because they are all cognate, on the contrary they carry a mass of comparative data within them, which are actually rather interesting and something we ought to be able to measure. (Of course there are always certain differences between dialects in lexis too, and one would ideally want to measure those as well, though that would require a separate technique dedicated to lexis proposed in Heggarty (in preparation). Such a method would have to overcome the much-discussed problems with the lexicostatistical method, and take analysis of similarity in lexical semantics to a level precise enough to give sufficiently distinct quantifications at the dialect level.) In any case, independently of this particular application in dialectology, there are very good reasons within phonetics and phonology for wanting to be able to measure phonetic similarity. That this has not really been achieved yet is an indication of just how slippery the very concept of phonetic similarity has been felt to be, and how hard to define. We shall return shortly to this question, and the approach we have taken to it that is behind the quantification method we propose. The simplest way to get a first grasp of what our method does, however, is to see the type of results it produces. Figure 2 shows our measures of phonetic similarity/difference between the cognates of castellum in a range of Romance varieties in which they have different pronunciations. The key to understanding these results is that we have obtained a more refined quantification of the degree of difference because our calculations are based upon scores of times more data per word than lexicostatistics uses, simply because we are working with more than a single data point per word. Whereas in lexicostatistics the method by its nature can ask only whether two items are cognate or not, giving a score of yes or no, one or zero, our method analyses and compares at a fairly deep level of phonetic detail. In the rest of this paper we shall explain just how our method comes up with these results, and quite what we claim for them. At this stage, it is enough to demonstrate that they are at least close enough to intuitive ideas about phonetic similarity to merit that further investigation.

Spanish: Madrid

Spanish: Bogota

Spanish: Caracas

Spanish: Buenos Air.

Spanish: S. del Estero

Portuguese: Rio

French: Paris

French: Poitou

Paul Heggarty, April Mc Mahon and Robert McMahon

Italian

46

81

72

70

68

66

68

76

34

33

Classical Latin

90

88

85

83

86

92

40

39

Italian

kastello

97

94

90

93

93

45

43

Spanish: Madrid

kastiXo

97

93

96

90

43

41

Spanish: Bogota

kastijo

90

93

88

43

41

Spanish: Caracas

kahtijo

97

86

37

36

Spanish: Buenos Aires kahtijo

89

39

37

Spanish: S. del Estero

kasti3o

43

41

Portuguese: Rio

kajtelu

97

French: Paris

Jato

French: Poitou

Jato

*kastellum

Figure 2. Phonetic similarity of cognates of Latin castellum in a selection of Romance varieties 100 = identity. Figures in bold face denote comparisons between varieties generally considered 'accents' of the same language, rather than different languages. Figures in italics denote tentative results for comparisons against assumed pronunciations for non-modern languages, for which phonetic forms cannot be certain - those used here for Classical Latin follow Allen (1978). As a second illustration of our technique, this time as applied not just to a single word but to many, Figure 3 below shows combined results for all the numerals from 'one' to 'ten' from a range of Romance varieties. Even for all of these forms together, lexicostatistics would again only have 100 in all these boxes, because all the numerals are cognate. Worse still, this would remain true even if Germanic languages were included alongside these Romance ones. The method presented here, in contrast, draws out of the very same words enough data to quantify degrees of similarity between them. We maintain that this is possible because

From phonetic similarity to dialect classification

47

phonetics simply supplies data inherently more suitable to quantification and comparison than does lexical meaning. These results were produced by one of a range of methods for language classification and comparison which are the subject of our current research project.1 Crucially, the methods are quantitative, and therefore computational, though it is most important to note that the one that produces the results shown above is strictly not a statistical method, since we are not subjecting our results at present to any tests of significance in a statistical sense, for instance. The method presented here involves a database and programme written in Microsoft Visual Basic for Microsoft Excel by Paul Heggarty (see also Heggarty 2000, 2001), to whom technical programming questions should be addressed. Before proceeding it is important to note that no claim is being made here that our figures somehow represent the definitive result, the perfect mathematical expression of the degree of phonetic difference between these forms; nor that a data set of just ten words (and a very particular lexicon subset at that) could be correctly representative of the languages as a whole. Results for a rather fuller data set will be presented later; the reason for presenting Figures 2 and 3 now is simply the better to demonstrate the main point: that these detailed results are obtained from precisely the same data corpus from which lexicostatistical methods could produce no differentiated figures at all - albeit a corpus considered at a different linguistic level, one more amenable to detailed and quantitative analysis. In this light, the extent of what might be claimed for these figures is that they do appear to have at least some relative validity; that is, they represent a considerable step towards capturing what we can tell even impressionistically - in fact we can go so far as to say we know - are the relative degrees of difference between these cognates in phonetics. Nonetheless, our confidence in how accurate, indeed how meaningful, these quantifications really are must ultimately hang on the nature of the methods used to produce them. It is to these that we now turn, before ending with a consideration of their potential applications in dialectology, and of possible extensions. It should be noted, however, that this short paper is intended as a guide to the principles underlying the method, and a discussion of its rationale and general structure, along with some preliminary and indicative results; it is not a complete outline of the method, far less a user's manual. Further details of the method and its operation can be found in Heggarty (2000a, 2000b); and a list of current and forthcoming papers on the topic can be found at: www.shef.ac.uk/english/language/quantling/papers.html.

Spanish: Madrid

Spanish: Bogota

Spanish: Caracas

Portuguese: Lisbon

Portuguese: Rio

Portuguese: Salvador

Port: S.P. State interior

French: Paris

French: Southern

Romanian

Paul Heggarty, April Mc Mahon and Robert McMahon

Italian

48

60

54

54

52

49

48

50

52

40

40

56 Classical Latin

68

69

66

57

61

60

60

43

43

61 Italian

98

94

66

69

70

69

47

47

54 Spanish: Madrid

94

66

69

70

69

47

47

54 Spanish: Bogota

62

67

66

66

45

45

54 Spanish: Caracas

89

88

94

56

56

53 Portuguese: Lisbon

93

91

54

54

53 Portuguese: Rio

91

54

54

54 Portuguese: Salvador

54

54

55 Port: S.P. State interior

93

45 French: Paris 45 French: Southern Romanian

Figure 3. Phonetic similarity of cognates of the numerals 'one' to 'ten' in a selection of Romance varieties 100 = identity. Figures in bold face and italics are as per Figure 2.

Of course, we are not claiming that this is the first attempt to measure the differences between dialects, or to quantify differences in phonetics. Goebl (1982, 1984), for instance, has pioneered work on dialectometry: his attempts to measure dialect differences are limited to lexical data, though he does introduce elements of weighting, which will be important in the illustration of our phonetic method below. Other methods that involve some form of phonetic quantification have been developed too, and there is a considerable literature on this topic (see, for instance, Kessler 1995, Nerbonne and Heeringa 1997). These alternatives tend to be based on the calculation of 'Levenshtein distances' (Levenshtein 1965), and differ systematically, in principle and practice, from the method outlined below.

From phonetic similarity to dialect classification

49

We return to a comparison between our method, and that of Kessler, Nerbonne and Heeringa, and their collaborators, in Section 5. For the moment, however, it is essential to demonstrate the operation of our own method more fully, so that this comparison can be more informed.

2. Measuring phonetic similarity The whole concept of phonetic similarity, and especially any suggestion of putting figures on it, is something phoneticians and phonologists have tended to steer well clear of. Lass (1984: 19) notes that phonetic similarity is "...a tricky criterion; but we can accept it as intuitively plausible, and in practice workable, even if not formally definable." However, to many nonlinguists - particularly the geneticists and archaeologists who are increasingly seeking to work together with historical linguists - this comes as a shock, accustomed as they are to the premium that their own disciplines put on reliable quantification. We side with them, and with Laver (1994: 391), in feeling that this is not really good enough: One of the most basic concepts in phonetics, and one of the least discussed, is that of phonetic similarity. An adequate general phonetic theory should be able to permit any two phonetic events to be compared and rated on a scale of relative similarity... Such a theory has not yet been completely developed. .. No explicit matrix exists to date which would allow such a graded comparison to be carried out across the full range of phonetic entities with any great sensitivity of comparison... Issues of phonetic similarity, though underlying many of the key concepts in phonetics, are hence often left tacit.

As just one instance of how phonetic similarity underlies a key concept in phonetics, consider the traditional definition of the phoneme as a collection of allophones in complementary distribution. This definition must crucially be tempered by a second requirement that allophones of a single phoneme be phonetically similar, to rule out notorious cases like English [h] and [η], which despite their complementary distribution are always regarded as different phonemes. The problem was, and is, that quite how much phonetic similarity was 'enough' to qualify was never made explicit, and indeed noone proposed a means of measuring it anyway. Yet after all the decades of work and advances in our analyses and models of phonetics, surely we should by now be in a position to start putting some meaningful figures on this notion that "underlies many of the key concepts in phonetics". Failing that, surely we should at least be concerned to try. It is in this spirit that the work reported here should be seen.

50

Paul Heggarty, April Mc Mahon and Robert McMahon

Perhaps the main reason so little work has been done in this direction hitherto is because the enterprise of putting meaningful figures on the precise degree of difference between language varieties can seem inherently rather daunting, for two fundamental reasons to do with the distinct tasks of comparison and quantification respectively. First, corresponding to the task of comparison is the cross-linguistic, or compatibility problem: how can we analyse and accommodate all the differences between individual languages within a single, unified model? Mallory (1989: 156) has a particularly helpful characterization of this problem: "Unfortunately, we must ensure with such a procedure that we are comparing like with like ... there is no recognized means of measuring degrees of conservatism and weighing each change in phonetics, grammar or vocabulary against another. We must ultimately trust to impressions that cannot help but be subjective." Secondly, we have the quantification problem: how can we reduce language structure to numerical expression, and put actual numbers on a concept such as 'phonetic similarity', or the phonological complexity of a language? Naturally, precise quantification means being as accurate and detailed as possible, so that any overall calculations must combine many individual data points for the various comparand features. Crucially, however, any overall numerical values compounded from more than one feature will be next to meaningless unless the features involved are also balanced, or 'weighted', against each other. Indeed, the failure to pay attention to doing so adequately can be seen as one of the greatest weaknesses of previous quantification methodologies which might seem superficially similar to the method being developed here (see, for example, Grimes and Agard 1959). Quantification cannot simply be a matter of measuring and counting a set of features; it also has to be about establishing the relative significance of those features. These two basic challenges for quantification and comparison also mean that one has to choose one's data appropriately, for however sophisticated the programming, the method can only work if the data themselves are inherently suitable to being quantified and compared. Our attempts to resolve the quantification and compatibility problems crucially involve building into the quantification technique itself different categories of detailed linguistic knowledge which we already have. In the case of the quantification problem, we build in our general knowledge of phonetics; and for the compatibility problem, we incorporate into the method our specific knowledge of the relationships between cognate forms and the processes of change by which they have diverged.

From phonetic similarity to dialect classification

51

3. The quantification problem How, then, can we make use of what we know about phonetics to help solve the quantification problem and assess numerically the phonetic similarity of any two phones? The main mechanism here involves a phonetic version of distinctive feature theory. Thus, in comparing a [p] and a [b], one finds two basic parameters in which they share the same feature, and one in which they differ: in LOCATION OF STRICTURE both are [bilabial], and in DEGREE OF STRICTURE both are [stop]; but they differ in their value for VOICING. If we were to compare by these three parameters alone, then assuming the parameters are assigned equal weight, the overall similarity rating for this pair of phones would be 2 out of 3, or 0.67. Note again that, although there are some similarities between this method and that of Grimes and Agard (1959), there are also very significant differences. In particular, the method put forward here compares phones rather than phonemes; it uses a far bigger range of parameters of comparison; and it does not limit analysis to segments but incorporates the concept of gestures as found in Articulatory Phonology (see Browman and Goldstein 1992). These and other emendations allow us to deal appropriately with issues of timing, secondary articulations, the similarity of consonants to vowels, and so on. As noted earlier, however, any figures we propose rely absolutely on an assessment of the relative significance of the different features analysed: we consider this weighting to be the key to successful quantification. For the purposes of weighting, we must assess the significance of phonetic differences, relative to each other. We have used three main types of weighting principle to help decide on these degrees of significance and weighting. Guiding principles are to be found firstly, and most crucially, in an appeal directly to language itself, specifically to what might be called 'natural principles' in the organizational structure of language. Consider the confident assertion with which Laver (1994: 592) concludes his authoritative Principles of Phonetics: "underlying the apparently extraordinary diversity shown by the thousands of mutually incomprehensible languages of the world, there is a remarkable, elegant and principled unity in the way that these languages exploit the phonetic resources of speech". Laver clearly regards phonetics as an inherently closed and highly structured system, whose various parameters can therefore largely be defined independently and objectively, to a high degree of detail. As such, phonetics represents a fairly suitable data type for quantification: it is this very nature of the field that can impart a high degree of objectivity to numerical assignments based directly upon its structural principles. The architecture of phonetic analysis thankfully provides objectively definable,

52

Paul Heggarty,

April Mc Mahon and Robert

McMahon

relatively independent parameters, in whose terms it is possible to be much more definitive and objective in analysing the degree of difference between, say, a phone [s] and the phones [z], [t] and [d], than that between word meanings such as Latin homö and English person, human or man. It is this nature of phonetics that permits an analysis more amenable to mathematical expression. The method put forward here duly strives to make use of, and indeed relies upon, this inherent suitability of phonetics for quantification, in particular in the key issue of assessing significance. The very architecture of phonetic classification is taken to stand as a 'ready-made' natural and objective guide to the relative significance of the various elements within that structure. As an illustration, consider what Laver(1994: 575) identifies as the 'default assumptions' for articulation: pulmonic egressive, a single stricture, articulators in a neutral (not displaced) configuration, airflow routing oral and central, and so on. These defaults are taken to establish a concept of a weighting 'core' made up of the three remaining parameters, i.e. those for which there is no default assumption and which must be specified for all sounds. For consonants these are the three discussed above in the comparison of [p] with [b]: LOCATION OF STRICTURE, DEGREE OF STRICTURE and GLOTTIS STATE. This 'core' serves in the quantifications as a basic weighting unit; all non-core parameters (covering laterals, trills and taps, ejectives, retroflex sounds, and so on) are assessed and weighted for their significance relative to the core. The phonetic analysis this model follows is generally that of the IPA chart and Laver(1994). The analysis of vowels follows the IPA system, itself something of a mixture of articulation and perceptual acoustics as to the relative 'phonetic distances' between the cardinal vowel points. For consonants, ideally the model would combine full articulatory and acoustic analyses, though as far as it has been possible to develop it to date it remains, with Laver and the IPA chart, essentially articulatory. As yet acoustic input has had to be limited to a number of ad hoc devices to balance the analyses of those sounds whose similarity in acoustic terms departs radically from their degree of similarity in articulation, and would therefore be seriously misrepresented by a purely articulatory analysis (e.g. [x] and [f], or alternative bunched and retroflex realisations of English III). The second general type of weighting principle used as an objective guide to relative significance is what we have called 'cross-linguistic norms'. If one studies any typological phonology database, a number of very clear cross-linguistic patterns are immediately apparent, not least that typically the LOCATION and DEGREE OF STRICTURE parameters bear more phonemic distinctions than does the VOICING parameter. Cross-linguistically, one generally finds three distinctions in LOCATION OF STRICTURE (for stops at

From phonetic similarity to dialect classification

53

least), the default being labial, denti-alveolar and velar; and likewise three distinctions (stop - fricative - approximant) in DEGREE. On the VOICING parameter, however, one typically finds just a two-way opposition of voiced versus voiceless. These clear cross-linguistic norms for phonemic distinctiveness can be taken as a guide to the relative significance of the phonetic differences on any given parameter. If one builds into the method a notion of how many phonemic distinctions a parameter generally bears, then it follows that our method does not in fact award equal weights to the three parameters in the comparison of [p] and [b]. It actually assigns 2 weighting points for DEGREE OF STRICTURE, and 2 for LOCATION, but only 1 for VOICING state, corresponding to the usual number of contrasts those dimensions carry in a typical language. This means the relative significance of the oral gesture to the voicing gesture is 4:1; [p] and [b] are therefore the same for 4 of those points of similarity, and different for 1, namely voicing. Four-fifths similarity is also statable as 0.8 (where 1 denotes identity and 0 the maximum difference possible on all the core parameters). Further illustrations for just these standard parameters of consonant comparison are shown in (1). (1)

[d] vs. [g] = 0.8 [d] vs. [s] = 0.6

[t] vs. [s] = 0.8 [b] vs. [s] = 0.4

[b] vs. [g] = 0.6 [p] vs. [x] = 0.4

Similar calculations can be made for vowels, given that again there are three basic 'default' dimensions - height, backness and lip-rounding which are also to be weighted differently according to the way they typically bear phonemic distinctions cross-linguistically. As is particularly appropriate for vowel height and backness, incremental steps along parameters that can bear more than one contrast are generally seen cumulatively in the quantification of degree of difference, applying to quantification a concept similar to the 'scalar' features of phonological analysis. Of course, as any phonetician or phonologist knows, sounds are related to each other in highly complex ways. Representing the degrees of difference between them at all meaningfully requires considerable refinement in the model used to weight those differences. So for instance, the various different phonetic 'dimensions' are not fully independent, but interact with each other significantly. In a fair proportion of languages, the denti-alveolar region supports rather more cross-linguistic phonemic distinctions in fricatives than in stops; likewise the vowel space tends to support rather more distinctions in backness in high vowels than in low ones. Our model incorporates appropriate adjustments to the numerical values assigned to the corresponding degrees of phonetic difference.

54

Paul Heggarty, April Mc Mahon and Robert McMahon

Similarly, the consonant vs. vowel opposition is in practice not at all as clear cut as it is often modelled for convenience ('semivowels'/ 'semiconsonants' and rhotic vowels are clear examples of intermediate status). The method here does not insist on treating the two as necessarily distinct, but allows vowel to consonant comparisons perfectly well, by calculations of difference relative to the most similar intermediate sound. (This applies the 'polar co-ordinate' analysis of the interaction between the vowel space and consonant description.) Finally, the model applies a number of purely logical principles, not specifically linguistic at all, which are particularly important for deciding on the mathematical proportions assigned to given relationships of significance. One important such principle is what we have called the 'simplest ratio' principle, which states that the mathematical ratio assigned to any given relationship is to be the simplest ratio within the greater than and less than bounds set by any relevant hierarchies. As an illustration of a specific linguistic application of this general principle, consider segment length. We can justifiably say that a long vowel or geminate consonant is somewhat shorter than two whole vowels or consonants, but somewhat longer than a single normal-length vowel or consonant. So we would wish to quantify the length of a long vowel or geminate as somewhere intermediate between two and one normal-length segments. Beyond that, however, there is no easy and justifiable way of calculating with any greater precision just how much closer it might be to one or to two. When we come up against these limits of what we can unequivocally justify linguistically, the simplest - and least subjective - way of reconciling the limits of our relevant (but unquantified) knowledge with the need for a specific figure for the model to use is to implement the simplest ratio principle, which calls on us to assign a weighting exactly intermediate between the two. Note, however, that the interpretation of 'intermediate' depends crucially on the fact that, in assessing and weighting relative significance, one has always to deal with proportions. So while the instinctive reaction might be that 'halfway' between 2 and 1 is 1.5 normal length segments, when one is working with proportions, this is not actually so. This is because the ratio of 1.5 : 2 is not the same as the ratio of 1 : 1.5. The correct value is actually V2, alias 1.4142..., since this is the value that gives a ratio (or proportion) equal in each case. Only a value of y/2 for χ allows us to rate one long segment at χ times the length of a single normal-length segment, and likewise weight two normal-length segments together at the same χ times the length of that long segment. A geminate [t:] compared to a [t] therefore comes out as a similarity rating of 0.7071, with the difference not in quality, but in length. The degree of difference between a geminate [t:] and a single [d], meanwhile, would be calculated as a compound of the

From phonetic similarity to dialect classification

55

differences both in quality and in length (two different dimensions in the model). Diphthongs and affricates are analysed in a similar way in overall length - they are modelled as y/2 times the length of a single segment. Here, however, the same quality is not maintained throughout, so each diphthong or affricate is modelled as two different parts: each half is weighted at jO.5, or 0.7071, and the two components together add up to the length of a geminate or long segment. (Different types of diphthongs, and combinations of semivowel plus full vowel, may be split differently as appropriate). It is crucial to remember that we are not making precise claims about actual phonetic length - indeed we actually prefer the notation 72, as a reminder that the model works with proportions, rather than 1.4142, which when expressed decimally appears an absurdly precise specification of how different a diphthong is from a single vowel. *J2 should more correctly be considered as the proportional expression that it is, as is entirely appropriate for a solution to the problem of weighting length for the purposes of quantification. Lest this article become bogged down in such numbers - essential though they are to illustrate important aspects of how the method works one point should be stressed about all these details of the quantification method. While these are issues that did indeed require very considerable thought and attention in developing the comparison, they are now built into the programme. All that it now demands of the linguist who wishes to use it for a given data-set of cognates is simply to enter their phonetic transcriptions into the database, together with basic numerical correspondence information on how the segments and gestures of each cognate map onto those of the common node skeleton used for those cognates (as we shall see below). From this input data, the programme calculates all the comparisons and weightings automatically.

4. The compatibility problem The method as illustrated thus far only lets us quantify the amount of phonetic difference between phones in isolation. Extending this to comparing the phonetics of entire words against each other brings us up against the second, compatibility problem. Turning again to lexicostatistics as a comparitor, we find that lexical meaning is used as the 'common denominator' to identify what to pair up between languages: the linguist chooses the lexical item that normally expresses a specific meaning in each language. The serious difficulties with

56

Paul Heggarty, April Mc Mahon and Robert

McMahon

trying to achieve cross-linguistic compatibility through lexical meaning have been much discussed in the literature on lexicostatistics. From our perspective, lexical meaning is a type of data inherently ill-suited to providing an objective solution to the compatibility problem, at least if one sticks to the traditional insistence on a single one-to-one identification of words in different languages. The reasons have to do with the inherent subjectivity of choosing meaning 'matches', and the other well-known question of whether the items in the usual lists, like the 'industry standard' 100 and 200 word lists (Swadesh 1952) really are at all as culture-neutral as is claimed. By not using lexical meaning, we free ourselves from all of these problems; but what, then, do we use instead to solve the compatibility problem? The more specific instantiation of this problem for us involves identifying which particular phones and articulations (or 'gestures') from a cognate in one language variety are to be compared against which in the cognate from a different variety. One cannot simply assume a sequential, linear analysis, since processes like metathesis, epenthesis, segment loss and coalescence frequently disrupt the original pairing of segments consider which segments the nasal gestures are found on in cognates such as Italian buono [bwono] and French bon [bo]. We need to be able to pair up the segments and gestures which ought to be compared, even if the linear structures of the present-day words no longer match straightforwardly slot-by-slot. As we saw in Section 3 above, the quantification problem was tackled largely by applying our existing knowledge of phonetics. Here too we apply our existing knowledge, this time our understanding of how cognates relate back to a common ancestral form, and as our 'common denominator' we make use of exactly that proto-form. That is, the common ancestor form of each word is used to link up its reflexes in the differedaughter languages. It is analysed as a 'template' of timing and gesture slots, which is used as a 'node' through which to assess which segments and gestures in one cognate are to be compared against which in the other. Figure 4 below offers an example of how this works for the sample castellum word, here showing how the sounds in Italian castello [kas'tello] are matched to those in French chäteau [fato]. Bear in mind that Figure 4 is a much simplified illustration, in a particularly easy case, of the full matching mechanism involved, which there is simply not space to lay out here in full detail (see Heggarty 2000 and in preparation). This example will serve to illustrate some issues, not least that it is hardly feasible to expect of languages that their cognates be of equivalent length. It is also clear that this is by no means a simple one-to-one, segment-to-segment comparison: where appropriate, more than one segment in one word can be compared against just one in another.

From phonetic similarity to dialect classification

ITALIAN

NODE

FORM

k

a

s

t

ε

1

I

0

I

ί

I

I

I

I

I

I

[kas'tello]

chateau

Lfato]

*k *a *s *t *ε *1 *1 *u *m I I

FRENCH

castello

57

J

a

I

I I

t

0

Figure 4. Segment Matching through an ancestor node representation

Examples include merging and splitting between vowels and diphthongs, simple stops, affricates and geminates. Indeed as suggested by the above figure, the single [1] in Norman French chastel could also appropriately be matched against the geminate from which it derived, which is preserved in Italian, and indeed all of these to the multiple realisations in different varieties of Spanish: [X], [j], [3], [d3] and [J]. Nor can one assume that vowels will always match up against vowels rather than consonants, and vice-versa. In a simple two-dimensional presentation as in the above figure, much of the sensitivity and complexity in the possible representation of relationships between the phonetics of cognates cannot easily be shown visually. It is most important to realise that in our model cognates are not conceived of as simple 'strings' of segments, as in some simple matching algorithms proposed. On the contrary, the analysis works on multiple different levels of phonetic structure. For Italian buono [bwono] and French bon [bo], for instance, the matching is necessarily more complex than in the illustration in Figure 4. To plot correctly from one word to the other, in this case it is not just the whole segments but the individual nasal and oral gestures that need to be matched up independently of each other. The analysis of nasals and nasalized vowels is therefore broken down into oral and nasal gestures, and these can be matched against each other across different timing slots, for the most appropriate quantification of the relationship of a sequence such as [on] with a single slot [δ]. (Obviously, the loss of the separate timing slot and the oral gesture in the [n] are central to the quantification of the phonetic difference here.) A similar type of matching analysis maps contractions of slot sequences onto contemporaneous primary and secondary articulations, such as palatalizations like [ti] to [tJ]· Further refinements allow for appropriate treatments and weightings of metathesis, epenthesis, morphophonemic variants, segment loss (and any

58

Paul Heggarty, April Mc Mahon and Robert McMahon

compensatory lengthening), and changes from consonant to semivowel to vowel or vice versa, such as any one or more of the steps [1] -» [w]-> [u]-> [o]. Standard principles are also applied to ensure consistent treatment from case to case and language to language of certain debatable matches, as for example in Figure 4 whether the [o] in jjato] above should be matched only to the reflexes of the skeleton slots for */ll/, or also to those of */ε/. Though not shown explicitly here, each slot is also given its own specification of phonetic length (as usual, according to dedicated principles for phonetic significance). This is particularly important in cases of one slot in one cognate being matched against more than one slot in the other cognate. Here the comparison process ensures that the relative lengths of the sounds being compared are calculated appropriately, no matter how many individual slots they are segmented into in traditional phonetic transcription, or indeed in the format of the data in the model itself. Our overall calculations for the similarity of the whole words are compounds not just of similarity in phonetic quality and relative length of the matched gestures, but also measure the crucial differences of segment loss or phonetic attrition, which of course contribute very significantly to the overall rating of phonetic similarity for each cognate pair. The quantification is a balance of the phonetic similarities between the cognates weighted against their phonetic differences. Furthermore, each of the 100 cognates within the list is simply taken as part of a phonetic sample of the language, and they too are weighted against each other in terms of the number and lengths of gestures and phones that they contribute to that overall data set. This stands to reason: measurements of phonetic difference are not to be weighted by the lexeme, a unit from a quite different linguistic level. Note that this model does not demand any 'rationalisation' or simplification of the transcription, as other quantification techniques do, for example to phonemic depth only, or limiting specification to just a selection of distinctive features taken as establishing important isoglosses. That there has only been space here to illustrate the main parameters in the phonetic comparison model should by no means be taken to imply that analysis is limited to them! All significant phonetic distinctions represented in the phonetic transcription are analysed, and any differences in any of them are quantified. Transcription is to the level of at least all major allophones, and follows a set of principles laid down to ensure as much standardisation as possible in the level of phonetic depth in the transcriptions of the various language varieties in the studies.

From phonetic similarity to dialect classification

59

Together all these refinements strive to allow as much as possible of our linguistic knowledge about the relationships between cognates to be harnessed directly to the task of quantifying how similar they are, to help take this measurement to as accurate a level as possible. It is above all in this that the technique proposed here is considerably more detailed and refined than most algorithms hitherto proposed for performing segment-matching between cognate forms, and for producing quantifications of phonetic similarity between languages. In fact, the above example also illustrates the important distinction that the matching here is not actually done by an automated algorithm at all. Some linguists have argued that an automated algorithm is preferable - and we will come back to the possibilities for that later - though it is by no means clear that the advantages of automation necessarily outweigh the specific gain from the matching system used here, that of greater detail in analysis to bring greater precision and reliability to our matches and quantifications. This approach was chosen deliberately in order to achieve that greater precision, by building that knowledge into the specification of our data. At the dialect level particularly, such knowledge is indeed normally available to the level of detail required. So for instance we know from the general development of French (in this case backed up by attested intermediate historical forms), that the final [o] of French chateau [fato] is not to be matched historically with that of Italian castello [kas'tello], but with the reflexes of original Latin /ll/ from which this [o] ultimately developed. At this point it is opportune to refine the characterization of what our technique is actually measuring. In any single phone-to-phone comparison it does indeed measure simply the 'surface' phonetic similarity between the phones. Our matching mechanism, however, ensures that only historically true phonetic matches are compared against each other. It is not an algorithm that searches for the particular match of phones between the two cognates in order to select the 'best case' matching that gives the highest rating for their current phonetic similarity. Most often this will be the effective result of our technique too, but in this case of the comparison of Italian castello to French chateau the two techniques do not give equivalent results. So our technique is most accurately conceived of as producing a measure of the significance of any relative differences between cognates in their respective phonetic reflexes of their original common ancestor form, arisen since the time of that common ancestor. It is in this sense that our use of the phrase 'quantification of phonetic similarity of cognates' should be taken. So the [o] segments in [jato] a n d [kas'tello] are not compared against one another; rather, the overall quantification includes the sum of the differences between: for original Latin */u/, in Italian [o] vs. in French total

60

Paul Heggarty, April Mc Mahon and Robert

McMahon

loss; and for original Latin */ll/, in Italian [11] vs. in French [o], the latter being the end result of a process of steps of simplification to single [1], velarization to [1], and vocalization probably through intervening [w] then [u] finally giving [o]. The loss of original final [m], meanwhile, is a change shared in both Italian and French, so obviously is not counted as a relative phonetic difference arisen between them, as there is no difference in their treatment of this original slot. It is crucial to understand, too, that beyond knowing the structure of the protoform in terms of its 'skeleton' of slots and gestures (primary and secondary oral gestures, nasal gestures, etc.), one does not actually need to know any further or more specific phonetic details for it. Indeed, no such phonetic details are present anywhere in the data entered (they are shown in Figure 4 for illustrative purposes alone, and likewise in the model are only there for ease of reference). At no point is a 'daughter' cognate compared directly against any phonetic transcription of the ancestor word itself, for there is none in the data. The phonetic comparison is made only through the ancestor template, in all cases directly against another daughter cognate. In other words - and at risk of labouring the point only to ensure it is clear the comparison for phonetic similarity is always made directly between the daughter cognates: the template is only there to help us work out which bits of the daughters to compare with which. So the vital element is not any phonetic transcription for the ancestor form, but just the overall sequence of gesture types and slots within it. The node template is thus purely a formulaic expression, and the data entered with each daughter cognate to link it to the node consist simply of a list of which gestures and slot numbers in the daughter cognate correspond to which in the ancestor template (whether in a simple one-to-one mapping, or in far more complex ways). Such 'mapping' data are all that is needed to enable the segments and gestures in one daughter cognate to be matched up, through the node skeleton, against those in any other daughter cognate. Effectively then, the structure of the ancestor 'skeleton' form, and the numerical data showing how the gestures and slots of any daughter cognate match up against those of the ancestor, are much like any reconstruction of a proto-form. For that too is at bottom simply a formulaic representation to encapsulate our knowledge of the relationship between descendant cognates and their common ancestor, whose form we reconstruct precisely on their evidence alone. All the same, in those cases where we do have reliable knowledge of the phonetics of the ancestral form (or one close to it, say Classical Latin in relation to the Romance languages, for instance), then we can also enter the ancestor language in phonetic transcription as if it were an 'unchanged daughter' language, to be compared in its own right against

From phonetic similarity to dialect classification

61

Romanian

French

Catalan

Galician

1

1 Port. Brazil (Esp. Santo)

Portuguese Lisbon

Span. Caracas

Spanish Madrid

Italian

!

all the other daughters. But this is a separate step, and not an essential one. Italics in the figures are for cases where we have entered transcriptions for Classical Latin, for instance, as just such a phonetic variety in its own right. The basic results this method produces are therefore quantifications of the degree of similarity between various language varieties, of the type seen above in Figures 2 and 3. In Figure 5 below are fuller results obtained from a weighted corpus of a full 100-cognate list for a range of Romance varieties, including in particular at the dialect level a number of Iberian dialect and accent varieties.

70.8 62.2 60.7 56.9 56.9 61.6 52.9 37.2 61.2 75.0 74.3 68.2 68.5 73.0 62.9 44.3 70.7

Classical Latin Italian

96.7 72.2 72.8 79.4 66.8 46.5 65.8 Spanish Madrid 70.7 71.1 77.2 65.6 46.9 64.7

Span. Caracas

87.2 75.5 69.2 52.6 64.5 Portuguese Lisbon 76.2 64.0 50.6 62.2 67.1 46.8 65.6 54.7 64.7 46.5

Port. Brazil (Esp. Santo) Galician Catalan French Romanian

Figure 5. Phonetic similarity of Romance varieties for a 100-cognate sample, with a particular focus on varieties of the Iberian peninsula 100 = identity. Figures in bold face and italics are as per Figure 2.

5. Comparison with other methods Readers working within dialectology may be familiar with an alternative method of dialectometry pioneered by Kessler (1995) in a classification of dialects of Irish, and developed subsequently by Nerbonne and Heeringa in

62

Paul Heggarty, April Mc Mahon and Robert McMahon

work primarily on dialects of Dutch (see Heeringa, Nerbonne and Kleiwig 2002, Nerbonne and Heeringa 1997, 2001a, b, Nerbonne, Heeringa and Kleiwig 1999, Nerbonne and Kretzschmar in press). Note that when discussing aspects of their approach common to most or all of these authors, we shall henceforth refer simply to 'Nerbonne et al.', giving specific references only when we cite particular works. At this point it is well to clarify an important distinction between what their method and ours seek to do. Theirs is focused on producing a classification, ours on producing measurements of similarity in phonetics, and it might be felt that the comparison between the methods is therefore not a direct one. That said, one of the main applications we see for our quantifications of similarity in phonetics between language varieties, is in offering an alternative - and for certain purposes equally or even more valid - type of 'classification', by similarity, rather than by traditional shibboleth, 'diagnostic' sound changes. This can then valuably be compared and contrasted with classifications founded on other criteria. We shall return to this issue later when we discuss Nerbonne et al.'s proposed 'calibration' of their model. Moreover, while it might be argued that Nerbonne et al. present their work primarily as a technique for classification rather than quantification, they do specifically describe their work as 'dialectometry', and 'measurement'. In any case, they do not seem to make a rigorous distinction between this and classification (see also section 6 below). In addition, the data that Nerbonne et al. use to make their classifications are of course none other than measurements of edit distance in phonetics between cognates (except where a cognate is missing, in which case they effectively replace phonetic with lexical semantic data, as we shall see). Effectively, their method is indeed classifying on the basis of an attempt at measuring phonetic difference. So whether their classification is meaningful at all necessarily hangs on whether the figures they come up with really are meaningful as putative numerical representations of the language data that they aim to base their classifications on in the first place. It is precisely this issue of principle from which our method starts out. And it is here that the key difference between the models lies: in our very different approaches to how phonetic distance can most meaningfully be expressed numerically. Nerbonne et al. calculate it in terms of edit distance between cognates, taking a computational concept developed for other applications and data types, while our technique takes a radically different approach to the task, starting from first principles in putting meaningful figures on phonetic difference. (Indeed, our research into these principles

From phonetic similarity to dialect classification

63

led us early on in the development of our own model to discard the option of a simple edit distance approach as inappropriate to the task). Even if we accept that the comparison between the models is not a direct one, given the differences between them in the relative roles of classification and quantification of phonetic distance, dialectologists already acquainted with Levenshtein distance models might well, due to the superficial similarities, approach our model as though it were necessarily also based on the edit distance approach. Not least for this reason, a direct comparison and contrast between the two techniques is called for here. To repeat, then: the most fundamental difference of all is that the method introduced here is very deliberately based on very different principles of quantification and comparison from Levenshtein distance approaches. Rather than seeking to take ready-made string-handling computational methods and fit them approximately to linguistic data, our method grew out of research specifically aimed at investigating the fundamental principles for quantifying language, to ensure that the figures that linguists strive to put on language similarity in fields such as phonetics are truly balanced, weighted and thereby meaningful as quantifications. Not surprisingly, then, the two methods were developed quite independently, and in parallel. Indeed, while Nerbonne and Heeringa have a strongly dialectological perspective, our project work has grown out of work in comparative linguistics, specifically on the quantification of similarity. It follows that we cannot simply ask why we need an alternative method, when one is already in play: both exist, and both are in use, with quite different approaches, histories, priorities, and rationales. The two methods are, however, applicable to overlapping data sets and types, and a future evaluation of the two against a common set of data is arguably a priority for validation and comparison of both techniques. This is beyond the scope of the present discussion, where only a more general and theoretical comparison is possible; but some comparison is clearly essential, to allow readers to evaluate the two methods.

5.1. The compatibility problem Let us begin this comparison by returning to the two essential problems raised and confronted above. For the compatibility problem, the solution advocated by Nerbonne et al. appears to be essentially a computational, automated matching algorithm working essentially at the level of segments. Our method, at the stage of inputting the phonetic data for the cognates, deliberately harnesses as much detailed linguistic knowledge as is available

64

Paul Heggarty, April Mc Mahon and Robert McMahon

in order to match up the phonetics of the cognates as precisely as possible, not just in segments but also in gestures and in highly sensitive weightings of length, multiple articulations, and so on. It might be objected that our method's current reliance on such information, and on cognate status, for matching phonetic forms is a limitation. These are 'limitations' we are working to remove in one particular extension to our method, as explained below, but in any case they are not particularly relevant to dialects, since at this level very large numbers of cognates are easily available, and with them linguistic knowledge to the level of detail required for the node form is mostly available and quite evident. The reverse argument is clear: by disregarding such abundantly available historical and phonetic data, and failing to make full use of it to refine their quantifications, in the interests of computational simplicity, Nerbonne et al. run the risk of sacrificing accuracy and meaningfulness in their quantifications. In any case, certainly in their applications so far, the same 'limitations' to the data effectively apply to the work of Nerbonne et al.. Their models too rely - though this fact is left tacit - on cognate status, for the very concept of edit distance as they apply it is based on an underlying assumption that the two forms are linked: in the linguistic reality, what their edit distances represent are relative changes since a common ancestor form. When this assumption fails - that is, in the absence of true cognates - the comparison is no longer in phonetics; and here the Nerbonne et al. model simply switches data type, to a quantification of distance in lexical semantics instead (a problem of consistency that we return to below). Moreover, Nerbonne et al. have limited themselves strictly to differences at the dialect level, while our method is deliberately aimed at applicability as widely as possible to much more different languages, as illustrated in the application here to the Romance languages as well as their dialects and accents, and to the Indo-European family much more widely (see Heggarty 2000, and section 6 below). In their approaches to the compatibility problem, then, the key difference between the models in fact lies primarily in how much linguistic knowledge is put to use; it follows that there is a difference also in the level of detail achieved in the quantifications.

5.2. The quantification problem We turn now to the quantification problem, of putting meaningful numbers on phonetic distance. It is in this that the dialectometrical method

From phonetic similarity to dialect classification

65

introduced by Kessler and developed by Nerbonne and Heeringa faces its most serious difficulties; and here we need further details on their method. To summarize, this is based on the calculation of Levenshtein distances or edit distances (Levenshtein 1965, Sankoff and Kruskal 1983, Navarro and Raffmot 2002). These are computationally simple measurements of the least costly way of converting one phonetic string into another, merely by counting up insertions, deletions and substitutions, and as such constitute a subset of approximate string-matching problems. 'Under edit distance, one difference equals one edit operation: a character insertion, deletion, or substitution' (Navarro and Raffmot 2002: 145), and therefore the edit distance between two strings is the minimum number of operations required to get from one to the other. The essential problem with this simple calculation is that not all phonetic differences (or historically, changes) are, in the real linguistic world, equal. Yet until their most recent work, Nerbonne and Heeringa (1997:1) seemed quite satisfied with, and even repeatedly and explicitly defend, a technique which produces 'quantifications' of phonetic distance that are as insensitive to the linguistic reality as this: "the pair [a,t] count as different to the same degree as [a,o]". It is instructive to look back to the starting point for the research by Kessler, Nerbonne et al.. They set out from Sankoff and Kruskal (1983), and its presentation of Levenshtein distance, mentioning various applications such as those "in computational biology to explore which DNA sequences are likely mutations of one another" (Nerbonne et al. 1999). They also note how the technique has been applied to various uses in linguistics, such as in speech pathology, aligning texts, particularly parallel translations, or even error corrections of keyboard input. For all of these applications, a string-based analysis focusing on orderreversals, deletions and insertions, is indeed quite appropriate. Indeed, speech pathology traditionally provides one of the classic illustrations of the reality of the segment as a unit, not least in spoonerism-type speech errors. Keyboard errors are an even clearer case of string-order reversals. Likewise, word-order in translations is yet another highly appropriate case for a string-like analysis. Outside linguistics, the same goes for DNA sequence mutations. However, the first problem occurs in assuming the suitability of the Levenshtein distance approach to a very different type of linguistic data. Compared with the above applications, quantifying the phonetic differences that have emerged over time starting out from a common ancestor immediately seems much less suitable for the application of the Levenshtein distance approach. DNA sequences are made up of a mere four possible discrete types of 'segment' (A, T, C and G), arranged in linear order. The differences in such

66

Paul Heggarty, April Mc Mahon and Robert McMahon

sequences that we seek to quantify are naturally expressible as questions of ordering, insertion and deletion. But this is not so at all in the phonetics of sets of cognates, where we have many hundreds of different types of phone, all different from each other in a host of different ways and to very different degrees. Phones can be composed of multiple overlapping gestures (or 'articulations'), and can differ also on other dimensions such as length. Quantifying phonetic distance in any meaningful way inescapably involves vastly more complex processes than simple insertion, deletion and re-ordering. Metathesis ('change in string order') does happen, true, but it is extremely rare relative to other types of change ('edits'), especially changes in sound 'quality'. In our 100-cognate sample for the Romance languages, metathesis represents a tiny fraction, under 1%, of all the phonetic differences between the cognates. Epenthesis (true 'insertion') is also relatively very rare, even if one includes processes which 'generate' segments in certain phonetically highly motivated contexts, such as the [d] between [n] and [r] in Spanish vendrd 'will come'. Segment loss ('deletion') is common, though in fact 'loss' is a misleadingly simplistic cover term for a number of complex processes, such as merging from a geminate to a simple stop, diphthong to single vowel, sequence of [Ci] to a palatalised [Cj], complementary lengthening, and so on. (Likewise for all the reverse processes to these, all simply 'insertions' in edit distance terms.) A much more refined analysis of the actual relationships between cognate forms than just 'deletion' or 'insertion' is called for if one is to hope to represent the degree of phonetic difference meaningfully in figures. Certainly, working with segment slots alone cannot achieve this, and one must analyse further down to the level of the gesture, since gestures may switch from one timing slot to another, as with the secondary [J] articulation in a historical shift of [ti] to [tj], or the nasal gesture in [on] to [5], for example. A very sensitive analysis of slot lengths, diphthongs, etc., is also called for, most particularly at the dialect level.

5.3. Quantifying differences in sound quality In any case, the majority of phonetic differences between cognates, especially in varieties as closely related to each other as dialects, are differences not of 'ordering', 'insertion' or 'deletion', but of sound quality. In their earlier work, as we have seen, Nerbonne et al. lump these together as 'replacements', yet this covers a vast range of possible changes from the tiniest difference in vowel quality such as [e] versus raised [e], to very significant sound changes, such as those that have given rise to varieties_of Spanish with different realisations of the grapheme : [X], [j], [3], [d3]

From phonetic similarity to dialect classification

67

and [fl. All such changes, irrespective of the phonetics, are given the same value in Ν & H's unweighted quantifications by Levenshtein distance, as 'replacements'. Further, Nerbonne et al. decide that any such replacement should count as a distance of 2, i.e. twice as 'costly' as either an insertion or a deletion, each rated at 1. No explanation is given of why, and on what principles, these numerical values were selected, though it is evident that they are computationally very convenient, as a replacement can be seen as a deletion and an insertion, viz. 1 + 1 = 2 . This may be computationally very convenient and simple, but here we really have to ask what we hope for from our 'quantification' in the first place. What makes for a valid quantification of phonetic distance must be rooted not primarily in computational convenience, but tied to the reality of the phonetic significance of the differences in question, according to objective principles. Nerbonne, Heeringa and Kleiwig (1999: viii-ix) provide an example comparing General American and Bostonian pronunciations of 'saw a girl', in which they explain that they rate the change of vowel quality from [i] to [0] (as per their transcriptions) as twice as significant numerically as the loss or gain of a whole segment! In their example the loss is that of '[r]', but the same would apply in a case where the vowel had been lost entirely: in their mathematics, the loss of a vowel segment would be rated as half as significant as the smallest change in that vowel quality. For any phonetician or historical linguist, it is hard to see this as anything but a travesty of the true significance of such phonetic differences. On what linguistic grounds (rather than convenient computational simplicity) is the slightest difference in sound quality to be rated as twice as significant phonetically as the loss of the entire segment? The defence that might be mounted is that the Nerbonne et al. technique is only intended to be a simple and rapidly applicable one, to come up with a rough estimation. Yet this is to misconstrue the essence of 'quantification'. It is not simply blunt to 'quantify' phonetic differences in this way; it is self-evidently counter-intuitive. Of two phonetic differences, rating the far more significant one as the smaller numerical difference jeopardises the meaningfulness of the entire quantification. Numerical values rooted in computational simplicity and string-type analysis, rather than in phonetics, do not constitute a numerical representation of phonetic distance. The task of finding a computational technique suited to the phonetic comparison of cognates must start from first principles (linguistic and other) for converting the concept of phonetic similarity into numerical expression, most importantly weighting phonetic differences for their relative significance.

68

Paul Heggarty, April Mc Mahon and Robert McMahon

In their latest work, of course, Nerbonne and Kretschmar (2003: 4) do themselves recognise explicitly that "linguistic variation is gradual". Indeed Nerbonne et al. have begun to follow this through to recognise that this calls for a quantification technique that allows for gradual, scalar measurement of variation in phonetics, rather than their discrete all or nothing concepts of same or different, and 'replacements'. This is clearly a crucial step in the right direction, but serious difficulties remain in their quantification of the significance of the phonetic difference between individual phones. We are now reassured that 'Replacement costs...vary depending on the basic sounds involved' (Nerbonne, Heeringa and Kleiwig 1999: ix). This is not illustrated specifically for the case in question, where the distance calculations are 'simplified...for clarity' (1999: ix). The lack of detail on the features used again means that the reader cannot make these more refined calculations with any confidence for him- or herself. Nonetheless, they do go on to give one, different illustration of the method they propose for comparing features rather than monolithic segments: 'To obtain a more sensitive measure, costs are refined based on phonetic feature overlap' (Nerbonne, Heeringa and Kleiwig 1999: ix). Unfortunately, no list of features or full definition of feature systems used is included in any of the publications we were able to consult (Heeringa, Nerbonne and Kleiwig 2002, Nerbonne and Heeringa 1997, 2001a, b, Nerbonne, Heeringa and Kleiwig 1999, Nerbonne and Kretzschmar in press). Reference is made in passing to work by Hoppenbrouwers and Hoppenbrouwers (1988) and by Vieregge (1984) in Nerbonne and Heeringa (2001b) and in Nerbonne, Heeringa and Kleiwig (1999). We are also told (2001b: 5) that 'Hoppenbrouwers' system is based on Chomsky and Halle's...and consists of 21 binary features which all apply for all phones...Vieregge's system consist of 4 multi-valued features only for vowels, and 10 multi-valued features only for consonants.' However, the example provided in Nerbonne, Heeringa and Kleiwig (1999: ix), which the text suggests is from Hoppenbrouwers and Hoppenbrouwers, in fact includes scalar features, in accord with the description of Vieregge's system. The lack of detail on the derivation of the values given (with [u] being '6(back)' and '3(short)', for instance) makes it difficult to evaluate these features or their application to the dialect data. Before moving on to a detailed look at the difficulties remaining with their model, it should also be noted that their features are still tied to units that are discrete segments. This can do nothing to address many of the difficulties inherent in the compatibility issue, so the Nerbonne et al. method still fails to analyze appropriately and solve the many problems in 'matching up the strings' in the cases discussed above. While there now appears to be some specification of length, this is analysed alongside other

From phonetic similarity to dialect classification

69

features of segments as if it were an issue of sound 'quality', where it is actually fundamental to the correct analysis of slot-matching too, as in compensatory lengthening for the 'loss' of a phone such as their '[r]' sound in 'girl'. In part this seems to be a confusion between the possible phonemic status of length, and its real nature in phonetic reality. Indeed, possible confusion between the phonemic and the phonetic appears repeatedly in their presentations. At no point in Nerbonne, Heeringa and Kleiwig (1999) is the issue clearly addressed of whether their features are actually truly "phonetic" as they describe them, or rather phonemic. The latter is suggested by the fact that "Hoppenbrouwers' system is based on Chomsky and Halle's", which was a phonemic not a phonetic system. Indeed the very term they have chosen to use, "(distinctive) feature", implies phonemic distinctiveness. Yet in their transcriptions they tend to use only phonetic square brackets, rather than phonemic slashes (or sometimes neither). Nor does their use of symbols clarify which level their model works at, though certainly their transcriptions seem far removed from any close phonetic level. For instance they give the Standard American and Bostonian pronunciations of 'saw a girl' as [soagirl] and [soragol] respectively (we have added the square brackets they normally use, since they cite without anything in this case). Even if one allows their use of '[r]' as shorthand (rather than as the alveolar trill it technically denotes), if their analysis is truly by features at a phonetic level as they say, one might expect their transcription to include a dark [1]. In any case it is hard to understand as either phonetic or phonemic their use of the symbols [o] for 'saw' in both Standard American and Bostonian, and especially [i] for Standard American 'girl'! This serves to illustrate another very significant difference between the methods. Again, in the development of our model, the prospect of comparisons on a phonemic level was discarded early on because of the enormous difficulties it poses in compatibility, given the differences in phonological systems between languages. (Those differences are of course of great interest, and something we would wish to quantify too, but this needs to be done by a different method, as outlined in Heggarty (in preparation).) Moreover, since phonemics is language-specific and thus 'language-subjective', for objective quantifications one needs to keep to the phonetic level. It is this level, too, which allows analysis to a greater degree of detail. Our model goes far beyond phonemic transcriptions to the true phonetic level of all major allophones. Moreover it is extremely sensitive to different 'string lengths', not just cognates with different numbers of segments, but segments of finely different lengths (from short to normal to half-long to long vowels, for instance).

70

Paul Heggarty, April Mc Mahon and Robert

McMahon

One issue that does seem clear in Nerbonne, Heeringa and Kleiwig (1999) is that the Vieregge (1984) feature system will allow only vowel to vowel and consonant to consonant comparison. This might in the first instance seem unproblematic for dialectometrical purposes: even if changes from vowel to consonant might happen between languages, we might not intuitively expect anything so dramatic among closely related varieties. Setting aside for the moment the question of whether we can so straightforwardly distinguish dialect from language differences in any case (our own method, for example, is designed specifically to deal with the blunt fact that history is continuous), we find that substitutions of vowels for consonants are in fact quite commonplace, even in Nerbonne et al.'s 'saw a girl'. Their own illustration exemplifies the problem that their system is unable to measure appropriately differences between consonants and vowels, for in many accents of English it is debatable whether their '[r]' in 'girl' is phonetically a vowel (as in the classic 'r-coloured schwa' of American English), or a consonant. The difference is minimal phonetically, but not measured as minimal in their model: instead it rates as a whole segment deletion and insertion. Our model very deliberately allows direct comparison between vowels and consonants, so important in cases like this or instances of semivowels/semiconsonants. Nor does Nerbonne et al.'s method have any clear way of factoring in compensatory lengthening of the vowel, also relevant to the question of phonetic similarity and therefore included in our model. Even when comparing two vowels or two consonants against each other, an even more fundamental problem in approach remains in Ν & H's feature analysis. We are now told (Nerbonne et al. 1999: ix) that the differences between the segments [i] vs. [e], and [i] vs. [u], are no longer both rated as equal differences (one replacement), as in their earlier model. Now they are rated as 1 and 5 respectively. This dramatic change in the figures for the same phonetic difference must raise concerns over Ν & H's approach to the quantification question. Again, there is no discussion whatever of how they came to those new figures, rather than any others, as meaningful numerical representations of the significance of those phonetic differences. The only clue lies in their table showing that [i] and [e] are rated as 2 for advancement, and [u] as 6; while [i] and [u] are rated as 4 for high, with [e] as 3. These appear to have been drawn from "Hoppenbrouwers' (SPE-like) features". The problem is not necessarily in the Hoppenbrouwers' feature system itself, of course. It is that that feature system appears not to be weighted for phonetic significance, which is what is needed if it is to be applied as a technique for quantification of phonetic distance. This raises the question of why - on the basis of which objective principles, linguistic or other - the difference in advancement between [i]

From phonetic

similarity

to dialect classification

71

and [u] should be rated as 4 times as significant in the figures as the difference in height between [i] and [e]? Indeed, this seems to go against the key principles we propose here, not least that of cross-linguistic evidence for the phonemic uptake of incremental phonetic differences. Languages do not normally show six phonemically contrastive degrees of advancement (backness), as Nerbonne et al.'s figures of "2(back)" and "6(back)" suggest. Certainly they do not generally use more phonemic contrasts in backness than in height - for which Nerbonne et al. allow only 4 heights, as opposed to 6 degrees of advancement. A quick glance at the IP A chart is effectively a confirmation by phoneticians' consensus built upon long cross-linguistic experience: the chart sees the need for, and provides, separate symbols for four vowel heights, but only three degrees of backness. In fact, the matter is more complex still, since at the lower vowel heights the backness parameter is 'compressed', generally bearing fewer distinctions (and with only two IPA symbols), another necessary refinement built into our model.

5.4. Weighting The issue here, of course, is fundamentally one of needing to weight one's quantifications of various phonetic differences against each other for their relative significance. Even though Nerbonne et al.'s feature-based approach is clearly an improvement, it still inspires little confidence when they fail to discuss any principles for attaching numerical values to differences. There is nothing whatever to convince us that the outcome of the Hoppenbrouwers feature system is necessarily a meaningful quantification, nor that the [i] vs [u] difference in backness and rounding should be rated as 4 times as significant in the figures as the difference in height between [i] and [e]. Do these numbers represent the true relative significance of those phonetic differences? On what grounds? On the other hand, what our method explicitly seeks to do is to analyze and weight all of these details in a principled way. The reference to weighting here raises another important issue, since in their whole approach to this, Nerbonne et al. do not appear to use weighting techniques consistently, and their assessment of weighting vacillates: we find one statement that 'Weighting of features mostly improves the results' (Nerbonne and Heeringa 2001b: 11), but another to the effect that 'Unweighted representations outperform representations to which weightings were added' (Nerbonne and Heeringa 1997), though they add the rider that "This is surprising". Again, disappointingly, the fact that these are simply presented as summary statements of results makes it difficult to assess what made the

72

Paul Heggarty, April Mc Mahon and Robert McMahon

difference. Moreover, the failure of weighting here appears less surprising when we consider what Ν & Η propose as "weighting", and how they attempt to validate its usefulness. It is revealing to see what is behind Nerbonne and Heeringa's (2002:1) bold conclusion that "weighting is not useful". Firstly, Nerbonne et al.'s interpretation of 'weighting' is a very limited and specialised one, apparently relating only to 'weighting each feature by information gain' (Nerbonne and Heeringa 2001b: 7). They note, for example, that 'it turned out that no positive value for the feature [flap±] occurred in any of the sounds of the words in the dialects examined' (2001b: 7). Yet remove the feature, and one cannot extend the comparison to include other language varieties that do have flaps. In any case, why should the numerical value to be assigned to precisely the same phonetic distinction feature be allowed to vary, and need to be re-weighted from one study to the next, depending on whether it "turns out that" it occurs in a particular set of language varieties? The phonetics do not change from one set of languages to the next, so why should the figures representing the phonetics? Quite on the contrary, the same phonetic difference ought to be weighted once and for all, in terms of objective linguistic principles. In a system to measure phonetic similarity, the type of weighting needed is not of features 'by information gain' in any particular subset of languages. This would suggest that quantification is only about differences, threatening to return us from objective quantification to focusing on unweighted 'shibboleth' isoglosses (on which see further below). Our view is that the whole point of weighting is quite the contrary: that differences need to be assessed for their significance, relative to the similarities, according to some objective principles for gauging significance in phonetic terms. It follows that our own method sees appropriate weighting of features, whether by linguistic and logical principles or on a cross-linguistic basis, as absolutely central to the success of phonetic comparison in producing quantifications that are meaningful. Nerbonne et al. do not seem to consider explicitly this most essential form of weighting. It is no longer a surprise that their own attempts at 'weighting by information gain' fail to gain them anything in their results. The sweeping conclusion that "weighting is not useful" relates only to this narrow interpretation of the term. Moreover, their 'calibration' of whether even that form of weighting gains anything is not based on any principled grounds, but reflects the failure of their interpretation of 'weighting' to give significantly differentiated results in the one data set they applied it to, when it is quite a different form of weighting that is needed in any case.

From phonetic similarity to dialect classification

73

Nerbonne and Heeringa (2002:1) do themselves note that 'feature representations are more sensitive'. This stands to reason, and hardly needs to be 'demonstrated' by them. And in fact, breaking down segments into component features is one initial step in weighting differences against each other (degrees of difference can now be distinguished relative to each other). So while their application of weightings is unprincipled, as noted above for vowel height and backness, by seeing the need for the move to features they are agreeing, without seeming to realise it, that the type of weighting analysis we propose is indeed more sensitive. By the same token, it follows that provided the principles are indeed sound, the much more detailed level of analysis proposed in our model will bring more of exactly the same advantages they see in feature analysis. The whole question needs to be approached in this principled, explicit way. One potential alternative system of weighting explored in earlier stages of our work was to try to weight not in terms of 'universal' phonetic significance, but by significance only at a phonemic level. This inescapably comes with the very significant difficulty, however, that phonemic significance is naturally highly language-specific, a fundamental problem of compatibility for comparisons between language varieties. It might be argued that within a single dialect area the problems in different phonemic systems between dialects might not be too significant, as in the Dutch dialects study by Nerbonne et al., and that their comparison is operating at a higher level of abstraction. Indeed as already noted, the few transcriptions they offer, and their approach to features, suggest that their analysis is actually in large part only at a roughly phonemic, not a phonetic level. They do not address the issue of the how close their phonetic analysis is, however, while they specifically and repeatedly call their features phonetic, not phonemic, and cite IPA symbols in phonetic square parentheses, not phonemic slashes. The lack of clarity here too leaves fundamental questions unanswered, and little confidence in the consistency of their vision, analysis and model.

5.5. Validation Of course, we hardly wish to claim that our approach has uniquely identified the only, truly 'correct' quantifications of the precise degree of phonetic difference between sounds. Put like that, such a goal is hardly feasible at all. But we do argue that the essence of quantification lies in attempting to achieve at least the least inaccurate approximation to that, and that for any quantification to be meaningful the figures must be founded not

74

Paul Heggarty, April Mc Mahon and Robert McMahon

on models devised for other purposes, but on objective principles for putting numbers on phonetic differences, to reflect the real phonetic significance of those differences. It is on the basis of some such first principles of quantification, and a discussion among phoneticians (and mathematicians) as to their reliability, meaningfulness, and accuracy in application to real phonetic data, that we can hope to progress to something of a consensus on how to quantify phonetic distance, and validate methods that strive to do so. This brings us to the issue of validation against linguistic reality, which is strongly foregrounded by Nerbonne et al. (see in particular Heeringa, Nerbonne and Kleiwig 2002), though they see it in more computational terms, as 'calibration'. It is, of course, vital that methods, especially new and essentially untested ones, should be validated, and that experimenters should consider ways of meeting this criterion. Nerbonne et al. propose two methods: first, they derived a dendogram, or tree representation, from their distance data, and compared this with a dialect map reflecting the opinions of 'expert dialectologists' (Nerbonne and Heeringa 1997: 15). Elsewhere, they refer to this type of supposed validation as 'a gold standard, based on the consensual classification of a well-studied area' (Heeringa, Nerbonne and Kleiwig 2002: 445). However, Nerbonne and Heeringa (1997: 16) themselves note an element of circularity in the validation they propose, since their work is based on Dutch dialects, calibrated on Dutch dialects, and tested on those same Dutch dialects: 'For more definitive results, the method should be tested on material for which it has NOT been calibrated'. Their 'gold standard', which asserts that 'dialectometric methods ought to agree with expert consensus in well-studied cases' (Heeringa, Nerbonne and Kleiwig 2002: 446), is also partial, since they observe that the two consensus maps they select for comparison 'are criticized (legitimately or not), but this does not disqualify them from use in our project' (2002: 447). Further difficulties arise from the data used in Nerbonne and Heeringa's analysis and validation. Firstly, their judgments of their own methods as producing 'reasonable' results are founded essentially only on their outline studies of Dutch dialects. Their distance matrices are themselves largely derived from transcriptions in the Reeks Nederlands(ch)e Dialectatlassen (Blancquaert and Pee 1925-1982), but there is an obvious issue of control over the sample. These transcriptions relate to speakers aged between 12 and 80 at the time of recording, which itself varied from the 1920s to the 1970s, covering a period of several generations during which changes might have begun or continued. The treatment of these data as essentially synchronic is therefore open to question. Secondly, the dialect maps used for 'gold standard' validation were themselves based in part on speaker

From phonetic similarity to dialect classification

75

judgements of distance between varieties, and hence on perceptual dialectology, a promising but itself largely unproven method. It might seem unduly optimistic to assume that establishing a correlation of the results of two untested methods can be regarded as validating both. Nerbonne and Heeringa's (1997) simplistic representation of diphthongs is a perfect illustration of how their supposed means of 'validation' is yet again founded upon no principled, linguistic grounds. On the contrary, they 'experiment' with results from two types of representations, either as a single segment or as two, and then simply tell us which gave results that happened to fit in with what they seem already to have decided as the right answer, even though that is based entirely on unweighted judgments, not objective quantifications. The one that happens to fit best in the one case study they present, Dutch dialects, is therefore 'proved' to be right. There is no discussion whatsoever of how or why this should necessarily be the case on principled grounds. Our particular concern here is that the judgements of speakers on how close or distant a certain variety is from their own might rely in large part on individual 'shibboleth' features, which have arguably undue prominence in folk linguistics - and indeed, potentially in historical linguistics too. Aside from these questions, it may well be that two of the characteristics of their analysis have had the chance effect of somewhat counterbalancing each other, so that their results in the one case study they present emerge roughly in line with standard analyses. We have seen how in their earlier model, any difference in phonetic quality between segments is overweighted as fully double the degree of difference of any complete segment insertion or deletion. Yet given that the level of detail of their transcriptions seems to be very far from a close phonetic transcription, most of the finer phonetic details of the differences at the dialect level will simply not register in their analysis to be counted as 'replacements' in the first place. The net effect of failing to detect many of the cases that their model overweights may well be to bring the results back to figures that fortuitously happen to approximate to the actual degrees of difference between the language varieties they study in the 'shibboleth' isoglosses the traditional classification is founded on. From the summary presentation of their method and results in the single case study they offer, one simply cannot tell, but questions remain. Either way, it is indicative that despite their initial confidence in their segment replacement model and the viability of its results, Nerbonne et al. nonetheless have chosen now to abandon it in favour of a more sensitive feature-based one. Nerbonne and Kretzschmar (2003: 4) do themselves identify the flaw of 'simply insisting on one's favourite features'. Yet they then go on to

76

Paul Heggarty, April Mc Mahon and Robert McMahon

assume that they can 'calibrate' their method and establish its accuracy precisely on the basis of the 'favourite features' that other linguists have picked on in the traditional classifications. We believe that much of what we have to gain from objective quantification, the whole point in it, lies precisely in challenging the traditional perceptions of dialect and language similarity, founded on a few subjective, unweighted, decontextualised differences. Measuring those differences objectively on the basis of principles for converting language into meaningful numerical expression, those principles being crucially independent of traditional impressions of language and dialect relationships, offers the great gain of seeing all the phonetic differences and similarities in the languages under comparison put into the true perspective of their real relative significance. One comment from Nerbonne and Kretzschmar (in press:2) well illustrates the difference in our approaches: "the vowels in Dutch huis, muis ('house', 'mouse') do not align with other linguistic distinctions, and thus do not determine dialect areas satisfactorily". Yet the fact that an isogloss does not match with others does not mean that it is therefore of no use in classifying dialects. Quite on the contrary, it indicates that the classification based only on the others is not the whole story, which needs to be seen in terms of all the features appropriately weighted against each other. In section 6 below we shall return in depth to this issue of contrasting classifications by traditional 'shibboleths', or by similarity ratings. In general, we do not consider traditional classifications as a suitable yardstick against which to judge a technique for quantification·, the two are not the same thing at all (though the distinction is not made explicit by Nerbonne et al.). We propose two alternative techniques for validation, one direct and one indirect. The first guide for validation involves testing for correlation between trees derived from our phonetic comparison method, and those derived from lexicostatistical results. (We say only a 'guide' since a correlation is not in itself fully conclusive, as language relationships can of course be different on different linguistic levels.) Phonetic similarity measures for 40 cognates in 11 Indo-European languages (Heggarty 2000a), show a striking linear correlation with the distance matrix expressing cognacy judgements taken from a database of 200 meanings in 95 Indo-European languages (Dyen, Kruskal and Black 1992): the value of Spearman's rho, a nonparametric test of correlation which does not assume underlyingly normal distribution in either variable, is 0.560, which is highly significant (p < 0.001). In this case, the lexical and phonetic data have been collected and analysed independently of each other. This option for validation does not,

From phonetic similarity to dialect classification

77

however, appear to be available on Nerbonne and Heeringa's edit distance approach, since they (2001b: 4) confuse the measurement of difference/similarity in phonetics and in lexis, and in doing so invalidate their results as a meaningful measure of either. This confusion also leads them to adopt simple word-by-word averaging of phonetic differences. That is, they end up using the lexeme as their basic weighting unit for what they claim is a measurement of phonetic similarity. The obvious outcome is that from one word to the next, precisely the same phonetic difference is measured differently and contributes differently to the overall results. Why should this be so? Nerbonne et al. seem unclear as to whether they are measuring similarity in lexis, which would indeed call for a sample weighted by lexical units; or similarity in phonetics, for which they need a sample of phonetics weighted on phonetic principles. The current inclarity has two consequences: there is a question mark over our confidence in their quantifications when it is not clear how far these reflect a measurement of difference in phonetics, and how much in lexis; and the option of validating their phonetic results by considering independent lexical evidence is unavailable. Of course, we are equally not arguing that lexical and phonetic measures will necessarily match in every case: one of the most interesting and informative aspects of the comparison of dialects and languages, especially in cases of contact, is precisely what can be revealed when language varieties are more or less similar on certain levels, say lexis, than on others, such as phonetics. This is why it is particularly important to consider a range of case studies and examples, and not restrict the analysis to a single one, potentially drawing unwarrantedly general conclusions. It also means that a further means of validation is necessary, and here we opt for a comparison of the trees derived from our phonetic comparisons, with conventional historical linguistic trees. There is, however, a crucial difference from Nerbonne et al.'s practice here: whereas they see a direct mapping between the trees as validating their method, we are particularly interested in cases where the two do not match, and focus on such cases, seeking explanations for the mismatches we find. A case of exactly this kind, which takes us further in our exploration of 'shibboleth' changes and features, is the subject of the next, final section. This difference, along with the detailed derivation of numerical comparisons in our own method, as against the rather higher-level assignment of values in Nerbonne et al.'s work, derives from a fundamental mismatch in rationale and methodology. It is beguiling to see the two methods as similar, since they both involve some form of segmentation and of feature analysis, and quantification of phonetic properties of the string.

78

Paul Heggarty, April Mc Mahon and Robert McMahon

This is misleading, however, for while our own method begins from the phonetics and attempts to analyse what we know phonetically in a computational, quantitative way, Nerbonne et al's work starts from the mathematics, and applies a predetermined computational system to the phonetics. Our work grapples with the issue of how phonetics is to be quantified as the first and most important step; Nerbonne et al. opt for a methodology tried and tested in other fields, but not necessarily appropriate to phonetics (whose 'strings' are not at all of the same nature as those of DNA and the other applications the method was initially designed for). As a result they can only ever measure distance, which is above all a relative concept, as opposed to difference, which can be absolute. In fact, these two radically different approaches may sometimes in clearcut cases produce the same results: one might expect a similar outline classification of varieties where the signal in the data is distinct enough for even the less sensitive method of Nerbonne et al. to come up with a clear classification. Nonetheless, there is a big difference in the possible extensions of the methods from the purely dialectometrical applications illustrated in Nerbonne et al.'s work, and in the interpretation we might place on the quantifications themselves that underlie that classification. We argue that the Nerbonne et al. method is not easily extendable beyond the dialect level, because they are working within a space defined purely relatively. Moreover, since they are not calculating phonetic similarity at a fine enough level of detail, it is unlikely that their method would be successful at finer levels of mere accent differences. There is also an inherent limitation on their work, since they require much more data, in the form of a word list of significant size: they cannot provide sufficient resolution over smaller amounts of data, since their method intrinsically abstracts away from many of the details of the surface phonetics. As we have shown, our own comparisons give sensible results for Romance when we only have 10 words, or in fact only 1. Each single cognate provides a great deal of comparative data, so a relatively small number of cognates already constitute a reasonably representative sample of phonetic differences between language varieties. This means that our method also offers the prospect of applicability in circumstances of endangerment or poor preservation of historical data (at the accent level, dialect level, or beyond), where only a small amount of data is available. This reflects the fact that our method crucially aims to get the largest amount of information out of the minimal amount of data, and that it involves the measurement of absolute rather than relative phonetic distance. Our method might be criticised for building in knowledge of linguistic relatedness, and it is true, as we point out ourselves above and in Section 6

From phonetic similarity to dialect classification

79

below, that this limits its current applicability to sets of systems which we know to be historically related (though Nerbonne et al.'s applications are in practice similarly limited, as noted above). On the other hand, our method of comparison itself does involve calculation of absolute phonetic distance and is therefore extendable in principle. Let us recap on the fundamental difference between our model and that of Nerbonne et al.. The single most serious weakness we see in their work lies in the apparent absence of, or at least lack of discussion of, any rigorous principles to inform how one might be able to express phonetic concepts and differences in numbers. They simply make no mention of the whole issue of how they come up with the numbers they attach to phonetic differences. They just tell us, for instance, that a deletion or insertion counts as 1 difference, a substitution 2, a relationship which seems ungrounded in any true linguistic significance. In their recent version too, Nerbonne et al. still do nothing to justify their rating of the difference in backness between [i] and [u] as 4, as against that in height between [i] and [e] as 1. One is bound to wonder how representative of phonetic reality is the classification they derive from such numbers. Certainly, that Nerbonne et al.'s method is computationally very simple is one of its initial attractions. On the other hand, their general approach, and their presentations, are perhaps too heavily focused on the 'computational' - rather than the real 'linguistics' that their formulae and the figures in their results are supposed to represent. They do not start out from first principles that might inform them of how phonetic differences can most meaningfully be converted into figures. They only attempt to justify their method empirically after the fact, nor do they succeed in doing this convincingly. The question is what is more important: sticking closely to the linguistics, and then trying to develop ways of solving the computational problems that brings, or idealising to a sometimes extreme degree in the phonetics because the computational solutions you are using will necessarily be crude? As already noted at various points, all this is far from saying that the method we propose here is somehow the definitive answer to the quantification of phonetic similarity, for use in the classification of dialects or to other ends. It is one proposal, and it certainly still has limitations. As indicated, for instance, its analysis is principally articulatory; clearly it would in principle be preferable to balance that more with acoustic phonetic analysis. And of course, there are plenty of difficult calls in the application of general linguistic and logical principles for quantification to the highly complex reality of differences and relationships between different sounds. Nor is phonetic similarity the be-all and end-all of the

80

Paul Heggarty, April Mc Mahon and Robert McMahon

relationships between dialects and languages: we also need meaningful quantifications for the relationships between whole phonotactic and phonological systems and inventories, not to mention measures on other linguistic levels. What we are offering, then, is in a limited field, and just one possible solution to each of the many questions involved. But the solutions we propose are at least based on objective principles for quantification. We see it as essential that progress towards producing measures of phonetic difference (and 'classifications' based upon them) that are truly meaningful should proceed through debate on such principles for how various aspects language might each best be quantified, rather than just by trying to apply computational techniques devised for other purposes. This is the only way to ensure that the numbers our methods come up with bear a meaningful, indeed ultimately the closest possible, relationship to the linguistic reality that they are intended to represent and quantify.

6. Applications and extensions In designing this method, a very deliberate decision was taken on the general approach to the quantification of linguistic distance, to ensure that the applicability of the method is not restricted to a particular 'domain' across which varieties may differ. It can be applied not just to cases such as the above, where the varieties differ in 'space' (i.e. by geographical region), but equally well to varieties of the same language through time, or - of particular interest to social dialectologists - to different registers and sociolinguistic varieties of a language. It follows that our method is intended to be applicable in a historical linguistic as well as a dialectological context, and to be usable all the way up, and down, the typical language family tree. Secondly, the method outlined here is designed to make maximal use of the phonetic data and linguistic knowledge available: that is, it operates at a considerable level of detail, with weightings for different gestures, features or parameters playing a central role in the calculations; and it builds into the comparisons a great deal of the detailed knowledge we have about the origins of the phonetic differences between the cognates compared. It might be argued that there is simply no need for such refinement, and that a method working at a more general level could give results for dialect variation that are just as good. We shall return soon to the issue of just what is a 'good' quantification, relative to traditional classifications. In any case, we see it as desirable in principle that a method which is to operate at both the dialect and language

From phonetic similarity to dialect classification

81

level should be as fine-grained as possible. Testing the results against those of other methods (such as those involving Levenshtein distances) to demonstrate the utility of these detailed measurements should constitute a later stage in the investigation: for the moment, we believe that there are easily enough difficulties and inclarities over the edit distance approach to warrant investigation of our alternative. From the same data set used to produce the cross-language comparisons, our programme also automatically calculates figures for various phonological indices, such as consonant-to-vowel ratios, syllable complexity and so on. The usefulness of these obviously depends on how representative the list of cognates is of the language's phonology as a whole, a question there is not space to discuss here. Yet to illustrate at least the kind of information the programme is able to retrieve from the data, Figure 6 below shows a few sample statistics calculated from the same corpus of 100 Romance cognates. Note that the average word length can also be taken in a more historical sense as (an inverse of) an index of phonetic attrition of the original Latin forms. As it currently stands, this method is workable only in cases where we are confident that languages are related, and have a reasonably plausible reconstruction of the general shape of the common ancestor forms. What this does mean is that besides being viable for even fairly distantly related languages such as English and French within as wide a family as Indo-European, the method is particularly well adapted to measuring degrees of linguistic distance at the level of dialect variation - hence the inclusion in the above results of a range of Romance varieties below the language level. Dialects are ideal data because we know they are related, and it is generally quite straightforward to establish a nodal skeleton form linking them - even though there might be considerably more debate as to the further details that the model does not need, namely the precise phonetic form the common ancestor might have had. As noted at the start of this paper, there has been a tendency in historical linguistics to ignore dialects, though in many ways this is at odds with the trend towards emphasizing connections between variation and change. With this method, we can now work towards measuring the relative closeness of dialects, along with the degree and type of change involved. On the other hand, where languages are not demonstrably or unquestionably related, we can scarcely aspire to reliable knowledge of how particular words are to be matched up against each other, if we are not even sure that they actually are historically cognate at all. As currently formulated, the technique cannot be extended to comparing words for which we do not have a known skeleton reconstruction. Work is therefore

82

Paul Heggarty, April Mc Mahon and Robert McMahon

COMPLEXITY

Classical Latin

5.89

2.26

1.42

0.97

0.88

Italian

5.08

2.17

1.06

0.96

0.00

Spanish Madrid

4.60

1.96

1.20

0.86

0.22

Span. Caracas

4.49

1.96

1.12

0.85

0.22

Portuguese Lisbon

4.03

1.92

1.28

0.89

0.15

Port. Brazil (Esp. Santo)

4.27

1.92

1.08

0.91

0.08

Galician

4.51

1.95

1.16

0.90

0.21

Catalan

3.83

1.47

1.43

0.92

0.61

French

3.07

1.15

1.38

0.91

0.68

Romanian

4.30

1.68

1.41

1.00

0.55

(in segments)

AVERAGE CODA

COMPLEXITY

(in segments)

AVERAGE ONSET

CONSONANT

TO VOWEL RATIO

SYLLABLES PER WORD

AVERAGE NUMBER OF

VARIETY

(long segments weighted more)

LANGUAGE

AVERAGE WORD LENGTH

underway to develop an extension to the method so that it can be applied to such cases of languages about whose genetic affiliations we are not so sure. This extension will be based on an algorithm to find automatically the 'best case' match between segments, as described above, and as such will bear a certain resemblance to the algorithms already developed by other researchers.

Figure 6. Sample phonological statistics from the Romance cognate sets in Figure 5

Nonetheless, a completely new automated algorithm will need to be devised, since those currently available do not work in a way compatible with this model, many because they do not analyse to the level of detail that its structure requires in order to match not only segments but also gestures against each other, and independently of each other; to match up segments of different lengths, consonants to vowels; and so on. Our new technique

From phonetic similarity to dialect classification

83

will then allow the same method to be applied not just at the levels of dialects and demonstrably related languages, but also to languages whose genetic affiliation remains in question. This in turn will allow us to quantify diversification, and the shift from variation to change, in terms of the continuum it really seems to be. In conclusion, we shall consider just one more promising avenue for this work on quantifying phonetic similarity. Another aspect of our current research involves a consideration of how computational tools from biology, and specifically from genetics, can be adopted and adapted to linguistic data (see also McMahon and McMahon 2003). This research currently focuses on the PHYLIP suite of programs (Felsenstein 2001). The tree in Figure 7 below, based on the numerals 'one' to 'ten' for Romance varieties, is the output of one of those programs, FITCH, which draws a vast range of possible trees and selects the best one - or in other words, the tree which is most consistent with the data. In the output of the Fitch program, branch lengths are significant, so that longer lines signal a greater degree of change; and it is perhaps no surprise here to see that French is the language which has changed most extensively. Note also that, although the PHYLIP programs produce unrooted trees, in a star conformation, this tree has been rooted (using Latin) to provide a diagram more easily comparable with typical linguistic family trees. The tree in Figure 7 may initially seem 'problematic', since not all New World varieties of Spanish or of Portuguese cluster together: the Spanish of Bogota and the Portuguese of the interior of Säo Paolo State cluster instead with their European counterparts. There are two issues to bear in mind here, however. First, the diagram in Figure 7 is a phenogram rather than a phylogram: it is drawn on the basis of the degree of similarity between varieties in terms of phonetics, rather than showing the order of historical branching. It is entirely possible that the two types of tree may not match. It is true that Bogota and Caracas Spanish, for instance, do share certain features that are taken as a common trait of Latin American Spanish, such as conflating as /s/ what in Madrid are still two distinct phonemes, /Θ/ and /s/. A 'genealogical' view of this would represent this as a historical split, of all Latin American varieties on the one hand, relative to Madrid Spanish on the other.

84

Paul Heggarty, April Mc Mahon and Robert McMahon

Classical Latin

Spanish Caracas

Spanish Madrid

Spanish Bogota

Italian

French Paris

French Southern

Portuguese Rio

Portuguese Salvador

Portuguese Lisbon

Portuguese Säo Paulo State interior

Figure 7.

Rooted Fitch tree for Romance varieties using the results for phonetic similarity for the numerals 'one' to 'ten' in as shown in Figure 3

Yet that one change is of course only part of the story of the actual degrees of similarity in phonetics between the three modern varieties in question. For since the time of any such putative split, Colombian highland Spanish (Bogota) has remained among the most conservative of the Latin American varieties, changing little and remaining very similar to Madrid Spanish in most other respects. Coastal Venezuelan Spanish (Caracas), meanwhile, has undergone certain striking changes, such as [s] before a consonant or word-flnally weakening to [h], a process that appears to be continuing to a weaker aspiration still, or simply loss of the [s] altogether. In terms of overall similarity in phonetics (if not also between their

From phonetic similarity to dialect classification

85

phonological systems), this and other 'innovative' features that differentiate Caracas Spanish from both Madrid and Bogota Spanish outweigh the significance of those other features, such as the loss of the /0/~/s/ contrast, that distinguish both Caracas and Bogota from Madrid Spanish. Outweigh' is indeed the key word here, since our measure of overall phonetic difference is specifically weighted: both by the phonetic significance of the contrasts of [Θ] vs. [s] vs. aspiration; and by the relative frequency of the various phones within the cognates sample, /Θ/ being a considerably rarer phoneme than /s/ even in the varieties that still have it. To express this relationship between languages as a more general principle, consider a language history in which a first split into varieties A and Β is followed shortly by a second whereby A splits in turn into varieties Ai and A2. Both Ai and Β then change little thereafter, while A2 changes rapidly, so much so as to become more different from Ai than Ai is from B. This is a perfectly possible scenario, but a phylogram alone cannot represent the full nature of this history. Thus the phenogram showing Bogota Spanish to be closer to Madrid than to Caracas Spanish can indeed be viewed as a valid representation of the relationship between those three varieties in terms of their synchronic phonetic similarity. So far from being 'wrong', the fact that this phenogram differs from the phylogram should be taken rather as an indication that the 'problem' can just as much be seen as the traditional insistence on conceiving of the relationships of language varieties uniquely or principally in terms of phylograms. This tradition tends to lead linguists to focus on certain 'shibboleth' sound changes and isoglosses felt to be more 'diagnostic', but which are not weighted again each other. As in this case, though, these favoured sound changes are often only part of the story, and represent only one among the many different types of relationship - in the wider sense of the word - that are possible between language varieties. Besides common origin, plenty of other factors can and do pull in radically different directions to the 'shibboleth' sound changes: not least different rates of change over time (slower or faster), parallel change, and contact. A similar analysis holds for the relationship that appears at a higher level in the tree, between Spanish, Portuguese and French. Although there are indeed certain 'Iberian' features common to the first two, the last two in turn share other features in which Spanish is not involved, such as nasalization and devoicing/loss of certain final vowels. Expressing this as a general type of relationship in the same terms as above, in these features it is languages A 2 (Portuguese) and Β (French) that have both changed, in parallel, such that they resemble one another more closely than either of them resembles the more conservative language Ai (Spanish). Such changes in parallel are normally seen as historically independent and, like

86

Paul Heggarty, April Mc Mahon and Robert

McMahon

differential rates of change, given relatively little consideration in classifying languages; but they show up significantly in phenograms. In any case, when a language breaks up initially into a dialect continuum, as happened with Romance, the varieties splinter into a complex constellation of different types of relationship - and the one just discussed, in which it is Spanish that is more conservative, is just one of the possible types that can contribute to the tree in Figure 7. The same similarity pattern can be produced whichever language is innovating. In some cases it is Spanish that has 'innovated' features that neither French nor Portuguese have: changing certain vowels to diphthongs, for instance, as in siete vs. sept, sete. Or all the languages may have changed, but Spanish in one way, and French and Portuguese in another. Many cases of the cluster [kt] turned to [tfl in Spanish, but French and Portuguese both simplified it to [t], as in ocho vs. huit, oito. Meanwhile original [k(w)] before high front vowels ended up as [Θ] in Spain, but as [s] in France and Portugal: cinco vs. cinq, cinco. Taken together, all of these features can indeed outweigh the 'common Iberian' ones. Note that the erosion of final vowels applies to many examples, and nasalisation also to the one and five cognates. In combination they unquestionably do make certain French and Portuguese cognates overall more similar phonetically to each other than to Spanish, their spellings notwithstanding: siete vs. sept, sete is actually [sjete] vs. [set], [set] (or [seBt]). This is precisely as is reflected in the weighted quantifications for phonetic similarity that our model produces. Bearing all these different features in mind, just for the numerals 'one' to 'ten' that make up the small and purely data set in Figure 7, that tree suddenly no longer looks so 'wrong' at all. As these cases in Romance show, such mismatches between similiarity and genealogy are more than simply possible, they are only to be expected in the history of real languages. It should come as no surprise to find different trees for language classification by history (with a focus on order of branching) and by comparison (measuring degrees of similarity in some parameter or parameters). And it is hardly the case that the phenograms are of no interest; on the contrary they should be of interest precisely because they validly challenge a phylogeny-only analysis and show it in a wider context. It has long been recognized that in many cases the phylogeny model is far from ideal as the sole model for classifying language varieties, and means of representing the relationships between them. Indeed nowhere is this more true than at the dialect level, particularly for dialect continua with their complex webs of overlapping and cross-cutting features which do not map neatly onto a single tree. The origins and relationship of Latin American to European Spanish is a case in point, and a much more

From phonetic similarity to dialect classification

87

complex issue than a simple binary split: southern Spain actually shares many of the so-called 'Latin American' traits. Yet many of the linguists who have rightly criticized the tree-model as not the whole story have then proceeded in the same breath to say that they use it anyway for lack of any practical alternative. We suggest that methods of quantifying similarity in phonetics (and at other levels), and producing phenograms from the results, do offer useful alternatives to complement the traditional - but selective and unweighted - family tree representation. So while there often will be a correlation between the phylogram and phenogram representations, we must also take into account the possibility that classical historical linguistic trees, drawn typically on the basis of particular shibboleth sound changes that are 'known' to have contributed to the splitting of the languages or varieties concerned, may differ from trees generated automatically, using a broad range of data which has not been sifted or hand-picked (one might even say 'edited') in advance by the linguist. In these cases, our programmes may pick up cumulative differences or similarities across the phonetic data that have not been prioritized by historical linguists hitherto, on the grounds that they are not so directly diagnostic of the branching history of the family. The disentangling of the various types of relationship is also an intriguing prospect, though there is not the space to attempt this here. The point of introducing them now is only to encourage readers not to write off the method simply because the trees produced may not match in every respect those which are familiar from historical linguistics. Finally, Figure 8 shows a subset of the results for our 100-cognate set displayed using another computational tool, this time the statistics package SPSS. This scatter plot compares the change between Latin and Caracas Spanish on the one hand, as opposed to that between Latin and Madrid Spanish on the other. Forms in the top right-hand corner are identical to one another and to their Latin ancestor, while those in the bottom left-hand corner show most difference from the Latin ancestor. If all present-day words in the two Spanish varieties were identical to one another, regardless of their distance from Latin, they would fall on a straight line; and deviations from this straight line in either direction show that the phonetic shape of the cognate in one variety of Spanish is more different from Latin than is its counterpart in the other. Points below the line indicate a more conservative form in Madrid Spanish, or at least a form closer to the Latin one; conversely, points above the line show cases where Caracas Spanish is closer to Latin than is Madrid Spanish. If the rate of change in both varieties had been identical, then we would expect a roughly equal scatter of points above and below the line.

88

Paul Heggarty, April Mc Mahon and Robert

McMahon

l.Oi .9' .8·

rt J "λ ο • I—I cc//55 C3

J?

.7' •D

.6'

o

ο rt t-l a U

-C [Λ

'2ea P. en

jr

„Ο

υ

ΪΛ «J

oD

.5·

a

0° Ο

ο

.4·

t

tP

tP

.3' .2'

.1·

.2

.4

.6

.8

1.0

Spanish Madrid / Classical Latin

Figure 8. Scatter plot of phonetic similarity between Latin and Caracas Spanish plotted against the similarity between Latin and Madrid Spanish for the 100-cognate sample used in Figure 5

In fact we find more points below the line, indicating that Caracas Spanish has been subject to more change than Madrid Spanish since the time when they shared a common series of phonetic forms. Clearly, this is relevant to the discussion on Figure 7 above, on the discrepancies between phenograms and phylograms that may be caused by different rates of change. The two diagrams taken together suggest that Caracas Spanish has indeed changed quite considerably in its phonetics, while Madrid and Bogota Spanish have both remained relatively conservative since the time when all three shared a common ancestor It is not possible here to develop and discuss the implications of these extensions and applications, which are necessarily only preliminary

From phonetic similarity to dialect classification

89

sketches. However, they should serve to show that we are seeking not only to develop a single method, but to integrate the results of our phonetic similarity comparisons with other work, involving the application of biological and statistical computational tools. We hope that extensions and developments of this research will allow us to move towards a means of establishing, quantifying and representing different types, extents and patterns of phonetic change, as well as simply synchronic distance, at the dialect level and beyond.

Notes 1. The project 'Quantitative Methods for Language Classification' is supported by the Arts and Humanities Research Board (grant number AN6720/APN 12 536, 2001-04). We gratefully acknowledge the support of the AHRB and of our coworker on the project, Natalia Slaska.

References Allen, William Sidney 1978 Vox Latina. Cambridge, Cambridge University Press Blancquaert, E. and W. Pee 1925-82 Reeks Nederlands(ch)e Dialectatlassen. Antwerp: De Sikkel. Browman, C., and L. Goldstein 1992 Articulatory phonology - an overview. Phonetica, 49: 155-180, 222-234. Dixon, Robert M.W. 1997 The Rise and Fall of Languages. Cambridge, Cambridge University Press. Dyen, Isidore, Joseph B. Kruskal and Paul Black 1992

An Indoeuropean classification: a lexicostatistical experiment. Transactions of the American Philosophical Society 82, Part 5. Data available at http://www.ldc.upenn.edu.

Felsenstein, J. 2001 PHYLIP: Phylogeny Inference Package. Version 3.6. Department of Genetics, University of Washington. Goebl, H. 1982 Dialektometrie: Prinzipien und Methoden des Einsatzes der Numerischen Taxonomie im Bereich der Dialektgeographie. Wien: Österreichischen Akademie der Wissenschaften.

90

Paul Heggarty, April Mc Mahon and Robert McMahon 1984

Dialektometrische Studien: Anhand italoromanischer, rätoromanischer und galloromanischer Sprachmaterialen aus AIS und ALF. 3 Volumes. Tübingen: Max Niemeyer. Greenberg, Joseph 1987 Language in the Americas. Stanford, Stanford University Press. Grimes, J.E., and F.B. Agard 1959 Linguistic divergence in Romance. Language, 35: 598-604. Heeringa, W., J. Nerbonne and P. Kleiwig 2002 Validating dialect comparison methods. In Classification, Automation, and New Media: Proceedings of the 24th Annual Conference of the Gesellschaft für Klassifikation, University of Passau, March 15-172000, W. Gaul and G. Ritter (eds.), 445-52. Berlin: Springer. Heggarty, Paul 2000 Quantifying change over time in phonetics. In Time Depth in Historical Linguistics, Colin Renfrew, April McMahon, and Larry Trask (eds.), 531-562. Cambridge: McDonald Institute for Archaeological Research. in prep. Measured Language: From First Principles to New Techniques for Putting Numbers on Language Similarity: Oxford: Blackwell. Hoppenbrouwers, C. and G. Hoppenbrouwers 1988 De indeling van noordoostelijke dialecten. TABU: Bulletin voor taalwetenschap 23: 193-217. Kessler, Brett 1995 Computational dialectology in Irish Gaelic. In Proceedings of the 7th Conference of the European Chapter of the Association for Computational Linguistics, 60-67. Dublin: European Chapter of the Association for Computational Linguistics. Lass, Roger 1984 Phonology. Cambridge, Cambridge University Press. Laver, John 1994 Principles of Phonetics. Cambridge, Cambridge University Press. Levenshtein, V.l. 1965 Binary codes capable of correcting spurious insertions and deletions of ones. Problems of Information Transmission 1: 8-17. Mallory, J.P. 1989 In Search of the Indo-Europeans. London, Thames and Hudson. McMahon, April, and Robert McMahon 2003 Finding families: quantitative methods in language classification. Transactions of the Philological Society 101 (1): 7-55. Navarro, G. and M. Raffinot 2002 Flexible Pattern Matching in Strings: Practical on-line search algorithms for texts and biological sequences. Cambridge: Cambridge University Press.

From phonetic similarity to dialect classification

91

Nerbonne, J., and W. Heeringa 1997 Measuring dialect difference phonetically. In Workshop on Computational Phonology, J. Coleman (ed.), 11-18. Madrid: Special Interest Group of the Association for Computational Linguistics. Nerbonne, J., with W. Heeringa and P. Kleiwig 1999 Edit distance and dialect proximity. In Time Warps, String Edits and Macromolecules: The theory and practice of sequence comparison, D. Sankoff and J. Kruskal (eds.), ν -xv. Stanford: CSLI. Nerbonne, J., and W. Heeringa 2001a Dialect areas and dialect continua. Language Variation and Change 13:375-400. Nerbonne, J., and W. Heeringa 2001b Computational comparison and classification of dialects. Dialectologia et Geolinguistica: Journal of the International Society for Dialectology and Geolinguistics 9: 69-83. Nerbonne, J., and W. Kretzschmar In press Introducing computational techniques in dialectometry. To appear in Computers in the Humanities. Sankoff, D. and J.B. Kruskal (eds.) 1983 Time Warps, String Edits, and Macromolecules: The theory and practice of sequence comparison. Reading, Massachusetts: AddisonWesley Publishing Company. Swadesh, Morris 1952 Lexico-statistical dating of prehistoric ethnic contacts. Proceedings of the American Philosophical Society 96: 452—463. Vieregge, W.H., A.C.M. Rietveld and C.I.E. Jansen 1984 A distinctive feature based system for the evaluation of segmental transcription in Dutch. In Proceedings of the 10th International Congress of Phonetic Sciences, 654-659. Dordrecht: Foris.

Inflectional variation in Belgian and Netherlandic Dutch: A usage-based account of the adjectival inflection Jose Tummers, Dirk Speelman, and Dirk Geeraerts

Introduction The aim of the present contribution is to report on a quantitative analysis of inflectional variation in the Dutch language area, viz. in the Netherlands and in Belgium. After the Treaty of Munster (1648), the Southern Netherlands, nowadays the Dutch-speaking part of Belgium, and the Northern Netherlands, nowadays the Netherlands, were separated. While a Dutch standard language was developed in the independent Northern part, the Southern part remained occupied by the Spanish, the Austrian and the French, all using French in formal contexts. This privileged position of French was maintained during the first century after the Belgian independence in 1830. At the beginning of the twentieth century, the Dutch-speaking part of Belgium - Flanders - adopted the Northern, Netherlandic norm. From the thirties of the twentieth century onward, an explicit language policy resulted in a converging tendency between Belgian and Netherlandic Dutch, Belgian Dutch moving toward Netherlandic Dutch. Nevertheless, this separation caused a still observable standardisation backlog of the Southern area, the Dutch-speaking part of Belgium, with respect to the Northern area, the Netherlands. In this paper, the synchronic geographical and stylistic variation in Dutch will be focussed on. The following research questions will be addressed. First of all, we will verify to what extent the model for lexical variation in Dutch (Geeraerts, Grondelaers, and Speelman 1999; Grondelaers et al. 2001) can be applied to an inflectional variable. Therefore an attempt will be made to project the quantitative model developed in Geeraerts, Grondelaers, and Speelman (1999) for measuring lexical variation to the domain of inflection. Second, the impact of lexical and inflectional variables on the overall language variation will be compared. This paper is organised in the following way. First, the research variables are described (1.). Next, the consulted corpus and data selection procedure will be presented (2.). After having tackled some methodological issues (3.), we will discuss the results of the empirical analysis (4). This

94

Jose Tummers, Dirk Speelman, and Dirk Geeraerts

contribution will be concluded with a brief overview of the findings obtained in the present investigation and we will touch on some problems to be addressed in further research (5.).

1. The dependent variable In this paper, one particular type of inflectional variation in Dutch will be focused on, namely the alternation between the declined and the undeclined attributive adjective with a singular neutral nominal head and a definite determiner, e.g. het sterke paard vs. het sterk paard ('the strong horse'). 1 The definite determiner can be a definite article {het 'the'), a demonstrative (die/dat 'this/that'), or a possessive pronoun (mijn/jouw/zijn/... 'my/your/his/...'). Although Standard Dutch prefers the declined adjective, i.e. the adjective taking the suffix -e {het/dit/dat sterke paard), the undeclined alternative is not prohibited in this particular syntagmatic context (het/dit/dat sterk-0 paard). The absence of clear rules regulating the use of both possibilities results in a language situation characterised by hesitation and hence variation. Even normative grammarians have to admit that the use of both inflectional forms often is a matter of personal appreciation by the language user (Verhasselt 1972). Although the choice between these two forms does not seem to be semantically motivated,2 different language internal and language external parameters may influence the alternation between declined and undeclined adjectives. The alternative without the suffix -e appears more often when the determiner is a possessive pronoun (Haeseryn et al. 1997: 411; de Rooij 1980a: 6, 13-25): in sequences with a possessive determiner, such as mijn sterk paard ('my strong horse'), the undeclined adjective is more widespread than in sequences where the definite article het ('the') is used as determiner, as in het sterk paard ('the strong horse'). Apart from the word class of the determiner, rhythmic characteristics of the adjective may alter the inflection. The longer the adjective, the easier the undeclined form is accepted, as in het ongedifferentieerd(e) corpus ('the undifferentiated corpus') (Booij 1992: 129-130; Haeseryn et al. 1997: 411-412; de Rooij 1980a: 7-8, 13-14, 22-23). This inclination is reinforced when the adjective ends in an unvoiced syllable, such as het ontzagwekkend(e) paard ('the awe-inspiring horse'). From a lexical point of view, a tendency to use the undeclined adjective can be observed when the adjective-noun sequence constitutes a lexical idiom (Booij 1992: 128; Haeseryn et al. 1997: 408; de Rooij 1980a: 7). Because of the idiomaticity of the expression, the declined adjective in het mondelinge examen ('the oral exam') is considered less suitable than the undeclined alternative in het

Inflectional variations in Belgian and Netherlandic

Dutch

95

mondeling examen. A lot of such adjective-noun idioms with an undeclined adjective refer to public entities, such as het Openbaar Ministerie ('the Public Prosecutor') and het Algemeen Nederlands ('Standard Dutch'). Since the predominance of the undeclined adjective in lexical idioms almost excludes any variation, these expressions will be excluded as much as possible from the empirical analysis. From a language external point of view, stylistic and geographical parameters modify the adjectival inflection. Stylistically, the undeclined alternative is considered to be more widespread in less formal contexts (de Rooij 1980a: 7, 15). Geographically, Belgian Dutch favours the undeclined adjective compared to Netherlandic Dutch (Haeseryn et al. 1997: 412; de Rooij 1980a: 15, 24—25). This geographical assumption is confirmed by dialectological evidence. Dialect charts clearly show that the undeclined adjective is restricted to Belgian - Flemish - dialects and some dialects in the Northern and North-eastern part of the Netherlands, whereas the declined counterpart is characteristic of Netherlandic dialects (de Rooij 1980b: 109-119; Taeldeman 1980: 224). In brief, the alternation between declined and undeclined adjectives in Dutch is far from being unambiguous. Instead, an intricate web of interacting tendencies results in a linguistic situation characterised by hesitation and variation. 2. The CONDIV corpus and the data selection The use of non-elicited language data is a very appropriate method for variational linguistics. First, the language user's awareness often reports less flexibility than exhibited in the actual language use (Geeraerts, Grondelaers, and Bakema 1994: 17). Second, the gradual nature of language variation excludes the use of binary evaluations based upon introspection (Hudson 1996: 148-150). The use of linguistic surveys, which is a very common method in dialectology, can be considered as collective introspection. Finally, since the perceptibility between the declined and the undeclined adjectival forms amounts to the presence or absence of the suffix -e, which is phonetically realised as a mute e, the use of introspection and elicited language judgements in surveys can be considered less suitable for the present investigation. Having justified the use of non-elicited language data, we will present the structure of the consulted CONDIV corpus.3 This is a geographically and stylistically controlled database of written Dutch containing about 45 million words. Geographically, the corpus is divided in a data set for Belgian Dutch and a data set for Netherlandic Dutch. In both

96

Jose Tummers, Dirk Speelman, and Dirk

Geeraerts

geographically defined corpus parts, the stylistic axis constitutes a cline ranging from very formal to very informal language situations. The formal pole of this continuum comprises materials extracted from newspapers. Quality newspapers as well as popular newspapers have been excerpted.4 Moreover, the language use in the newspapers is subjected to a double editorial control, a first one by the journalist and a second one by the editorial board of the newspaper. The less formal data have been downloaded from the internet: data from offline Usenet discussion groups and materials produced in online Inter Relay Chat (IRC) sessions have been gathered. The IRC materials are the most informal and colloquial register because of the online production of the discussions. The language use of this communication medium is often referred to as written spoken language (Grondelaers et al. 2000: 358). Most IRC channels are centred around the inhabitants of a particular city or region and hence regionally defined. Moreover, Belgian channels consequently "ban" Netherlandic chatters, and vice-versa. The offline production of the usenet discussions, participants add an e-mail to a "thread", results in a more controlled and less colloquial style compared to the IRC data. In brief, the following parameters define the stylistic continuum: the editorial control, the size of the intended audience, and the adaptation to the degree of education of the intended audience. The first parameter fits the labovian view on register variation (Labov 1994: 157), attributing variation to the attention paid to the production, while the other two parameters answer to the register conception outlined by Bell (1991: 104—125), linking stylistic variation to the speaker's adaptation to the intended audience. The data used in the empirical analysis have been collected by means of a syntagmatic search.5 The use of the syntagmatic search strategy is justified because of the untagged nature of the corpus and the nature of adjectival inflection which is based upon agreement. Both circumstances require a strict delimitation of the syntagmatic context. On the one hand, the syntagmatic search compensates for the absence of part-of-speech tags by means of a lexical delimitation strategy, that searches for specific lexical items in the first stages of the corpus query. On the other hand, since the adjectival inflection is based upon agreement, the grammatical context determining the variation must be defined: the adjective must follow a definite determiner and precede a singular neuter noun. The former circumstance can be considered a practical restriction, the latter a grammatical one. More in particular, the corpus search consists of three consecutive steps, satisfying both the practical and the theoretical requirements. First, the corpus has been queried for the most informative part of the syntagmatic context, viz. the singular neutral nominal head. The following twenty-two nominal heads have been selected in the corpus frequency lists: beeld ('image'), beleid ('policy'), boek ('book'), einde

Inflectional variations in Belgian and Netherlandic

Dutch

97

('end'), geweld ('violence'), hoofd ('head'), karakter ('character'), kind ('child'), land ('country'), lichaam ('body'), loon ('wages'), materiaal ('material'), meisje ('girl'), nummer ('number'), oog ('eye'), probleem ('problem'), publiek ('public'), scherm ('screen'), spel ('game'), verhaal ('story'), verschil ('difference'), water ('water'). Proceeding this way, nouns figuring in a lot of (registered) idiomatic expressions such as examen ('exam') can be eliminated from the data base at the outset (see section 1). The results obtained in the first step are refined in the second stage. Only the observations with a definite determiner as leftmost element, followed by an underspecified word position, i.e. a string of alphabetic characters, and one of the twenty-two head nouns, have left. The aim of this step is twofold. First, after having fixed the rightmost element of the syntagmatic context, the leftmost element, the determiner, is defined. Moreover, the corpus being untagged, the adjective position remains underspecified during this step. The final stage of the selection procedure specifies the adjective position, eliminating invariable adjectives such as het pima spel/*het primae spel ('the excellent strategy'). Furthermore, different observations that formally match the sequence constituted by a definite determiner and a string of alphabetic characters followed by a singular neutral noun such as dat een boek ('that a book') are discarded because they do not fulfil the linguistic prerequisites. The dat ('that') in dat een boek ('that a book') cannot be a determiner. This element is a complementiser and een boek ('a book') is the first constituent in the subordinate clause introduced by dat, as in ik denk dat een boek een goed geschenk is ( Ί think that a book will make a great gift'). After these three consecutive stages 4087 observations remain: in 3454 cases the adjective is declined, in the 633 remaining observations the adjective does not take the suffix -e. This proportion between both inflectional variants corroborates the status of the declined adjective as the preferred alternative.

3. Methodology In this section, we will define the measures used to quantify inflectional variation: the inflectional profile and the degree of uniformity between two language varieties. The inflectional profile adapts the onomasiological profile to the domain of inflectional morphology. The onomasiological profile of a lexical concept collects the various lexical items expressing this concept, regardless of conceptual or categorial variability (Geeraerts, Grondelaers, and Bakema 1994: 1-16; Geeraerts, Grondelaers, and Speelman 1999: 5-6, 37). According to this approach, the onomasiological profile of the concept JEANS is composed of the lexical items jeans and blue jeans, which are conceptually equal. By contrast, the alternation between

98

Jose Tummers, Dirk Speelman, and Dirk Geeraerts

jeans and trousers is conceptually differentiated: the first alternative is a hyponym, while the second one is a hyperonym. That is, these two items do not constitute an onomasiological profile. Projected onto the domain of inflectional morphology, the inflectional profile collects conceptually equal alternatives to express an inflectional category, viz. the declined and undeclined adjectives as a case of agreement within the noun phrase.6 To put it in a quantitative definition, the inflectional profile expresses the ratio between the relative frequencies of the inflectional varieties in a data set. In the research reported upon in this paper, the inflectional profiles are defined within the syntagmatic context determined by the 22 head nouns (see section 2). Consider the following example. In the syntagmatic context defined by the head noun nummer ('number'), 55 adjectives take an inflectional -e and 10 remain undeclined in Belgian Dutch. In the Netherlandic data set, these numbers are respectively 49 and 3. Converted into relative data, this amounts to 0.85 declined forms and 0.15 undeclined forms in the Belgian data set, and 0.94 declined forms and 0.06 undeclined forms in the Netherlandic counterpart. These proportions constitute the inflectional profiles of nummer in the Belgian (0.85:0.15) and in the Netherlandic (0.94:0.06) data sets. For every geographically and stylistically defined data set in the corpus, the inflectional profile can be assessed. The inflectional profiles in two sources constitute the numeric input for the computation of the uniformity degree between those sources. The uniformity measure computes the degree of overlap between two data sets, i.e. between two inflectional profiles. The result ranges from 0 to 1: a result of 0.00 indicates the absence of overlap; a score of 1.00 indicates a total overlap between the concerned data sets; a score of 0.50 indicates that one half of the observations in the confronted data sets overlaps and that the other half does not. The uniformity score is computed in two stages. During the first stage, the uniformity Uz between two individual inflectional profiles is measured. During the second stage, the individual uniformity measures are integrated in a global uniformity measure U' between two data sets. In order to compute the uniformity degree between two inflectional profiles, the following formula is conceived in Geeraerts, Grondelaers, and Speelman (1999: 44): (1)

Uz (X,, Y2) = £ min(FZJ ((x,), Fzje 2 (x,)) i=1

The formula in (1) sums up (Σ) the smallest proportions of the adjectival alternatives Xj to xn in the inflectional profiles7 Ζ for the concerned data sets Yi and Y2: min(FZYi(xi),Fz>Y2(Xi)). Since there are two inflectional

Inflectional variations in Belgian and Netherlandic Dutch

99

alternatives, viz. the declined and the undeclined form, η is equal to 2. We will illustrate how this formula works looking at the inflectional profiles for nummer (Z). Computing the uniformity for nummer between Belgian Dutch (data set Yi) and Netherlandic Dutch (data set Y2), this formula adds up the smallest proportion of the declined forms in the Belgian and the Netherlandic data sets and the smallest proportion of the undeclined forms in these data sets, respectively 0.85 (the lowest score out of 0.85 and 0.94) and 0.06 (the lowest score out of 0.15 and 0.06). This amounts to a uniformity rate, UNUMMER, of 0.91. At this point, it becomes clear why the inflectional profiles use relative frequencies. Working with relative frequencies, we reduce both distributions to the same denominator, i.e. 1.00. It should be mentioned that the uniformity degree does not indicate any degree of standardisation or normativity.8 Since uniformity quantifies the degree of overlap between the language use in two data sets, a high uniformity score does not necessarily imply a high standardisation degree. In this research, the normativity can be indirectly observed in the inflectional profiles, since the declined form is considered more normative than the undeclined alternative (see section 1). In the case of nummer, the language use in Netherlandic Dutch would be more normative than the language use in Belgian Dutch, due to the higher proportion of declined forms in the former variety. The second stage in the computation of the uniformity degree between two data sets integrates the individual uniformity measures U z for the inflectional profiles Zj into a global uniformity measure U' for all profiles. The underlying idea is that a selected sample, viz. the 4087 observations defined by the 22 neuter head nouns, is representative for a particular variable, viz. adjectival inflection. The computation of the global uniformity U' is performed using the formula in (2) (Geeraerts, Grondelaers, and Speelman 1999: 45). (2)

U'W^YJ-YUzMJi^WtPiVYJ i=l

This formula consists of two parts. The first part sums up (Σ) the uniformity scores U z for the different inflectional profiles Zj to Z„. Given the 22 nominal heads defining the inflectional profiles in the database, η is equal to 22. This U z result - which is the output of the first formula computing U z - is multiplied by the weighting factor G z for that inflectional profile Ζ in the concerned databases YiuY 2 . This weighting factor is the quotient of the frequency of the implied inflectional profile and the total frequency of all inflectional profiles summed up. Again, we will discuss the case of nummer to exemplify this technical definition. The score of the inflectional profile

100 Jose Tummers, Dirk Speelman, and Dirk Geeraerts U z defined by nummer is 0.91. This uniformity score is multiplied with the weighting factor W 2 for the inflectional profile of nummer. This factor is obtained dividing the number of tokens in the inflectional profile of nummer by the total number of tokens in the 22 inflectional profiles included in the database. Doing so, GNUMMER equals 0.029 (= 117/4087, i.e. all tokens defined by the inflectional profile nummer divided by all tokens included in the database). Next, this weighting factor has to be multiplied by UNUMMER. Doing so, 0.91 multiplied by 0.029 equals 0.026. The procedure for nummer has to be repeated for every inflectional profile, and the sum of these 22 scores is U'. This overall U' rate between Belgian and Netherlandic Dutch equals 0.8861.

4. Empirical analysis The empirical investigation falls into two parts: the proper analysis of the inflectional variation (4.1.) and the comparison of the inflectional and the lexical variation (4.2.).

4.1. Variation of the adjectival inflection As indicated in the introduction, one of the goals of this paper is to verify to what extent the inflectional variation matches the model of lexical variation in the Dutch language area outlined in Geeraerts, Grondelaers, and Speelman (1999) and in Grondelaers et al. (2001). Geeraerts, Grondelaers, and Speelman (1999) use an ad hoc corpus in order to measure geographical and stylistic similarities for near-synonyms expressing 15 football and 15 clothing concepts. One of the thirty concepts is TROUSERS, that can be conveyed either by broek or by pantalon. Geographically, Belgian and Netherlandic sources are distinguished; stylistically, the clothing materials are gathered from - formal - fashion magazines and informal - shop windows. Grondelaers et al. (2001) replicate this analysis for the clothing concepts using the CONDIV corpus, with three stylistic strata : IRC, Usenet and the newspaper materials. The synchronic model for variation in the Dutch language area that emerged from the quantitative analyses of the lexical variables can be turned into the following hypotheses. First, there exists a positive correlation between the degree of similarity between the national varieties of Dutch on the one hand and the formality of the stylistic register on the other hand. Second, the differences between the formal and the informal

Inflectional variations in Belgian and Netherlandic

Dutch

101

registers are smaller in Netherlandic Dutch than the corresponding differences in Belgian Dutch. This global picture seems to be a general tendency in so-called pericentric languages (Clyne 1992). The first hypothesis states that the degree of similarity between Belgian and Netherlandic Dutch is higher in the formal than in the informal registers. This can be implemented using the equation in (3), predicting growing uniformity results moving from the informal end to the formal end of the stylistic continuum. In the formulas, Β and Ν respectively stand for Belgian Dutch and Netherlandic Dutch. The characters in subscript identify the register: IRC represents Inter Relay Chat, USE the Usenet materials, and NP the newspaper materials. (3)

U'(BIRC,NIRC) < U'(B use ,N use ) < U'(B np ,N np )

This hypothesis is corroborated by the empirical findings. The data in (4) show a progressive increase of U' between the Belgian and the Netherlandic data sets from IRC over USE to NP. (4)

U'(B irc ,N irc ) < U'(B use ,N use ) < U'(B np ,N np ) 0.7066 NUSE)

0.7330 lamb-es, kalb-es > lemb-ir, kelb-ir The stem building suffix thus turned into a plural suffix. From the fourteenth till fifteenth century on (Wegera 1987:210s), this suffix was extended to other neuters {Band 'ribbon', Land 'country', Brett 'board', Feld 'field', Grab 'grave', Haus 'house', Holz 'wood', Kind 'child', Kleid 'dress', Kraut 'herb', Loch 'hole', Rad 'wheel', Wort 'word'), as well as to some masculines {Ball id., Mann 'male person', Mund 'mouth'). The speakers thus had a new plural marker available which allowed for distinct plural marking even in apocopating dialects, the more so as this -ir suffix triggered umlaut, originally the consequence of a vowel harmony rule. A further advantage of the -er plural consisted in its facilitating biunique plural forms, whereas the -e forms were homonym with the dative singular: dem Lande - die Lande, but dem Lande - die Länder. Recall that in former times NPs were used more frequently without articles. Thus we find unequivocal dative singular forms in zu Hause ('at home'), zu Wasser und zu Lande ('by sea and land'), zu Grabe tragen ('carry to the grave'), bei Lichte besehen ('on closer examination', lit. 'seen at light'), zu Buche schlagen ('show favorably in the books') which differ from the new -er plural forms Häuser, Länder, Gräber, Lichter, Bücher. In some cases, concurring forms coexist until today in place names (5a), as variants with meaning differentiation of the doublets (5b), and as variants in apocopating southern dialects (5 c):

212

Heide Wegener

(5) a. archaic form name common noun Lande ('lands') Niederlande ('Netherlands') Länder ('countries') Eberswalde (place name) Wälder ('forests') Burghausen6 (place name) Häuser ('houses') Baden (place name) Bäder ('baths') b. Worte - Wörter ('utterances' - 'words'), Gesichte - Gesichter ('visions' - 'faces'), Bande - Bänder ('ties' - 'ribbons'), Lande Länder ('lands' - 'countries'), Schilde - Schilder ('shields' 'signs'), Tuche - Tücher ('cloths' - 'scarfs'), Orte - Örter ('places' - 'coalfaces'), Lichte - Lichter ('candles' - 'lights'). c. Standard Steine, Beine, Stücke, Reste, Brote ('stones', 'legs', 'pieces', 'rests', 'breads') Betten, Hemden, Hefte, Dinge, Rosse ('beds', 'shirts', 'note-books', 'things', 'horses')

Southern dialects Steiner, Beiner, Stücker, Rester, Bröter Better, Hemder, Hefter, Dinger, Rösser

The formerly high productivity of the -er plural is documented by forms used in the eigteenth century, where this plural marker was applied to native nouns (6a) as well as to loans (6b), see Paul (1917: 24, 32, 134): (6) a.

Beeter ('beds'), Beiler ('axes'), Blecher ('steel sheets'), Ender ('ends'), Flösser ('rafts'), Jöcher ('yokes'), Kleinöder ('gems'), Kreuzer ('crosses'), Röhrer ('pipes'), Wichter ('gnomes'), Gesänger ('singings'), Scheusäler ('ugly mugs'), Seiler ('ropes'), Törer ('gates'), Ungetümer ('monsters'), Werker ('works'), Zelter ('tents'), Zieler ('goals'), Zeuger ('stuffs')

b. Hospitäler ('hospitals'), Regimenter ('regiments'), Kabinetter ('cabinets'), Kameler ('camels'), Lokäler ('pubs'), Elementer ('elements'), Publikümer ('audiences'), Lazaretter ('MASHs'), Billetter ('tickets'), Buketter ('flower bunches'), Skandäler ('scandals'), Testamenter ('last wills'), Complimenter ('compliments') (see Wegera 1987:196s for further examples) Later on, some of the -er forms were substituted by -e forms (Kreuzer > Kreuze 'crosses') when the final schwa became re-articulated since the sixteenth to the seventeenth century (Wegera 1987:196s), others became outdated, see below and Wegener (2002:295). Thus, the -er plurals were diminished. During the twentieth century their number decreased considerably, see table 1 in the appendix.

Development and motivation of marked plural forms in German

213

In the framework of NM and Optimality Theory (OT), -er plural forms can be explained with the following Naturalness Principles NP (Wurzel 1984, Dressler 1999) or well-formedness constraints (cf Golston and Wiese 1996, Fery 2001): (7)

FinSon:

Plurals end in a sonorant: Hunde ('dogs'), Frauen ('women'), Länder/ *LandFeet are trochaic: Länder/ *Land- (NP of optimal word length). The morphological form should be unambiguous: dem Lande - die Länder vs. dem Lande - die Lande (NP of bi-uniqueness).

TrFoot: *Hom:

The -er plural can be explained functionally by its uniqueness and as a substitute for the inaudible -e plural (expressed in T1 through -*e). In Southern German dialects with general schwa apocope, the -er plural still has a compensatory function, especially for nouns with a front vowel that can not apply the umlaut plural, see (5c). T1 Sg: Land stN Land-*e Länd-*e •"»-Länd-er

FinSon

TrFoot j *Hom

*

*

*

*

! * I

1 1 1

It is therefore not by chance that the loan word Ski (id.), which, just like its referent, is mostly used in the southern parts of the German language area, developed an -er plural: Skier. If skies were predominantly used in the north, we would probably find the plural form Skie, analogous to Knie ('knees'). To my knowledge, Skier is the last creation of an -er plural in Standard German, where this plural marker is now unproductive (Wurzel 1984, Eisenberg 1998); in dialects, however, it still continues to be productive. The diminution of the -er plural today is less due to change in inflection class (e.g.: Denkmäler > Denkmale 'monuments'), but should rather be seen as a result of the extinction of nouns such as Bälger ('brats'), Wämser ('doublets'), Gemächer ('chambers'), Mäkler ('meals'). The -er plural vanishes together with these nouns which are being substituted by more modern ones {Bälger ~ Kids). The -er plural can thus be explained as an opportunity to satisfy the relatively high-ranked demands for distinct, overt, iconic, and unambiguous plural marking. Since the -e plural was, in Early NHG, not unambiguous and insufficient due to schwa apocope, and the ~(e)n plural would not be IC-specific, the -er plural form constituted the optimal solution.

214

Heide Wegener

The decline of the dative singular marker -e (dem Lande > dem Land 'to the country'), however, renders one of the two motivations for the -er plural superfluous, and in non-apocopating varieties, the second motivation does not apply either. Insofar, the -er plural is afunctional in today's Standard German. This gives a functional explanation for its present unproductivity.

3. The umlaut plural Similar conditions hold for the umlaut plural -"(e), which is also more prevalent in Southern German than in Standard German, originates at the same time and is motivated in the same way. Umlauted -(e) plurals originate in the OHG -i class, where the umlaut was phonologically triggered by the following -i {gest-i, plural of gast ('guest')), and then spread to nouns of other classes which never had the umlaut triggering environment. In this way, umlaut was reanalysed and morphologizised to a plural marker, see Wegera/Solms 2000. When used in its syllabic form as -"e, the result is a combination of affix + vowel fronting, hence a redundant plural form that contradicts basic assumptions both of MT and OT concerning principles of economy. Nevertheless, this plural allomorph was not only retained in many nouns mostly of high token frequency (see Kopeke 1994), but it was also newly created with borrowings (see Paul 1917:133): Kardinäle ('cardinals'), Pläne ('plans'), Märsche ('marches'), Chöre ('choirs'), Strände ('beaches'), Choräle ('chorals'). In both cases it changes the stem vowel and leads to forms not corresponding to their bases. Thus we have to explain why preference was given to the less corresponding form for certain nouns. There are two factors relevant for the fronting of the vowel. First, umlaut is the only plural marker audible in apocopating dialects which inhibit suffixation hence fronting is really necessary. This assumption is supported by the observation that umlaut occurs - in all plurals of non-derived strong feminines (die Kühe, Mütter 'cows', ' mothers V, - more often in varieties in the South of the German-speaking area where schwa apocope is general:7 Täg, Arm. The reason for both usages is a functional one: Feminines share the same article with plurals, in apocopating varieties -e is truncated, hence the -e plural is inaudible without umlaut. The result would be non-distinct plural forms: die *Hand ('hands'), *Arm ('arms'), *Tag ('days'). The plural forms of feminines would be completely indistinguishable from their singular, masculines and neuters only by the article. Dialectal umlauted forms Arm-, Täg- (Standard: Arm-e, Tag-e) are a compensation of the not

Development and motivation of marked plural forms in German

215

realisable schwa affix. It is not by chance that it is the Alemannic plural variant for Park (id.) that surfaces as Park-.8 Second, non-umlauted -e plural forms are identical with the forms for dative singular of strong masculines and neuters: dem Hunde / die Hunde ('to the dog / dogs'), but dem Wolfe / die Wölfe ('to the wolf / wolves'). Umlaut plurals hence are unambiguous, biunique, whereas pure -e forms are not. Recall again that in former times, many more dative singular nouns without article were used: zu Grunde gehen ('perish', lit. 'go to the ground'), zu Kopfe steigen ('go to someone's head'). These forms were disambiguated by the umlaut plural forms Gründe, Köpfe. (Bi)Uniqueness is a well known principle of NM, i.e. the principle that "one and the same form has always the same meaning (and vice versa)" (Dressier 1999:137). In sum, umlaut is retained and was newly applied in order to make the plural forms distinct and unique. Like the -er plural, the umlaut plural can be explained functionally by its uniqueness and as a substitute for the inaudible -e plural. T2 Sg: Bach gtM +-e Bach-e wBäch-e Bäch (Swabian)

FinSon

TrFoot ; *Hom

:* 1

*

*

:

It is generally accepted that it was exactly the non-distinctness of the forms of the weak inflection class which made the entire case marking disappear from the singular of weak feminines (see Wegera 1987:75, 254). 9 The same factor motivates the ongoing loss of the -e suffix from the dative singular of strong nouns (dem Hunde > dem Hund 'to the dog'). However, this decline of the dative singular -e renders the umlaut plural, too, afunctional in non-apocopating dialects. It is therefore likely that, in the future, borrowings may not front the vowel. Since the dative singular form (dem ?Parke) is obsolete, the -e plural is no longer ambiguous and its umlauted form could, except for apocopating dialects, become as afunctional as the -er plural. In fact, umlaut is rare in assimilated borrowings (for statistics see table 1 in the appendix, also Kopeke 1993:144,148). My claim that umlaut is a compensatory device, seems contradicted by the statistical distribution of umlaut in nouns (with an umlautable stressed vowel) applying the syllabic -e plural compared to nouns ending in -el, -er, -en and thus applying its zero variant. Whereas 50,1 % of the former do front the stressed vowel, only 3,24 % of the latter do so. 10 The reversed would be expected. This striking non-occurrence of the umlaut in nouns where it is needed, can be explained by three factors:

216

Heide Wegener

1. Most of the nouns ending in -er are deverbal agent nouns {Fahrer 'driver', Zuschauer 'spectator', Besucher 'visitor', Raucher 'smoker') in which an umlaut plural would blur the identity between the verbal base and the derived noun. 2. Most of the nouns ending in -el are deverbal instrument nouns {Paddel 'paddle' - paddeln, Hobel 'plane' - hobeln) and / or the base for denominal verbs {Handel 'trade' ~ handeln 'to trade', Wandel 'change' - wandeln, Jubel 'exaltation' - jubeln , Tadel 'rebuke' - tadeln, etc/ Here, an umlaut plural would blur the identity between the base and the derivative. 3. The nouns ending in -en only recently changed from the weak inflection class (never applying umlaut) to the strong one {Balken 'beam', Gedanken 'thought', Funken 'spark'), from which only very few developed an umlaut plural {Gärten 'gardens'), more in Southern dialects {Wägen, Bögen ('carriages', 'arches'). For the first two groups, it is obvious that an analysis of the pluralization has to take into account not only the singular stem but other related stems as well. The third group of rather new strong nouns shows the importance of the date of a noun's appearance in a given inflection class.

4. The -s plural Since the seventeenth century, German has created a further plural marker, the -s, and so developed a new plural class. Why did German develop an additional plural marker? -s plural forms penetrated the German plural system from Low German in the sixteenth, from French in the eighteenth, and from English in the nineteenth century. In the case of proper names, the -s plural originates in the genitive-s: 's Meiers (Swabian, 'The Meier's (farm, house and family) '), see Curme (1922:93). Today, the -s plural is not only applied to loan words from French or English (8a), where it could have been imported together with the noun, but it also occurs with italianisms, latinisms (8b), and neologisms ending in a full vowel (8c), as well as with native words under the condition that these are phonologically or lexically special or marked, i.e. with onomatopoeia and interjections (9a) and with proper names (9b).

Development

and motivation of marked plural forms in German

217

(8) a.

anglicisms, gallicisms: die Balkons ('balconies'), Parks (id.), Lifts ('elevators'), Jobs (id.), Hits (id.), Laptops (id.)

b.

latinisms, italianisms: die Pizzas (id.), Pizzerias ('pizza restaurants'), Mensas ('university lunchrooms'), Kontos ('bank accounts')

c.

neologisms with final vowel: die Opas ('grandpas'), Unis ('universities'), Autos ('cars'), PKWs [pekaves] ('cars')

a.

onomatopoeia: die Uhus ('eagle owls'), Kuckucks ('cuckoos'), Wauwaus ('bow-wows'), die Achs und Wehs ('continual lamentations')

b.

proper names: die beiden Berlins ('the two Berlins'), die zwei Deutschlands ('the two Germanys'), die Kochs, Bachs, Beckers ('the Koch, Bach, Becker family'), die (Opel)Kadetts (car name).

(9)

The following questions need to be addressed: Is this new plural marker a substitute plural (just like the -er and the umlaut plural were in Early NHG), is it an emergency plural for cases where native pluralization is impossible, or is it even the German default plural (as claimed by Marcus et al. 1995)?

4.1 The -s plural = an emergency plural? In most cases, the -s plural is not an emergency plural, since the productive native suffixes -(e) and -(e)η are being used for the homonyms of the proper names: die Köche ('the cooks'), Bäche ('brooks'), Bäcker ('bakers'), Kadetten ('cadets'), but not for the proper names or product names, see (9b). The emergency plural explanation will thus only apply to nouns ending in a full vowel. Here, native pluralization would lead to a trisyllabic word with a hiatus which, in German, is impossible for unstressed syllables: *'Opae ('grandpas'), *'Unien ('universities'), *'Autoe ('cars'), *'Pizzaen (id.), for details see 4.4.1 and 4.5.2. This is the only case where the -s plural is phonologically conditioned, i.e., -s is the emergency plural marker only for nouns with final unstressed full vowel.

218

Heide Wegener

4.2 The -s plural = a substitute plural? In some cases, the -s plural may be considered a substitute plural, namely when used in Northern German varieties for nouns applying the -0 plural in MSG, i.e. strong masculines and neuters ending in a schwa syllable: (10)

die Mädels ('the girls'), Kumpels ('buddies'), Onkels ('uncles').

In this function, Southern dialects especially in Austria employ -n: die Onkeln, Madeln, or the umlaut plural: die Wägen,, Bögen, see 3. The -s plural has been used as a substitute for -0 since it became available in the seventeenth century, see Paul 1917: 131. Nouns ending in a schwa syllable can not create syllabic plurals because such a form would violate the constraint prohibiting adjacent homonymic unstressed syllables, in OT, the OCP (Obligatory Contour Principle). They can neither apply an -n plural, as this form would violate the requirement for plurals to be inflection class conform. The umlaut plural is available only for nouns with a back vowel and is avoided for newly borrowed nouns for reasons that will be explained below. Consequently, these nouns can not overtly mark the category plural. Just like the -er plural and the umlaut plural in apocopating dialects, the -s plural has a compensatory function here and constitutes a repair strategy for the -0 plural. Contrary to the -er and the umlaut plural forms described in section 2 and 3, however, none of these -5 plural forms have been able to prevail in Standard German, so that the -s plural can be considered a substitute plural only in Northern German varieties of colloquial speech.

4.3 The -s plural = the correspondence plural It is an interesting fact that the -s plural applies to two phonologically respectively semiotically special word classes, namely onomatopoeia and proper names, although native plural forms are possible and even do exist for the names: (11)

a. die Kuckucks ('cuckoos'), Wauwaus ('bow-wows'), Uhus ('eagle owls'), b. die Bachs, die Vogels, die Kochs ('the Bach, Vogel, Koch family'), die zwei Deutschlands ('the two Germanys'), die beiden Berlins ('the two Berlins').

In these cases, the -s plural is motivated communicative-grammatically rather than phonologically. For both onomatopoeia and proper names, the

Development and motivation of marked plural forms in German

219

non-syllabic -s suffix allows for highly corresponding, i.e. structure preserving, "conservative" forms similar to the singular by adding only a segment, not a syllable to the stem. By contrast, a native plural with a syllabic suffix alters more or less severely the phonological shape of the base. E.g. Bücher [by:.9s] ('books') differs from its singular Buch [bu: x], Länder [len.de] ('countries') from Land [lant] 1. 2. 3. 4.

in the fronting of the vowel, in the palatalization resp. the sonority of the final consonant, in the additional syllable, i.e., in the prosodic structure, the foot, in the syllabification, separating morpheme from syllable boundary:

(12)

Buch - Bü.cher, Land - Län.der

The alterations yield a continuum of similarity depicted in the following diagram: (13) Sg: PI:

Ecke Ecken

Frau Frau.en

Boot Boo.te

Hund [t] Hun.de [d]

Hand [t] Hän.de [d]

Sg: PI:

Berg

Riss

Bank

Burg [k]

Buch [χ]

Bergs

Ris.se

Ban.ken

Bur.gen [g]

Bü.cher [ζ]

0,17

0,82

0,91

2,29

3,94 10,14

The extremely high token frequency of -er and the relatively high token frequency of the umlaut plural do not only explain that these forms have

Development and motivation of marked plural forms in German

229

been preserved but also how they are learned: it is their great lexical strength that allows these forms to be learned by rote (Bybee 1995: 428, 433). The high token frequencies result from the fact that the nouns applying -er or the umlaut plural belong to the central lexicon, i.e. they refer mostly to entities of the 'sphere personnelle', see Kopeke 1994: 83 who demonstrates that the nouns with umlaut plural refer to humans, higher animals and body parts rather than to plants or parts of plants. Umlauted forms are less transparent but more salient than non umlauted forms, in combination with the other alterations of the base they represent more or less suppletive forms. It is a well known fact (see Werner 1989, Wurzel 1990) that irregular and even suppletive forms likewise problematic in NM bear certain advantages in communicative-pragmatic regard, that morphological irregularity is thus not only a by-product of phonological development. For words of the core lexicon a high degree of differentiation within the forms of the paradigm represents an advantage since the possibility of misunderstanding is reduced. Hence we can consider the -er and the umlaut plural as unmarked within marked subdomains of the lexicon, as cases of markedness reversal. By contrast, -s plural forms are of extremely low token frequency, they constitute the opposite pole on the token:type-ratio scale in (23). This corresponds to their position on the similarity scale in (13). Obviously, there is a connection between token frequency and non-correspondence or saliency. As shown in section 4, -s is a plural marker that facilitates structure preserving plural marking without violation of the identity constraints. The common characteristic of the three word classes that apply the -s plural (onomatopoeia, names and loans) is their phonological or semiotic markedness. The phonological structure of onomatopoeia, nouns formed completely following phonological criteria, is motivated by their onomatopoeic function, hence their plural form is particularly dependent on a high degree of similarity or correspondence with the singular, that of proper names, nouns that identify their referents, is sacrosanct, they are linguistically and even legally protected against modification. Loan words can only be altered in their phonological structure after they have gained a certain degree of familiarity. I will therefore propose to consider the three word classes that prefer the -s plural as a 'similarity domain'. The new -s plural can likewise be explained communicative pragmatically: Correspondence with the base is an advantage for the reconstruction of the singular from the plural, for the maintenance of specific functions, i.e. the imitating function of onomatopoeia and the identifying function of proper names, and for the spreading of borrowings into the new speech community. In contrast to suppletive forms and the irregular plural classes with -er and umlaut, the nouns that take the -s plural

230

Heide Wegener

are mostly of extreme low token frequency, see (23). Due to the special phonological and semiotic structure of these word classes, correspondence becomes a necessity that dominates all other constraints. The -s plural for nouns with special lexical features can thus be considered another case of markedness reversal. The identity constraints are generally dominated in German by the principles of iconicity, of optimal word length and of optimal syllable structure, or, put in OT, by the prosodic and phonological wellformedness constraints TrFoot, FinSon (and Onset, BiMor etc). Common native nouns follow the constraint hierarchy (21)

H c n : Wellformedness constraints »

identity constraints.

By contrast, the special, "anomalous" nouns follow the constraint hierarchy for special nouns (22)

HSn: Identity constraints » wellformedness constraints.

Hence these nouns constitute exceptions within the German plural system, they are marked. Their markedness results from the special lexical features of the nouns they apply to. The native schwa plural formation for common nouns is rooted in the pressure for trochaic plural forms to be easy to articulate and for inflection class differentiation. The -s plural formation of special nouns, by contrast, is rooted in the need for preservation of the sound shape of onomatopoeia and of proper names, and in the need for listener-oriented ease of recognition of unassimilated borrowings, nouns not yet fully established in the speech community. The way the different plural forms are retained, used, and reproduced in Modern Standard German thus exhibits useroptimality constraints (see Haspelmath 1999).

Development and motivation of marked plural forms in German

231

Appendix: Table I. Statistical development of the plural classes during the twentieth century, type frequencies for regular and irregular classes in MSG, Ν = 2307 (Mugdan 1977:97, based on Kaeding 1898), resp. Ν = 6505 (Pavlov

weak: -(e)n 1898 1975

Regular strong:

55.78 47.5

Σ %1898 Σ % 1975

-(e) 27.35 35.8 85.73 92.3

special: -s 2.6 9.0

mixed: -(e)n 1.04 1.7

Irregular strong: -"(e) 9.23 3.8 12.57 % 6.2 %

strong: -er 2.3 % 0.7 %

List of abbreviations C F Gen Sg IC MHG MSG NHG NM Nom PI NP OHG stN stM Fem V

Consonant Foot Genitive Singular Inflection class Middle High German Modern Standard German New High German Natural Morphology Nominative Plural Naturalness Principle Old High German strong Neuter strong Masculine Feminine Vowel

List of constraints FinSon: TrFoot: *Hom: Onset: Max-ΙΟ: Max-Lex: Max-Sf: BiMor: *Syll: Ident-SP:

Plurals end in a sonorant Feet are trochaic The morphological form should be unambiguous Syllables have an onset. Every segment of SI has a correspondent in S2 (no deletion). Every lexical segment of SI has a correspondent in S2. Every suffix segment of SI has a correspondent in S2. Unaccentuated syllables are maximally bimoraic. Do not add a syllable. Corresponding forms of singular and plural have identical values for V (vowel of the stem), C (final consonant), F (foot structure).

232

Heide Wegener

Notes 1. I would like to thank Barbara Menzel for helping to translate this article. 2. I will use the term "umlaut plural" for the -(e)+ umlaut plural in contrast to the er plural which exhibits obligatory fronting of the vowel. 3. Strong feminines have disappeared except for the plural forms. 4. Weak neuters have disappeared except for the plural forms: Ohren ('ears'), Augen ('eyes'). 5. Statistically, the -(e)n plural applies to 100% of weak nouns, the -(e) plural to 72,57 % of strong masculines and neuters (type frequencies, see Pavlov 1995:45s). 6. * indicates that there are no neuters in that IC. 7. This is a dative plural form, as is Baden. 8. Hände, Füße ('hands', 'feet') are in Swabian die Händ-, die Füß-, 9. In a test conducted at different universities, 33% of the students from Zurich chose the form Parke compared to maximally one person from the other universities. I would like to thank Peter Gallmann for his help. 10. The former genitive singular in -en is conserved only in compounds: Frauenkirche, Marienkirche ('Our Lady's church', 'Mary's church'). 11. The statistics were calculated on the base of the indications given in Pavlov 1995:44. Since not all anglicisms and gallicisms apply -s (die *Gangsters, die *Oranges, die *Bosses, die *Faxes), the -s plural must be considered as created by Germans. For the non-occurrence of -s, see Wegener i.p. 13. Both forms coexist today, see Wegener 2003 for statistics. 14. Some speakers create mit den Auton. 15. During the twentieth century their number decreased considerably, see table 1 in the appendix. The ratio is calculated on the base of the indications given in Pavlov 1995:45-48.

References Bartsch, Renate, and Theo Vennemann 1983 Grundzüge der Sprachtheorie. Tübingen: Niemeyer Bybee, Joan L. 1995 Regular Morphology and the Lexicon. Language and Cognitive Processes 10: 425-455. Casali, Roderic F. 1997 Vowel Elision in Hiatus Contexts: Which Vowel Goes? Language 73: 493-533. Curme, George O. 1922 A Grammar of the German Language. New York: Unger

Development and motivation of marked plural forms in German

233

Dressler, Wolfang U. 1999 What is Natural in Natural Morphology? In Prague Linguistic Circle Papers 3, Eva Hajiqova e.a. (eds.), 135-143. Amsterdam/Philadelphia: Benjamins. Duden 2 1989, 42001 Deutsches Universalwörterbuch. Mannheim: Duden Eisenberg, Peter 1998 Grundriss der deutschen Grammatik, Vol. 1: Das Wort. Stuttgart: Metzler. Fery, Caroline 2001 Phonologie des Deutschen, Eine optimalitätstheoretische Einführung, Vol. 1. Potsdam UB Golston, Chris, and Wiese, Richard 1996 Zero morphology and constraint interaction: subtraction and epenthesis in German dialects. In Yearbook of Morphology 1995, Geert Booij and Jaap van Marie (eds), 143-159. Dordrecht: Kluwer Haspelmath, Martin 1999 Optimality and diachronic adaptation. Zeitschrift für Sprachwissenschaft 18.180-205. Kopeke, Klaus-Michael 1993 Schemata bei der Pluralbildung im Deutschen, Tübingen: Narr 1994 Zur Rolle von Schemata bei der Pluralbildung monosyllabischer Maskulina. In Funktionale Untersuchungen zur deutschen Nominalund Verbalmorphologie, Klaus Michael Kopeke (ed.), 81-96. Tübingen: Niemeyer, Mayerthaler, Willi 1981 Morphologische Natürlichkeit. Wiesbaden: Athenaion. Marcus, Gary F., Ursula Brinkmann, Harald Clahsen, Richard Wiese, and Steven Pinker 1995 German inflection: the exception that proves the rule. Cognitive Psychology 29: 189-256. McCarthy, John J., and Alan S. Prince 1995 Faithfulness and reduplicative identity. In Papers in Optimality Theory, J.N. Beckman, L. Walsh Dickey, and S. Urbanczyk (eds.), 249-384. University of Massachusetts Occasional Papers 18. Pavlov, Vladimir 1995 Die Deklination der Substantive im Deutschen. Synchronic und Diachronie. Frankfurt/M.: Lang Paul, Hermann 1917 Deutsche Grammatik. Vol. 2: Flexionslehre. Tübingen: Niemeyer. Uspensky, B.A., and Zhivov, V.M. 1977 Center-Periphery Opposition and Language Universals, Linguistics 196: 5-24

234

Heide Wegener

Vennemann, Theo 1988 Preference laws for syllable structure and the explanation of sound change. Berlin: de Gruyter. Wahrig, Gerhard 1968 Deutsches Wörterbuch, Gütersloh/Berlin: Bertelsmann Wegener, Heide 2002 Aufbau von markierten Pluralklassen im Deutschen - eine Herausforderung fur die Markiertheitstheorie, Folia Linguistica 36: 261-295. 2003 Normprobleme bei der Pluralbildung fremder und nativer Substantive, Linguistikonline 15/3-03 i.p. Pizzas und Pizzen - die Pluralbildung (un)assimilierter Fremdwörter im Deutschen. Wegera, Klaus-Peter 1987 Flexion der Substantive. (Vol. 3 of Grammatik des Frühneuhochdeutschen, Moser, Hugo, Hugo Stopp, and Werner Besch (eds), Heidelberg: Winter Wegera, Klaus-Peter, and Hans-Joachim Solms 2001 Morphologie des Frühneuhochdeutschen. In Sprachgeschichte. Ein Handbuch zur Geschichte der deutschen Sprache und ihrer Erforschung, Besch, Werner, Anne Betten, Oskar Reichmann and Stefan Sonderegger (eds), 1542-1554. Berlin/New York: Mouton de Gruyter. Werner, Otmar 1989 Sprachökonomie und Natürlichkeit im Bereich der Morphologie. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 42: 34-47. Wurzel, Wolfgang U. 1984 Flexionsmorphologie und Natürlichkeit. Ein Beitrag zur morphologischen Theoriebildung. Berlin: Akademie-Verlag. 1990 Gedanken zu Suppletion und Natürlichkeit. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 43: 86-91.

Not arbitrary, not regular: The magic of gender assignment Marcin Kilarski and Grzegorz Krynicki

1. Introduction In this paper we would like to report on a study of gender assignment of English loanwords borrowed into Danish, Swedish and Norwegian.1 Gender in the Scandinavian languages has proved to be a lively area of research, which is demonstrated by a range of recent publications.2 Several important questions remain to be answered: most importantly, to what degree gender assignment can in these languages be regarded as regular. If it is regular, then we would like to know what formal and semantic features of the nouns are relevant in gender assignment, and how important they are with regard to one another. We want to suggest that given proper analytical tools gender assignment, while not completely regular, can to a large degree be attributed to individual criteria. Therefore we would like to validate statistically the contribution of these criteria, and measure their relative importance in the three languages. We begin with a brief description of the gender systems in the analysed languages and of the postulated gender assignment criteria. This is followed by an overview of the corpus of English loanwords and the types of analysis used in the study. The paper concludes with a discussion of the results for the three languages.

2. Gender in English and Scandinavian English appears to be an ideal candidate as the lending language as it lacks grammatical gender; in addition, it continues to be a rich source of borrowed lexicon. The choice of the borrowing languages was motivated by the presence of subtle differences between these typologically and genetically close languages. Danish and Swedish share a two-gender system of neuter gender and common gender formed through a merger of masculine and feminine.3 Norwegian presents a more complex case as it may be treated as a two- or a three-gender system, depending on dialectal, structural and stylistic considerations. In Norwegian Bokmäl, feminine

236

Marcin Kilarski and Grzegorz

Krynicki

gender enjoys an unstable position, which is demonstrated by the small number of nouns with obligatory feminine forms and a weak status of feminine agreement 4

3. Gender assignment criteria In line with previous studies, two broad types of criteria were postulated as significant in the assignment of gender: semantic and formal, where among the latter we further distinguish between phonological and morphological criteria. Semantic rules include four semantic criteria: personal, nonpersonal animate, abstract and concrete, as well as associations between loanwords and synonymous native nouns. The principle of animacy, according to which animate nouns are assigned to a non-neuter gender, reflects what Greenberg (1966), Aksenov (1984) and Corbett (1991) have referred to as the semantic basis or core of gender. Phonological rules involve the number of syllables, CV structure, selected final sequences of segments, and finally associations with native homophones. And lastly, we deal with four types of morphological assignment rules: inflectional, based on plural declension,5 derivational, based on suffixation (here 32 suffixes were analysed) and the derivation of deverbal nouns of the type run and take-off, and finally compounding.

4. Description of the corpus The study is based on 3,796 English nouns, borrowed into the three languages in the following numbers: 2,728 in Danish, 2,037 in Swedish and 2,527 nouns in Norwegian. The corpus includes nouns borrowed from or through English; in vast majority these are words borrowed in the post-war period. A few early loans from the Old English period have been excluded due to a different structure of all the languages involved. As regards types of loanwords, the only type included are direct loans, and so excluded are hybrids, caiques and semantic loans. The nouns were checked against a selection of six monolingual dictionaries and dictionaries of anglicisms.6

5. Analysis Two types of analysis have been applied. In the first place, a quantitative analysis was applied to account for the share of individual genders within the three languages, and to measure the contribution of individual assignment rules within the genders. Secondly, a discriminant function

Not arbitrary, not regular, the magic of gender assignment

Til

analysis made it possible to build a model that allowed us to classify loanwords with respect to their gender assuming the postulated criteria. Further, this procedure was used to determine the relative contribution of the criteria to the assignment of gender to loanwords. Discriminant function analysis is one of the extensions of Multivariate Analysis of Variance - a test for significant differences between the means of several groups. In the variant we applied, discriminant function analysis consists in a linear delimitation of the spaces constituted by the features of the classified loanwords. In the course of the analysis, the coefficients of the functions defining these spaces are adjusted to minimize the misclassification rate of new occurrences.

6. Discriminant analysis - predictor variables and levels of grouping variable In our study, we postulated that 19 factors might influence the assignment of gender to a word borrowed from English into any of the three Scandinavian languages. These 19 factors constituted a grammatical specification of each loanword. This specification was used in the analysis as a set of independent variables on the basis of which each loanword was classified with respect to the gender it assumed in one of the target languages. In our study, the set of potentially significant independent variables was the same for each of the languages we considered (see Table 1). Nevertheless, some of these variables, different in each language, had to be eliminated in the course of the analysis for technical reasons (see section 6.2.1).

The dependent (grouping) variable was one and the same for each language: the gender of a noun borrowed from English into that language. The values it assumed for Norwegian, however, were different than the values it assumed for Danish and Swedish as we distinguish three genders in Norwegian in contrast with two genders in Danish and Swedish.

23 8

Marcin Kilarski and Grzegorz Krynicki

Table 1. Predictor variables Question asked to find values of the Variable name variable Is the noun POLYSYLLABICITY polysyllabic? Is the penultimate PENULTIMATE PHONEME LENGTH phoneme long? rt LAST PHONEME Is the last phoneme υ 00 LENGTH ο "ο NUMBER OF How many a 1 FINAL consonants are there word-finally? CONSONANTS Does the Danish FINAL GLOTTAL noun end in a glottal STOP IN DANISH stop? Is the borrowed DEVERBAL NOUN noun a deverbal WITH A PARTICLE noun with a particle,

Variable classification

Admissible answers in decreasing order of frequency yes, no no, yes no, yes

1,0, 2 , 3 , 4

no, yes

no, yes

e.g., take-off? DEVERBAL MONOSYLLABLE

Η Ο

Is the noun a deverbal monosyllable, e.g.,

no, yes

run?

PL,

Λ •Ο rH

£bO "o

SUFFIX

What derivational suffix does the noun contain?

PLURAL

What plural form does the noun build?

BASE GENDER

Assuming the noun is a compound, what is the gender of its base?

Λ& Ο

0, -er, -ing, -ie, -ion, -ism, -ist, -an, -ine, -al, -ette, -ance, -man, -ment, -ics, -ery, -ity, -ness, -um, -eme, -age, -ure, -ant, -ate, -ive, -phone, -logy, -ship, -us, -ade, -scope, -graph, -graphy unassigned, -er, -s, 0, -e, -ar, -r, -or, -n, -men, -man, -es, -a, -menn, -im NO GENDER, common, masculine, neuter, common OR neuter, masculine OR feminine

Not arbitrary, not regular, the magic of gender assignment

239

Table 1. cont.

PERSONAL

NON-PERSONAL ANIMATE

a e

(U 00

CONCRETE

ABSTRACT

Influence of native equivalent

Does the noun belong to the semantic category of PERSONAL? Does the noun belong to the semantic category of NON-PERSONAL ANIMATE? Does the noun belong to the semantic category of CONCRETE? Does the noun belong to the semantic category of ABSTRACT? What is the gender of the native equivalent of the noun in the form of a homonym or synonym?

no, yes

no, yes

yes, no

yes, no

NO GENDER, common, neutral, masculine, feminine

7. Methodological aspects of the study 7.1. Assumptions behind the discriminant function analysis Before discriminant analysis can be employed certain assumptions about the variables have to be met. Technically speaking, in order for the Discriminant Analysis to be performed, the data must comply with the following requirements (Chatfield and Collins 1980: 125ff.; Krzysko 1982: 9): a) The number of classes (groups) distinguished in the analysis must be at least 2. In the case of the present study, for Danish and Swedish we distinguish 3 classes ( "no gender", "common" and "neuter") and for Norwegian we distinguish 4 classes ("no gender", "feminine", "masculine" and "neuter").

240

Marcin Kilarski and Grzegorz

Krynicki

b) Each class must be represented by at least 2 cases. In our case each class is represented by 5066 cases in Danish, 4732 cases in Swedish and 3704 cases in Norwegian. c) The number of distinct variables must be less than the total number of cases minus the number of groups. For Danish maximally 11 distinct variables were accepted in the model, for Swedish 12, and for Norwegian 7. In all cases the number of variables is much smaller than the total number of cases minus the number of groups. Among the assumptions of discriminant analysis there is also a requirement that the classes must have a multivariate-normal distribution and that the within-group variabilities (expressed as within-group correlation matrices) must be approximately equal. The non-categorial variables on which this study was based showed minor violations in the normal distribution (especially the number of syllables in a noun). It has been assumed, however, that minor violations of normality requirement are not fatal to the results of the analysis, as is commonly agreed upon among statisticians. No significant violation of homogeneity of covariances was observed. Another assumption of discriminant function analysis is that the variables that are used to discriminate between groups are not completely redundant. As part of the computations involved in discriminant analysis, the variance/covariance matrix of the variables included in the model are inverted, which requires that no variable in the matrix is completely redundant with any other variable. In our study, this problem made it impossible to include, e.g., the variables ABSTRACT and CONCRETE together with the INANIMATE variable since the latter variable was just a sum of both ABSTRACT and CONCRETE variables. A final assumption of the discriminant function analysis is that the means for variables across groups are not correlated with the variances (or standard deviations). To guard against this problem we inspected the means and standard deviations and variances for such a correlation removing any outliers in non-categorial variables.

7.2. Nominal case classification instead of noun classification In order to classify nouns borrowed from English into the Scandinavian languages with respect to criteria that may have alternative forms for a single noun (e.g., the English noun baby when borrowed into Swedish may take three plural endings: -ar, -er and -5), the number of cases submitted to classification had to be greater than the number of the nouns themselves. A

Not arbitrary, not regular, the magic of gender assignment

241

noun with several alternative forms at a given field of its grammatical specification had to be instantiated by a corresponding number of cases. Each case of the same noun had the same specification as its counterparts, except those elements of the specification that included alternative forms. If the given noun contained yet another element of the specification that had several alternative forms (e.g., variant pronunciations for baby), the total number of cases was a product of the number of alternative forms at the first field of specification and the number of alternative forms at the second field of specification. If the noun contained any more alternative fields in its grammatical specification, the procedure of multiplying cases was iterated until all alternative fields were considered. Thus, the classification we conducted was not so much a classification of the borrowings themselves. It was rather a classification of grammatical configurations these borrowings occurred in.

8. Results of the quantitative analysis A general result of the study is that common gender in Danish and Swedish and masculine gender in Norwegian are overrepresented in comparison with the native lexicon (see Table 2).7 Table 2. Distribution of nouns in the corpus, in comparison with the native lexicon

Danish Swedish

Norwegian

loan native loan native

loan native

common 85.3 75 90.4 75 masculine 91.1 65

neuter 14.7 25 9.6 25 feminine 0 24

neuter 8.9 11

The data for Danish and Swedish show a marked increase in the percentage of common nouns (85.3% and 90.4% vs. 75%), together with a decrease among neuter nouns (14.7% and 9.6% vs. 25%).8 In Norwegian 91.1% of single-gender nouns appear as masculine, as opposed to only 8.9% for neuter nouns. Importantly, there are no feminine-only nouns, and 11 nouns vacillate between masculine and feminine. While a comparison with the native lexicon is here difficult on account of the instability of feminine gender, the tendency away from the feminine and neuter is clear.

242

Marcin Kilarski and Grzegorz Krynicki

The expansion of common and masculine genders is confirmed by the results of the discriminant function analysis. 9. Results of discriminant function analysis As we have already mentioned, discriminant analysis has two major applications: to include many potentially significant predictor variables in order to determine what is the relative discriminating power and to predict a correct classification of cases after a discriminant function model has been built. In our study, discriminant function analysis was applied, firstly, to find out what is the relative importance of different criteria in the assignment of gender to nouns borrowed from English into the three Scandinavian languages and, secondly, to check to what degree the grammatical specification of these borrowings determines their gender in the target languages.

9.1. Classification results Once the discriminant function model has been built, we may show how well we can predict to which gender group a particular case belongs. The fact that many variables were rejected in the classification was caused by their violation of at least one of the assumptions of discriminant analysis (see section 6.2.1 above). The post-hoc classification results we obtained for each language are given below in Table 3. Table 3. Results of discriminant analysis classification for Danish, Swedish and Norwegian9 percentage of correctly classified cases Danish Swedish

Norwegian

common

neuter

unassigned

total

95.34556 98.4953

19.46524 2.719665

13.82253 3.195489

71.91077 67.39222

neuter 0

unassigned 11.82635

total 67.65659

masculine 97.08

feminine 0

The overall percentage of correctly classified cases ranges from 67.4% in Swedish and 67.6% in Norwegian to 71.9% in Danish. It has to be admitted that this result is rather modest and suggests that we are only dealing with a certain degree of regularity in gender assignment in the analysed languages. As regards the individual genders, the model classifies

Not arbitrary, not regular, the magic of gender assignment

243

common and masculine cases more successfully than neuter and feminine ones - compare the figures for Danish, with 95.3% for common as opposed to 19.5% for neuter gender. The results for common gender in Swedish and masculine in Norwegian are 98.5% and 97.1% respectively, compared with only 2.7% and 0% for neuter gender. Interestingly, no feminine and neuter cases in Norwegian have been classified correctly; these nouns are instead grouped in the model together with masculine nouns. It appears then that there are only a few or no features at all which positively characterize neuter and feminine genders.10

9.2. Relative discriminatory power of predictor variables based on a forward stepwise discriminant function analysis In forward stepwise discriminant analysis, a model of classification proceeds step-by-step. At each step all the variables still not in the model are evaluated to determine which one has the greatest discriminating power. The variable with the greatest discriminating power is then included in the model and the analysis is iterated until all the variables are considered. The results for the three languages are shown below in Tables 4—6. Plural inflection was shown to have the greatest discriminant power, with a comparable value of Wilks' Lambda in Danish and Norwegian (0.81 and 0.82) and a higher one in Swedish (0.93). These results match the contrasts between the two comparable declensional systems in Danish and Norwegian and the more complex system in Swedish. As regards semantic rules, perhaps surprisingly, the variables PERSONAL, NON-PERSONAL ANIMATE and ANIMATE appear only in Danish and Swedish, in contrast with the variables CONCRETE and ABSTRACT which appear in all the three languages with average values. More significant is the contribution of formal variables, including other morphological variables than PLURAL, i.e. DEVERBAL MONOSYLLABLE and SUFFIX together with DEVERBAL NOUN WITH PARTICLE, and two phonological variables: POLYSYLLABICITY and LAST PHONEME LENGTH. The relative importance of the variable GENDER OF HOM/SYN appears to be surprisingly different across the three languages, varying from the third rank position in Norwegian to the last but one rank in Swedish. Likewise, the weak discriminating power of the variable BASE GENDER does not confirm the results of the quantitative analysis which indicates that there is a strong correlation between the gender of the compound and the gender of its base.

244

Marcin Kilarski and Grzegorz Krynicki

Table 4. Danish

ρ

entr/rem11

P"^ eve ^' 2

1

575.4286

0

0.814792

575.4286

0

2

198.4331

0

0.755555

380.7828

0

3

55.11076

2.10E-24

0.739451

274.8252

0

LENGTH°NEME

4

32 78526

7.13E-15

0.729992

215.579

0

DEVERBAL NOUN WITH PARTICLE

5

18.21457

1.31E-08

0.724773

176.6853

0

HOMSYN^

6

16 44407

7.61E-08

0.72009

150.4226

0

ABSTRACT BASE GENDER SUFFIX POT Y-

7 8 9

12.27034 8.810967 10 32715

4.83E-06 0.716613 0.000151 0.714124 3.34E-050.711218

130.9714 115.8772 104.3381

0 0 0

SYLLABICITY

10

3 991998

0-018521

0.710096

94.35886

0

CONCRETE

11

3.688877

0.025067

0.709061

86.16161

0

Step F entr/rem p-level 1 185.7887 0 2 27.27944 1.66E-12

Lambda 0.92715 0.916573

F-value 185.7887 105.2434

p-level 0 0

PLURAL DEVERBAL MONOSYLLABLE ANIMATE

·

·

·

Lambda13 F-value14

p-level15

Table 5. Swedish PLURAL SUFFIX POLYSYLLABICITY

3

29 07405

2.82E-13

0.905435

80.23929

0

4

21.36452

5.80E-10

0.897322

65.76724

0

·

CONCRETE DEVERBAL MONOSYLLABLE -STPHONEME

5

9.043933

0.00012

0.8939

54.51035

0

6

7 349089

0 .000651

0.891128

46.71016

0

PERSONAL

7

3090564

0.045568

0.889963

40.49634

0

PERSONAL

8

5 275509

0.005146

0.887979

36.12542

0

9 10

12.86873 3.017261

2.67E-06 0.049029

0.883164 0.882036

3161999 30.57253

0 0

11

2 033453

0.130997

0.881277

27.98413

0

0.193137

0.880663

25.79267

0'

ABSTRACT BASE GENDER fiFNOHR OF HOM/SYN DEVERBAL NOUN WITH PARTICLE

12

1.644929

Not arbitrary, not regular, the magic of gender assignment

245

Table 6. Norwegian

PLURAL CONCRETE GENDER OF HOM/SYN POLYSYLLABICITY ABSTRACT SUFFIX LAST PHONEME LENGTH

Step F entr/rem p-level 1 276.8458 0 2 26.97395 2.98E-17

Lambda 0.81668 0.799196

F-value 276.8459 146.2288

p-level 0 0

3

16.19723

1.85E-10

0.788831

102.3736

0

4

8.337372

1.60E-05

0.78353

78.73171

0

5 6

7.833744 2.838859

3.29E-05 0.036602

0.77858 0.776789

64.55461 54.23858

0 0

7

2.492129

0.058338

0.77522

46.83486

0

10. Conclusions In conclusion, a number of patterns have been observed. As regards the degree of arbitrariness of gender assignment, while it cannot be regarded as completely regular, a substantial proportion of the data was covered by the selected assignment criteria. The results of the discriminant analysis, together with the quantitative analysis of the individual assignment criteria have shown that the analysed criteria cluster around common or masculine gender, as a result of which neuter gender is typically characterized in negative terms. It is only in a few isolated cases that neuter gender can be described as being defined positively. A comparison of the distribution of gender and the status of selected assignment rules in the two areas of the lexicon, native and borrowed, shows that in the latter there is a marked change towards common or masculine genders in terms of type frequency, and towards the assignment rules associated with these genders. Diachronically, these synchronic patterns suggest a continuing expansion of common and masculine genders, which happens at the expense of the weakening rules for neuter gender in the three languages, and feminine gender in Norwegian.

Notes 1. For a more exhaustive treatment of gender assignment in the three languages see Kilarski (forthc.); Kilarski (in press) deals with the quantitative aspects of the assignment of English loanwords. 2. See, e.g., Braunmüller (2000), Hansen (1995) for Danish; E. Andersson (2000), Dahl (2000), Fraurud (2000), Kuhn (1985), Källström (1995, 1996) on Swedish; Enger (2001), Graedler (1996) and Trosterud (2001) on Norwegian.

246 3.

4.

5.

6.

7.

8.

Marcin Kilcirski and Grzegorz Kiynicki We hereby simplify the picture by not accounting for the differences in adjectival and pronominal agreement. For example, Swedish gender has been interpreted in terms of four traditional genders (Tegner 1892; Källström 1996), two separate gender systems (grammatical vs. semantic) (Teleman 1987) and individual gender distinctions (Dahl 2000). This nicely shows that different models can be applied - more or less successfully - to a description of the problem. A more exhaustive treatment is needed, in a comparison with Danish and Norwegian (but cf. Enger 2001). Initially the study aimed at a comparative analysis of the assignment process in four Scandinavian languages, including Icelandic, together with a possible comparison with Faroese. The two insular languages, Icelandic and Faroese, together with Norwegian Nynorsk, possess the traditional Old Norse threegender systems. The following discussion will be based however on Danish, Swedish and Norwegian Bokmäl due to the comparable size of the collected data. As regards plural inflection, it is an open question whether we can establish an assignment rule according to which the gender of a loanword is assigned on the basis of the indef. pi. ending. According to Corbett (1991: 65), "[djeclensional type must be included in the lexical entries of Russian nouns and it therefore makes sense to claim that gender specification can be derived from it". However, it equally makes sense to claim the opposite: while gender has been viewed as the primary criterion for German inflection (e.g., Bittner 1994, 2000; Zubin and Kopeke 1981) drew attention to the role of the frequency with which a noun appears in the plural. This last point is especially relevant in the case of loanwords where the plural may not be assigned at all, or when assigned, may get several competing endings, or most importantly, may be assigned the plural after being assigned the gender. It appears then that we should allow for two directions of motivation (cf. Doleschal 2000; Enger 2000; Zubin and Kopeke 1981), and as a result, the status of this assignment rule should be treated with caution. For Danish: Sorensen (1997) and Retskrivningsordbogen (1996); for Swedish: Selten (1993), Svenska akademiens ordlista (1996) and Nationalencyklopedins ordbok (1999); and for Norwegian: Graedler and Johansson (1997). The figures are based on Hansen (1995) and Serensen (1995) for Danish; A.B. Andersson (1992) and Källström (1996) for Swedish; and Trosterud (2001) for Norwegian. In order to facilitate the comparison with native nouns excluded are nouns which vacillate in assignment and nouns without an assigned gender. Typically for borrowings, they constitute a sizeable group, here ranging from between 19.9% in Danish to 22.7% in Swedish and 25.3% in Norwegian. Interestingly, parallel figures are given by Jarvad (1995) in her dictionary of neologisms in Danish for the period 1955-1998. The share of common nouns is larger both among native nouns (80% vs. 20%) and loanwords (90% vs. 10%).

Not arbitrary, not regular, the magic of gender assignment 9. 10.

11.

12. 13. 14. 15.

247

See Tables 4-6 for a list of the variables that were accepted by the model in the three languages. While we here describe common and masculine genders as positively characterized, these genders can also be interpreted as default. This should be done with caution, however, as the notion has been used in a number of often contradictory ways. Nevertheless, it may certainly be useful in accounts of gender assignment and agreement - cf. the recent Corbett and Fraser (2000), Fraser and Corbett (1995), Rice and Steinmetz (2000). The effect of the stepwise discriminant analysis depends on the value of F to enter and F to remove values. The greater the F value for a given variable the greater its statistical significance in the discrimination between genders. This means that F reflects the extent to which the variable contributes to the prediction of group membership. P-value for F to enter / F to remove. Wilk's lambda; the test statistic reflecting the relative discriminating power of the variable. F-value for Wilk's lambda. P-level for the that F-value.

References Aksenov, A.T. 1984 Κ probleme ekstralingvisticeskoj motivacii grammaticeskoj kategorii roda [On extralinguistic motivation of the grammatical category of gender], Voprosyjazykoznanija 33: 14-25. Andersson, Anders-Börje 1992 Second Language Learners' Acquisition of Grammatical Gender in Swedish. (Gothenburg Monographs in Linguistics 10.) Göteborg: University of Göteborg. Andersson, Erik 2000 How many gender categories are there in Swedish? In Gender in Grammar and Cognition. Vol. 1. Approaches to Gender. Vol. 2. Manifestations of Gender, Barbara Unterbeck, and Matti Rissanen (eds.), 545-559. Berlin: Mouton de Gruyter. Bittner, Dagmar 1994 Die Bedeutung der Genusklassifikation für die Organisation der deutschen Substantivflexion. In Funktionale Untersuchungen zur deutschen Nominal- und Verbalmorphologie, Klaus-Michael Kopeke (ed.), 65-80. Tübingen: Max Niemeyer. 2000 Gender classification and the inflectional system of German nouns. In Gender in Grammar and Cognition. Vol. 1. Approaches to Gender. Vol. 2. Manifestations of Gender, Barbara Unterbeck and Matti Rissanen (eds.), 1-23. Berlin: Mouton de Gruyter.

248

Marcin Kilarski and Grzegorz Krynicki

Braunmüller, Kurt 2000 Gender in North Germanic: A diasystematic and functional approach. In Gender in Grammar and Cognition. Vol. 1. Approaches to Gender. Vol. 2. Manifestations of Gender, Barbara Unterbeck and Matti Rissanen (eds.), 25-53. Berlin: Mouton de Gruyter. Chatfield, Christopher, and Alec Collins 1980 Introduction to Multivariate Analysis. London: Chapman & Hall. Corbett, Greville G. 1991 Gender. Cambridge: Cambridge University Press. Corbett, Greville G., and Norman M. Fraser 2000 Default genders. In Gender in Grammar and Cognition. Vol. 1. Approaches to Gender. Vol. 2. Manifestations of Gender, Barbara Unterbeck and Matti Rissanen (eds.), 55-97. Berlin: Mouton de Gruyter. Dahl, Osten 2000 Elementary gender distinctions. In Gender in Grammar and Cognition. Vol. 1. Approaches to Gender. Vol. 2. Manifestations of Gender, Barbara Unterbeck and Matti Rissanen (eds.), 577-594. Berlin: Mouton de Gruyter. Doleschal, Ursula 2000 Gender assignment revisited. In Gender in Grammar and Cognition. Vol. 1. Approaches to Gender. Vol. 2. Manifestations of Gender, Barbara Unterbeck and Matti Rissanen (eds.), 117-165. Berlin: Mouton de Gruyter. Enger, Hans-Olav 2000 On the relation between gender and declension: A diachronic perspective from Norwegian. Paper presented at the conference Historische Linguistik und Grammatiktheorie 5, Berlin, December 2000. 2001 Genus i norsk bor granskes grundigere [Gender in Norwegian should be studied more thoroughly], Norsk Lingvistisk Tidsskrift 19: 163-183. Fraser, Norman M., and Greville G. Corbett 1995 Gender, animacy and declensional class assignment: A unified account for Russian. Yearbook of Morphology 1994: 165-183. Fraurud, Kari 2000 Proper names and gender in Swedish. In Gender in Grammar and Cognition. Vol. 1. Approaches to Gender. Vol. 2. Manifestations of Gender, Barbara Unterbeck and Matti Rissanen (eds.), 167-219. Berlin: Mouton de Gruyter. Graedler, Anne-Line 1996 Morphological, Semantic and Functional Aspects of English Lexical Borrowings in Norwegian. (Acta Humaniora 40.) Oslo: Universitetsforlaget. Graedler, Anne-Line, and Stig Johansson 1997 Anglisismeordboka. Engelske lanord i norsk [The Dictionary of Anglicisms. English Loanwords in Norwegian], Oslo: Universitetsforlaget.

Not arbitrary, not regular, the magic of gender assignment

249

Greenberg, Joseph H. 1966 Language universals. In Current Trends in Linguistics. Vol. 3. Theoretical Foundations, Thomas E. Sebeok (ed.), 61-112. The Hague: Mouton. Hansen, Erik 1995 Genus i nye fremmedord [Gender in new loanwords]. Sprog i Norden 24-31. Jarvad, Pia 1995 Nye ord - hvorfor og hvordan? [New Words - Why and How?]. Copenhagen: Gyldendal. Källström, Roger 1995 Om svenskans genussystem: En diskussion av nägra analysalternativ [On the Swedish Gender System: A Discussion of Selected Approaches], (Meddelanden fran Institutionen for Svenska Spräket.) Göteborg: Institutionen för Svenska Spräket. 1996 On gender assignment in Swedish. In Proceedings of The Ninth International Conference of Nordic and General Linguistics, University of Oslo, January 11-11, 1995, Kjartan G. Ottosson, Ruth V. Fjeld, and Arne Torp (eds.), 151-167. (The Nordic Languages and Modern Linguistics 9.) Oslo: Novus. Kilarski, Marcin in press Gender assignment in Danish, Swedish and Norwegian: A comparison of the status of assignment criteria. In Proceedings of the 19th Scandinavian Conference of Linguistics, Tromso, 2002. forthc. Gender assignment of English loanwords in Danish, Swedish and Norwegian. Frankfurt am Main: Peter Lang. Krzysko, Miroslaw 1982 Analiza Dyskryminacyjna [Discriminant Analysis], Poznaii: Wydawnictwo Naukowe UAM. Kuhn, Hans 1985 Genustilldelning hos främmande ord [Gender assignment of loanwords]. Svenskans beskrivning 15: 291-302. no editor 1 1999 Nationalencyklopedins ordbok. (Version 2.0 on Cd-rom.) Höganäs: Bra Böcker. no editor 2 1996 Retskrivningsordbogen. (Version 1.0 on Cd-rom.) Copenhagen: Aschehoug Dansk Forlag. Rice, Curt, and Donald Steinmetz 2000 The evolution of gender in English. Paper presented at the Workshop on Gender and Inflectional Class, University of Tromso, 27 May 2000. Selten, Bo 1993 Ny svengelsk ordbok [New "Svengelsk" Dictionary"]. Lund: Studentlitteratur.

250

Marcin Kilarski and Grzegorz Krynicki

Serensen, Knud 1995 Engelsk i dansk. Er det et must? [English in Danish. Is it a must?]. Copenhagen: Munksgaard. 1997 A Dictionary of Anglicisms in Danish. (Historisk-filosofiske Skrifter 18.) Copenhagen: Munksgaard. no editor 3 1996 Svenska akademiens ordlista. (11th edition, electronic version.) Stockholm: Norstedt. Tegner, Esaias 1892 Om genus i svenskan [On gender in Swedish]. Stockholm: Svenska Akademiens Handlingar. Teleman, Ulf 1987 Hur manga genus finns det i svenskan? [How many genders are there in Swedish?]. In Grammatik pä villovägar, Ulf Teleman (ed.), 106-114. (Skrifter utgivna av Svenska spräknämnden 73.) Stockholm: Esselte Studium. Trosterud, Trond 2001 Genustilordning i norsk er regelstyrt [Gender assignment in Norwegian is regular], Norsk Lingvistisk Tidsskrift 19: 29-58. Zubin, David Α., and Klaus-Michael Kopeke 1981 Gender: A less than arbitrary grammatical category. Papers from the Seventeenth Regional Meeting, Chicago Linguistic Society 439-449.

Future time reference: English and Dutch compared1 Griet Beheydt

1. Introduction Both the English and the Dutch tense systems contain various verbal means of referring to the future. This article is meant to provide a survey of the tense forms2 whose function it is (or may be) to locate a situation in the post-present (future). The focus will be on absolute future tense forms, which means that they locate the time of a situation posterior to the temporal zero-point (to) and thus relate it directly to t0.3 In (la) and (lb) for instance, the future tense form locates the situation of 'repeating' later in time than the temporal zero-point, here the moment of speech. (1)

a. I will repeat this once again. b. 'k Zal dit nog een keertje herhalen.(Cgn\)4 I will this still one time repeat Ί will repeat this once again'.

In Declerck's terminology (1991b: 20), locating the time of a situation in the post-present by means of an absolute tense form is called establishing a post-present domain. A temporal domain is "the time interval taken up by a situation or by a number of situations which are temporally related to each other by means of special tense forms". In both Dutch and English various means of establishing post-present domains are available: the future tense will/shall vs. zullen + infinitive, as in example (1), the originally allative5 tense forms be going to vs. gaan + infinitive and the present tense. We will discuss these absolute tense forms with future time reference in both English and Dutch, with respect to their uses and interpretations. It is a theoretical preliminary that the semantics (temporal meaning) of these three tense forms is the same: viz. locating the time of a situation in the postpresent. The restrictions the context imposes on their use are, however, not identical for the different tense forms and they are to some extent languagespecific. These restrictions are not inherent in the semantics of the tense. Rather, a particular connotation arises because of contextual hints. The

252

Griet Beheydt

distributional properties of the various post-present tense forms are more a question of pragmatics than of semantics.6 The research presented in this article is partly based on a parallel corpus consisting of two originally English detective novels and their (northern) Dutch translations: Dexter, Colin 1994 The Daughters of Cain. London: Macmillan. (Cain)7 Dexter, Colin 1995 De Dochters van Καϊη. Baarn: Centerboek. (Kai'n) James, Phyllis D. 1994 Original Sin. London: Faber and Faber. (OS) James, Phyllis D. 1995 Erfzonde. Amsterdam: De Boekerij. (EZ)

From these novels 398 and 392 examples of future tense forms (will/zullen + infinitive, be going to/gaan + infinitive, present tense) were extracted for English and Dutch respectively. As the novels are written in the past tense, the examples are taken from the dialogues. For Dutch we further examined 2470 examples from the Corpus Gesproken Nederlands [Corpus Spoken Dutch] from the Nederlandse Taalunie. 1031 of them (labelled Cgnl) are northern Dutch, 1439 (labelled Cgn2) southern Dutch. This corpus consists of both spontaneous and nonspontaneous speech. Finally, we also extracted 321 examples from the Flemish newspaper Het Nieuwsblad, May 5, 6, 9, 2001.

2. Be going to vs. gaan + infinitive First, we will describe the pragmatic interpretations of be going to and gaan. We will come to the conclusion that the various factors promoting or restricting their uses can be subsumed under one general pragmatic interpretation (2.1). Further, be going to and gaan have undergone a grammaticalization process from lexical, allative to grammatical, future (temporal) meaning. In 2.2 we will argue that this process has not yet come to an end. In 2.3 we will compare the frequency of this tense form in English and Dutch as it can be observed in the corpora at our disposal. We will end this section by pointing out that there is some regional and sociolinguistic variation with respect to the frequency and the use of be going to and gaan (2.4).

Future time reference

253

2.1. Roots in the present Most linguists ascribe two interpretations to the English form be going to + infinitive (cf. amongst many others: Declerck 1991a: 112-116; Leech 1971: 54-55; Swan 1998: 211): (a) future actualization of a present intention, as in example (2), and (b) future situation that is presently predictable, as in (3). (2) (3)

I was never much good with words, but I'm going to try. (Cain 96) I don't somehow think it's going to be of much importance to us, what time the murderer made his entrance. (Cain 40)

Various other interpretations have been attributed to be going to, and a wide range of terms have been used to capture them: "direct present evidence" (Van Brederode and Koopman 1990: 140), "present orientation" (Wekker 1976: 124), "current orientation" (Palmer 1979: 121), "present contextualization" (Haegeman 1983a: 155), "reference to both the present indications and the resulting future situation at the same time" (Close 1977: 148). Binnick (1971: 45) makes the same remark and lists three interpretations: "immediate or proximal futurity", "inceptive present" and "intentionality or intentive present". Fleischman (1982: 18-19) adds "imminence", "ulterieur", "present relevance" and "prospection or prospective present" to Binnick's list. In our view, they can all be subsumed under one overall interpretation or be derived from it, viz. the future situation is represented as presently predictable, because it is in some way rooted in, or linked with some state of affairs that already holds at t0. The nature of this link may be quite diverse (intention, present signs, cause, preparation, initiation, etc.) and therefore the exact interpretation depends on the context and is thus a pragmatic matter. Our claim is that basically the same holds for Dutch gaan. Apart from its lexical and inchoative (aspectual) meanings, gaan is used as a tense form. Temporal gaan is basically used for referring to a future situation that is somehow rooted in the present. There is, however, no agreement on this issue in the existing research, since many linguists characterizing the use of future gaan follow the Algemene Nederlandse Spraakkunst [Comprehensive Grammar of Dutch] (Haeseryn et al. 1997: 975-978). 8 The ANS sums up quite a number of contexts for, and restrictions on the use of gaan. They are quite diverse. Firstly, gaan is said to be typically used in statements about the weather. Secondly, the ANS mentions that gaan is used for referring to the semantic context 'change of state', with verbs like trouwen 'marry', verhuizen 'move (house)', etc. And finally, there are some

254

Griet Beheydt

restrictions with respect to the verbal complement: gaan does not occur in combination with hebben 'have', zijn 'be', gaan 'go' or composed verbs containing gaan, modal auxiliaries and with (be)horen 'be to', dienen 'be to/have to' and durven 'dare'. What is absent in the ANS is a generalization showing that all these restrictions and contexts can be captured under the overall interpretation of gaan given above. This is precisely what we will argue for in the rest of this section. Firstly, the claim in the ANS that gaan is often used in statements about the weather, as in (4), follows from the fact that these predictions are mostly made on the basis of present signs (in reality, on a weather map, etc.). (4)

Morgen gaat het hard waaien. Tomorrow goes it hard blow 'Tomorrow the wind is going to blow hard'.

Sentences like (5), in which these verbs do combine with zullen make clear that this is not a lexical matter, i.e. that gaan does not form a collocationlike structure with verbs like regenen 'rain', sneeuwen 'snow', etc. The future situation is not in any way grounded in the present. Hence there is no stringent reason to use gaan (cf., however, 2.2).9 (5)

(Op het einde van de vakantie zal ik een groot tuinfeest At the end of the holidays will I a big garden-party geven.-) Maar misschien zal het wel regenen. give But maybe will it modal-particle rain '(At the end of the holidays I will give a big garden party. -) But maybe it will rain'.

A similar line of reasoning can be followed with respect to the second type of verbs "preferably" combined with gaan, namely those referring to a change of state, working environment, place of residence, etc. (Haeseryn et al. 1997: 978). (6)

Wat Staat alles hier vol. Gaan jullie verhuizen? What stands everything here full go you move 'What a mess. Are you going to move?'

This, too, is totally in keeping with the above description of the use of gaan. These verbs refer to rather drastic changes in a human life. Therefore it is quite plausible that one has already made up one's mind or that one has

Future time reference

255

already started the preparations when predicting that such a drastic change is going to happen. However, there are again contexts that do not promote the interpretation that the future situation is linked with the present, and accordingly, in such contexts verbs like verhuizen can easily combine with the alternative future tense forms zullen or the present tense. Both categories (verbs referring to the weather and verbs referring to a change of state) are thus not really relevant as such. That gaan often combines with these verbs follows from the general rule that gaan is used when the future situation is somehow rooted in, or linked with the present. Thirdly, the ANS and a number of other authors stress the ungrammaticality of the co-occurrence of gaan and a number of auxiliaries. De Rooij (1986: 17-18) observes that gaan preferably co-occurs with two semantic verb categories: action verbs and verbs with the ingressive interpretation 'to move/pass/go on to, to proceed to'. This is said to be linked with two basic readings of gaan, viz. 'to move' and 'to pass on to' respectively. De Schryver (1979: 196) and De Rooij (1985: 99) also observe that the infinitival complement of gaan is often an action verb, which is quite understandable when gaan has intentional connotations (cf. to intend to perform an action). However, this cannot be stated as a rule, since, like English be going to10 (cf. Swan 1998: 212), gaan can combine with verbs that belong to neither of these categories, but which refer to a state. Example (7) illustrates this point. The alleged incompatibility of gaan with hebben, zijn and the modal auxiliaries (or, more accurately, the low frequency of this combination) can therefore not only be explained by the fact that the latter verbs are neither action verbs nor transitional verbs. (7)

Ga je daarvan later geen spijt Go you of this later no regret 'Won't you regret it later?'

hebben? (Cgn2) have

This combination, which the ANS marks as "unacceptable", is attested in the corpora, especially in Flanders (cf. also Colleman 2000b: 61). It is extremely rare in the north (only one example in Corpus Spoken Dutch). Also, it is largely restricted to spoken language (cf. Van Es and Van Caspel 1974: 109). The ANS, Ten Cate (1991: 28) and De Rooij (1985: 101), who speak of a clear and absolute restriction, are thus too strict in this respect. It is, however, true that gaan is rarer with these verbs than the alternatives zullen and the present tense and that it is not accepted by every one and in all regions. In 2.2 we will further elaborate on what factors other than the motivation 'roots in the present' may play a role.

256

Griet Β eh eydt

The remaining three verbs that the ANS mentions, viz. (be)horen 'be to', dienen 'be to/have to' and durven 'dare', seem to be somewhat arbitrarily chosen: the grammar does not provide us with any reason for this choice. Moreover, (be)horen and dienen are very infrequent in Dutch anyhow, because they are typical of formal, written language. Gaan, on the other hand, is more common in spoken language (cf. 2.4) and this may explain its infrequent combination with these verbs. Future (be)horen and dienen are not attested once in the examined part of the Corpus Spoken Dutch, durven only twice, once in the present tense, once with zullen.

2.2. Further grammaticalization: Pure futurity In some present-day spoken Dutch regiolects, gaan has started to adopt a pure future meaning, devoid of any traces of its original lexical meaning and of the above mentioned 'roots in the present' interpretation. This can be described as (the continuation of) a process of grammaticalization.11 Gaan is on its way to becoming a tense auxiliary, only expressing future time. In some West-Flemish varieties (or should we say dialects?) this grammaticalization process has already reached its terminal point, as it is the normal marker of neutral future in spoken language. The future tense form zullen is not even part of the paradigm. Even in (standard Dutch) news bulletins, we have heard pure future gaan used in combination with auxiliaries, like zijn, hebben or the modals, a combination marked as unacceptable in the ANS (cf. 2.1), but illustrating a highly grammaticalized use of gaan. (8) is an example. (8)

En die 10%, kan dat meteen worden ingevoerd of gaat dat And this 10% can that at-once be introduced or goes that ook op termijn zijn? (Radio 1 ,Actueel, 12/03/01) also on term be 'And this 10%, can it be introduced at once, or will it also be in the long term?'

English is travelling along the same grammaticalization path. Occasionally, a weather forecaster makes predictions such as that in (9), a context in which the future tense will + infinitive is typically used. (9)

Tomorrow it is going to rain.

Future time reference

257

And sporadically, be going to is even attested in combination with open and hypothetical conditions, a use which is not (yet?) generally accepted in educated English (cf. Declerck 1991a: 115-116). (10)

How the hell do you think we're going to get to the bottom of this case unless we make the odd hypothesis here and there? (Cain 7172)

The grammaticalization of a linguistic item is, amongst others, characterized by phonological and morphological reduction, semantic bleaching and specialization. Firstly, be going to has an informal and colloquial variant gonna (/gana/, sometimes even further reduced to /η/), which is especially used in American English. Also, as a future tense auxiliary, be going to has a defective conjugation (always progressive, no perfect tense forms, etc.). Secondly, be going to, like gaan, gradually loses its intentional or 'roots in the present' interpretation. What remains is pure futurity (cf. Brisard 2001: 254). And thirdly, specialization refers to the tendency for an originally optional form to become obligatory. Be going to and gaan occur in more contexts and have become the 'normal' markers of pure future in some regiolects.

2.3. Comparing frequencies As we saw, gaan and be going to are basically used in the same pragmatic contexts in Dutch and English. When actually comparing the frequencies of be going to and gaan in our parallel corpus consisting of two detective novels (cf. above), we see that these future tense forms are relatively rare (table 1) in written Dutch (dialogues). Remember that the Dutch in question is northern Dutch. English be going to is somewhat more frequent than gaan, but this sample is not heterogeneous enough to be fully representative and further research will be needed. In spoken Dutch, gaan is considerably more frequent than in written Dutch (cf. 2.4). Table I. Frequency of be going to vs. gaan + infinitive used to refer to situations having absolute future time reference in The Daughters of Cain and Original Sin and their Dutch translations English be going to Dutch gaan

10.6% 6.6%

258

Griet Β eh eydt

Be going to is still more restricted to contexts in which the future situation is rooted in the present than gaan. English be going to can therefore always be translated by Dutch gaan, though this is not actually the case in our corpus: be going to is mostly translated by zullen (38.1%) or by the present tense (31%). Only 21.4% of the English sentences containing be going to is translated by Dutch gaan. Conversely, Dutch gaan cannot always be translated by English be going to, because grammaticalized uses are rarer in English than in Dutch. Still, in 34.6% of the sentences in the Dutch novels gaan is the translation of English be going to. In 23.1% and 19.2% it is the translation of the English present progressive and the future tense respectively.

2.4. Sociolinguistic and regional variation There is some sociolinguistic variation, which again seems to be similar in Dutch and English. Be going to is more frequent in spoken and informal language than in written and more formal language. Wekker (1976: 124) calls it a "typically conversational" variant (cf. also Close 1977: 147; Brisard 2001: 254). The same holds for Dutch. This may be partly due to the stigmatization of (pure) future gaan in many normative books (cf. De Schryver 1979: 192). The figures in table 2 illustrate this point. The corpora are ordered according to register, starting from spontaneous speech, over interviews, meetings, speeches, radio news and dialogues in novels (Kai'n and E Z and library for the blind) to written newspaper language: Table 2. Frequency of gaan + infinitive used to refer to situations having absolute future time reference according to register Spoken language (Corpus Spoken Dutch) - spontaneous speech - interviews - meetings - speeches - radio news - library for the blind (dialogues in novels) Dialogues in novels (Καϊη and EZ) Newspaper, Ηet Nieuwsblad

23.6% -

-

-

-

-

30.7% 38.2% 16.8% 25.9% 12.9% 6.4%

6.6% 2.9%

Though the sociolinguistic variation is far more significant, there is general agreement 12 that there is some regional variation between the north

Future time reference

259

and the south of the Dutch speaking area: gaan is more frequent in Flemish.13 Table 3 shows the percentages for the Corpus Spoken Dutch: Table 3. Frequency of gaan + infinitive used to refer to situations having absolute future time reference in the Corpus Spoken Dutch according to region Regional variation Corpus Spoken Dutch

Northern Dutch 18.1%

Southern Dutch (Flemish) 27.4%

This regional and register variation have to do with the frequency of gaan, but also with the pragmatic contexts licensing its use. The use of gaan is less severely restricted (more grammaticalized) in spoken language and in the south.

3. The present tense with future time reference In both English and Dutch, the use of the present tense is not restricted to locating a situation at or around the temporal zero-point. The present tense may have pre-present,14 past15 or future time reference. It is the latter intepretation of the present tense that concerns us here. In English both the present simple (11a) and the present progressive (12a) may have absolute future time reference. (11)

(12)

a. Either we do business with Skolling or we go bankrupt. (OS 94) b. Of we doen zaken met Skolling, of we gaan failliet. (EZ91) Or we do business with Skolling or we go bankrupt 'Either we do business with Skolling or we go bankrupt'. a. I am jacking the job in next year. b. Ik stop volgend jaar met werken. I stop next year with working. Ί am jacking the job in next year'.

In spite of the fact that the present tense is multi-referential and can refer to all times, we believe that locating the situation at the temporal zero-point is the basic function of the present tense in the two languages. The use of the present tense with future or past time reference is a metaphorical extension.16

260

Griet Beheydt

3.1. Factuality The mechanism underlying the use of the present tense with future time reference is the same in Dutch and English. For some reason or other (cf. below), the speaker uses the present tense to locate a particular situation in the (immediate or more remote) future. Declerck (1991b: 24-25) uses the term 'shift of temporal perspective' for English. It is perfectly applicable to Dutch, too. This means that the speaker shifts his temporal perspective from the future to the present, because he wants to represent the future situation as if it were actualizing at the temporal zero-point. It follows that there is only one absolute present tense, viz. the one locating a situation at t0. The present tense with future time reference in sentences like (11) and (12) is a metaphorical extension of this use, because it is the result of a shift of temporal perspective. The main argument in favour of treating the present tense with future time reference as a genuine present tense is a semantic-pragmatic one. The speaker shifts his temporal perspective from the post-present to the present, because there are clear reasons in the linguistic or extralinguistic context for treating the future situation as a present one. This motivation holds for the two languages, and can in very general terms be described as 'representing the situation as if it were a present fact'. In spite of the parallelisms between English and Dutch, the two languages differ considerably with respect to the actual contexts in which the present tense is used. The use of the English present simple with future time reference is restricted to contexts in which the future situation is represented as scheduled and completely predetermined. The situation is seen as completely unalterable and uncontrollable by the speaker, because it is fixed on a calendar (13) or a timetable, because it results from a definite plan or agreement (14), etc. (cf. Declerck 1991a: 91). (13) (14)

And I'm eighty-three in December. (Cain 34) Tomorrow we go over to Triebschen ... where Richard Wagner spent some of his life. (Cain 30)

The difference with the progressive counterpart is not a purely aspectual one: the present progressive is used to represent a future situation as resulting from a present plan or arrangement. The referent of the subject noun phrase 17 is felt to have control over the situation, which is not seen as completely unalterable. The speaker wants to stress that the subject really intends to carry out his plan and therefore he shifts his temporal perspective

Future time reference

261

to the present and, using the progressive, he represents the situation as already ongoing, as having started at the moment of speech. (15) Actually he's publishing his memoirs. bringing them out next spring. (OS 28)

The Peverell

Press

are

As there is reference to a plan or arrangement, this use of the present progressive is only possible with intentional situations. Apart from some well-known exceptions, such as sit, stand, wait, etc., Stative verbs, which are not intentional, do not have progressive forms. The corpus confirms this: all cases of the present progressive in The Daughters of Cain and Original Sin refer to an intentional situation. In (16) to see functions as a dynamic verb, meaning 'to meet'. (16)

What time are you seeing Armstrong-Jones? (Cain 234)

For Dutch, various motivations for the use of the present tense with future time reference have been proposed in the literature on the topic. They are not all clearly distinct from each other and the terminology is not always well-defined. The present tense is sometimes said to be used to refer to the future in order to represent the post-present situation more vividly. Others say that sentences characterized by the semantic feature [+probable] contain a present tense significantly more often than sentences referring to a situation whose future occurrence is more doubtful. One step further than probability is the idea of certainty. The speaker uses the present tense because he sees the situation as inevitable and completely certain. Said differently, the speaker thinks of the future situation as present reality or actuality. Or the present tense is used when the future situation is in some way already relevant at the present moment, etc. We believe that the above descriptions are not mutually exclusive. They can be captured under, or be derived from, a more general motivation: 'representing the situation as if factual, as if actualizing at to'. As we saw, this holds for English too, but the English present tense with future time reference is considerably more restricted than the Dutch one: in Dutch the speaker can shift his perspective more easily. When the speaker of English has a reason (factuality) to shift his temporal perspective to the present, the speaker of Dutch has one, too. The reverse does not necessarily hold. There are far more contexts in Dutch where the speaker can represent the future situation as a factual one, as (17a) and (17b) show.

262

Griet Be heydt

(17)

a. Tomorrow will be rainy and warm. / *Tomorrow is rainy and warm. b. Morgen zal het regenachtig en warm zijn./Morgen is Tomorrow will it rainy and warm be / tomorrow is het regenachtig en warm. it rainy and warm 'Tomorrow will be rainy and warm'.

3.2. Future contextualization A basic condition for the correct use and interpretation of the present tense with future time reference is that it must be clear from the linguistic or extralinguistic context that the situation is to be interpreted as lying in the post-present. Future contextualization is thus a necessary condition for a correct understanding of the present tense in both English and Dutch. The reason is that without indication to the contrary, the present tense is understood as a form locating a situation at t 0 or referring to a habitual situation or general truth (including to), since t 0 is the unmarked and most relevant time to the speaker. Examples (18a-d) show that the present tense may be ambiguous between a present and a future reading. The context has a disambiguating function. (18)

a. She is giving a lecture there. That is why she cannot be disturbed. b. Ze geeft daar een lezing. Daarom kan ze niet gestoord She gives there a lecture therefore can she not disturbed worden. be 'She is giving a lecture there. That is why she cannot be disturbed'. c. She is giving a lecture there. Only the exact hour is still to be fixed. d. Ze geeft daar een lezing. Alleen Staat het exacte uur nog She gives there a lecture only stands the exact hour yet niet vast. not fixed 'She is giving a lecture there. Only the exact hour has not been fixed'.

As a consequence, sentences with present tense forms referring to the future often contain a time adverbial. This may also be explained as

Future time reference

263

follows: the present tense is used in English when the future situation is represented as completely predetermined and fixed or as planned. This presupposes that its time and/or place is already known to the speaker, and that this information is relevant to the hearer. Of course clear future time reference is also possible without an explicit time adverbial in the sentence itself. Various other elements in the linguistic or extralinguistic context may take over the function of a time adverbial (cf. Van Campenhout 1974: 652). By implicature, the unmarked interpretation is the near future when there is no adverbial, as in (19) and (20). (19) How are you getting home, Ellie? (Cain 193) (20) About a month after they had become lovers he had said: "Why don't we get married? " (OS 66) A clear future contextualization is, however, not a sufficient condition. The present tense is not always possible, not even when the sentence is clearly contextualized as referring to the future. This is quite obvious for the English present simple and progressive, as (21a) and (22a) show. This is less obvious in Dutch, where in most cases the present tense is an alternative to the future tense, as the corresponding b-examples show. a. *The sun shines tomorrow. / *The sun is shining tomorrow.18 b. Morgen schijnt de zon. Tomorrow shines the sun 'The sun will shine tomorrow'. (22) a. *In a few weeks the whole construction is coming down. b. Over een paar wehen stört de hele constructie in. In a few weeks comes the whole construction down 'In a few weeks the whole construction will come down'. (21)

Sentences like (23) prove that a clear future contextualization is not the only factor licensing the use of the present tense as an alternative to the future in Dutch either. (23)

*Ben ikzo lelijkals oma, als ik volwassen ben? Am I as ugly as grandma when I grown-up am 'Will I be as ugly as Grandma, when I have grown up?'

Subsequently, the question arises why the latter example is not grammatical, whereas the following example, apparently having the same syntactic structure, is.

264

Griet Beheydt

(24)

Op welk uur ben ikin Leeds als ik de trein van 6.33u At what houram I in Leeds if I the train of 6.33h 'What time will I be in Leeds, when I take the 6.33 train?'

neem? take

The motivation of factuality provides us with an answer. In (23), the situation is so uncertain and unpredictable that one can hardly think of a context in which it would make sense to represent 'being ugly' as a present fact. Shifting the temporal perspective is not functional. In (24) it is functional and even plausible in order to express that the future situation is perceived as absolutely certain, as a present fact.

3.3. Comparing frequencies The corpus confirms that the use of the present tense with future time reference is stricter in English than in Dutch, where it can occur in more contexts. In English, 'factuality' is to be interpreted in a very strict sense. Speakers of Dutch shift the temporal perspective more easily. Table 4 substantiates this claim. Table 4. Percentage of present tenses used to refer to situations having absolute future time reference in The Daughters of Cain and Original Sin and their Dutch translations English - present simple - present progressive Dutch

11.8% - 6.5% - 5.3% 59.9%

Interestingly, English present tense forms can always be translated by Dutch ones. This actually happens in 81.1% of the cases in our parallel corpus (see examples [11] and [12]). The reverse does not hold: only 16.8% of the presents in Dutch are translations of English presents. English uses the future tense in most cases: 59%, as in (25). (25)

a. Een fijne dag en ik zie u morgenochtend weer. (Ka'in 13) A nice day and I see you tomorrow-morning again 'Have a lovely day and I will see you in the morning', b. Have a lovely day and I will see you in the morning. (Cain 10)

Future time reference

265

3.4. Sociolinguistic and regional variation With respect to Dutch, Haeseryn et al. (1997: 120; 980-981) mention some sociolinguistic variation in the use of the present tense with future time reference. This tense is said to be more frequent in spoken language and in an informal register. In written and formal language zullen is more often used. This claim was not corroborated by Colleman's corpora (2000b: 60). He compared journalistic writing and chat language. Table 5 shows that our corpora do not corroborate this claim either. Table 5. Percentage of Dutch present tenses used to refer to situations having absolute future time reference according to register Corpus Spoken Dutch, spontaneous speech Corpus Spoken Dutch, in general Dialogues in detective novels - De dochters van Καϊη - Erfzonde Newspaper: Ηet Nieuwsblad

53.6% 45.3% 60.2% - 50.9% - 66.8% 51.6%

Apart from this variation according to register, Haeseryn et al. make mention of regional variation: the present tense referring to the future is more frequent in the Netherlands than in the Dutch-speaking part of Belgium. De Rooij (1985: 106-107) confirms this claim, but Colleman (2000b: 60) did not find significant differences. In our corpora there seems to be quite a difference between the north and the south, when only spontaneous speech is taken into account (cf. table 6). When other kinds of spoken language are considered too, the difference is less outspoken. Comparing larger corpora of both northern and southern Dutch should give a decisive answer to this question. Table 6. Percentage of Dutch present tenses used to refer to situations having absolute future time reference according to region Regional variation Corpus Spoken Dutch Corpus Spoken Dutch, spontaneous speech

Northern Dutch 47.9% 60.9%

Southern Dutch 42.7% 46.4%

266

Griet

Beheydt

4. Wilt/shall vs. zullen + infinitive The hottest point of discussion with respect to the future tense in Dutch and English is whether there is a future tense at all. Many authors treat will/shall and zullen as modal auxiliaries (cf. for instance Kirsner 1969). We cannot examine this issue in more detail here for reasons of space (for references, cf. note 2), but we would like to claim that these verbs can have both modal and temporal meanings. We are concerned with the temporal ones. 19

4.1. Pure futurity In both English and Dutch the future tense form expresses pure futurity. The speaker merely indicates that he thinks that a situation will take place in the future. This is the basic (purely temporal) meaning of will/shall vs. zullen + infinitive. This claim has an empirical basis: sentences like (26a) and (26b) refer to a pure, neutral future (similar examples are given by Haegeman 1983b:20). (26)

a. At some time I will have to retire. b. Ooit zal ik met pensioen moeten gaan. Ever will I into retirement have-to go 'At some time I will have to retire'.

Modal connotations may arise: obligation, willingness, supposition, hypothesis, uncertainty, promise, etc. These are, however, not inherent. They simply arise because of certain contextual indications that make these connotations plausible. Moreover, future time reference by nature involves some modal connotations, because the future is by definition less certain, as it is yet to come. But that does not mean that one cannot make statements about the temporal location of a future situation, without implying modality. Concluding that will and zullen are modal auxiliaries on the basis of modal connotations is one step too far. The pure futurity interpretation of will/shall vs. zullen makes it the unmarked future tense form, not quantitatively, but as far as 'pure temporal meaning' is concerned. In principle, these tense forms can therefore always be used in future contexts (instead of be going to/gaan or the present tense).20 However, the Dutch example (27) shows that this statement needs some modification.

Future time reference (27)

267

We hebben jullie lets te melden: wij gaan / ?zullen trouwen. We have you something to announce we go / will marry 'We have some news: we are going to / will marry'.

The semantics of some verbs (such as trouwen 'marry', verhuizen 'move', etc.) in combination with a particular context may make the use of gaan more likely than that of zullen (cf. supra). That is why gaan will not so easily be replaced by zullen in (27). The choice for gaan in preference to zullen may have to do with the desire to be more specific (in casu, expressing intentionality and a 'roots in the present' interpretation). When not used in a context that fits in well with the interpretation of gaan, zullen is equally possible and even preferred to gaan. (28)

Later zal ik trouwen met een Later will I marry with a 'Later, I will marry a very rich man'.

hele very

rijke man. rich man

The same holds for English: when there is a specific context promoting one of the alternative future tense forms (e.g. scheduled future), this specific form (i.e. the present simple) is preferably used, not the more neutral form will + infinitive, as example (29) illustrates. (29)

Next week is / *will be my twentieth

birthday.

4.2. Comparing frequencies Table 7 shows that English will/shall is far more frequent that zullen in Dutch when there is reference to the future. This follows directly from the more severe restrictions on the use of be going to and the present tense in English. Table 7. Percentage of future tenses will/shall or zullen + infinitive used to refer to situations having absolute future time reference in The Daughters of Cain and Original Sin and their Dutch translations English will/shall Dutch zullen

73.6% 33.4%

Only 31.7% of the English future tenses are translated by Dutch ones. As we have seen, the present tense is the most frequent one in Dutch. 51.3% of the English future tenses are converted into presents by the translator. As

268

Griet Beheydt

may be expected, Dutch present tenses, on the other hand, are mostly (71.5%) translations of English future tenses.

4.3. Sociolinguistic and regional variation According to the Dutch grammar ANS (Haeseryn et al. 1997: 126), the main difference between the use of the future tense and that of the present tense or gaan + infinitive (which express "dezelfde tijdsverhouding" [the same temporal relation]), is that zullen is especially used in written language. The alternatives gaan + infinitive and the present, on the other hand, are said to be more prominent in spoken language. De Schryver (1979: 335) concludes that zullen is extremely frequent in written language (94%), but this picture is completely distorted, because he only compares the auxiliaries zullen and gaan, leaving the present tense out of consideration. We come to the following (less extreme) picture on the basis of our corpora: zullen is somewhat less frequent in spontaneous, spoken Dutch than in more formal or written Dutch (table 8). Table 8. Frequency of the future tense zullen + infinitive used to refer to situations having absolute future time reference according to register Spoken language (Corpus Spoken Dutch) - spontaneous speech - interviews - meetings - speeches - radio news - library for the blind (dialogues in novels) Dialogues in novels (Καϊη and EZ) Newspaper, Het Nieuwsblad

28% 33.3% 45.4%

17.2% 21.3% 49.5% 40.8% 29% 40.2%

To our knowledge, similar sociolinguistic variation is not mentioned in the literature on English. Further corpus research will have to confirm or refute our intuitions that the present tense replaces the future tense more often in spoken (informal) than in written (formal) language in English as well. There is no significant regional variation in the use of Dutch zullen in the Corpus Spoken Dutch. The percentages for the north and the south are comparable: 28.5% and 27.7% respectively.

Future time reference

269

5. Conclusion Table 9 summarizes the relative frequencies of the various future tense forms discussed in this article on the basis of the above mentioned parallel corpus, which consists of written language, though all future corpus sentences are extracted from the dialogues. Table 9. Relative frequency of the various tense forms with absolute future time reference in The Daughters of Cain and Original Sin and their Dutch translations

Will/shall vs. zullen + inf. - future simple - future progressive Present tense - present simple - present progressive Be going to vs. gaan + inf.

Dutch 33.4% -

-

59.9% -

6.6%

English 77.6% - 73.6% - 4.0% 11.8% - 6.5% - 5.3% 10.6%

In this corpus of written English and Dutch be going to and gaan are relatively infrequent: 10.6% and 6.6% respectively. This is no doubt due to the colloquial nature of this tense form: although most tenses referring to the future are taken from the dialogues in our corpus of written language, the percentage (31.9%) is still significantly higher in the spontaneous language and in the interviews in the Corpus Spoken Dutch. It has been argued that will/shall vs. zullen + infinitive is the unmarked future tense form both in Dutch and in English, with respect to neutral or pure temporal meaning. The percentages in table 9 reveal that, in English, the future tense is the unmarked form in a quantitative sense as well: 73.6% (future simple). In Dutch, however, the present tense (59.9%) is the most common means to establish future time reference.

Notes 1. I would like to thank prof. R. Declerck, the supervisor of my doctoral project, and prof. I. Depraetere for their valuable comments on an earlier draft of this article. The terminology used in the present article is borrowed from Declerck (1991b) and will be explained along the way wherever necessary. 2. Modal auxiliaries and lexical verbal means (e.g. be on the point of be about to, intend to in English and op het punt staan te, van plan zijn te in Dutch) will be disregarded here. In the literature there is no agreement on the grammatical

270

Griet Beheydt

status of will and zullen, which are often called modal auxiliaries. Although we cannot repeat all the arguments here (cf. Niekerk 1972:129-131; Wekker 1974:497; 1976: 18-19; Declerck 1991b: 10-11, and further references there), we adhere to the view that will and zullen can have modal as well as temporal meanings. 3. Relative tenses are outside the scope of the present article. They relate a situation only indirectly to the temporal zero-point, i.e. via one or more intermediate stages; they express a temporal relation with the time of another situation than to. Four relative tenses can have the function of temporally relating the time of a situation to the time of a post-present situation. They are the same as the set of absolute tenses: the preterit and the present perfect for anteriority, the present tense for simultaneity and the future tense for posteriority. The present tenses in the subclauses of (ia) and (ib), for instance, express a simultaneity relation with the time of the head clause situation (referred to by the absolute future tense will come and the absolute present tense komt in English and Dutch respectively). The relation to to is not made explicit, (i) a. The happiness will come ifyou 're lucky. (OS 43) b. Als je boft, komt hetgeluk vanzelf. (EZ 46) If you are-lucky comes the happiness by itself 'If you are lucky, the happiness will come by itself.' 4. The abbreviations within brackets indicate the corpus in which the example was attested. More information on the corpora used is given at the end of the introduction. 5. Allative means "expressing the meaning of motion 'to' or 'towards' a place" (Crystal 1997: 15). 6. Actually, there is quite some agreement in the literature on this point (cf. Binnick 1971: 48, 1972: 8; Declerck 1991b: 370; Mcintosh 1966: 303; Haegeman 1989: 291; 292-293). We will use the term "meaning" with regard to semantics and "interpretation" with regard to pragmatics. 7. The corpus examples in this article are labelled by means of the abbreviations within brackets, followed by the page number. 8. Van Bree (1997), Decroos (1997, 2000), Colleman (2000a, 2000b), De Rooij (1985, 1986), Abraham (1989) and Ten Cate (1991) all start from (part of) the description given in the ANS (Haeseryn et al. 1997 or previous editions). 9. De Rooij (1986: 22), Abraham (1989: 370-371) and Colleman (2000b: 62-63) give similar examples. 10. Declerck's statement that be going to is seldom possible with stative verbs (Declerck 1991a: 114) is somewhat in contradiction with the examples he gives, especially the ones containing stative be (That musical is going to be a great success). 11. Cf. Van Bree (1997: 75) for a similar analysis. 12. Cf. Haeseryn et al. (1997: 978), Brisau (1975: 24), De Rooij (1986: 5), Decroos (1997: 15, 2000: 115), Van Bree (1997: 74), Colleman (2000b: 60-61). 13. De Schryver (1979: 192-193) comes to the opposite conclusion. The author himself explains this unexpected result by the fact that this (endogenous

Future time reference

271

Flemish) form is often stigmatized as an incorrect Belgianism, which should be avoided, esp. in written language. His corpus only consists of written journalistic prose. 14. In (ia) and (ib) the present tenses in the main clause have the value of the present perfect Ik heb gehoord/I've heard. (i) a. I hear that your daughter is pregnant again. b. Ik hoor dat je dochter weer zwanger is. I hear that your daughter again pregnant is Ί hear that you daughter is pregnant again.' 15. In historical present contexts, the present tense is used to refer to a situation that is completely over, for instance in narrative. (i) a. I was walking along in Hyde Park, when suddenly there comes a man. b. Ik was in Hyde Park aan het wandelen.wanneer plots een man I was in Hyde Park on-the-walking when suddenly a man afkomt. comes. Ί was walking along in Hyde Park, when suddenly there comes a man.' 16. Apart from this basic function, the present tense with future reference is also used as a relative tense expressing simultaneity with the time of another future situation. This is for instance the use of the present tense in - amongst others conditional and temporal clauses and sometimes in restrictive relative clauses (cf. also footnote 3). 17. Sporadically, it is not the referent of the subject noun phrase who made the arrangement. It may sometimes be an (implied) human agent, as in Examinations are beginning next week (Declerck 1991a: 93). 18.Bybee, Perkins, and Pagliuca (1994: 251) give a similar example for English: *It rains tomorrow. 19. For reasons of space, the future progressive tense cannot be treated here. It is, however, a very interesting topic with regard to the comparison between English and Dutch, since this tense form does not exist in the latter language. The difference with the future simple in English is more than purely aspectual. The future progressive may refer to a pure future, often with the connotation that the future situation is expected, because it is (or is expected to be) part of a routine or of the normal pattern of events. In Dutch the present tense (53.3%) or the future tense (33.3%) are mostly used in this context. 20. Some linguists argue that it is the other way around: in their view, the present tense is the unmarked form to refer to the future, whereas the use of zullen needs to be motivated, because it expresses additional shades of meaning. It is for instance claimed that, when used in an unambiguously future context, zullen is more emphatic than the "neutral" present tense, or conveys a modal reading (Ten Cate 1991: 29). A similar idea of zullen being more emphatic than the present tense is found in the ANS (Haeseryn et al. 1997: 120-121). Cf. also Droste (1956: 42).

272

Griet Beheydt

References Abraham, Werner 1989 Futur-typologie in den germanischen Sprachen. In Tempus Aspekt - Modus: Die lexikalischen und grammatischen Formen in den germanischen Sprachen, Werner Abraham, and Theo Janssen (eds.), 133-154. (Linguistische Arbeiten 237) Tübingen: Niemeyer. Binnick, Robert 1971 Will and be going to. Papers from the Seventh Regional Meeting. Chicago Linguistic Society, 40-52. 1972 Will and be going to II. Papers from the Eighth Regional Meeting. Chicago Linguistic Society, 3-9. Brisard, Frank 2001 Be going to: an exercise in grounding. Journal of Linguistics 37 (2): 251-285. Brisau, Andre 1975 A Note on going to and Netherlandic gaan. Taal, Taalkunde en Vertaalkunde [Language, Linguistics and Translation Studies]: 1924. Bybee, Joan L., Revere D. Perkins, and William Pagliuca 1994 The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago: The University of Chicago Press. Close, R.A. 1977 Some observations on the meaning and function of verb-phrases having future reference. In Studies in English Usage: The Resources of a Present-Day English Corpus for Linguistic Analysis, WolfDietrich Bald, and Robert Ilson (eds.), 125-156. (Forum Linguisticum 6.) Frankfurt: Lang. Colleman, Timothy 2000a De toekomstaanduiders van het Nederlands [Future markers in Dutch]. Nederlands van Nu 48 (5): 86-88. 2000b Zullen, gaan of presens. Een verkennend corpusonderzoek naar de toekomstaanduiders in het (Belgische) Nederlands [Zullen, gaan or the present. A preliminary study of the future markers in (Belgian) Dutch]. In Nochtans was Scherp van Zin. Festschrift Hugo Ryckeboer Veronique De Tier, Magda Devos, and Jacques Van Keymeulen (eds.), 51-64. Gent: University of Gent. Crystal, David 1997 A Dictionary of Linguistics and Phonetics. Oxford: Blackwell Publishers. Declerck, Renaat 1991a A Comprehensive Descriptive Grammar of English. Tokyo: Kaitakusha. 1991b Tense in English: Its Structure and Use in Discourse. London: Routledge.

Future time reference

273

Decroos, Bram 1997 Wat is er met gaan aan de hand in het zuidwestelijke deel van ons taalgebied? [What is the matter with gaan in the south-western part of our language area?]. Unpublished paper, Department of Linguistics, University of Leuven. 2000 Wat is er met gaan aan de hand...? (een aanvulling op Van Bree 1997) [What is the matter with gaan...? (an addition to Van Bree 1997)]. In Nochtans was Scherp van Zin. Festschrift Hugo Ryckeboer, Veronique De Tier, Magda Devos, and Jacques Van Keymeulen (eds.), 111-116. Gent: University of Gent. De Rooij, J. 1985 De toekomst in het Nederlands I [The future in Dutch I] Τ aal en Tongval 37: 96-123. 1986 De toekomst in het Nederlands II [The future in Dutch II] Taal en Tongval 38: 5-32. De Schryver, Johan 1979 Het gebruik van de hulpwerkwoorden gaan and zullen in hedendaags Nederlands joumalistiek proza [The use of the auxiliaries gaan and zullen in present-day Dutch journalistic prose]. Unpublished lie. thesis, Department of Linguistics, University of Gent. Droste, Frederik G. 1956 Moeten, een Structured Semantische Studie [Moeten, a structural semantic study]. Groningen: J.B. Wolters. Fleischman, Suzanne 1982 The Future in Thought and Language: Diachronic Evidence from Romance. Cambridge: Cambridge University Press. Haeseryn, Walter, K. Romijn, Guido Geerts, J. De Rooij, and M.C. van den Toorn 1997 Algemene Nederlandse Spraakkunst [Comprehensive Grammar of Dutch], Groningen: Wolters Noordhoff. Haegeman, Liliane 1983a Be going to, gaan, and aller, some observations on the expression of future time. International Review of Applied Linguistics in Language Teaching 21: 155-157. 1983b The Semantics of Will in Present-day British English. Brüssel: Koninklijke Academie voor Wetenschappen, Letteren en Schone Künsten van Belgie. 1989 Be going to and will: a pragmatic account. Journal of English Linguistics 25: 291-317. Kirsner, Robert S. 1969 The role of zullen in the grammar of modern standard Dutch. Lingua 24: 101-154. Leech, Geoffrey N. 1971 Meaning and the English Verb. Longman: London.

274

Griet Beheydt

Mcintosh, Angus 1966 Predictive statements. In In Memory of J.R. Firth, J.R. Firth, C.E. Bazell, and J.C. Catford (eds.), 303-320. London: Longmans. Niekerk, P.K. 1972 L 'expression du Futur en Frangais et en Neerlandais. Etude Synchronique sur les Syntagmes Verbaux Susceptibles d'exprimer la Futurite. VRB: Groningen. Palmer, Frank R. 1979 Modality and the English Modals. London: Longman. Swan, Michael 1998 Practical English Usage. Oxford: Oxford University Press. Ten Cate, Abraham P. 1991 Bemerkungen zum deutschen und niederländischen Futur. In Betriebslinguistik und Linguistikbetrieb, Linguistisches Kolloquium 24, Eberhard Klein (ed.), 23-31. (Linguistische Arbeiten 260-261.) Tübingen: Niemeyer. Van Brederode, Tom, and Willem Koopman 1990 A key to Better English 1: An Advanced English Grammar for Dutch Speakers. Muiderberg: Coutinho. Van Bree, Cor 1997 Wat is er met gaan aan de hand [What is the matter with gaan]. In Leven in de Oudgermanistiek [Life in Old Germanistics], Han Nijdam, M.L. Gerla, and Klazina van Dalen-Oskam (eds.), 72-77. Leiden: Vereniging van Oudgermanisten. Van Campenhout, F. 1974 Zinnen met zullen en zinnen zonder zullen met futurische betekenis [Sentences with zullen and sentences without zullen with future meaning]. Tijdschrift voor Levende Talen/Revue de Langues Vivante 40 (6): 650-670. Van Es, G.A., and P.Ρ J . Van Caspel 1974 Syntaxis van het Moderne Nederlands [Syntax of Modern Dutch] I (41) 'De verbale groepsvorming in de zin V' [Verbal grouping in the sentence]. Groningen: University of Groningen. Wekker, Herman C. 1974 Enkele suggesties voor de syntaktische analyse van het hulpwerkwoord zullen in het Moderne Nederlands [Some suggestions for a syntactic analysis of the auxiliary zullen in modern Dutch], De Nieuwe Taalgids 67: 493-499. 1976 The Expression of Future Time in Contemporary British English. Amsterdam: North-Holland.

Cleft constructions in French and Spanish1 Katleen Van den Steen

Introduction Although the phenomenon of cleft constructions2 exists in many languages, I want to restrict myself in this paper to the constructions used in French and Spanish. In French we have cleft sentences in the narrow sense of the word (1) ("les phrases clivees") and pseudo-cleft sentences (2) ("les phrases pseudo-clivees"). The Spanish language for its part presents clefts (3), pseudo-clefts (4) and inverted pseudo-clefts (5) ("las hendidas, pseudohendidas y pseudo-hendidas inversas"). The sentences in (1) to (5) exemplify the five constructions: (1)

C'est le travail qui est important. 'It is work that is important.'

(2)

Ce qui est important c'est le travail. 'What is important is work.'

(3)

Es el trabajo lo que es importante. 'It is work that is important.'

(4)

Lo que es importante es el trabajo. 'What is important is work.'

(5)

El trabajo es lo que es importante. 'Work is what is important.'

In the first section of this paper, I will briefly summarise the different theoretical descriptions which have been proposed for these constructions: they range from syntactic, strictly formal to pragmatic approaches. I will also compare the Spanish cleft constructions to their French equivalents. The second section is devoted to the translation of French clefts into Spanish. The point of departure of my study was the impression that these kinds of sentences are used more often in French than in Spanish, and,

276

Katleen Van den Steen

more importantly, that these constructions may sound rather "bad" to Spanish-speaking people. I verified this intuition with a quantitative analysis and I investigated a number of factors susceptible of facilitating or hampering the use of a cleft construction in Spanish.

1. Definitions In this section, I discuss the French cleft sentences (in the narrow sense of the word), the French pseudo-cleft sentences and the Spanish cleft sentences.

1.1. The cleft construction in French A first point to make is that the cleft construction extracts a constituent of the sentence. In example (6), one can extract the group de lä 'from there', move it to the beginning of the sentence, enclose it with c'est 'it is' on the left and a relative pronoun on the right, thus obtaining the cleft (7). (6)

Et leur eternelle jeunesse, leur inhumaine beaute viennent de lä. 'And their eternal youth, their inhuman beauty come from there.'

(7)

Et c'est de lä que viennent leur eternelle jeunesse, leur inhumaine beaute. (Tournier 1972: 232) 'And it is from there that come their eternal youth, their inhuman beauty.'

Most grammarians do not give a functional definition of the "phrase clivee", they just describe the structure of such sentences. According to Moreau (1976: 14) for instance, a cleft sentence corresponds to [C'est Κ QU R], where Κ is a variable in the context C'est/ {de lä in the example (7)) QU is a relative pronoun or a conjunction {que), and R is a sentence amputated of one of its syntagms {viennent leur iternelle jeunesse, leur inhumaine beaute), like in relative clauses. Besides, it is possible to combine Κ and R in one sentence (cf. (6)). The cleft structure allows to focalise an element, giving it the guarantee and the exclusivity of the forefront. The extracted element is the rheme of the sentence, the focus, the topic3; and the effect of the structure is what is called in French "la mise en relief'.

Cleft constructions in French and Spanish

277

According to Vikner (1973), the use of a cleft sentence supposes the existence of two different and not empty entities: a positive entity and a negative one. In example (8), le chapitre XXXIX de la Genese constitutes the positive entity: it is the case that chapter XXXIX of genesis resounded this time through the voice of Robinson. At the same time, there are other things (other chapters, other books) -constituting the realm of the negative entity- that did NOT resound through his voice. (8)

C 'est le chapitre XXXIX de la Genese qui retentit cette fois par la voix de Robinson. (Tournier 1972: 178) 'It is chapter XXXIX of genesis that resounds this time through the voice of Robinson.'

1.2. The pseudo-cleft construction in French To my knowledge, the French pseudo-cleft construction has not been the object of thorough investigation yet. Some grammarians do not even mention it. This can probably be attributed to the low frequency of the construction: only 12,2% of the cleft sentences (in the broadest sense of the word) of my corpus are pseudo-clefts. Many grammarians take the pseudo-cleft construction to be the inversion of the cleft construction. The structure of the pseudo-cleft is indeed the same as the cleft, provided that the order is changed —the focalised element Κ is now located at the end of the sentence— and provided the morpheme ce 'it' is added at the beginning: Ce QUR, c 'est K. However, the mechanism is not that simple: there are more restrictions for the pseudo-cleft construction. For instance, the combination of Κ and R holding in one sentence is not always possible and only two types of constituents can be extracted (the element K) by the pseudo-cleft construction: the grammatical subject and the direct object. On the semantic level, the relation between the two parts of the pseudocleft sentence can be described in terms of specification (cf. Valli 1981: 204). First, something is presupposed (cf. example (9): "there is something that has changed the most"), but the meaning remains vague until the specification introduced by the second element ("c'est l'ecoulement du temps, sa vitesse et meme son orientation"). At the same time, this meaning contrasts with other possible referents. In this case, it can be la solitude 'the solitude' ou I'absence de luxe 'the absence of luxury', but that depends on the context, of course.

278

Katleen Van den Steen

(9)

Ce qui a le plus change dans ma vie, c 'est I 'icoulement du temps, sa vitesse et meme son orientation. (Tournier 1972: 218) 'What has changed most in my life is the course of the time, its fastness and even its orientation.'

1.3. The cleft constructions in Spanish Just like the French ones, the Spanish cleft constructions consist of three parts: the emphasised entity (the "focus"), a form of the verb ser 'to be' and a relative clause. The usual classification depends on the order of the three constituents of the construction 4 . The distinction is threefold: (i) If the focus is situated between the verb ser 'to be' and the relative clause, we have a cleft sentence. (ii) If the emphasised element is situated at the end of the sentence, we have a pseudo-cleft sentence. (iii) If it is situated at the beginning of the sentence, we have an inverted pseudo-cleft sentence. Sentences (10) to (12) exemplify these definitions. (10)

Era en ese momento cuando nacia en έΐ un hombre nuevo, completamente ajeno al administrador. (Tournier 1992a: 134) 'It was at that moment when a new man in him was born, totally strange to the administrator.'

(11)

Lo que huyo era un cuerpo sin alma, un cuerpo ciego, como esos patos que se salvan batiendo las alas despues de que se les ha cortado la cabeza. (Tournier 1992a: 157) 'What flew there was a body without a mind, a blind body, like those ducks who run away flapping their wings after their head has been cut.'

(12)

La mujer es quien cierra las puertas de una casa. (Tournier 1992b: 102) 'The woman is who closes the doors of a house.'

From the semantic point of view, the cleft constructions of Spanish play the same role as the French ones: they yield a structure of emphasis (cf.

Cleft constructions in French and Spanish

279

Martinez 1994: 44—49.). Their function is to put forward a particular linguistic entity, which contrasts with other possible entities. Broadly speaking, Spanish clefts behave like their French counterparts. However, there are also some differences. I mention the most important ones: (i) In the Spanish clefts almost any relative pronoun can be used, as opposed to the French sentences where only qui 'who' and que 'which' can appear. (13)

a. C'est alors que nous I'avons decouvert, notre unique mendiant de ce Noel ä Calcutta. b. Fue entonces cuando descubrimos nuestro ünico mendigo de aquella Nochebuena en Calcutta. (Tournier 1992b: 162) 'It was then when we discovered our only beggar of that Christmas Eve in Calcutta.'

However, it is worth mentioning that instead of the expected relative the use of the relative que —like in French— becomes more and more frequent. This tendency is known as a grammatical "gallicism". (ii) If the extracted element begins with a preposition, this preposition can be repeated (14c), which is not possible in French (14a-b). (14)

a. C'est dans cette embarcation [0] que Robinson et son compagnon regagnerent I 'ile comme le soir tombait. (Tournier 1972: 247) b. *C'est dans cette embarcation dans que Robinson et son compagnon regagnerent I 'ile comme le soir tombait. c. Fue en aquella embarcacion en la que Robinson y su companero regresaron en la isla al caer el sol. (Tournier 1992a: 258) 'It was in that vessel that Robinson and his companion returned to the island around nightfall.'

(iii) In Spanish there is no invariable neuter pronoun which occupies the subject position of the construction, such as the French ce 'it'. As a consequence, the copula can come in different persons5 (15b). (15)

a. C'est/*Ce suis moi qui suis responsable. b. *Es/Soyyo el que es responsable. 'It is I who am responsible.'

280

Katleen Van den Steen

(iv) The verb of the relative clause concords with the extracted subject in French (16a), but not in Spanish (16b). (16)

a. (...) c'est moi tout entier qui agis etpense en lui. (Tournier 1972: 154) b. (...) soy yo enteramente quien actüa y piensa a traves suyo. (Tournier 1992a: 163) 'It is I who act and think in him.'

(v) In Spanish, the temporal concord of c'est 'it is' with the rest of the sentence is obligatory, whereas the concord in French often is absent. (17)

a. Ce η'est pas ä mon miroir que j'ai pose la question, c'est ä ma mere. (Tournier 1989: 15) b. No fue a mi espejo a quien le plantei la pregunta, fue a mi madre. Tournier 1992b: 13) 'It was not to my mirror I asked the question, it was to my mother.'

In spite of these differences, it is beyond doubt that the French and Spanish constructions are equivalent when it comes to their meaning structure. But that doesn't mean that French clefts are translated automatically by a Spanish one.

2. Translation of French cleft sentences into Spanish First I want to discuss how cleft and pseudo-cleft sentences can be translated from French into Spanish. Subsequently I will give an overview of the factors that affect the translation of cleft sentences.

2.1. The translation possibilities The pseudo-cleft sentences of my French corpus are translated nearly systematically by a Spanish pseudo-cleft6. The translation of the clefts in the narrow sense of the word, on the other hand, turns out to be much more heterogeneous. The typology presents the following seven translation categories:

Cleft constructions

in French and Spanish

281

(i) Not even half of them (42,2%) are translated by a Spanish cleft sentence. (18)

a. C'est alors que le soleil Ιαηςα ses premieres fleches. (Tournier 1972: 253) b. Y fue entonces cuando el sol Ιαηζό sus primer os dar dos. (Tournier 1992a: 266) 'It was then when the sun dropped its first rays.'

(ii) 4,6% of the French clefts are translated by a Spanish sentence. (19)

pseudo-cleft

a. Mais c'etait surtout l'amenagement des communs qui m'excitait. (Tournier 1989: 73) b. Pero lo que mäs me excitaba era el arreglo de las dependencias. (Tournier 1992b: 58) 'But it was particularly the arrangement of the goods what excited me.'

(iii) The inverted pseudo-cleft construction represents 8,1% of the translations. This means that only 54,9% of the French clefts are translated by one of the cleft constructions in Spanish. (20)

a. C'est la presence de ce frere qui a incline ma vie. (Tournier 1989: 9) b. La presencia de este hermano fue lo que condiciono mi vida. (Tournier 1992b: 9) 'It is the presence of my brother that has set my life.'

(iv) In a quarter of the cases, the translators change word order. A first strategy, used in 4% of the cases, consists in postponing the grammatical subject. That way the subject is presented as the new element, the rheme, and the effect is the same as what Tournier expressed through the cleft construction. (21)

a. Une ere nouvelle dibutait pour lui — ou plus precisement, c'etait sa vraie vie dans I 'lie qui commengait apres des defaillances dont il avait honte et qu 'il s 'efforgait d'oublier. (Tournier 1972: 45) b. Una era nueva comenzaba para el —o, mäs exactamente, comenzaba su verdadera vida en la isla despues de la etapa de

282

Katleen Van den Steen debilidad que ahora le producta vergüenza y se esforzaba por olvidar—. (Toumier 1992a: 53) Ά new era began for him —or, more exactly, it was his real life in the island that started after a period of debility which he was ashamed of and which he tried to forget—.'

(v) Another possibility that has to do with the order of the syntagms is the following one: a syntagm (excepting the grammatical subject) is placed before the verbal predicate because it receives more communicative "weight" in initial position. This occurs in 21,4% of the translations. (22)

a. C'est comme cela qu'il m 'a eue. (Tournier 1989: 16) b. Asi me consiguio. (Tournier 1992b: 14) 'It was so that he obtained me.'

(vi) Sometimes the translator uses another construction (5,8%); these are marginal "sui generis" translations. (23)

a. Car si le spectacle de ces brutes dechainees accaparait toute son attention, ce η 'etaient ni les arbres stupidement mutiles ni les betes massacrees au hasard qui le retenaient, c'etait le comportement de ces hommes, ses semblables, a la fois sifamilier et si etrange. (Tournier 1972: 237) b. Porque si el espectäculo de aquellos hombres desenfrenados acaparaba su atencion, no era porque le preocuparon los ärboles estüpidamente mutilados ο los animales masacrados sin ton ni son, sino por el comportamiento de aquellos hombres, sus semejantes, a la vez tan familiar y tan extrano. (Tournier 1992a: 248) 'Because if the spectacle of those unrestrained brutes absorbed all his attention, it were neither the stupidly mutilated trees nor the animals massacred at random which worried him, it was the behaviour of those men, his congeners, at the same time so familiar and so strange.'

(vii)

In the remaining cases (13,9%), the cleft construction is not translated.

(24)

a. Ce serait egalement a Plozevet qu'on les marierait. (Tournier 1989: 279) b. Ytambien se casarian en Plozevet. (Tournier 1992b: 220) 'It would also be in Plozevet where they would marry.'

Cleft constructions in French and Spanish

283

Upon closer examination of these different treatments, the choice does not seem to be haphazard and different factors seem to play a role. In order to find out whether there exists some systematicity and how the different tendencies interact with one another, a parametrized analysis has been applied to my contrastive corpus.

2.2. Factors affecting the translation of cleft sentences Nine formal parameters were applied to the sentences of the French corpus: the grammatical function and the category of the extracted constituent, the information done by this element and the contrast created by the cleft construction, the type of focalisation markers7, the syntactic status (principal or subordinate clause) and the modality of the sentence (affirmative, negative, interrogative or negative-interrogative sentences), and the concord in tense and in number of c'est 'it is'. The analysis shows that some of them do not have a clear impact on the translation of the clefts (in the narrow sense of the word), or that their impact could not be measured because of the disproportionate distribution of the sentences over the different factors. In what follows, I am only dealing with the parameters that do have a clear influence. The parameters that turned out to be too disproportionate to measure their impact are the following ones: the focalisation markers, the concordance in number of c'est 'it is', the syntactic status (principal or subordinate clause) and the modality of the sentence. I would like to emphasize that I am not describing "absolute rules", I am speaking of tendencies. (i) Two parameters that certainly have an impact on the type of translation of the cleft sentence are the grammatical function and the category of the extracted constituent K. Let me start with the function of Κ. I distinguished seven factors in this parameter: grammatical subject, direct object, indirect object, supplement8, adverbial adjunct, attribute and complement of the predicative adjective. As mentioned in section 1.2, only the two first functions are possible with a pseudo-cleft sentence. The principal results can be summarized as follows: clefts in which the grammatical subject or the direct object is extracted, are translated proportionally more often by a cleft (25) or a pseudo-cleft sentence (26). Besides, for the first ones, lineal readjustment, viz., postposition of the grammatical subject, is also a possible translation (27).

284

Katleen Van den Steen

(25)

a. C'est apparemment un plaisir ego'iste que poursuivent les amants (...). (Tournier 1972: 131) b. Aparentemente es un placer egoista el que persiguen los amantes (...). (Tournier 1992a: 140) 'Apparently it is an egoistic pleasure that lovers pursue (...).'

(26)

a. Mais c 'etait surtout I 'amenagement des communs qui m 'excitait. (Tournier 1989: 73) b. Pero lo que mäs me excitaba era el arreglo de las dependencias. (Tournier 1992b: 58) 'But it was particularly the arrangement of the goods which excited me.'

(27)

a. A ce moment, c'est toute Valienation de la condition feminine qui leur est tombee sur les epaules. (Tournier 1989: 14) b. En ese momento les cae encima toda la alienacion de la condicion femenina. (Tournier 1992b: 13) 'At that moment it is the whole alienation of the feminine condition that falls on them.'

The clefts which extract an adverbial adjunct are the sentences that are translated proportionally the least by a cleft or pseudo-cleft. Conversely, the translation by an inverted pseudo-cleft sentence (28) and the collocation of Κ in initial position (29) are frequent with this type of clefts. (28)

a. C'est la quej'ai rencontre Oudalle. (Tournier 1989: 17) b. Allifue donde conoci a Oudalle. (Tournier 1992b: 15) 'It was there where I met Oudalle.'

(29)

a. C'est alors qu'eut lieu une apparition fantastique dont ma vie reste marquee ä tout jamais. (Tournier 1989: 169) b. Entonces tuvo lugar una aparicion fantdstica que ha marcado mi vida para siempre. (Tournier 1992b: 132) 'It was then when a fantastic apparition which has marked my life for ever took place.'

(ii) The grammatical category of Κ is another parameter which influences the translation of the cleft construction. However, it has to be pointed out that this parameter is often related to the previous parameter, the function of K. There are nine possibilities: nominal group, proper name, pronoun,

Cleft constructions

in French and Spanish

285

prepositional group, gerund, infinitive, past participle, subordinate clause and adverb. I summarize the most important results. The clefts that extract a nominal group, a proper name or a pronoun are more often translated by one of the Spanish cleft constructions (30). They are also the only ones that can be translated by means of lineal readjustment (31), which is logical, since Κ has to be the grammatical object of the sentence (and as a result Κ can not be a prepositional group, for instance). (30)

(31)

a. Ce fut Nicolas qui me surprendit dans I'atelier. 1989:162) b. Fue Nicoläs quien me sorprendio en el taller. 1992b: 127) 'It was Nicolas who surprised me in the studio.'

(Tournier (Tournier

a. C 'est lui qui t 'envoie ? (Tournier 1989:142) b. iTe manda el? (Tournier 1992b: 115) 'Is it he who sends you?'

If the extracted element is a verbal form, it is put in clause initial position (32). (32)

a. J'ai lu et relu ces lignes, et c'est en les recitant encore que je fus me coucher. (Tournier 1972: 168) b. He leido y releido estas lineas y recitdndolas todavia fui a acostarme. (Tournier 1992a: 179) Ί read and reread these lines and I was still reciting them when I went to bed.'

The fact that Κ is an adverb favours both the translation by an inverted pseudo-cleft sentence (33) and the translation with Κ in clause initial position (34). (33)

a. Mais c 'est la que tout a change. (Tournier 1989: 66) b. Pero ahifue donde cambio todo. (Tournier 1992b: 53) 'But is was there where everything changed.'

(34)

a. C'est alors que des coups timides frappes ä la porte reveillent la Colombine vivante. (Tournier 1989: 274) b. Entonces unos timidos golpes en la puerta despiertan a la Colombina viva. (Tournier 1992b: 216)

286

Katleen Van den Steen 'It is then when a few timid knocks on the door wake up the lively Colombine.'

(iii) Concerning the informational structure of the construction, there is one conspicuous result: the clefts in which Κ transmits already mentioned information9 bear a large proportion of translations by one of the cleft constructions. (35) a. Ces silences etait combles par le bavardage du second, Joseph, tout entier tourne vers la vie pratique et les progres techniques de la navigation, et qui eprouvait visiblement ä l'egard de son superieur une admiration renforcee par I'incomprehension la plus totale. Apres le dejeuner, ce fut lui qui entraina Robinson sur la passerelle, tandis que le commandant se retirait dans sa cabine. (Tournier 1972: 244) b. Aquellos silencios eran llenados por el parloteo del segundo, Joseph, volcado completamente a la vida präctica y α los progresos tecnicos de la navegacion, y que experimentaba visiblemente con respecto a su superior una admiracion reforzada por la mäs total incomprension. Al terminar el almuerzo fue έΐ quien condujo a Robinson a la cabina de mandos, mientras que el capitän se retiraba a su camarote. (Tournier 1992a: 256) 'These silences were filled by the chatter of the second officer, Joseph, who was strongly inclined towards the practical life and the technical progress of navigation, and who obviously held his superior in the highest regard, increased by utter incomprehension. After lunch it was he who led Robinson to the bridge, while the captain retired to his cabin.' (iv) The parameter designed to measure the type of contrast created by the cleft construction, leads to the conclusion that it especially influences two types of translations. If the contrast is explicit, it seems to be impossible to put Κ in clause initial position. The absence of translation, on the other hand, is very frequent with these sentences. (36) a. Mais tandis que la souille me faisait hanter principalement ma soeur Lucy, etre ephemere et tendre —morbide en un mot—, c 'est ä la haute et severe figure de ma mere que me voue la grotte. (Tournier 1972: 111) b. Pero mientras que la cienaga me hacia obsesionarme con mi hermana Lucy, ser tierno y efimero —morbido, en una palabra—, la

Cleft constructions

in French and Spanish

287

gruta me lleva hacia la figura elevada y severa de mi madre. (Tournier 1992a: 120) 'But whereas the quagmire made me principally be haunted by my sister Lucy, an ephemeral and weak being —morbid in one word—, the cave destines me to the elevated and strict figure of my mother.' (v) If the cleft construction is used in a past or future context, c'est 'it is' need not concord in tense. The results of the analysis show that the sentences without temporal concord are less often translated by one of the cleft constructions. The absence of temporal concord seems to facilitate the translation by anteposition of K. (37)

a. Ce η'est qu'avec le recul et la reflexion que sa complexite m'apparut. (Tournier 1989: 81) b. Tan solo con la distancia y la reflexion aparecio toda su complexidad. (Tournier 1992b: 65) O n l y by the distance and the reflection, its complexity appeared.'

After the analysis of the cleft sentences in the narrow sense of the word, the hypotheses were checked in two ways. First, the characteristics of the pseudo-clefts of the corpus were submitted to a close scrutiny. It turned out that many results were corroborated by the properties of the French pseudocleft construction: the factors that seem to favour the translation of a French cleft by a Spanish pseudo-cleft are, indeed, also well represented in the French pseudo-clefts. The second way to check some of the hypotheses was a kind of poll with native speakers of Spanish. They received a list of cleft, pseudo-cleft and inverted pseudo-cleft sentences and had to judge the degree of acceptability of the sentences. On the whole, the results of this poll confirmed my hypothesis.

Conclusion This study not only confirms that in Spanish cleft and pseudo-cleft constructions are less frequently used than in French; it also reveals certain tendencies in the mechanisms of translating French cleft constructions into Spanish. Only in a bit more than half of the cleft sentences of my corpus, the translator uses a Spanish cleft construction. In the other cases, he appeals to the order of syntagms, uses another construction or does not translate the construction.

288

Katleen Van den Steen

My quantitative analysis has allowed me to sort out the parameters that have an impact on the translation: the grammatical function and the category of the extracted constituent, the informational structure of the construction, the type of contrast and the temporal concord of c'est 'it is'. The grammatical function of the extracted element can clearly be retained as the most important parameter: indeed, many of the other observations can be explained by this parameter.

Notes 1. This paper is based on my unpublished MA thesis, which I wrote under the guidance of Prof. Dr. N. Delbecque and presented in June 2000 (Van den Steen, 2000). For the empirical part of this study I used a French corpus —two novels by Michel Tournier— and its Spanish translation (Tournier 1969, Tournier 1989, 1992a, 1992b). The French version of this corpus contains 173 cleft and 24 pseudo-cleft sentences. I am grateful to Prof. Dr. Nicole Delbecque and to Peter De Groof for revising this paper. 2. In Spanish the term construcciones ecuacionales includes the different constructions, but since the English equivalent equational constructions is not commonly used with this sense, I prefer not to use it. Instead, the term cleft will be used either to refer to the different kinds of clefts in general, or more specifically, to the cleft in the narrow sense of the word, e.g. (1) and (3). In case of ambiguity, I specify which sense is the intended one. 3. The notion of topic is used here as defined by Lambrecht (1994: 117-131). 4. Different linguists use this distinction: e.g. Alcina Franch and Blecua (1975: 1136), Guitart (1989: 133), Moreno Cabrera (1999: 4251) and Sedano (1990: 14; 1995: 149). 5. In certain cases, the concord is not obligatory. See Moreno Cabrera (1999: 4255) for the conditions. 6. Indeed, 23 of the 24 cases are translated by a pseudo-cleft sentence, i.e. 95,8%. 7. I examinated the impact of three focalisation markers types: metalinguistic (e.g. precisement 'precisely', tout entier 'entirely'), temporal (e.g. presque toujours 'nearly always', surtout 'especially') and epistemic ones (e.g. apparemment ' apparently', peut-etre ' mayby'). 8. The term supplement is understood to mean a prepositional complement whose presence and preposition are selected by the verb itself. C'est de cette comparution que dependent le räengagement de tout l'equipage et le sien. (Tournier 1989 : 12) 'It is on this appearance that the reemployment of the whole crew and of him depends.'

Cleft constructions in French and Spanish

289

9. The information is considered as "already mentioned" if Κ is a pronoun, designates one of the interlocutors, or if it repeats what has been said in the previous context (either literally or anaphorically).

References Alcina Franch, Juan and Jose Manuel Blecua 1975 Gramatica espanola. Barcelona: Ariel. Guitart, Jorge M. 1989 On Spanish cleft sentences. In Studies in Romance Linguistics. Selected Papers from the Seventeenth Linguistic Symposium on Romance Languages (XVII. LSRL) Rutgers University, 17-29 March 1987\ Carl Kirschner and Janet Decesaris (eds), 129-137. Amsterdam: Benjamins. Martinez, Jose A. 1994 Cuestiones marginadas de gramätica espanola. Madrid: ISTMO. Moreau, Marie-Louise 1976 C'est. Etude de syntaxe transformationnelle. Universite de Möns. Moreno Cabrera, J.C. 1983 Las perifrasis del relativo. VVAA, Sita Philologica F. Läzaro Carreter: 455-467. Madrid: Catedra. 1999 Las funciones informativas: las perifrasis de relativo y otras construcciones perifrasticas. In: Gramätica descriptiva de la Lengua Espanola vol. 3; Ignacio Bosque and Violeta Demonte (eds.), 42454302. Madrid: Espasa. Sedano, Mercedes 1990 Hendidas y otras construcciones con ser en el habla de Caracas. Caracas: Litopar (Cuaderno del Institute de Filologia "Andres Bello"). 1995 Las construcciones hendidas desde dos perspectivas teoricas y metodologicas. Lingiiistica 7: 143-152. Tournier, Michel 1972 (1969) Vendredi ou les limbes du Pacifique. Editions Gallimard. 1989 Le midianoche amoureux. Editions Gallimard. 1992a (1986) Viernes ο los limbos del Pacifico. Madrid: Alfaguara. [Translated by Loudes Ortiz] 1992b (1991) El medianoche enamorado. Madrid: Alfaguara. [Translated by Santiago Martin Bermiidez] Valli, Andre 1981 Notes sur les constructions dites 'pseudo-clivees' en fransais. In: Recherches sur le frangais parle 3; G.A.R.S. (eds.), 195-211. Van den Steen, Katleen 2000 Las construcciones ecuacionales en frances y en espanol. Unpublished thesis, Katholieke Universiteit Leuven.

290

Katleen Van den Steen

Vikner, Carl 1973 Quelques reflexions sur les phrases clivees en fransais moderne. Actes du 5ieme

Congres des romanistes scandinaves: 4-19.

How to express indifference in Germanic: Towards a functional-typological research programme Torsten Leuschner

Introduction The ability to express indifference appears to be universal in languages, yet the expressions used for this purpose have so far met with complete indifference on the part of linguists. The purpose of the present paper is to outline a project for the study of expressions of indifference in a functionaltypological vein. The research programme I envisage should eventually be capable of being generalized to, and refined against, a large, methodologically sound language sample from around the world. In the present paper, however, I will restrict myself to a preliminary exploration of the basic parameters likely to be involved in such an undertaking, and to a cursory overview of a small sample of languages that are both well-documented and easy to compare. These languages are the Germanic ones. Expressions of indifference have many forms and uses, but for my pre-sent purposes I will provisionally define them as predicates that assert the proposition 'X does not matter', with the additional restriction that X be expressed as a clausal complement, more precisely an embedded interrogative. Beginning with a basic (and greatly simplified) analysis of the meaning of predicates of indifference, I discuss a few patterns of use of such predicates (section 1) and then set out a synchronic typology of expressions of indifference in English, German and Dutch, with reference to structural patterns and the types of lexical elements employed (section 2). Next I review the processes involved in the evolution of the present-day typology (lexical renovation, lexicalization, idiomatization), and also discuss the borrowing of German predicates of indifference into Continental Scandinavian and their further development there between the sixteenth and nineteenth centuries (section 3). In the Conclusion, the sensitivity of expressions of indifference to word-play, register and taboo-breaking on the one hand and crosslinguistic borrowing on the other is highlighted (section 4). Most examples are taken from computerized text corpora and historical gram-mars and dictionaries; cf. the abbreviations listed at the end of the text, pre-ceding the endnotes and the references.

292

Torsten Leuschner

1. Patterns of use 1.1. Basic analysis Let us begin by trying to explicate our intuition what it means to express indifference with regard to the content of some embedded interrogative. Consider the following example: (1)

a .It doesn't matter who comes to the party.

Let us assume some version of Hamblin (1976) and Karttunen (1977) such that questions are defined as functions which for a given situation pick out the set of propositions that jointly constitute a complete (and, for Karttunen: true) answer in that situation; and let us furthermore assume that a constituent interrogative expresses an open sentence in which the WH word represents a variable: (1)

b. χ will come to the party.

Expressing indifference as to who comes to the party amounts to a refusal to pick out the correct proposition(s) from (l)b. as defined by different instantiations of Λ:. Predicates of indifference thus qualify as "non-venturing", i.e. as predicates which cannot be used to ask the question encoded by the interrogative (Horn 1986: 180f.). A very similar analysis is possible for alternative interrogatives: (2)

a. It doesn't matter whether John comes to the party (or not). b. John will either come to the party or he won't come to the party.

Again, (2) expresses a refusal to choose between the propositions implied by the embedded interrogative, except that this time only two values are taken into consideration. Predicates of indifference in Germanic come in two basic structural types: either with an element of negation or with an element denoting iden-tity or equality. Each type can be formed in the same two ways: with the copula 'be' plus a predicate nominal (prototype in English: it is no matter) or with a main verb construction. Main verb constructions with negation can be subdivided into personal constructions (prototype: I don't care) and impersonal constructions (prototype: it makes no difference)·, 'be' + predicate nominal expressions are always impersonal. The resulting structural possibilities are listed in Table 1.

How to express indifference in Germanic

293

Table I. Structure types of expressions of indifference in Germanic Type negation

'identity / equality'

'be' + nominal

impersonal

main verb construction

impersonal

Prototype in English it is no matter

personal

it makes no difference, it doesn't matter I don't care

'be' + nominal

impersonal

it's all the same

main verb construction

impersonal

it comes (down) to the same (thing)

An elegant explanation for the neat division into patterns with negation and 'identity / equality' can be derived from Baker's account of the function of what he calls "/natfer-predicates" (1968: 113-119). This class is roughly identical to Karttunen's class of "predicates of relevance" (1977: 6), i.e. the affirmative versions of predicates like matter, make a difference, care, give a damn / hoot / etc. plus a few others (Baker 1968: 167).1 Concentrating on make a difference, Baker asks what a speaker would have to do to challenge another speaker's statement involving a negated version of make a difference. The answer is that they would have to make a series of statements as in (3) (based on 1968: 115): (3) A: It doesn't make any difference to Johnson who the Republicans nomi-nate. B: (Yes, it does make a difference.) If the Republicans nominate Xh then S,. If the Republicans nominate X2, then S2. If the Republicans nominate X3, etc.

The challenger would have to point out that different choices for Republican nominee will in fact lead to different consequences for Johnson. So what predicates of indifference do is "deny the existence of a relation of dependency between some set of antecedents and a set consisting of two or more distinct consequences" (based on Baker 1968: 114): (4)

(It doesn't make a difference to Johnson who the Republicans nominate.) If the Republicans nominate X\, then S. If the Republicans nominate X2, then (also) S. If the Republicans nominate X3, then (still) S. If the Republicans nominate X4, etc.

This analysis is easily compatible with our earlier description of expressions of indifference as a refusal to choose between propositions that are defined by the interrogative and differ with regard to the instantiation of

294

Torsten Leuschner

some variable in them. Such a refusal will be felicitous iff the writer / speaker believes that the relevant set of propositions leads to one and the same consequent S in a conditional schema in which they function as antecedents.2

1.2. Partitioning and predicator-complement order Corpus data confirm the analysis given above. In the following example, the alternative interrogative embedded in it doesn't matter raises two different values Xj and X 2 for the salient variable 'type of wearer', and the whole is followed by a statement spelling out S. As if to confirm that the instantiation of X will really lead to the same S under any conceivable circumstance, an elaboration is added mentioning a particularly extreme instantiation for Xj: (5)

No other manufacturer makes such a wide variety of shirts as the CWS. Society shirts are made to please all types of wearer, from the artisan to the executive, and for all occasions. It doesn't matter either whether you are a giant or a dwarf, your Co-operative society can fit you out with a CWS shirt. The Broughton was recently called upon for a shirt with a 22 1/2 in. collar and a 66 in. chest measurement. (LOB)

(5)'

If you are a giant (- XJ, your Co-operative society can fit you out with a CWS shirt (= S). If you are a dwarf (= Χτ), your Co-operative society can fit you out with a CWS shirt (= S).

As the paraphrase (5)' shows, (5) is a clear instance of Baker's schema. But the example also makes an interesting point concerning the typical use of predicates of indifference: it suggests that speakers / writers often conceive of S first and then introduce the issue of dependency deliberately as a way of strengthening S. In (5), this is based on a semantic mechanism that Couper-Kuhlen and Thompson (2000: 398) call "partitioning": the embedded interrogative sets up a distinction between several kinds of instantiations for the variable 'types of wearer'. By means of the predicate-complement construction, the writer introduces into the text a pragmatic presupposition if X, then S which suggests that different instantiations Xi, X 2 etc. for the variable could lead to different consequents Si, S2 etc., and simultaneously denies that this is the case. S is thus strengthened by appearing as valid regardless of conceivable circumstance. A slightly different case is (6). Again, a possible parameter for the partitioning is introduced ('ownership of the papers'), but this time the comple-

How to express indifference in Germanic

295

ment is a constituent interrogative and no specific instantiations are spelled out: (6)

Der in Ihrer Wohnung weilende Polizeibe[am]te darf zwar in sämtliche Ecken und Winkel sehen, Schränke, Schreibtische und Tresore öffnen, er darf aber nicht Ihre Papiere durchsehen! Darunter ist nun alles mögliche zu verstehen, so die Liebesbriefe Ihrer Frau, Ihre Privataufzeichnungen über den künftigen Hausbau, Tagebücher, alle Arten von Akten, Urkunden und Rechnungen sowie Handels- und Rechnungsbücher usw. Wem die Papiere gehören, ist vollkommen gleichgültig, entscheidend ist allein, daß sie sich in Ihrem Besitz befinden. (MK) 'The policeman searching your home may look into all nooks and crannies, he may open wardrobes, desks and safes, but he may not browse through your papers! This refers to every conceivable stuff, like your wife's love letters, your private notes concerning the house you are going to build, diaries, all sorts of dossiers, documents and invoices, accounts concerning merchandise and finance, etc. Whose papers they are makes no difference at all, the sole decisive factor is that they are in your possession.'

(6)'

If the papers in your possession belong to person X b the policeman may not browse through them. If the papers in your possession belong to person X2, the policeman may not browse through them. If the papers in your possession belong to person X3, ...

Note also that the sentence following the predicate-complement construction does not state S here; rather, S was mentioned two sentences earlier. Linked to this is another regularity of use in predicates of indifference: the distinction between extraposition and non-extraposition of the complement. It is not a coincidence that the interrogative complement precedes the predicate in (6), where S also precedes, whereas it follows the predicate in (5), where S also follows. In the impersonal constructions, where the complement has the grammatical function of subject, the complement is in cases like (5) represented in the subject slot by the cataphoric pronoun it (German es, Dutch het)\ alternatively, the complement may remain in its canonical position, as in (6). A general rule for the choice between extraposition and non-extraposition is the "principle of complementary weight" (Erdmann 1990: 137f.): the lighter the predicate in relation to the complement, the greater the probability of extraposition. More specifically, however, extraposition of complements with predicates of indifference may be motivated as a strategy to facilitate the interpretation of the interrogative (e.g. to clarify as early as possible that it is not to be processed as an actual question) and / or as a focusing strategy for the predicate of indifference itself. But above all, examples like (5) and (6) point to the importance of

296

Torsten Leuschner

the textual function of the complement. As Erdmann (1990: 127-140) has shown, non-extraposed complements tend to have an exclusively anaphoric function, whereas extraposed complements are neutral in this respect and are often used as coherence bridges to facilitate thematic transitions between stretches of text. This is precisely what distinguishes the non-extraposed complement in (6), which looks back to a previously mentioned S, from the extraposed one in (5), which looks ahead to an as yet unmentionedS.

1.3.

Rhetorical Dialogues

Especially when the predicate of indifference with its complement follows the statement of S as in (6) or is inserted into it parenthetically, the impression can arise that the writer is conducting a kind of oblique conversation with the addressee. Consider the following example: (7)

Voor de belangenbehartigers, of die laatsten nu individuen of mammoetorganisaties zijn is om het even, is ... geen rol weggelegd. (38MC) 'Representatives of [extra-parliamentary] interests, whether they are individuals or powerful organisations doesn't matter, have no part to play.'

(7)'

If the representatives are individuals (= X/), they have no part to play (= S). If the representatives are powerful organisations (= Χχ), they have no part to play (= S).

Like in the other examples, the embedded interrogative sets up a partition between several kinds of instantiations for the relevant variable, in this case individuals and powerful organisations as instantiations of representatives, and the predicate of indifference declares this distinction irrelevant for S. What is special this time is that the predicate-complement construc-tion is related to the surrounding text by nothing else than the partitioning itself, making no contribution to its further thematic development (note again the backward-looking complement-predicator order). This is also the case in the following example, where the inserted construction is actually surrounded by brackets that indicate its thematic isolation: (8)

Construed as an internal impression which is thought to function as a cause that issues in some item of so-called overt behaviour (whether this be some bodily movement or an action is of no matter for our present purposes), the impression must be describable without reference to any event or object distinct from it. (LOB)

How to express indifference in Germanic

297

Such parentheses raise an interesting question: if the issue that is expressed in them is irrelevant, why don't the maxims of quantity and relevance prevent the writer / speaker from mentioning it in the first place? Or in other words: how can a parenthesis be cooperative that blantantly flouts these maxims? The answer is: it is cooperative if the issue raised is Hearer-old (term from Prince 1992), i.e. if the parenthesis is interpreted as responding to something that is already on the addressee's mind. Parentheses as in (7) and (8) can thus be regarded as "Rhetorical Dialogues" (Popovici 1981, cf. Herring 1991 and Leuschner 1998), viz. as a strategy of forward defense with which the writer / speaker can anticipate and forestall some potential objection that might put S in doubt in the addressee's mind given some relevant aspect of world knowledge. The anticipating process consists in the writer / speaker thinking of some expression in the prior stretch of text as a variable and looking for at least one instantiation of the relevant variable which the addressee might believe supports ~S as consequent in a conditional sentence schema. For (7) this works out as follows: (7)"

If the representatives are individuals (= Xj), they have no part to play (= S). But if the representatives are powerful organisations (= Χτ), they may well have a part to play (= ~S).

The relevant aspect of world knowledge that might lead the addressee to reason in this way could be expressed by the following regularity: (7)"'

normally (powerful organisations are better able to make their influence felt than individuals)

In order to make her statement S safe from the potential challenge (7)", the writer / speaker inserts the parenthesis into her text as an instruction to the addressee to update her knowledge store such that (7)"' is invalidated, i.e. that X 2 leads to S rather than ~S, in this particular case.

1.4. Origo-identification We should also note the very common use of expressions like to me, German mir (dative), Dutch (voor) mij etc. in (certain types of) predicates of indifference. The function of these expressions can be described as relativizing the indifference to a specific point of view which I will call its "origo". One function of origo-identifying expressions is to distinguish the origo of the indifference from the speaker. In the following example, ihm

298

Torsten Leuschner

'him' locates the indifference concerning the woman's identity in the mind of the main character: (9)

Im Grunde war ihm gleichgültig, wer sie wirklich war, wenn sie nur bei ihm blieb, wenn er sie nur nicht verlor. (MK) 'Basically it didn't matter to him / he didn't care who she really was, if only she stayed with him, if only he didn't lose her.'

But such uses appear to be only a special case of a more general pattern which can be described in terms of felicity conditions. In its original context, the non-origo-identifying sentence in (10) was written as part of a discussion of presumable limits to the expansion of markets within the world economy. In order to back it up, the speaker could point to dependencies in the real world which show that different instantiations of X lead to the one and the same S: (10)

Die Wirtschaft hat heute die Tendenz, ihre besonderen Interessen für absolute Lebensbedingungen zu erklären. Es ist der Stop der Expansion, der das Schicksal und die Wende der Wirtschaft ist. Ob die Begrenzung durch den Umfang des Erdballs oder durch den halben Erdball gegeben wird, ist grundsätzlich einerlei. (MK) 'The economy has the tendency nowadays to present its specific interests as if they were absolute conditions of life. It is the end of the expansion which marks the fate and the turning-point of the economy. Whether the limit is set by the earth's circumference or half the earth, is basically all the same.'

Any origo-identifying expression changes this drastically. The sentence could now be used by, say, a teacher advising his students on how to solve some mathematical equation: (10)'

Ob die Begrenzung durch den Umfang des Erdballs oder durch den halben Erdball gegeben wird, ist mir grundsätzlich einerlei. 'Whether the limit is set by the earth's circumference or half the earth, is basically all the same to me.'

Crucially, the speaker could back up this version by doing nothing (e.g. by assessing all students the same, regardless of how they chose to solve the equation). So what is at issue in origo-identifying sentences is not any objective dependency between states of affairs but the question whether people allow circumstances to influence their actions; the predicate of indifference is subjectified and becomes similar to the Ί don't care' type (in which the origo is expressed as the subject). To test this hypothesis, let us

How to express indifference in Germanic

299

apply Baker's method and see what a challenger could say in response to a statement as in (11): (11)

A: It doesn't make a difference to me what the weather is like. I go jogging every day (= S). B: Yes, it does make a difference to you (what the weather is like). If the sun is shining (= Xi), you go jogging (= X2). But if it's raining (= X2), you stay at home (= S2).

If A asserts that the weather has no influence on her actions, then this claim can be invalidated by, e.g., pointing out instances when she changed her mind given a certain type of weather. This is also the impression given by another version of the same exchange: (12)

A: It doesn't make a difference to me what the weather is like. I go jogging every day (= S). B: It does make a difference to me (what the weather is like). If the sun is shining (= Χι), I go jogging (= Si). If it's raining (= X2), I stay at home (= S2).

The origo-identiiying expression makes it possible to juxtapose two patterns of decision-making, either by comparing two people's actions under similar circumstances, as in (12), or by comparing a speaker's description of her behaviour with reality, as in (11) above. More research is obviously needed, and I will not speculate any further on this intriguing issue here.

1.5. Ellipsis and grammaticalization A special feature of the use of predicates of indifference is their proneness to ellipsis, as in the following examples: (13)

"Was will sie?" fragte die alte Dame gereizt. "Sie ist eine Bürgerliche, nicht wahr?" "Egal, Großmama", gab ihr Enkel scharf zurück. "Sie ist unser Gast." (MK) '"What does she want?" the old lady asked, irritated. "She's a commoner, isn't she?" "No matter, Grandma", retorted her grandson sharply. "She's our guest.'"

(13)'

[Es istJ egal[, ob sie eine Bürgerliche ist oder nicht], Großmama. Sie ist unser Gast.

Note that in the absence of any S, the main clause following the predicatecomplement complex can be read as an account for why the social status of

300

Torsten Leuschner

the referent of sie is irrelevant. Like other disaligning negatives (Ford 2001), predicates of indifference are often followed by such accounts or elaborations (cf. the entscheidend sentence in example (6) above) which, in spoken interaction, tend to justify the disaligning speech-act as cooperative behaviour. Under the right circumstances, 'be' + nominal predicates of indiffe-rence plus their interrogative complements can become grammaticalized, through ellipsis, into a type of adverbial clause known as "concessive con-ditionals" (Haspelmath and König 1998). Such concessive conditionals can be analysed as a conditionals with several antecedent values instead of one: (14)

if {a, b, c, ...}, then q

Alternative and constituent interrogative clauses combined with more or less reduced predicates of indifference are typical means for evoking the set of antecedent values in such constructions: (15)

Egal, wie die Verhandlung ausgeht, ich will mit Löfflad nichts mehr zu tun haben. (MK) 'No matter how the court case ends, I want to have nothing more to do with Löfflad.'

(15)'

Egal, ob die Verhandlung gut oder schlecht ausgeht, ich will mit Löfflad nichts mehr zu tun haben. 'No matter whether the court case ends positively or negatively, I want to have nothing more to do with Löfflad.'

The semi-grammaticalized character and great formal variability of concessive conditionals in many languages provide an unusual opportunity to observe the emergence of a subordinating construction from discourse (König 1992, Leuschner 2003). As in sentence-combining generally (Hopper and Traugott 1993: 170), the relevant mechanisms of change are the unification and bonding of the clauses concerned and the hierarchical downgrading of one of them as satellite to the other. The hierarchical downgrading of the predicate-complement construction to protasis proceeds via the ellipsis of the dummy subject 'it' and the verb 'be'. This means that only predi-cates of the 'be' + nominal type (English it doesn't matter, German es ist egal, Dutch het is om het even etc.) can be affected, because only in this type of predicate is the information expressing the indifference concentrated solely in the predicate nominal, whereas the verb is only responsible for the asserting and hence dispensible from a purely semantic point of view. As a result of the grammaticalization, the reduced predicate expression comes to

How to express indifference in Germanic

301

function as a quantificational prefix on the protasis (see Nishigauchi 1990: 177-201 for English no matter). With alternative interrogative clauses, the prefix can even be dropped altogether, so that the former disjunctive complementizer 'whether ... or' now marks a semantic relationship between protasis and apodosis: (16)

Whether fan of the Pagnol films or stage show, whether partial to music or no, you can't help but derive joy from this picture if you have a sense of humor and a heart. (BRO)

Note that ellipsis continues to affect these more grammaticalized constructions inside the protasis: Whether (you are) a fan of the Pagnol films or stage show, whether (you are) partial to music or no, ....

2. Lexical types As to the linguistic means involved in the formation of predicates of indifference, the refusal to choose from a set of propositions as defined by different instantiation of a variable can essentially be expressed in two ways: from the point of view of S, asserting that S is identical or the same (equal) irrespective of X, or else from the point of view of X, denying that each X leads to a different S. These two different conceptualizations of the same situation can either be expressed in a more personal, subjective way (hence the personal constructions with verbs expressing involvement like 'care') or more objectively (with 'it is'). Which concrete lexical elements are used is contingent on individual languages and their history (cf. section 4), but they will be compatible with one of the basic schemata, possibly mediated by metaphor, e.g. in terms of MATERIAL (English matter < Old French matere; OED, matter sb.l), DIRECTION (either DOWNWARD: Dutch het komt op hetzelfde neer 'it comes down to the same', or OUTWARD: German es läuft auf dasselbe hinaus 'it comes out onto the same'), and so on.

2.1. The 'don't care' type As far as the three languages in question are concerned, the English expression I don't care is the only representative of the personalized structure and meaning type to express indifference in a relatively neutral way; a more emphatic, colloquial version of it is I couldn't care less. One can of course say ich kümmere mich nicht darum or ich schere mich nicht darum in Ger-

302

Torsten Leuschner

man (the latter is also mentioned by Wunderlich 1976: 188) or ik bekommer me er niet om in Dutch, but these phrases express a degree of personal involvement more akin to benign disinterestedness. An interesting syntactic difference is that the English expressions can embed interrogatives directly, whereas German sich um etwas kümmern / scheren and Dutch zieh om iets bekommeren 'to care about s.th.' both require a pronominal adverb or equivalent phrase that incorporates the preposition um / om (German darum, Dutch er ... om) and takes the interrogative clause as complement, as e.g. in ich kümmere / schere mich nicht darum, wer kommt and ik bekommer me er niet om wie komt Ί don't care who comes'. Note that English also has I don't mind, with mind in the sense of 'care' which is also preserved in, e.g., child-minder.

2.2. The 'no matter' type This type exists only in English. Its two main varieties are it doesn't matter and the slightly archaic it is (of) no matter, both of which occurred in earlier examples. Among negated impersonal main verb constructions, it doesn't matter is a special case because it consists only of a negated verb without any nominal or other component. It is no matter is the older of the two expressions, and nowadays it mostly occurs as a quantificational prefix {no matter) in concessive conditionals or, in a further step of grammaticalization, as a preposition whose meaning can be described as concessive-conditional: (17)

He was no longer able to relax in the presence of his parents and found it difficult to keep up a conversation with his mother or father, no matter the subject. (BRO)

Another variant of it is (of) no matter that is now somewhat archaic is it makes no matter. The verb matter in it doesn't matter is a simple conversion of the noun which dates from the sixteenth century (OED, matter ν 2). An expression incorporating a negated adjective based on the noun mate-rial itself also exists, though it, too, is somewhat archaic: (18)

To me it'd be immaterial whether you'd retaliate or not. (LOB)

All these expressions embed interrogatives directly.

How to express indifference in Germanic

303

2.3. The 'no difference' type This is the most varied and cross-linguistically most interesting type. It is also the fuzziest one, and admittedly the label 'no difference' is somewhat misleading because many of its representatives do not literally involve the noun 'difference'. Prototypical examples are English it makes no difference (and its synonym it doesn't make a difference) and German es macht keinen Unterschied: (19)

It makes no difference what part of the world is involved, what form of regime, what particular issue. The US cannot take the initiative against the Left. (BRO)

(20)

Ob Sie nun weiter wie bisher in Deutschland umherfliegen und üben oder über den Mittelmeerländern, das macht doch keinen Unterschied. (MK) 'Whether you continue flying over Germany for your training or over the Mediterranean countries, that makes no difference, does it.'

Phrases like these show a relatively simple main verb + direct object structure that also appears in German es spielt keine Rolle / Dutch het speelt geen rol 'it doesn't play a part (as in drama)'. But expressions like German es tut nichts zur Sache / Dutch het doet er niet toe it doesn't add to the matter' can also be subsumed under this type, and so can no doubt others: (21)

Dabei spielt es keine Rolle, ob diese Ermüdung von dem Polizisten absichtlich herbeigeführt wurde oder nicht. (MK) 'In such a situation it plays no part whether this fatigue was brought about by the policeman intentionally or not.'

(22)

Het doet er niet toe of je op een schip goed kunt roeien of zeilen of navigeren. (5MC) 'It doesn't matter whether you are good at rowing, sailing or navigating a ship.'

More instantiations could easily be added, such as Dutch het kan mij niet schelen (lit. 'it can't differ to me'), het geeft niet (lit. 'it doesn't give') and het maakt niet uit (lit. 'it doesn't make out', cf. the related German es macht (mir) nichts aus 'it makes me nothing out'): (23)

Mij maakt het niet uit tegen wie ik moet speien. (5MC) 'It doesn't matter to me who I have to play against.'

Dutch onverschillig (lit. 'indifferent') also belongs here:

304

Torsten Leuschner

(24)

Het aardige is dat de gebruiker altijd kan bellen of gebeld kan worden, onverschillig in welk toestel hij zijn kaart steekt. (5MC) 'The nice thing is that the user can always ring or be rung, regardless of which phone he places his card in.'

Onverschillig is derived by ellipsis from the predicate het is onverschillig (lit. 'it is indifferent') and is nowadays mainly used as a quantificational prefix in concessive conditionals.

2.4. The 'identity / equality' type Predicates that contain an element denoting 'identity' or 'equality' come in two subtypes, neither of which is very prominent in English. One subtype involves the verb 'be' and as predicate nominal some adjective that can be paraphrased as 'equal' or contains an element of similar meaning, such as German gleich 'equal' (cf. Belgian Dutch gelijk), egal (< French egal), einerlei 'of one kind', gleichgültig 'equally valid' and gleichviel 'equally much': (25)

Da hatte der Museumsdirektor Mühe, einen dritten Wärter zu finden - oder war es schon der elfte, den er suchte? Gleichviel, welcher er war! (MK) 'The museum's director now ran into trouble finding a third guard - or was it already the eleventh? No matter (lit.: Equally much) which one it was!'

Dutch even (lit. 'equal, smooth, regular, undisturbed'; WNT, even I A) in het is (mij) om het even also belongs here. It's all the same (to me) is another instantiation of 'identity / equality' type, as is dialectal Dutch het is (mij) eender, the eender being derived from a genitive of een 'one' (WNT, eender). The other subtype contains a lexical verb and usually expresses the notion of equality through a (nominalized) adjective or a pronoun, as in English it comes (down) to the same thing, German es läuft auf das gleiche hinaus and es kommt auf das gleiche heraus (lit. 'it runs / comes out on the same'), and Dutch het komt op hetzelfde neer (lit. 'it comes down on the same') and variants thereof: (26)

Ob nun Gorbatschow in seinem Urlaub selbst zum Zurechtrücken seiner Worte gekommen ist, oder ob er dem Druck des Politbüros ausweicht, in der Sache kommt das auf das gleiche heraus: es gibt keinen wirklichen Kompromißzwischen Ost und West... (H87)

How to express indifference in Germanic

305

'Whether Gorbachev, during his holiday, decided to amend his wording out of his own initiative, or was evading pressure from the politbureau, essentially it comes down to the same thing: there is no genuine compromise between East and West.' (27)

Of men nu behavioristisch denkt dat een jong kind zo veel mogelijk onaangenaamheden vermijdt, of de verbondenheid met vader en moeder niet wil verliezen of een oplossing zoekt voor de angst vanwege verboden oedipale wensen, steeds komt het erop neer dat het zieh identificeert met de wensen en gedragingen van de ouders .... (5MC) 'Whether one thinks behavioristically that a young child avoids unpleasant experiences as much as possible, or seeks not to lose the bond with Father and Mother, or looks for a solution for the angst arising from repressed oedipal desires, it always comes down to an identification with the parents' wishes and behaviour patterns.'

It is characteristic of these expressions to express some metaphorical motion, which is outward in German {hinaus / heraus 'out', cf. the English noun outcome 'result'), but downward in Dutch {neer 'down') and English. Interestingly, there are systematic differences in usage between the two subtypes of 'identity / equality' expressions: the embedded interrogative clause in the main-verb subtype is invariably an alternative interrogative, the order is always complement-predicate, and yet the predicate always contains a dummy subject 'it'. 3. Diachronic processes 3.1. Lexical renovation The main precondition for the creation of the present-day typology of predicates of indifference was a large-scale process of renovation in the relevant parts of the lexicon of the Germanic languages. In German, this process took place in the transition from Middle to Early Modern High German. The great majority of the Middle High German expressions of indifference listed by Koller, Wegstein and Wolf (1990, s.v. gleichgültig) are derived from one of three productive bases, none of which play any role in the present-day typology any longer, viz. the verb ruochen 'to care' and the adjectives mcere 'interesting, important' and tiure 'expensive, dear'. For example, 'something doesn't matter to me' could be expressed (with negation) as mir ist etw. untiure 'something is un-dear, un-valuable to me' (Lexer 1872-78, s.v. untiure) and (with an expression of equality) as daz ist mir also mcere 'that is equally interesting / important to me' (1872-78, s.v. mcere adj), but also using a negated form of ruochen and an embedded in-

306

Torsten Leuschner

terrogative clause. In this case one and the same simplex verb could be used to express indifference in a personal and an impersonal construction ( Ί don't care' vs. 'it doesn't matter'): (28)

doch enruoche ich wem du'z sagest (Lexer 1872-78, ruochen) 'Yet I don't care whom you tell it.'

(29) ez enruochet mich, swaz darumbe dir geschiht (Lexer 1872-78, ruochen) 'It doesn't matter to me what happens to you as a consequence.'

Interestingly, while ruochen is still prefaced by the negative prefix en- in (28) and (29), the indifference meaning was apparently becoming lexicalized in the verb itself, for it was sometimes used without the negative and still meant 'not care' (Benecke, Müller and Zarncke 1854-66, ruoche 11). Simplex expressions of indifference containing neither overt negation nor any element denoting identity or equality are extremely rare and unattested in all the modern languages in question. Middle High German also had alein(e), which was often used as a concessive subordinating conjunction ('though') and has been explained as a grammaticalization of daz ist al ein 'that is all one' (Behaghel 1928: 64, Schieb 1952: 276f.; cf. modern Polish wszystko jedno 'no matter < all one'). The development, which is as plausible as (unfortunately) unattested, would have run something like this: (30)

alein er si des guotes bloz, doch ist er von gebürte fri (Behaghel 1928: 63) 'even if / no matter if he (may) lack property, he is still a freeman by birth'

< [it is] all one: he (may) lack property [or not], he is still a freeman by birth

Even if this theory is correct, alein(e) remains the only grammaticalized form of predicates of indifference that I am aware of in the medieval languages in question. It disappeared from the German lexicon at the end of the Middle High German period along with ruochen and mcere, which left only a few unproductive traces (ruch-los, Mär-chen), and the same happened to the cognates of ruochen in Dutch (roeken, cf. modern roeke-loos) and English (recan, cf. modern reck-less). So it seems that in the long term the basic structural patterns of predicates of indifference remain the same but the actual elements that make them up are renovated. This process is at least partly linked to changes in the lexicon at large: as certain roots become unproductive, their derivatives disappear from predicates of indifference and are replaced with derivatives of newly productive roots.

How to express indifference in Germanic

307

3.2. Lexicalization The first mechanism of change following lexical renovation that contributes to the formation of predicates of indifference is lexicalization. Lexicalization (cf. Blank 2001, Lehmann 2002) makes available lexical elements to take the place of the older, obsolescent component parts. Of the regular processes of semantic change that drive it, I will mention only one here, viz. subjectiflcation (in the sense of Traugott 1995). Subjectiflcation is already in evidence in the change from the original meaning of matter 'material' to the expression it is no matter (to me), and it is in the same subjectified meaning that the noun became converted to the verb matter. Another good example is the development of German gleich 'equal' (and Dutch gelijk) and its compounds gleichgültig 'equally valid' and gleichviel 'equally much'. These compounds (which do not occur in either Middle High German or Dutch) were at first both spelled in two words. Even after univerbation, gleichgültig could at first still denote equal material value: (31)

cautiones, die durch guter bestalt sein, können auf den notfall durch gleichgültige guter ausgewechselt werden (1603; DWB, gleichgültig A l a ) 'Deposits for goods can in case of emergency be exchanged for goods of equal value.'

But before such expressions could be used in predicates of indifference, the assertion of equality (possibly of value, as in gleichgültig, or amount, as in gleichviel) had to become a matter of subjective judgement. As a re-sult, something could now be gleich / gleichgültig / gleichviel to someone, and if necessary, the origo of the judgement could be expressed by the dative discussed earlier, e.g. X ist mir (etc.) gleich / gleichgültig / gleich-viel. The subjectified meaning of gleichgültig began to show up in literary texts during the early decades of the eighteenth century: (32)

damit ein allgemein gleichförmigs einerley / dem herzen nicht gleichgültig sey (Brockes, 1727; DWB, gleichgültig Β 1 b α.) 'so that the heart may not remain indifferent to a generally undifferentiated uniformity'

Note that the metre in (32) requires stress on -gültig, apparently reflecting the stress pattern of the original adjective phrase gleich GÜLtig\ in presentday usage this has become GLEICHgültig, with regular word-initial stress. The same has not happened to gleichVIEL, which is still stressed on the final syllable. Gleichgültig has also moved further than gleich and gleichviel in that it has developed a personal construction: being gleichgültig

308

Torsten Leuschner

came to be predicated on the subject making the judgement, i.e. one can now be gleichgültig (and show Gleichgültigkeit) towards something. Note that Dutch onverschillig has undergone the same evolution and can now be used as an adjective in the sense of 'indifferent'.

3.3. Idiomatization The second mechanism in the evolution of the present-day typology is idiomatization. In a nutshell, idiomatization is the process creating new idiomatic material (complex predicates of various kinds) from existing lexical items; like grammaticalization it involves loss of paradigmatic variability, i.e. a gradual narrowing-down of choice among alternative forms (Traugott 1999). Idiomatization thus has the effect of a clean-up operation: it creates the present-day typology by removing a certain amount of earlier structural variation. A good example from the history of Dutch is the reduction of variants of the phrase (het is) onverschillig. In the eighteenth century, we still come across data like the following which are impossible in modern usage: (33)

Het is ook niet veel verschillig, of deze oorzaak aan 't lichaam uytwendig, inwendig, natuurlyk, toevallig, gezond, ziekig of dodelyk zy. (1745; WNT, verschillig 4) 'It also doesn't make much difference whether this bodily cause be external, internal, natural, accidental, healthy, pathological or fatal.'

In present-day Dutch, only onverschillig survives, whereas niet verschillig is excluded. Similarly, a significant amount of variation has been lost in German, where the verb gelten ('be valid') used to be common in predicates of indifference with gleich and its compounds. Thus, apart from gleich (-gültig, -viel) sein, there was at first also gleich gelten 'be equally valid' and gleichviel gelten 'be valid equally much': (34)

aujf diese weis gilts gleich, ob sich einer wol hält, oder nicht (JJ.C. von Grimmelshausen, seventeenth century.; DWB, gleich II C 1 b) 'as a result it doesn't matter whether a person continues to be well or not'

(35)

ihm gaits gleichviel, wer kam und ihm sein brenz absoff (U. Bräker, 1789; DWB, gleichviel 2 b) 'it was all the same to him who came and drank his brandy'

In time, however, gelten disappeared from expressions of indifference in German just as make has all but disappeared from phrases with matter in

How to express indifference in Germanic

309

English (it makes no matter alongside it is no matter). And predicates with the verb matter, too, have seen their variability significantly reduced. For matter had a 'personalized' variant synonymous with 'not care' from about the mid-seventeenth to the late nineteenth century. Here is one of the oldest attestations in the OED: (36)

I matter not what careless abuses there may be put upon a word. (1664; OED, matter v. 3)

The loss of this use has clearly contributed to the predominance of impersonal 'it' predicate types which are 'personalized' by means of additional origo-identifying expressions like to me. One of the effects of idiomatization is thus to systematically restrict cer-tain kinds of lexcial expressions to certain structure types. For instance, predicates with or based on 'equal' now only occur as predicate nominals with sein (type iv) in German and no longer in predicates with main verbs (type v). This is important, for it is precisely the structures based on 'be' that are liable to ellipsis and hence to becoming involved in the grammaticalization of concessive conditionals (cf. above). Interestingly, corpus evidence (see Leuschner 2003) suggests that there is at least one predicate of indifference per language that is now used exclusively (or almost exclusively) as a quantificational prefix in concessive conditionals: German gleichviel, Dutch onverschillig and English no matter. Since the permanent ellipsis of 'be' implies another reduction of paradigmatic variation, the creation of such prefixes and the concomitant grammaticalization of concessive conditionals can be regarded as a further step in the process of idiomatization.

3.4. Borrowing Inventories of predicates of indifference may also be shaped through language contact. We already came across two well-known borrowings from French, one into English, viz. matter (< materel, ca. 1300?), and one into German, viz. egal (< egal, early eighteenth century). As with gleich and its com-pounds, the meaning of egal in das ist mir egal (< French cela m 'est egal 'that doesn't matter to me') can be described as a subjectification of the earlier meaning 'equal', which in turn appears to be a metaphorical exten-sion of meanings like 'smooth, regular' (DWB, second ed., egal 3). These are isolated examples, however. Overall, expressions of indiffe-rence in Germanic have evolved independently of non-Germanic models, and the

310

Torsten Leuschner

most significant borrowing of expressions of indifference happen-ed amongst the Germanic languages themselves. From the first half of the eighteenth century, several expressions of indifference show up in the Continen-tal Scandinavian languages that are based on (High) German compounds with gleich (see Leuschner 2001 for more details on the following). Exam-pies based on gleichgültig are Swedish det är (mig) likgiltigt and Danish det er (mig) ligegyldigt, whereas Swedish det är (mig) lik(a)mycket and Danish det er (mig) lige meget are based on gleichviel. The following are the first attested examples of the Danish versions: (37)

Enten en Hund staaer, gaaer eller legger sig, kan i sig selv vcere meget ligegyldigt. (C.B. Tullin, 1728-65; ODS, ligegyldig 1) 'Whether a dog stands, walks or lies down is (lit.: can be), taken by itself, no matter at all.'

(38)

Det kan ellers vcere mig ligemeget, om der bliver Comoedie eller ej. (L. Holberg, 1684-1754; ODS, lige VIIIA 2.4) 'For the rest it is (lit.: can be) all the same to me whether a comedy will be given or not.'

Apart from the native lik(a)- / lige- 'equal' that translates gleich, these expressions are so precisely modelled on the German originals3 that one could assume that these two compounds were taken over individually, pos-sibly along with the noun derived from gleichgültig meaning 'indifference' (Swedish likgiltig-het, Danish ligegyldig-hed, cf. German Gleichgültigkeit). In fact, however, Continental Scandinavian did not borrow the gleichbased expressions of indifference one by one but wholesale as a template {Phraseologiemodell). An indication of this is the fact that addi-tional expressions were formed, especially in Danish but initially also in Swedish, which did not have German models. Examples are lige godt 'equally good / well', which continues to exist in Danish today, and ligefuldt 'equally full', which has died out: (39)

For Resten er det lige got, hvad de kaldes .... (1724; ODS, god 4.5) 'For the rest it is all the same (lit.: equally good) what they are called.'

(40)

... er det lygefuldt, huad... der brugis dertiil. (Letters of Christian IV, 163640; Kalkar 1976, ligefuld) '... is it all the same (lit.: equally full) what is used for this purpose'

The realizations of the template then developed in sync with their German counterparts for about a century, i.e. for precisely the time that German influence on Continental Scandinavian continued to last. Just as in German,

How to express indifference in Germanic

311

there were initially also versions with a full verb meaning 'be valid' in Continental Scandinavian (Swedish gälla, Danish gelde), and exactly as the corresponding German combinations of gleich and its compounds with the full verb gelten 'be valid', they were lost during this period: (41)

sä gäller dem lika, om diefvulen och hans moder ... ock skulle hielpa dem (1744; SAOB, gälla 3 a γ) 'so it is all the same (lit.: valid equally) to him if the Devil and the Devil's mother are also going to help him'

(42)

det künde gielde dem lige meget, hvem der var Ejere af Italien (1734; ODS, gcelde Β 5.1) 'it could be all the same (lit.: it could be valid equally much) to him who was the owner of Italy'

In other words, the same idiomatization took place simultaneously in Continental Scandinavian and in German. After the end of German influence, the loss of paradigmatic variability continued, but with slightly different results: in Swedish most of the German-based expressions (except likgiltigt) have more or less died out and in Danish it is lige meget (idiosyncratically spelled in two words), not ligegyldig, that has become the most current expression of indifference based on 'equal'. The history of German-predicates of indifference in Continental Scandinavian is thus not only clearly contingent on the history of German-Scandinavian language contact generally (see Naumann 1989), the differences between Swedish and Danish, especially the much greater survival rate of German-based expressions in the latter, also reflect differences in medium and intensity in the original contact situation: contact with High German in Sweden was more limited to the written language than in Denmark, where the court and military and even significant parts of the population were German-speaking until at least the late eighteenth century (Winge 1992). Needless to say, the borrowing relations between German and Conti-nental Scandinavian should not obscure the existence in Scandinavian of many expressions of indifference that are quite independent of West Ger-manic models, as in the following examples from Swedish and Icelandic: (43)

Om han [dvs. kronhjorten] hade nägon hind att bröla för ma vara osagt.... (SUC) 'Whether the prize stag was roaring for some female doesn't matter [lit:, may be unsaid].'

312

Torsten Leuschner

(44)

£>ad mä liggja milli hluta hvort heldur hefur verid. (Kress 1982: 250) 'It doesn't matter [lit.: It may lie between the parts] which of the two things has happened.'

On the other hand, there are also expressions of indifference which have more or less close equivalents in the other Germanic languages, such as Swedish det gör (mig) detsamma 'it does (me) the same' and det spelar ingen roll 'it plays no part' (cf. German and Dutch), Swedish det kommer pä ett ut / Danish det kommer ud pä et 'it comes out on one', Danish det g0r ingen forskel / det er (mig) uden forskel 'it makes no difference / it is (for me) without difference', Icelandic pad er sama 'it is equal', pad stendur ά sama 'it stands to the same', pad skiptir ekki mdli 'it doesn't change the affair', pad gerir ekki til 'it doesn't add to' (see Böövarsson, ed., 1994 and Jonsson 1994 on Icelandic, Andersen 1998 on Danish, Schottmann and Pettersson 1989 on Swedish). Swedish det är (mig) egalt was borrowed either from French directly or via German during the eighteenth century (SAOB, s.v. egal), Danish det er (mig) egalt is more recent. Note also Swedish det kvittar 'it doesn't matter', a shortening of det kvittar (mig) lika (SAOB, s.v. kvitta), which is unusual because it appears to be the only predicate of in-difference in modern Germanic that makes do with just a simplex verb. For most of these expressions it will be hard or impossible to identify any bila-teral borrowing relations, and some are probably best described as inter-nationalisms or Europeanisms.

4. Outlook The purpose of this paper has been to explore a functional-typological approach to expressions of indifference. Clearly, a lot of research indeed remains to be carried out on all the issues covered here: the meaning and uses of expressions of indifference, their synchronic and diachronic typo-logy and variability, and their role as sources of linguistic change. The lat-ter aspect in particular is likely to provide an alternative perspective to Haspelmath and König's claim that "languages hardly manifest any systematic typological variation in [the more] weakly grammaticalized patterns" of concessive conditionals (1998: 581). As the present papers suggests, crosslinguistic variation in the more weakly grammaticalized patterns is in fact quite systematic, and the study of such patterns is likely to enable us better than the study of the more grammaticalized patterns alone to pinpoint the discourse basis of concessive conditionals and other processes of

How to express indifference in Germanic

313

grammaticalization in which predicates of indifference may turn out to be involved. Besides the synchronic and diachronic perspectives, the functionaltypological study of expressions of indifference should also focus on some cultural parameters that inform them. One likely candidate is register: although expressions of indifference are found in all registers of language use, those used in informal (familiar, colloquial, vulgar) styles differ markedly from those in the written register. A good example are such German untranslatables as schnurz and wurscht. The latter is derived from Wurst 'sausage' and, as worst, is also common in Dutch: (45)

Ob wir uns einen Schnupfen holen oder die Schwindsucht, kann dir wurscht sein, mein ich! (MK) 'Whether we catch a cold or tuberculosis should be no concern of yours [lit.: should be sausage to you], that's my opinion!'

(46)

Is het nu een muziekfestival of een dansfeest? Het zal de festivalgangers worst wezen. (VRT Flemish television news, July 29, 2000) 'Is it a music festival or a dance party? To those who attend, it will be all the same. [Lit.: To those who attend, it will be sausage.]'

In German at least, emphasis, word-play and innovation occasionally combine in long compounds like schnurzpiepegal; there are also (quasi-)tautologies as German das ist gehupft wie gesprungen and das ist Jacke wie Hose 'that makes as much difference as hopping or jumping / as jacket or trousers', and some that display in addition an ostensible lack of semantic motivation, e.g. Danish hip som hap / Swedish hipp som happ. Another interesting culturally determined parameter has to do with taboo-breaking, cf. such German phrases as ich scheiß' drauf Ί shit on it' and es geht mir am Arsch vorbei 'it passes by my ars'. In the English expression I don't give a damn (etc.), lexical elements taken from diverse taboo-sensitive parts of the lexicon are used as minimizers in "zero-semantics" (Postma 2001). Dutch expressions like het kan mij geen fluit / barst schelen 'it can't make a flute's / a chink's difference to me' may be similarly motivated.

Abbreviations [5MC:] [38MC:] [BNC:] [BRO:]

5 Million Word Corpus of written Dutch 38 Million Word Corpus of written Dutch British National Corpus (on-line demo version) Brown Corpus of written American English

314

Torsten Leuschner

[DWB:] [H87:] [LOB:] [MK:] [ODS:] [OED:] [SAOB:] [SUC:] [WNT:]

Deutsches Wörterbuch (Grimm) Handbuch-Korpus of written German, year 1987 Lancaster-Oslo/Bergen Corpus of written British English Mannheim Corpus of written German Ordbog over det danske Sprog [Dictionary of the Danish language] Oxford English Dictionary Svenska Akademiens Ordbok [Swedish Academy Dictionary] Stockholm-Umea Corpus of written Swedish Woordenboek der Nederlandsche Taal [Dictionary of the Dutch language]

Notes 1. There has of course been much debate on interrogative-embedding predicates since Baker's study (1968), but it has practically ignored predicates of indifference, and of the few authors who include such predicates in their discussion at all, Baker remains the only one who so much as mentions the negated versions. Lahiri (2002: 41-43) has recently proposed an explication of the semantics of the affirmative versions in terms of universal quantification, but since it is unclear how it can be extended to the negated versions, I will ignore it here. 2. This schema is one version of what is known as mouvement discursif conclusif in the Geneva school of discourse analysis (Moeschler 1985: 133f.). 3. The spelling of giltig in Swedish is that of a German variant of gültig with -i-.

References Andersen, Stig Toftgaard 1998 Talemäder i dansk. Ordbog over idiomer [Phrasal expressions in Danish: Dictionary of idioms]. Kopenhagen: Munksgaard. Baker, Carl L. 1968 Indirect questions in English. Ph.D. diss., University of Illinois (Urbana). Behaghel, Otto 1928 Deutsche Syntax. Eine geschichtliche Darstellung. Band III: Die Satzgebilde. Heidelberg: Winter. Benecke, Georg, Wilhelm Müller, and Friedrich Zarncke 1854-66 Mittelhochdeutsches Wörterbuch. 3 vols. Leipzig: Hirzel. Blank, Andreas 2001 Pathways of Lexicalization. In Language typology and language universals; Martin Haspelmath, Ekkehard König, Wulf Oesterrei-cher and Wolfgang Raible (eds.), 1596-1609. (HSK 20.) Berlin / New York: de Gruyter.

How to express indifference in Germanic

315

Böövarsson, Ärni (ed.) 1994 Islensk oröabok [Icelandic dictionary], Reykjavik: Bokaütgafa Menningarsjoös. Couper-Kuhlen, Elizabeth, and Sandra A. Thompson 2000 Concessive patterns in conversation. In Cause - condition - concession - contrast: Cognitive and discourse perspectives', Elizabeth Couper-Kuhlen and Bernd Kortmann (eds.), 381-410. Berlin / New York: Mouton de Gruyter. Erdmann, Peter 1990 Discourse and grammar: Focussing and defocussing in English. Tübingen: Niemeyer. Ford, Cecilia E. 2001 At the intersection of turn and sequence: negation and what comes next. In Studies in interactional linguistics·, Margret Selting and Elizabeth Couper-Kuhlen (eds.), 51-79. Amsterdam / Philadelphia: Benjamins. Hamblin, C.L. 1976 Questions in Montague English. In Montague grammar; Barbara Partee (ed.), 247-259. New York: Academic Press. Haspelmath, Martin and Ekkehard König 1998 Concessive conditionals in the languages of Europe. In Adverbial constructions in the languages of Europe·, Johan van der Auwera (ed.), 563-640. Berlin / New York: Mouton de Gruyter. Herring, Susan C. 1991 The grammaticalization of rhetorical questions in Tamil. In Approaches to grammaticalization, vol.1: Focus on theoretical and methodological issues; Elizabeth Closs Traugott and Bernd Heine (eds.), 253-284. Amsterdam / Philadelphia: Benjamins. Hopper, Paul J. / Traugott, Elizabeth Closs 1993 Grammaticalization. Cambridge: Cambridge University Press. Horn, Laurence 1986 Presupposition, theme and variations. In, Papers from the parases-sion on pragmatics and grammatical theory at the twenty-second regional meeting·, Anne M. Farley, Peter T. Farley and Karl-Erik McCullough (eds.), 168-194. Chicago: Chicago Linguistics Society. Jonsson, Jon H. 1994 Oröastadur. Oröabok um islenska mälnotkun [The place of words: Dictionary of Icelandic usage]. Reykjavik: Mai og Menning. Kalkar, Otto 1976 Ordbog til det celdre danske Sprog (1300-1700) [Dictionary of the Older Danish Language, 1300-1700]. 6 vols. Copenhagen: Thiele, [first published 1881-1918] Karttunen, Lauri 1977 Syntax and semantics of questions. Linguistics and philosophy 1: 3-44.

316

Torsten Leuschner

König, Ekkehard 1992 From discourse to syntax: the case of concessive conditionals. In Who climbs the grammar tree; Rosemarie Tracy (ed.), 423-434. Tübingen: Niemeyer. Koller, Erwin, Werner Wegstein and Norbert Richard Wolf 1990 Neuhochdeutscher Index zum mittelhochdeutschen Wortschatz. Stuttgart: Hirzel. Kress, Bruno 1982 Isländische Grammatik. München: Hueber. Lahiri, Utpal 2002 Questions and answers in embedded contexts. Oxford: Oxford University Press. Lehmann, Christian 2002 New reflections on grammaticalization and lexicalization. In New reflections on grammaticalization; Ilse Wischer and Gabriele Diewald (eds.), 1-18. Amsterdam / Philadelphia: Benjamins. Leuschner, Torsten 1998 At the boundaries of grammaticalization: what interrogatives are doing in concessive conditionals. In The limits of grammaticalization·, Anna Giacalone Ramat and Paul J. Hopper (eds.), 159-187. Amsterdam: Benjamins. 2001 Hochdeutsch-nordischer Phraseologietransfer in areallinguistischer Sicht. Gleichgültigkeitsausdrücke zwischen Sprachkontakt- und Sprachstrukturgeschichte. In: Sprachkontaktforschung und Areallinguistik: Europa, Theme issue 8 (no. 1/01) of Linguistik online-, Torsten Leuschner (ed.), http://www.linguistik-online.de/l_01/Leuschner .html. 2003 Between discourse and syntax: the syntacticization of concessive conditionals in English, German and Dutch. Ph.D. diss., Freie Universität Berlin. Lexer, Matthias 1872-78 Mittelhochdeutsches Handwörterbuch. 3 vols. Leipzig: Hirzel. Moeschler, Jacques 1985 Argumentation et conversation. Elements pour une analyse pragmatique du discours. Paris: Hatier-Credif. Naumann, Hans-Peter 1989 Typen niederdeutsch-nordischer Interferenz im Bereich der Phraseologie." In Niederdeutsch in Skandinavien II. Akten des 2. nordischen Symposions 'Niederdeutsch in Skandinavien' in Kopenhagen, 18.-20. Mai 1987; Karl Hyldgaard-Jensen, Vibeke Winge and Birgit Christensen (eds.), 241-259. (= Zeitschrift für deutsche Philologie, Beiheft 5), Berlin: E. Schmidt. Nishigauchi, Taisuke 1990 Quantification in the theory of grammar. Dordrecht / Boston / London: Kluwer

How to express indifference in Germanic

317

Popovici, Vasile 1981 Dialogues rhetoriques. Degres 9: il-i6. Postma, Gertjan 2001 Negative polarity and the syntax of taboo. In Perspectives on nega-tion and polarity items·, Jack Hoeksema, Hotze Rullmann, Victor SanchezValencia and Ton van der Wouden (eds.), 283-330, Amster-dam / Philadelphia: Benjamins. Prince, Ellen F. 1992 The ZPG letter: subjects, definiteness, and information-status." In Discourse description: Diverse linguistic analyses of a fund-raising text\ William C. Mann and Sandra A. Thompson (eds.), 295-325. Amsterdam/Philadelphia: Benjamins. Schieb, Gabriele 1952 Al, alein(e), an(e) im Konzessivsatz. Ein Beitrag zur Veldekekritik. Beiträge zur Geschichte der deutschen Sprache und Literatur 74: 268285. Schottmann, Hans and Rikke Petersson 1989 Wörterbuch der schwedischen Phraseologie in Sachgruppen. Münster: LIT Verlag. Traugott, Elizabeth Closs 1995 Subjectification in grammaticalization." In Subjectivity and subjectivization: Linguistic perspectives·, Dieter Stein and Susan Wright (eds.), 31-54. Cambridge: Cambridge University Press. 1999 A historical overview of complex predicate types. In Collocational and idiomatic aspects of composite predicates in the history of English·, Laurel J. Brinton and Minoji Akimoto (eds.), 239-260. Amsterdam/Philadelphia: Benjamins. Winge, Vibeke 1992 Dänische Deutsche - deutsche Dänen. Geschichte der deutschen Sprache in Dänemark 1300-1800 mit einem Ausblick auf das 19. Jahrhundert. Heidelberg: Winter. Wunderlich, Dieter 1976 Fragesätze und Fragen. In id., Studien zur Sprechakttheorie, 181-250. Frankfurt: Suhrkamp.

The lexicalization of speech act evaluations in German, English and Dutch Gisela Harras and Kristel Proost

1. Introduction This article is concerned with the way in which different types of speech act evaluations are lexicalized by speech act verbs and speech act idioms. We first distinguish different types of explicit and implicit evaluations which may be lexicalized by speech act verbs. The meanings of speech act verbs in German, English and Dutch are compared to examine which types of evaluations are lexicalized in each of these languages. Having established an inventory of evaluation types lexicalized by speech act verbs, we compare the evaluations lexicalized by speech act verbs with those lexicalized by speech act idioms. Particularly, we shall ask ourselves whether certain types of evaluations may be lexicalized by idioms rather than by verbs, and if so, whether this phenomenon also holds crosslinguistically. We shall also examine whether those evaluations typically expressed by speech act idioms are the same in German, English and Dutch.

2. Two specific properties of the semantics of speech act verbs The representation of the meaning of speech act verbs requires semantic categories different from those needed for that of other verb classes such as causatives and verbs of vision or motion. This is due to the following two properties specific to the semantics of speech act verbs. (l)The use of speech act verbs involves two types of situations. The first is one in which a speaker uses a speech act verb to refer to a speech act performed by another speaker. This situation type is called the "discourse situation (DS)" and contains three situation roles: a speaker, a hearer and an utterance which, in the prototypical case, contains a proposition. The second situation type is the one the discourse situation speaker refers to by using a speech act verb. The latter type of situation is called the "resource situation" (RS). Like the discourse situation, the resource situation contains the roles of a speaker, a hearer and an utterance containing a proposition (cf. Barwise/Perry 1983, Harras/Winkler 1994):

320

Gisela Harras and Kristel

Discourse situation

Proost

Resource situation

Figure I. Two types of situations

When a speech act verb is used performatively, the discourse and the resource situation coincide. (2) Speech act verbs lexicalize configurations of attitudes of the RS speaker. These include propositional attitudes as well as speaker intentions and speaker presuppositions (cf. Searle/Vanderveken 1985, Vanderveken 1990, Harras 2001). The propositional attitude of a speaker is the attitude of that speaker towards the propositional content of his utterance. In the case he asserts p, he takes ρ for true (epistemic attitude), in the case he promises ρ or asks someone to do p, he wants ρ (attitude of wanting), or in the case he praises or blames someone for p, he evaluates ρ positively or negatively (evaluative attitude). On the whole, the attribute of the RS speaker's propositional attitude may be assigned one of the following values:

The lexicalization

1

""\

epistemic

ν

f

\

S takes for true: (not) ρ S knows: (not) ρ assertive speech act verbs

assert inform mitteilen

grading

^

J

act evaluations

321

Λ

r

wanting

of speech

emotive

evaluative

J

v.

S wants: (not) ρ

S finds: ρ

S finds: ρ good/ bad

S feels joy, sorrow, anger

directive and commissive speech act verhs

verhs of grading

expressive speech act verhs

expressive speech act verbs

ask, promise, bitten, versprechen

grade einordnen

praise, blame, loben, tadeln

complain, klagen

Figure 2. Values for the RS speaker's propositional attitudes

The intentional attitude of the RS speaker is related to the RS hearer's reaction. The possible values for the attribute of the speaker's intention are: S wants Η to do or not to do ρ {ask, prohibit), S wants Η to take ρ for true {assert) or false {deny), S wants Η to know ρ {inform), S wants Η to recognise that S evaluates ρ positively {praise) or negatively {criticize), S wants Η to recognize that S is in a particular mental state {complain). Apart from these two attitudes, speech act verbs lexicalize presuppositions of the RS speaker. These correspond to the preparatory conditions which have to be fulfilled if a given speech act is to be performed successfully and non-defectively (cf. Searle and Vanderveken 1985, Vanderveken 1990). The presuppositions of the RS speaker concern the expectability of p, the field of interest of ρ and the presupposed attitude of H:

322

Gisela

Harras

and Kristel

Proost

expectable: (not) ρ

in the interest of S: (not) ρ in the interest of H: (not) ρ

Η takes for true: (not) ρ Η knows: (not) ρ

warnen, warn verbieten, prohibit

request, verbieten promise

deny, bestätigen ask, inform

Figure

3. V a l u e s for the R S speaker's presuppositions

For some speech act verbs, the propositional content may be constrained in several ways: the use of the verb announce ist restricted to future events; the use of request is restricted to a future action of the RS hearer, the use of promise is restricted to a future action of the RS speaker and so on.

3. The lexicalization of evaluations by speech act verbs Of the three types of attitudes mentioned above (i.e. the RS speaker's propositional attitude, the RS speaker's intention and the RS speaker's presuppositions), two are relevant to the lexicalization of speech act evaluations: (1) the propositional attitude of the RS speaker (2) the presuppositions of the RS speaker

The lexicalization of speech act evaluations

323

Speaker intentions are not relevant, because they involve propositional attitudes. For example, the verbs ask and prohibit lexicalize the intentional attitude 'S wants Η to do ρ /not to do p)', which contains the propositional attitude 'S wants ρ /not p ) \ The evaluation of ρ as something which is desirable/ undesirable for S is a genuine propositional attitude and only an indirect part of the speaker's intention. In those cases where the RS speaker's propositional attitude is evaluative, the evaluations expressed by the corresponding verbs are explicit. This is true of expressive speech act verbs like German loben!tadeln/kritisieren,

English

praise! blame! criticize

and

Dutch

prijzenlberispenlkritiseren (cf. Zillig 1982). Explicit evaluations can also be part of the discourse situation: they are evaluations by a DS speaker of a RS speaker's speech act. Evaluations like these are lexicalized by the following verbs: German angeben!sich aufspielen (English boast, Dutch opscheppen!bluffen)

and German schwindeln

and flunkern (English fib,

Dutch jokken) as opposed to lügen (English lie, Dutch liegen). Speech act verbs may also lexicalize implicit evaluations. These are expressed by utterances containing directive or commissive speech act verbs like German auffordern (English request, Dutch vragen) or German versprechen (English promise, Dutch beloven) (cf. Zillig 1982; Fries 1991). auffordern {request! vragen) is used to refer to situations where the speaker considers the action ρ desirable for him- or herself, while versprechen (promise, beloven) is used to refer to situations where the speaker considers the action ρ desirable for the hearer. Although these implicit evaluations are part of the preparatory conditions of the corresponding speech acts, they do not constitute lexical presuppositions. However, the verbs warnen! warn! waarschuwen are both exceptions to this general fact, because sentences involving these verbs presuppose a negative evaluation of an expectable event. On the whole, we get the following possiblities for evaluations lexicalized by speech act verbs:see figure 4. The semantic difference between speech act verbs lexicalizing evaluations as elements of a discourse situation and those lexicalizing evaluations as elements of a resource situation is that evaluations which are elements of a resource situation can be cancelled by a DS speaker, while such which are elements of a discourse situation cannot. Sequences of utterances such as the following seem to be quite coherent: (1) Hans hat Anton dafür getadelt, dass er ins kalte Wasser gesprungen ist, aber bei der Hitze war das das einzig Richtige. (2) John blamed Anthony for having jumped into the cold water, but in this heat it was just the right thing to do.

324

Gisela Harras and Kristel Proost

(3) Hans hat Anton dafür getadelt, dass er ins kalte Wasser gesprungen ist, dabei hat Anton nur ein Fußbad genommen. (4) John blamed Anthony for having jumped into the cold water, but he only took a foot bath.

The examples show that the DS speaker can - from his point of view cancel the supposed evaluation (in this case: 'jumping into the cold water is bad') as well as the presupposed event of Anthony's jumping into the cold water. We can conclude that, like the cognitive verbs glauben, believe, geloven, these expressive speech act verbs have two uses: a de re use where the DS speaker shares the attitudes of the RS speaker and a de dicto use where the DS speaker only reports the attitudes of the RS speaker without sharing them (cf. Lakoff 1970; Fillmore 1973; Wunderlich 1973).

Evaluations

Figure 4. Different kinds of evaluations

It is quite clear that supposed evaluations which are elements of the discourse situation cannot be cancelled by a DS speaker without getting into a kind of Moorean paradoxon. The following utterances are defective in much the same way as an utterance like The cat is on the mat, but I don't believe it is:

The lexicalization of speech act evaluations

325

(5) *Hans hat mächtig mit seinem Auto angegeben, aber ohne zu übertreiben (6) *John boasted enormously about his car, but without exaggeration

Apart from some rare examples such as those instantiated by warnen (English warn, Dutch waarschuwen), evaluations which are part of a RS speaker's presuppositions do not constitute lexical presuppositions. Verbs like German auffordern and bitten, English request and ask or Dutch vragen en verzoeken lexicalize the RS speaker's propositional attitude ,S wants: p' with the implicit positive evaluation ,p is in the interest of S', but the negation test for the corresponding lexical presuppositions fail, cf.: (7) Hans hat Anton nicht aufgefordert, die Tür zu schließen (8) John did not ask Anthony to close the door (9) Hans heeft Tony niet gevraagd de deur te sluiten

In all these cases, there is nothing left to be presupposed concerning the attitudes of a RS speaker: attitudes of speakers do not exist in the world before the corresponding speech act is performed. Rather, they are introduced by the utterance itself, and in the case that there is no request at all, there are no speaker attitudes either. As Searle has pointed out: speech acts create social facts, and this is reflected by the semantic behavior of speech act verbs. One definition of possible non-lexical presuppositions of sentences containing speech act verbs has been proposed by Chierchia and McConnell-Ginet. These authors define non-lexical presuppositions of sentences containing speech act verbs in terms of context-dependency (cf. Chierchia and McConnell-Ginet 1990): A sentence S presupposes Β relative to a set of premises Ρ iff S can be felicitously uttered in a context c in which premises Ρ hold only if c entails B. Ρ includes premises about human behaviour, elements of Gricean practical reasonings and other more specific premises about a state of affairs (cf. Kadmon 2001). The definition covers cases where a pragmatic, i.e. conversationally triggered, presupposition may arise in some contexts and fail to arise in others. The difference between conversationally triggered and lexical presuppositions is that lexical presuppositions such as those contained in sentences with verbs like aufhören (English stop, Dutch ophouden) contribute to the truth value of utterances, while conversationally triggered presuppositions do not. The utterance (10) John promised Bill to come to the party has the conversational implicature in (10'):

326

Gisela Harras and Kristel Proost

(10') Bill wants John to come to the party. This conversational implicative arises in such contexts in which (11) holds: (11) People perfom speech acts only if they take it for granted that the preparatory conditions of these hold true. (cf. Kadmon 2001). (11) determines the contexts in which (10') constitutes a pragmatic presupposition. The interpretation of (10) as containing (10') is a preferred, i.e. default, interpretation, but (10) may be true if (10') is not the case. (10') constitutes a generalized implicature, which is part of the presumptive meaning of the utterance (cf. Levinson 2001). The pragmatic account clearly shows that the preparatory conditions of speech acts do not manifest themselves as lexical presuppositions of the corresponding speech act verbs. The rare cases where the negation test for lexical presuppositions does not fail are the verbs warnen (English warn, Dutch waarschuwen) and ermahnen, cf.: (12) (13) (14) (15)

Hans hat Anton nicht vor dem Unwetter gewarnt John did not warn Anthony of the thunderstorm Hans heeft Tony niet voor het onweer gewaarschuwd Hans hat Anton nicht ermahnt, das Manuskript heute abzugeben

In the examples (12)-(14) the referent of the prepositional phrases vor dem Unwetter, of the thunderstorm, voor het onweer constitutes an existential presupposition, and therefore it cannot be cancelled by negation. In (15), Anton's obligation of doing something {Manuskript abgeben) is lexically presupposed as a result of a past request of Hans or some other person, and consequently, an implicit positive evaluation (,p is in the interest of Hans or some other person') is presupposed as well. Both the natural event of a thunderstorm and the social event of a past request are facts of the external world. They are independent of the performance of any act and therefore constitute real lexical presuppositions.

4. The lexicalization of evaluations by idiomatic speech act expressions Having shown that the evaluations expressed by speech act verbs are either explicit or implicit, we will now take a closer look at explicit evaluations, especially those lexicalized by idiomatic speech act expressions. We shall be concerned with the question of whether the evaluations lexicalized by idiomatic speech act expressions of the category 'verb' differ from those expressed by speech act verbs. Particularly, we shall ask ourselves whether there are specific types of evaluations which only idiomatic expressions can express and, if so, whether these are identical in different languages.

The lexicalization

of speech act evaluations

327

It has often been suggested that idiomatic expressions (i.e. lexical expressions showing a certain degree of idiomaticity such as idioms and collocations) express speaker attitudes, often including an evaluation, to the situation referred to (cf. Cernyseva 1984 and 1989, Dobrovol'skij 1988, Schindler 1993, Fleischer 1997). As we have shown in the first part of this paper, speech act verbs differ from most other semantic classes in that they express speaker attitudes, often including speaker evaluations, as an essential part of their meaning. If idiomatic expressions in general typically express speaker attitudes and evaluations as well, it is not clear how the meanings of such idiomatic expressions which refer to speech acts differ from those of speech act verbs. In spite of the wide-spread view that idioms and collocations typically express speaker attitudes, there are quite a few idiomatic speech act expressions which do not lexicalize any such attitude. This is especially true of such idioms and collocations making reference to conversational structure (e.g. broach the subject, canvass the idea (that)), a particular mode of communication (e.g. drop sb. a line) or a specific manner of speaking (e.g. German in Raten sprechen - meaning 'stutter'). Still, the majority of speech act idioms and collocations do express a certain speaker attitude, doing justice to the idea that the meaning of idiomatic expressions is characterized by a certain degree of "subjectivity". In the light of the fact that this holds true of speech act verbs too, the claim that idiomatic expressions typically express speaker attitudes at least requires some specification as far as speech act expressions are concerned. We shall illustrate what is specific about the meaning of idiomatic speech act expressions (as opposed to speech act verbs) by means of the lexicalization of the concepts (TO) LIE and (TO) BOAST.

4.1. Verbs and idiomatic expressions lexicalizing the concept (TO) LIE The evaluations expressed by idiomatic speech act expressions are primarily elements of the discourse situation. Examples are provided by the German word field which centres around the verb lügen {lie). The verb lügen {lie) does not by itself lexicalize any type of evaluation, though it does, of course, express a particular attitude on the part of the speaker. Specifically, the verb lügen!lie is used where a RS speaker does not take the proposition of his utterance to be true, while at the same time wishing that the RS hearer does take it to be true. This means that the attitude of the speaker towards the proposition is epistemic in nature, and not evaluative, as in the case of tadeln {reprimand). Neither does the meaning of lügen/lie involve any evaluative presupposition, the only presupposition of the speaker being that the hearer does not know ρ (cf. Figure 5):

328

Gisela Harras and Kristel

Proost

lügen (lie)

-

PROPOSITIONAL ATTITUDE: S does not take to be true: ρ SPEAKER'S INTENTION: S wants: Η take to be true: ρ SPEAKER'S PRESUPPOSITION: Η does not know: D

Figure 5. speaker attitudes lexicalized by liigen/lie (S-Speaker', H-Hearer',

p-proposition') The word field centring around the German verb lügen (lie, liegen) includes other verbs like flunkern and schwindeln (both equivalents of English fib) and may be expanded by the presence of idiomatic expressions. All of these inherit the speaker attitudes lexicalized by lügen, but unlike the verb lügen itself, they seem to lexicalize some kind of evaluation by the DS speaker of the speech act performed by the RS speaker. Particularly, these verbs and complex expressions lexicalize the DS speaker's evaluation of the gravity of the lie. When the DS speaker wants to emphasize the RS speaker's insincerity, he can choose from a wide range of options including lügen wie gedruckt ('lie as if it were printed'), lügen, dass sich die Balken biegen ('lie until the beams bend') and jmdm. die Jacke/die Hucke voll lügen ('lie until somebody's pockets are filled'). Additionally, it is possible for the DS speaker to specify the source of the gravity of the lie by focusing on single components of the idiom's meaning, such as the directness of the act of lying (jmdm. ins Gesicht lügen - lie sb. straight into the face) or the impossibility for the proposition to come true (das Blaue vom Himmel herunter lügen- 'lie the blue down from heaven'). Though attributes like 'specification of the proposition' or 'manner of speech act performance' are not by themselves evaluations, the values attributed to them (e.g. 'direct' or 'unrealistic') are. The images used in these idioms serve to emphasize the intensity of the act of lying. Apparently, acts of lying which are judged to be particularly serious cannot be referred to by verbs. However, in such cases where the DS speaker considers the lie to be of a rather more facetious nature, his choice of expressions to refer to the corresponding act ranges from verbs like flunkern and schwindeln (both meaning 'fib') to true idioms like jmdn. auf den Arm nehmen ('to take sb. onto one's arm'). In these cases too, the proposition may be specified by the use of a particular image which emphasizes the innocence of the lie (e.g. ein Xfür ein U - 'X instead of U'), but apparently, only idioms and not verbs are used to express this kind of specification. Whenever the act of lying is judged as neither particularly severe nor particularly innocent, both verbs and idiomatic expressions may be used (e.g. lügen - lie and jmdm. einen Bären aufbinden).

The lexicalization of speech act evaluations

329

Summarizing what has been said so far, a DS speaker describing an act of lying can basically choose between words and idiomatic expressions, except in such cases where he believes the RS speaker to be particularly insincere. In such cases, the DS speaker can only choose from a range of idiomatic expressions (cf. Figure 6). lügen (lie) and related idiomatic expressions (German)

Intensification

Neutral

Attenuation

{lügen wie gedruckt, lügen, dass sich die Balken biegen, jmdm. die Jacke/die Hucke voll lügen}

{lügen, jmdm. einen Bären aufbinden, jmdm. etw. weismachen}

{flunkern, schwindeln, jmdn. auf den Arm nehmen)

[+Manner] (directly)

^Specification of ρ] (ρ unrealistic)

[+ Specification of ρ] (ρ unrealistic)

{jmdm. ins Gesicht lügen}

{das Blaue vom Himmel herunter lügen}

{jmdm. blauen Dunst vormachen, jmdm. ein Xfür ein U vormachen}

Figure 6. Degrees of intensity lexicalized by lügen (lie) and related idiomatic expressions (German) - (p-proposition1; H-Hearer') A comparison of Figure 6 with Figures 7 and 8 shows that, on the whole, the lexical field around lie yields the same picture in English and Dutch as it does in German: intensifications of the meaning of lie are lexicalized by idiomatic expressions, while such lies which are judged to be less innocent or neither innocent nor serious may be referred to by verbs and idiomatic expressions alike. Single conceptual components, such as the manner in which the act of lying is performed or some kind of specification of ρ (the proposition), may be lexicalized as well. In spite of these similarities, the three languages differ with respect to the values assigned to attributes like 'Manner' or 'Specification of p \ German, for example, does not seem to allow the lexicalization of the value 'harmless' for the parameter 'Specification of p'. However, the various degrees of intensity are lexicalized in all three languages, and intensifications of the meaning of lie are consistently lexicalized by idioms, not by verbs. This is also true of

330

Gisela Harras and Kristel Proost

specifications of the proposition, no matter whether these emphasize the severity or the innocence of the lie (cf. Figures 7 and 8). lie and related idiomatic expressions (English)

Intensification

Neutral

{lie, tell sb. a lie}

[+Manner]

Attenuation

{fib, tell sb. a fib}

[+Specification of p] (p harmless)

{lie sb. flat/straight into the face, tell sb. a barefaced lie, lie one's head o f f )

{tell sb. a white lie}

Figure 7. Degrees of intensity lexicalized by lie and related idiomatic expressions (D-Dr0D0siti0n': H-Hearer') liegen (lie) and related idiomatic expressions (Dutch)

Intensification

{liegen dat je barst/ dat je zwart ziet/ dat je het zelf gelooft/ alsof het gedrukt Staat}

Neutral

Attenuation

{liegen, iem.iets wijsmaken, iem. iets op de mouw Speiden}

Figure 8. Degrees of intensity lexicalized by liegen {lie) and related idiomatic expressions (Dutch)

{jokken}

The lexicalization of speech act evaluations

331

Bearing this pattern in mind, we shall turn the concept BOAST and the way it is lexicalised in German, English and Dutch. 4.2. Verbs and idiomatic expressions lexicalizing the concept (TO) BOAST Unlike the verb lügen (lie, liegen), verbs like angeben, prahlen, protzen and aufschneiden (English boast, brag and vaunt) do lexicalize speaker evaluations. Some of these are elements of the resource situation. Verbs like angeben/boast are used to refer to the behavior of a RS speaker who evaluates his own past actions positively and also wishes that the RS hearer recognize this. Thus, a sentence containing a verb like angeben/boast supposes a positive evaluation of the proposition (p) by the RS speaker. At the same time, the RS speaker presupposes ρ to be the case, reflecting the fact that angeben is a factive verb (see Figure 9). angeben (boast) -

PROPOSITIONAL ATTITUDE: S evaluates: ρ good SPEAKER'S INTENTION: S wants: Η recognize: ρ good SPEAKER'S PRESUPPOSITION: ρ is the case

Figure 9. Speaker attitudes lexicalized by angeben/boast (S-Speaker', H-Hearer', p-propositiori)

Other evaluations lexicalized by verbs like angeben are elements of the discourse situation. A A DS speaker uses these verbs to indicate that he considers the RS speaker's positive evaluation of his own past actions to be exaggerated. Like the RS speaker's positive evaluation, the negative evaluation by the DS speaker is part of the lexical meaning of boast. All of the idiomatic expressions lexicalizing the concept (TO) BOAST inherit both types of evaluations expressed by verbs like angeben. As Figures 1 0 - 1 2 show, English, Dutch and German all possess idiomatic expressions which may be seen as synonyms of verbs like boast and do not lexicalize a degree of intensity significantly different from that expressed by the corresponding verbs. Moreover, attenuations of the concept (TO) BOAST do not seem to allow for lexicalizations in any of these languages. German permits the lexicalization of intensifications of (TO) BOAST, while intensifications of (TO) BOAST are not as clearly marked in English and Dutch. In German, clear intensifications of the meaning of angeben are

332

Gisela Harras and Kristel Proost

expressed only by idiomatic expressions, meaning that these German idiomatic expressions lexicalize an evaluation of degree apart from the evaluations which they inherit from the verb angeben itself. Clearly, evaluations of degree are also elements of the discourse situation: the DS speaker considers the RS speaker's evaluation of his own past actions to be far too positive. angeben {boast) and related idiomatic expressions (German)

Intensification

{angeben wie ein Wald voll Affenlwie eine Tüte Mücken, kräftig/mächtig ins Horn stoßen}

Neutral

Attenuation

{angeben, prahlen, protzen, aufschneiden, sich in die Brust werfen, den Mund (zu) voll nehmen, große Reden schwingen, eine Stange angeben}

Figure 10. Degrees of intensity lexicalized by angeben and related idiomatic expressions (German) boast and related idiomatic expressions (English)

Intensification

Neutral

Attenuation

{boast, brag, show o f f , blow one's own trumpet, give oneself airs}

[+Specification of P] (Contents)

{air/parade one's knowledge} Figure 11. Degrees of intensity lexicalized by boast and related idiomatic expressions (English)

The lexicalization of speech act evaluations

333

opscheppen {boast) and related idiomatic expressions (Dutch)

Intensification

-

Neutral

Attenuation

{bluffen, opscheppen, zieh op de borst slaan/ kloppen, een hoge borst opzetten, dik doen}

-

Figure 12. Degrees of intensity lexicalized by opscheppen and related idiomatic expressions (Dutch)

Here again, intensifications of evaluative meaning components are lexicalized by idiomatic expressions if they are lexicalized at all. This means that, on the whole, the lexicalization of the concept (TO) BOAST seems to follow the same pattern as that of (TO) LIE. This pattern surfaces in other speech act word fields, especially in the sets of lexical expressions lexicalizing speech acts involving a negative evaluation by the RS or the DS speaker. Negative evaluations are obviously involved in concepts like REPRIMAND, CRITICIZE and BOAST, but they may also come into play when a DS speaker comments on acts of lying, promising or requesting. In cases like these, the DS speaker may express negative evaluations like his estimation of the gravity of the act of lying, the improbability that S will keep his promise, or the obtrusiveness of requesting acts. Idiomatic expressions typically lexicalize intensifications of such negative evaluations. Concepts of communication involving a positive evaluation (e.g. PRAISE) do not seem to be affected by this pattern (cf. Proost 2001a and 2001b). This does not mean though that verbs cannot lexicalize intensifications or evaluations by a DS speaker. However, when this is possible (e.g. German ausfragen- ausquetschen (wie eine Zitrone), English question - pump, Dutch uitvragen), there are idiomatic expressions lexicalizing an even higher degree of intensity (e.g. German jmdn. in die Mangel nehmen - jmdm. Daumenschrauben anlegen, Dutch iemand op de roosterlde pijnbank leggen - iem. de duimschroeven aandraaien, English put the srews on sb.). In cases like these, idiomatic expressions seem to cluster around the extreme negative end of the evaluation scale. This pattern holds for those idiomatic speech act expressions which can be classified as belonging to one of the traditional speech act classes.

334

Gisela Harras and Kristel Proost

5. Conclusion Our observations concerning the lexicalization of speech act evaluations in English, Dutch and German suggest that all types of explicit and implicit evaluations are represented in each of these languages. Both types of evaluations may be lexicalized by speech act verbs. Moreover, the set of expressions we can choose from to lexicalize a particular speech act can be expanded by idiomatic expressions. Idiomatic expressions lexicalize explicit evaluations. In each of the languages under consideration, idiomatic expressions as well as verbs lexicalize degrees of intensity, while intensifications, specifications of ρ and modes of speech act performance are primarily if not uniquely lexicalized by idiomatic expressions. Intensifications and attribute values like 'unrealistic' and 'direct' are specific to idioms. The three languages considered here differ in the way in which ρ is specified or in the particular manner of speech act performance. Not all types of specifications and modes of performance which are possible in principle are realized in each of the three languages investigated. The same is true of degrees of intensity: not all degrees which may be lexicalized in one language need automatically be lexicalized in other languages as well.

Abbreviations S Η Ρ

RS DS SRS HRS PRS SDS HDS PDS

Speaker Hearer Proposition Resource Situation Discourse Situation Resource Situation Speaker Resource Situation Hearer proposition of the RS speaker's speech act Discourse Situation Speaker Discourse Situation Hearer proposition of the DS speaker's speech act

References Barwise, Jon and John Perry 1983 Situations and Attitudes. Cambridge/MA: The MIT Press. CernySeva, Irina J. 1984 Aktuelle Probleme der deutschen Phraseologie. Deutsch Fremdsprache 21 (1): 17-22.

als

The lexicalization of speech act evaluations

335

Cernyseva, Irina J. 1989 Strukturtypologische Phraseologieforschung in der sowjetischen Germanistik: Überblick und Ausblick. In Europhras 88: Phraseologie Contrastive. Actes du Colloque International Klingenthal-Strasbourg. 12.-16. Mai 1988, Gertrud Greciano (ed.). 4 3 ^ 9 . Strasbourg: Universite des Sciences Humaines. Chierchia, Gennaro and Sally McConnell-Ginet 1990 Meaning and Grammar. Cambridge/MA: The MIT Press. Dobrovol'skij, Dmitrij 1988 Phraseologie als Objekt der Universalienlinguistik. (Linguistische Studien). Leipzig: VEB Verlag Enzyklopädie. Fillmore, Charles 1973 Verbs of Judging: An Exercise in Semantic Description. In Präsuppositionen in Philosophie und Linguistik - Presuppositions in Philosophy and Linguistics, Jänos S. Petöfi and Dorothea Franck (eds.), 261-283. Frankfurt/Main: Athenäum Verlag. Fleischer, Wolfgang 1997 Phraseologie der deutschen Gegenwartssprache. 2., durchgesehene und ergänzte Auflage. Tübingen: Niemeyer. Fries, Norbert 1991 Bewertung: Linguistische und konzeptuelle Aspekte des Phänomens. Sprache und Pragmatik. Arbeitsbericht 23. Lund. 1-31. Harras, Gisela 1995 Eine Möglichkeit der kontrastiven Analyse von Kommunikationsverben. In Von der Allgegenwart der Lexikologie: Kontrastive Lexikologie als Vorstufe zur zweisprachigen Lexikographie, Akten des internationalen Werkstattgesprächs zur kontrastiven Lexikologie, 29 — 30.10.1994, Hans-Peder Kromann and Anne Lise Kjasr (eds.), 102113. (Lexicographica Series Maior; 66.) Kopenhagen. 2001 Performativität, Sprechakte und Sprechaktverben. In Kommunikations verben: Konzeptuelle Ordnung und semantische Repräsentation, Gisela Harras (ed.), 9-31. (Studien zur deutschen Sprache 24.) Tübingen: Narr. Harras, Gisela and Edeltraud Winkler 1994 Α Model for Describing Speech Act Verbs: The Semantic Base of a Polyfunctional Dictionary. In EURALEX 1994: Proceedings. Willy Martin et al. (eds.), 440-448. Amsterdam. Kadmon, Nirit 2001 Formal Pragmatics. Oxford: Blackwell. Lakoff, John 1970 Linguistics and Natural Logic. Ann Arbor. Levinson, Stephen C. 2001 Presumptive Meanings'. The Theory of Generalized Conversational Implicature. Cambridge/MA: The MIT Press.

336

Gisela Harras and Kristel Proost

Proost, Kristel 2001 a Zum Lexikalisierungsraum kommunikativer Konzepte. In Kommunikationsverben: Konzeptuelle Ordnung und semantische Repräsentation, Gisela Harras (ed.), 77-129. (Studien zur deutschen Sprache 24.) Tübingen: Narr 2001b Warum LÜGEN & Co., aber nicht WAHRSAGEN? Asymmetrien im Wortschatz. Sprachreport: Informationen und Meinungen zur deutschen Sprache. 2001 (1): 14-17. Searle, John R. and Daniel Vanderveken 1985 Foundations of Illocutionary Logic. Cambridge: Cambridge University Press. Schindler, Wolfgang 1993 Phraseologismen und Wortfeldtheorie. In Studien zur Wortfeldtheorie - Studies in Lexical Field Theory. Peter Rolf Lutzeier (ed.), 87-106. Tübingen: Niemeyer. Vanderveken, Daniel 1990 Meaning and Speech Acts. Cambridge: Cambridge University Press. Vanparys, Johan 1996 Categories and Complements of Illocutionary Verbs in a Cognitive Perspective. (Duisburger Arbeiten zur Sprach- und Kulturwissenschaft 26.) Frankfurt/Main: Lang. Verschueren, Jef 1980 On Speech Act Verbs. (Pragmatics & Beyond 4.) Amsterdam: John Benjamins. 1985 What People Say They Do with Words: Prolegomena to an EmpiricalConceptual Approach to Linguistic Action. (Advances in Discourse Processes 14.) Norwood/NJ: Ablex Publishing Corportion. Wunderlich, Dieter 1973 Präsuppositionen in der Linguistik. In Präsuppositionen in Philosophie und Linguistik - Presuppositions in Philosophy and Linguistics, Janos S. Petöfi and Dorothea Franck (eds.), 467-484. Frankfurt/Main: Athenäum Verlag. Zillig, Werner 1983 Bewerten: Sprechakttypen der bewertenden Rede. Tübingen: Niemeyer.

Index

absolute, 251, 257-260, 264, 265, 267-270 accounts or elaborations, 300 adverbial, 262 Albanian, 12 Alfonzetti, Giovanna, 19, 36 Andersen, Stig Toftgaard, 312, 314 Andersson, Anders-Börje, 245-247 Andres, Juan, 42 angeben, 323, 331 anger, 195-208 anger is fire, 195, 196, 200-206 anger is heat, 195-198, 203, 205, 206 anger is the heat of a fluid in a container, 195-197, 200, 202, 203, 205, 206 Anglo-Saxon, 178, 183, 192, 193 animacy, 236, 248 ANS, 253-256, 268,270, 271 anteriority, 270 Arabic, 9, 33 Aragonese, 13, 14, 19, 30 Armstrong, Nigel, 104, 109 Arnason, Kristjan, 36, 37, 42 Asturian, 30 Auer, Peter, 7, 19, 22, 36, 37, 40-42, 124,125 Austria, 20, 24, 25, 218 auxiliaries, 254-256, 266, 268, 269, 273 auxiliary, 256, 257, 274

Bakema, Peter, 109 Baker, Carl L., 293, 294, 299, 314 Balearic Islands, 13 Bartsch, Renate, 224, 225, 232 Barwise, Jon, 319, 334 Basque, 8, 11, 13 Bavarian, 12, 13

be going to, 251, 252, 253, 257, 258, 266, 267, 269, 270, 272 Behaghel, Otto, 306, 314 Beheydt, Ludo, 251 Belemans, Rob, 111, 125 Belgian Dutch, 304 Belgianism, 271 Bellmann, Günter, 22, 37 Belorussian, 30 Benecke, Georg, 306, 314 Bergin, Thomas G., 155, 172 Berruto, Gaetano, 19, 20, 22, 32, 33, 35,37 Bible, 168, 175 Bierwiaczonek, Boguslaw, 170, 172 Binnick, Robert, 253, 270, 272 Bittner, Dagmar, 246, 247 Blancquaert, Edgard, 114, 126 Blank, Andreas, 307, 314 blood, 195, 196, 201-204 Blount, Thomas, 157, 159, 173 bluffen, 323 boast, to, 327, 331, 333 Böövarsson, 312, 315 Bokmal, 15, 17, 20, 24, 234, 235, 246 Booij, Gert, 94, 108, 109 borrowing, 291. see language contact Bosnian, 26 Braunmüller, Kurt, 245, 248 Bree, Cor van, 130, 141 Bretonic, 11 Brisard, frank, 257, 258, 272 Brisau, Andre, 270, 272 Brown, Penelope, 131, 141 Bücherl, Rainald, 35, 37 Bulgaria, 12, 14, 25, 26, 33 Bulgarian, 13, 14,21,32,33,42 Bullokar, John, 157, 173 Burnley, J. David, 154, 173 Bybee, Joan, 229,232, 271, 272

338

Index

Byzantium, 14 C'est, 275-277, 279, 281, 282, 284, 285,288,289 capital, 18, 20,21 Casali, Roderic F., 223, 232 Castilian, 13, 14, 20, 21, 24 Catalan, 13, 21, 24, 25 Celtic, 11 CernySeva, Irina J., 327, 334, 335 change historical, 147 Chase, Thomas, 158, 173 Chatfield, Christopher, 239, 248 Cheshire, Jenny, 34, 37, 39 Chickasaw, 196 Chierchia, Gennaro, 325, 335 Chinese, 196, 197, 208 cleft, 275-278, 280-289 clivees, 289, 290 Close, R.A., 253, 258, 272 Clyne, Michael, 101, 104, 109 Coates, Jennifer, 125, 126 Cockeram, Henry, 143, 173 code-switching, 15, 16, 19, 20, 23, 36, 37 coherence bridge, 296 Coleman, Julie, 143-146, 148-150, 152-157, 159-161, 163-169, 171-174, 176 Colleman, Timothy, 255, 265, 270, 272 Collins, Alec, 239, 248 conceptualization, 143, 195, 196, 198, 199, 200, 201,202, 203, 204, 205, 206, 207 concessive conditionals, 300, 302, 304 concord, 280, 283, 287, 288 conditional, 271 conditional schema, 294 Continental Scandinavian, 291, 310 contrast, 283, 286, 288 convergence, 111, 120, 124, 125 Corbett, Greville G., 236, 246-248

corpus, 252, 257, 258, 261, 264, 268-271 Corpus Gesproken Nederlands, 252 Corpus Spoken Dutch, 252, 255, 256, 258, 259, 265, 268, 269 Coseriu, Eugenio, 7, 30, 38 Croatian, 12, 26 cross-cultural, 197, 207 cross-linguistic, 195, 196, 197, 205, 206 Crystal, David, 270, 272 Csernicsko, Istvan, 36, 38 cultural influence, 195, 197-199, 203-206 Curme, George G., 216, 225, 232 Curry, Walter C., 203, 204, 207 Cyprus, 11, 17, 25,32,39 Czechia, 11, 14 Dahl,Osten, 245, 246, 248 Danish, 15, 17, 18, 20, 34, 35, 39, 41,235-246, 249, 250,310 Day, Mabel, 200, 201, 207 De Rooij, J., 255, 265, 270, 273 De Schryver,Johan, 255, 258, 268, 270, 273 De Schutter, Georges, 115, 122, 126 Declerck, Renaat, 251, 253, 257, 260, 269-272 Decroos, Bram, 270, 273 Delveroudi, Rea, 35, 38 demotiki, 18, 35 Denmark, 17, 27, 34, 36, 39 Depraetere, Ilse, 269 diachronic, 195, 198, 203-206 dialect, 7, 8, 10, 12, 13, 15-17, 1923,25,27-36, 39,41,42 loss, 27 primary dialects, 30 dialect change, 111, 124, 127 dialect loss, 111, 124, 127 diglossia, 9, 12, 14, 15, 17, 19, 20, 22, 23, 29, 35 Außendiglossie, 13 Binnendiglossie, 13, 33

Index Dingeldein, Heinrich, 36, 38 discourse, 319, 320, 323, 324, 327, 331,332 discriminant function analysis, 237, 239-243 Dobrovol'skij, Dmitrij, 327, 335 domain, 144, 251 Donne, John, 156, 173, 175 Dressier, Wolfgang U., 209, 213, 215,233 Droste, Frederik G., 271, 273 Duden, 222, 225, 233 dummy subject 'it', 305 Dutch, 129, 301-305, 319, 323, 325, 326, 329, 331,333,334 Early Modern High German, 305 East-Flemish dialects, 116, 117 Eisenberg, Peter, 213, 233 Ekman, Paul, 196, 197, 205, 207 elicited speech, 114 ellipsis, 299 embedded interrogative, 291 embodiment, 195-198, 201-206 emergence, 300 emotion mild, 163 strong, 163 emphasis, 278 endoglossic, 8, 9, 11-14, 17, 24, 31, 33,34 Enger, Hans-Olav, 245, 246, 248 England, 7, 8, 14, 17, 25, 29 English, 11, 12, 14, 17, 20, 33,34, 37, 39, 40, 177-181, 183-185, 187-189, 192, 195-200, 202, 205-208, 292, 301,302,304, 305,319, 323,325,326, 328, 329,331,333,334 epistolary, 129, 131, 132, 135, 140 Erdmann, Peter, 295, 315 Estonia, 21 Europe, 7-11, 14, 15, 17, 19, 20, 22, 24, 30,31,40,41 Europeanism, 312

339

exoglossic, 7, 9-14, 17, 24, 33, 34 experience, 143 extract, 276, 284, 285 extraposition, 295 Fabiszak, Malgorzata, 206, 207 factual, 261 factuality, 260, 261, 264 Ferguson, Charles, 9, 12, 38 Fery, Caroline, 213, 222, 233 Fillmore, Charles, 324, 335 Finland, 11, 14, 25, 34,41 Finnish, 11, 15, 17,34,41 Flanders, 11, 27 Fleischer, Wolfgang, 327, 335 Fleischman, Suzanne, 253, 273 Flemish, 11,24 focalisation markers, 283, 288 focus, 276, 278 folk theory, 197, 203, 206, 207 foot structure, 219, 220, 223, 231 Ford, Cecilia E., 300, 315 forms of address, 129, 130, 132, 133, 137 France, 8, 11,20, 25,27,36,38 Fräser, 247, 248 French, 9, 11, 16, 19, 20, 27,36, 183, 188,275-277, 278-281, 283,287, 288,301,304 Fries, 323, 335 Friesen, Wallace V., 196, 197, 205, 207 Frings, Theodor, 10, 38 Fromm, Erich, 149, 173 future tense, 251, 252, 255-258, 263, 264,266-271 gaan, 251-259,266-269, 272-274 Gadet, Franfoise, 27, 32, 36, 38 Galicia, 26 Galician, 13, 21 Geeraerts, Dirk, 38, 177, 182, 185, 190, 192, 195, 197, 198, 207 Geerts, Guido, 111, 126, 273

340

Index

gender, 235, 236-239, 241-243, 245-250 German, 8-16, 18, 21, 24, 25, 29, 33, 34, 37, 209,210,213,214, 216-219, 221-228,230-233, 302-305,319, 323,325,327329,331,333,334 Early Modern High, 305 Middle High, 305, 306 Germanic, 200, 202, 291 Germany, 7, 17, 18, 20, 21, 23, 24, 25, 27, 29 Gevaert, Caroline, 195, 198, 202, 206, 207 Gibraltar, 11 Gilles, Peter, 9, 32, 38 Gilman, Α., 131, 141 Goeman, Anton, 125, 126 Golston, Chris, 213, 233 Goossens, Jan, 115, 117, 122, 126, 206, 207 Graedler, Anne-Line, 245, 246, 248 grammatical category, 284 grammatical function, 283, 288 grammaticalization, 252, 256, 257, 299, 308 Gray, Andrew, 156, 173 Great Britain, 11 Greece, 8, 11, 14, 29, 36 Greek, 10, 12, 14, 17, 18, 20, 25, 31, 35,38,39 Greenberg, Joseph, 236, 248 Grondelaers, stefan, 25, 38, 195, 197, 198, 207 Guitart, Jorge M., 288, 289 Haegeman, Liliane, 253, 266, 270, 273 Haeseryn, Walter, 253, 254, 265, 268,270, 271, 273 Hägen, Toon, 35, 38 Hamblin, Charles Leonard, 292, 315 Hansen, Erik, 245, 246, 249 Harras, Gisela, 319, 320, 335

Haspelmath, Martin, 233, 300, 312, 314,315 Haugen, Elmar, 8, 38 hearer, 319, 321-323, 327, 331 Hearer-old, 297 Hellemans, G., I l l , 126 hendida, 275 hendidas, 289 Herrgen, Joachim, 25, 38 Herring, Susan C., 297, 315 hierarchical downgrading, 300 Hinskens, Frans, 123-126 Hippocrates of Kos, 202 historical present, 271 Historical Thesaurus of English, 198 Holland, 129, 131, 133 Hooft, P.C., 132, 134, 135, 140, 142 Hopper, Paul J., 300, 315, 316 Horecky, Jan, 41 Horn, 292,315 Hornblower, Simon, 155, 156, 173 Hronek, Jiri, 41 Hughes, Geoffrey, 170, 173 humoral doctrine, 198, 202-204, 206 Hungarian, 11, 18, 36, 38, 39, 196 Hungary, 30 Icelandic, 312 idiomatization, 291, 308 indifference, 291 indifference, predicate of, 291, 292 inflection, 209-211,213,215,216, 218,220, 225-227, 230, 233 information, 289 informational structure, 286, 288 intention, 253 intentional, 255, 257, 261 intentionality, 253, 267 interdialectal exchanges, 123 internationalism, 312 interrogative, embedded, 291 inverted pseudo-clefts, 275 Ireland, 8, 12, 14, 39 Irish, 12, 14, 33, 36, 39

Index Italy, 7, 12, 17, 19, 20,21,25,35, 37,41 Ivars, Anne-Marie, 26, 38 Japanese, 196, 197, 207 Jarvad, Pia, 246, 249 Jaspaert, Koen, 111, 126 Johansson, Stig, 246, 248 Johnson, Marki 147, 149, 155, 170, 171, 174, 175 Jones, W.T., 171, 173 Jonsson, Jon H., 312, 315 Jutland, 27 Kadmon, Nirt, 325, 326, 335 Kalkar, Tho, 310, 315 Kallen, Jeffrey L„ 32, 33, 39 Källström, Roger, 245, 246, 249 Karnein, Alfred, 154, 174 Karttunen, Lauri, 292, 293, 315 Karyolemou, Marilena, 17, 25, 39 katharevousa, 14, 18, 34 Kay, Christian J., 146, 147, 170, 173,174,176 Keevallik, Leelo, 21, 39 Keinänen, Nely, 154, 174 Kilarski, Grzegorz, 235, 245, 249 Kirsner, Robert, S., 273 Kloss, Heinz, 8, 13, 33, 39 koine, 9, 20, 24,31,35 koineisation, 10 Koivisto-Alanko, Paivi, 168, 174 Koller, Erwin, 305, 316 König, Ekkehard, 300, 312, 314-316 Kontra, Miklos, 18, 32, 36, 39 Koopman, William, 253, 274 Kopeke, Klaus-Michael, 214, 215, 229, 233 Kövecses, Zoltan, 147, 148, 152, 153, 161, 163, 164, 166, 172, 174, 195-198, 205, 207 Kress, Bruno, 312, 316 Kristensen, Kjeld, 27, 34, 35, 39 Krynicki, Grzegorz, 235 Krzysko, Miroslaw, 239, 249

341

Kuhn, Hans, 245, 249 Lahiri, 314, 316 Lakoff, John, 146, 147, 149, 155, 165, 170, 171, 174, 195, 196, 198, 206, 207, 324, 335 language contact, GermanScandinavian, 309 Lanstyak, Istvan, 36, 39 Latin, 9, 17, 199, 200, 206 Leech, Geoffrey N., 253, 273 Lehmann, Christian, 307, 316 Leonese, 13, 14, 19, 30 Letzebuergesch, 9, 11, 13 Leuschner, Torsten, 291, 297, 300, 309,310,316 levelling, 111, 121, 123 Levenson, Robert W., 196, 197, 205, 207 Levinson, Stephen C., 131, 141, 326, 335 Lewis, C.S., 144, 149, 154, 163, 170, 171, 174, 192 Lexer, E.R., 305, 306, 316 lexical fields, 190 lexical renovation, 291, 305 lexicalization, 291, 307, 319 lexicon, 93, 95 lie, to, 327, 333 liegen, 323,328,331 loanwords, 235-237, 245, 246, 249 loss of paradigmatic variability, 308 love, 143 concept of, 143 intensity of, 163 of animals, 160 Romantic, 145 Spiritual, 146 lügen, 323, 327, 328,331 Luxembourg, 8, 11, 20 Macedonia, 26 Macfarlane, Alan, 162, 175 Mackridge, Peter, 18, 39 Maddalon, Marta, 19, 41

342

Index

Malta, 11 Marcus, Gary F., 217, 233 marked plural forms, 209, 228 Marie, Jaap van, 16, 39 Martin Zorraquino, Maria Antonia, 19, 33, 39, 40 Martinez, Jose Α., 279, 289 Matsuki, Keiko, 197, 207 maxims, 297 Mayerthaler, Willi, 209, 233 Mazur, Jan, 21, 24, 36, 40 Mbense, Thandi G., 196, 208 McCarthy, John, 219, 233 McConnell-Ginet, Sally, 325, 335 Mcintosh, Angus, 274 merger, 177, 178, 180, 182-187, 189, 190 Meritt, Herbert Dean, 192 metaphor, 195-197, 199-207, 301, 305 metaphorical, 259, 260 metaphtonym, 201, 203 metonym, 195-197, 199-206 metonymy, 167 MHG, 209,211,228, 231 Middle English, 183, 185, 192, 198, 199 Middle High German, 305, 306 migration, 11,31 Mihm, Arend, 23, 24, 34,40 Millet, Bella, 200, 208 Milroy, Jim and Leslie, 31, 34, 37, 40 Minangkabau, 197, 207 minority languages, 8 modal, 254, 255, 266, 269, 270, 271 modality, 283 Moeschler, Jacques, 314, 316 Moldavia, 11, 26 Molina, Clara, 177, 192 Moosmüller, Sylvia, 35, 36, 40 Moreau, Marie -Louise, 276, 289 Moreno Cabrera, Juan Carlos, 288, 289 Moser, Hugo, 33,40

Müller, Wilhelm, 306, 314 Multivariate Analysis of Variance, 237 narrative, 271 nation-building, 8, 20 native speakers, 287 Natural Morphology, 209, 231, 233 Naumann, Hans-Peter, 311,316 Nekvapil, Jiri, 32, 33, 40 Netherlands, 129, 132, 133 Nevalainen, Terttu, 170, 175 NHG, 209, 211, 213, 217, 228, 231 Nickisch, Reinhard M.G.142 Niekerk, P.K., 274 Nishigauchi, Taisuke, 301, 316 NM, 209, 213, 215, 228, 229, 231 Nordberg, Bengt, 32, 35, 40, 41 Norway, 14, 15, 17, 20, 24, 31, 41 Norwegian, 11, 17, 18, 34, 235-237, 239-243, 245, 246, 248-250 novel, 252 Noyes, Gertrude E., 143, 175 Nynorsk, 15, 34 Obligatory Contour Principle, 218 Old English, 177-179, 184, 185, 187-200 onomasiology, 177 Onset, 222-224, 227, 228, 230, 231 opscheppen, 323 Optimality Theory, 209, 213, 233 oral language, 129, 140 Oral language, 130 Origo-identification, 297 Osborn, 175 OT, 209, 213, 214, 218, 219, 221, 230 Oxford English Dictionary, 145, 175 Pagliuca, William, 271, 272 pain, 183-191 Pajusalu, Karl, 21, 39 Palmer, Frank R., 253, 274 paradigmatic variability, loss of, 308

Index Parry, John Jay, 154, 175 participant, 143 partitioning, 294 Paul, Hermann, 210, 212, 214, 218, 225,233 Paunonen, Heikki, 34, 41 Pavlov, Vladimir, 231, 232, 233 Pedersen, Inge Lise, 17, 32, 34-36, 41 Pee, Willem, 114, 126 Perkins, Revere D., 271, 272 Perry, John, 319, 334 Perryman, Judith, 201, 208 personal pronouns, 112, 116, 126, 127 Petersson, Rikke, 317 Phraseologiemodell, 310 Pinch, Adela, 171, 175 Plato, 169, 171, 175 plural marker, 210-214, 216, 217, 223, 225, 228, 229 Poland, 21,24, 25, 27, 36 Polish, 306 Popovici, Vasile, 297, 317 Portugal, 26 Portuguese, 13, 21 posteriority, 270 Postma, Gertjan, 313, 317 pragmatic, 252, 253, 257, 259, 260, 270,273 predicate of relevance, 293 predicates of indifference, 292 predicator-complement order, 294 predictable, 253 predictions, 254, 256 present perfect, 270, 271 present tense, 251, 252, 255, 256, 258-271 presupposition, 325-327 Prince, Ellen F., 219, 233, 297, 317 principle of complementary weight, 295 progressive, 257-261, 263, 264, 269, 271 Proost, Kristel, 319, 333, 336

343

proposition, 319, 327-329, 331, 334 prototype theory, 177, 192 pseudo-cleft, 275-278, 280, 281, 283-285, 287, 288 pseudo-clivee, 275 pseudo-hendida, 275 pseudo-hendida in versa, 275 quantificational prefix, 301, 302, 304 Radden, Günter, 172, 175 real time, 114 regiolect, 22, 23, 25, 27, 36 regional, 252, 258, 259, 265, 268 register, 258, 259, 265, 268, 291 Reigersberch, Maria van, 134, 136— 140, 142 relative, 270 relative clause, 271, 278, 280 renovation, lexical, 291 Reto-Romance, 8,11 Rhetorical Dialogues, 296 Rice, Curt, 247, 249 Roberts, Jane, 174 Rogge, H.C:, 142 Romani, 11 Romania, 8, 11, 26, 33, 36, 37 Romijn, K., 273 Reyneland, Unn, 30, 32, 34, 41 Russia, 25, 29, 33 Russian, 11, 13, 20, 30, 33 Ryckeboer, Hugo, 122, 126, 127 Salenius, Maria, 168, 172, 175 Sandey, Klara, 17, 34-36, 41 Saxonian, 10, 18 Saxony, 18, 27, 30 Scandinavian, 235, 237, 240, 242, 246, 249 Scandinavian, Continental, 310 Schieb, Gabriele, 306, 317 Schindler, Wolfgang, 327, 336 Schmid, Hans-Jörg, 170, 176 Schottmann, Hans, 312, 317 Searle, John, 320, 321, 325, 336

344

Index

Sedano, Mercedes, 288, 289 Selten, Bo, 246, 249 semasiological profiles, 177, 178, 183, 184 semasiology, 177 Serbia, 11 Serbian, 26 Sgall, Peter, 33,34,41 Shannon, Laurie, 151, 168, 175 Siebenhaar, Beat, 34, 41 simultaneity, 270, 271 Skeat, Walter, 204, 208 Slavic, 10, 30, 32 Slavonic, 9, 11, 14, 30,32,33 Sobrero, Alberto, 19, 41 sociolinguistic, 252, 258, 265, 268 Solms, Hans-Joachim, 214, 234 sore, 177, 179, 180, 184-190 S0rensen, Knud, 246, 249 sorrow, 177, 178, 182, 183, 185-190 Spanish, 10, 11, 13, 14, 20, 30, 275, 276, 278, 279-281, 285, 287289 Spawforth, Anthony, 155, 156, 173 Speake, Jennifer, 155, 172 speaker, 319-325, 327-329, 331334 specification, 277 speech act, 319-328, 333, 334 speech act idioms speech act, idioms, 319, 327 speech act verbs speech act, verbs, 319, 325, 326, 327, 334 speech community, 11 Speelman, Dirk, 25, 38 standard, 7-37, 39, 42 Standard German, 213, 214 standard language, 111, 122-124, 127 standardisation, 10, 11, 14, 21, 23, 26, 36 Starnes, de Witt, Talmage, 175 Steinmetz, Donald, 247, 249

stem, 210, 211, 214, 216, 219, 220, 223,227, 231 Stratmann, Francis Henry, 192 subjectification, 307 Sundgren, Eva, 35, 40 superimposition, 23 Swan, 253, 255, 274 Sweden, 17,25,35,39,41 Swedish, 11, 20, 25, 30, 34, 40, 235237, 239-250,310 Switzerland, 15, 25, 34 syntactic status, 283 Szabomihaly, Gizella, 36, 39 taboo-breaking, 291 Taeldeman, Johan, 125, 126 Taylor, Henry Osborn, 171, 175, 196, 207, 208 Tegner, Essaias, 246, 250 Teleman, Ulf, 246, 250 temporal clause, 271 temporal perspective, 260, 261, 264 Ten Cate, Abaham P., 255, 270, 271, 274 terms, 145 derogatory, 165 for caressing, 166 of Endearment, 145 theological, 157 Thelander, Mats, 27, 32, 35, 36, 39, 41 Thompson, Sandra Α., 294, 315, 317 Jjräinsson, Höskuldur, 36, 42 Tissari, Heli, 143, 144, 146, 147, 150, 160, 161, 163, 169, 170, 172, 176 Toronto Corpus, 198 translation, 258 Traugott, Elizabeth, 191, 193, 300, 307,308,315,317 Tricht, Hendrik Willem van, 135, 142 Trosterud, Trond, 245, 246, 250 Trumper, John, 19, 41

Index Turkish, 11, 12, 35 Turku, 15 Tyrol, 11 Tzitzilis, Christos, 36, 42 Überschichtung, 23 Ukrainian, 13, 30 umlaut, 209-211, 213-218, 220, 228, 229, 232 Ungerer, Friedrich, 170, 176 unification and bonding, 300 universal, 196-198, 205, 206 urban dialects, 111, 112 Uspensky, B.A., 221, 233 Valli, Andres, 277, 289 Van Brederode, Tom, 253, 274 Van Campenhout, F., 263, 274 Van Caspel, P.P.J., 255, 274 Van den Steen, Katleen, 275, 288, 289 Van Es, G.A., 255, 274 Vandekerckhove, Reinhild, 111, 123-125, 127 Vanderveken, Daniel, 320, 321, 336 variation, 119, 124, 125 Velde, Hans Van de, 42 Vennemann, Theo, 224, 225, 232, 233 vernacular, 8, 9, 10, 12, 14, 18, 21, 22,27, 30,31,34, 39, 42 Videnov, Michail, 22, 25,42 Vikner, Carl, 277, 290 Villena, Juan, 10, 24,35,42 Villena Ponsada, Juan Andres, 32, 42 Wahrig,Gerhard, 222, 231, 234 Wayne, Valerie, 154, 176 Wegener, Heide, 209, 212, 226, 227, 232, 234

345

Wegera, Klaus-Peter, 211, 212, 214, 215,234 Wegstein, Werner, 305 Wekker, Herman C., 253, 258, 270, 274 wellformedness, 220, 223, 230 Werner, Otmar, 229, 234 Westermarck, Edvard, 162, 176 West-Flemish dialects, 115-117, 121, 124, 126, 127 Widmark, Gün, 17, 19,42 Wiese, Richard, 213, 233 will, 251, 252, 256, 266, 267, 269, 270, 273 Willemyns, Roland, 27, 36, 42, 111, 115, 116, 127 Winge, Vibeke, 311, 316, 317 Winkler, Edeltraud, 319, 335 Wogan-Browne, Jocelyn, 200, 208 Wolf, Norbert Richard, 305 Wolof, 196 Woolhiser, Curt, 36, 42 word order, 281 word-play, 291 words of honour, 130, 131, 133 Wotherspoon, Irene, 174 Wunderlich, Dieter, 302, 317, 324, 336 Wurzel, Wolfgang U., 209, 210, 213, 229, 234 Young, Edward, 159, 176 Yu, Ning, 196, 208 Zarncke, Friedrich, 306, 314 zero-semantics, 313 Zillig, Werner, 323, 336 Zubin, David Α., 246, 250 zullen, 251, 252, 254-256, 258, 265271,273,274 Zulu, 196