French anticausatives: A diachronic perspective 9783110251340, 9783110251357

Table of contents :
Frontmatter
Contents
List of Abbreviations
1. Introduction
2. Anticausatives
3. The Emergence of the French reflexive anticausative
4. The Spread of the French reflexive anticausative
5. The Semantic relation between French reflexive and unmarked anticausatives
6. Auxiliary selection in unmarked anticausatives and the spread of the reflexive anticausative
7. Conclusion

Citation preview

Linguistische Arbeiten

537

Herausgegeben von Klaus von Heusinger, Gereon Müller, Ingo Plag, Beatrice Primus, Elisabeth Stark und Richard Wiese

Steffen Heidinger

French anticausatives A diachronic perspective

De Gruyter

ISBN 978-3-11-025134-0 e-ISBN 978-3-11-025135-7 ISSN 0344-6727 Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.d-nb.de abrufbar. © 2010 Walter de Gruyter GmbH & Co. KG, Berlin/New York Gesamtherstellung: Hubert & Co. GmbH & Co. KG, Göttingen ∞ Gedruckt auf säurefreiem Papier Printed in Germany www.degruyter.com

Contents

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VII List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VIII 1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

2 2.1 2.1.1 2.1.2 2.1.3 2.2 2.3 2.4 2.5

Anticausatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preliminary remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valence change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valence alternations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Definition(s) of the notion anticausative . . . . . . . . . . . . . . . . . Restrictions on anticausative formation . . . . . . . . . . . . . . . . . . Anticausative formation and anticausativization . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4 4 4 7 7 9 11 17 23

3 3.1 3.2 3.3

The Emergence of the French reflexive anticausative . . . . . . . . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chronology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zooming in: Anticausatives and reflexive constructions in 12th century Old French . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Course of the emergence . . . . . . . . . . . . . . . . . . . . . . . . . Reflexive constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . True reflexives, reflexive psych verbs and reflexive anticausatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Mechanism of language change . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24 24 25

The Spread of the French reflexive anticausative . . . . . . . . . . . Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data basis and data extraction . . . . . . . . . . . . . . . . . . . . . . . . . Quantitative observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Increase of the relative frequency of RAC from Old to Modern French . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Lexical diffusion of RAC from Old to Modern French . . . . . . . 4.4 Qualitative observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Regularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.2 Semantic verb classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.2.1 General points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.2.2 Diachronic course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.3 Animacy of the subject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

66 66 68 72

3.4 3.4.1 3.4.2 3.4.3 3.5 4 4.1 4.2 4.3 4.3.1

32 43 43 46 52 65

72 74 76 76 79 79 81 87

VI 4.4.3.1 4.4.3.2 4.4.4 4.5 4.6 4.7

General points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Diachronic course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Possibility of use in UAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 The Temporal course of the increase of RAC . . . . . . . . . . . . . . 96 Consequences for UAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

5

The Semantic relation between French reflexive and unmarked anticausatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Aspectual structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Zribi-Hertz (1987) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Labelle (1992) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.3 Bassac (1995) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.4 Telicity, perfectivity and the resultant state . . . . . . . . . . . . . . . 5.3 Causal structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Causes in anticausatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Markedness theory and the spontaneity of events . . . . . . . . . . . 5.3.3 Proposals on French: RAC vs. UAC . . . . . . . . . . . . . . . . . . . . 5.4 Case studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.2.1 Quantitative development of reflexive and unmarked anticausative uses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.2.2 Qualitative differences between reflexive and unmarked anticausatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.1 Resultant state or telicity as the relevant feature? . . . . . . . . . . . 5.5.2 The Resultant state in RAC . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.3 The Cause in French RAC . . . . . . . . . . . . . . . . . . . . . . . . . . .

110 110 111 111 118 121 126 136 136 140 144 151 151 157 157 162 172 172 175 184

6

Auxiliary selection in unmarked anticausatives and the spread of the reflexive anticausative . . . . . . . . . . . . . . . . . . . . . . . . . 187

7

Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

Acknowledgements

This book is a revised version of my doctoral dissertation which I wrote as a member of the Graduiertenkolleg "Sprachliche Repräsentationen und ihre Interpretation" (Universität Stuttgart) and the École doctorale "Cognition, langage, interaction" (Université Paris 8) and which I defended in October 2008. My work on this dissertation was supported by grants from the Deutsche Forschungsgemeinschaft and the Deutsch-Französische Hochschule. I would like to thank my advisors Achim Stein and Anne Zribi-Hertz for their interest in my work and their continuous support during the preparation of the dissertation. I also wish to thank Artemis Alexiadou, Mario Barra Jover, Martin Becker and Martin Hummel for accepting to be part of my committee and for their valuable comments on different aspects of this work, Florian Schäfer for kindly sharing with me his research results in the study of anticausatives, and Bernhard Hurch for fuelling my interest in linguistics during my undergraduate studies at the Karl-Franzens-Universität Graz. I am also grateful to the editors of Linguistische Arbeiten for accepting this work for their series.

List of Abbreviations

SMALL CAPS

= abbreviations for interlinear glosses

ABL ablative AC anticausative AG agent (semantic role) APP appearance APPL applicative ART article ASP aspectual COP change of position COS change of state DAT dative DegP degree phrase DEM demonstrative DEOBJ deobjective dO direct object DP determiner phrase ED eventuality description ERS expression of resultant state EXP experiencer (semantic role) FUT future GEN genitive IMP imperative INTR intransitive IRR irrealis LSR lexical semantic representation NEG negation NP noun phrase NOM nominative O object

OBL oblique plural possessive PP prepositional phrase PQC PP quantizing a change PRS present PSC PP specifying a change PST past QDA quantized degree adverb R Latin suffix -r (with allomorphs) RAC reflexive anticausative REFL reflexive RF relative frequency RLD relative lexical diffusion RS resultant state RPSY reflexive psych verb S subject SC small clause SG singular t (point in) time TH theme (semantic role) TR true reflexive TR transitive UAC unmarked anticausative V verb v little v VP verb phrase VPP past participle PL POSS

1

Introduction

How do new ways of encoding valence alternations emerge, how do they spread and why do they spread, and what are the consequences of their emergence and spread for already existing patterns? These are the questions that will be addressed in this book and that will be discussed on the basis of a concrete example of valence alternation. The valence alternation that shall be analyzed is the French causative-anticausative alternation. The new type or way of encoding is the reflexive anticausative (RAC), as illustrated in (1a.), and the old type is the unmarked anticausative (UAC) in (1b.).1 (1)

a. b.

La branche s'est cassée. 'The branch broke.' La jupe a séché dehors. 'The skirt dried outside.'

reflexive anticausative (RAC) unmarked anticausative (UAC)

This book is thus concerned with the following three topics: 1. the emergence of the French RAC, 2. the spread of the French RAC, and 3. the consequences that these processes have had for the UAC which existed already prior to the emergence of the RAC. In chapter 2, I will provide a general background for the analysis of the development of French anticausatives. I will introduce the relevant concepts and notions, such as anticausative, valence, valence alternation and valence change. Furthermore, I will discuss the syntax and semantics of anticausatives, two issues which have been at the heart of the recent literature on anticausatives. Of course, many specifications and completions on the issues discussed in chapter 2 will be addressed again throughout the book, when the developments of French anticausatives will be considered in detail.

_________ 1

Note that in my use the terms marked anticausative and unmarked anticausative only refer to differences on the level of form. No statements are made as far as the semantic or functional level is concerned. That is, the fact that French reflexive anticausatives are labeled marked does not imply that they are semantically or conceptually more complex. (The question of whether the asymmetry between types of anticausatives as far as formal complexity is concerned also reflects differences in conceptual complexity will be discussed in detail in chapter 5.)

2 The emergence of the French RAC will be described in chapter 3. This process is closely linked to the question of how reflexive morphology is able to acquire the function of marking anticausatives. In this book I would like to contribute to a better understanding of this question, which I will approach from (at least) three different perspectives. These, and the resulting claims I will make, are outlined in what follows. Firstly, I will describe the situation in 12th century Old French and show that already before the emergence of the RAC, the reflexive construction was underspecified with respect to (i) the semantic role of the subject and (ii) the referentiality of the reflexive. Crucially, there is an intermediate reflexive construction between true reflexives (se laver 'wash oneself') and reflexive anticausatives (cf. (1a.), namely, reflexive psych verbs (se fâcher 'become angry'). I will argue that this intermediate reflexive construction was a necessary condition for the emergence of the French reflexive anticausative. Secondly, as far as the mechanism of language change whereby the French RAC emerged is concerned, I will argue that the mechanism cannot be determined, neither on empirical nor on theoretical grounds. The two mechanisms of language change that are candidates for the emergence are (i) reanalysis and (ii) analogical extension. I will show that among the first French reflexive anticausatives there are cases which can be accounted for with analogical extension as the relevant mechanism as well as cases which can be accounted for with reanalysis. Thirdly, I will relate the above-mentioned conclusions and show that the emergence of RAC can be modeled with both analogical extension and reanalysis, based on the assumption that the existence of an intermediate reflexive construction is a necessary condition for the emergence of the RAC. As far as the relation between reanalysis and analogical extension is concerned, I will argue that the difference is much smaller than commonly assumed. In fact, in the case of the emergence of the French RAC reanalysis could be considered a subtype of extension. The spread of the RAC will be described in chapters 4 and 6, and the consequences of the emergence of the RAC for the UAC will be discussed in chapters 4 and 5. In chapter 4 I will look at both the quantitative and the qualitative aspect of the spread of the French reflexive anticausative. The two parameters by means of which I analyse the quantitative side of this process are (i) the relative frequency of RAC and (ii) its relative lexical diffusion. The first parameter is simply the frequency of occurrence of RAC relative to the size of the respective corpus; the latter parameter is the number of verbs (as types, not tokens) forming anticausatives, which are, again, correlated to the size of the respective corpus. I will describe the results of my investigation of the diachronic development of these two parameters from Old to Modern French in a corpus study, and I will show that both the relative frequency of RAC and the relative

3 lexical diffusion of RAC strongly increased from Old to Modern French. Crucially, the increase of the relative frequency begins very slowly and accelerates dramatically after Middle French. As far as the qualitative aspect of the spread is concerned I will analyse whether the quantitative increase of the RAC correlates with a spread of the RAC to new types of anticausatives. To this end, I will consider semantic verb classes, animacy of the subject and the possibility to use the respective verb as an UAC. The strong increase of the RAC leads to the question of the consequences of the emergence and the spread for the older pattern (the unmarked anticausative), which will also be addressed at the end of chapter 4. In chapter 5, the discussion of the consequences of the spread of the RAC will be approached primarily from the angle of the semantic relation between RAC and UAC. If a language possesses two ways of formally encoding one function, the question immediately arises whether there exist semantic differences between the two strategies, despite the obvious shared property of marking the same function. With respect to this question I will investigate the aspectual and causal structure of the two types of French anticausatives. The choice of these two semantic parameters will be motivated in sections 5.2 (aspectual structure) and 5.3 (causal structure), and, of course, the parameters will be described in detail in these sections. The method I chose to detect a possible semantic difference between French RAC and UAC with respect to these two parameters is a corpus study of the distribution of certain aspectual and causal indicators in the context of RAC and UAC. I conducted case studies for six French anticausative verbs (augmenter 'increase', durcir 'harden', empirer 'worsen', enfler 'swell', gonfler 'swell', grossir 'grow, gain weight'). In these case studies a difference between RAC and UAC emerged with respect to the distribution of these aspectual and causal indicators. In chapter 6, I try to reconcile two findings from chapter 4 that may seem contradictive at first sight, namely, the strong increase of the RAC on the one hand and the lack of a change in the semantic properties of the RAC during this change on the other hand. The account for the increase of the RAC with a change in its semantic properties relies on an observation concerning the semantic difference between RAC and UAC, which becomes obvious in the case studies (chapter 5): the RAC focuses more than the UAC on the resultant state of the event. Given this semantic difference between the RAC and the UAC, I will argue that the decrease and loss of the use of être 'be' as a perfect auxiliary with unmarked anticausative verbs (which also creates a construction that focuses on the resultant state of the event), should be interpreted as a factor favouring the use of the RAC.

2

Anticausatives

2.1

Preliminary remarks

2.1.1

Valence

The term valence has been introduced into linguistic theory by Tesnière (1965). In his book Éléments de syntaxe structurale he lays out the first description of a dependency grammar. The basic assumption of this syntactic model is that sentences are hierarchically structured entities with dependency relations between the constitutive elements (cf. Tesnière 1965: 11). Tesnière abandons the subject-predicate asymmetry of traditional grammars and the special status of the subject as opposed to the object that these grammars assume. He rejects asymmetrical structures as an unjustified adoption from formal logic that cannot be motivated by the linguistic facts of any given language (cf. Tesnière 1965: 104). The crucial role that the verb plays as the highest head in the structure of a sentence is linked to the notion of valence, which he illustrates in the now famous comparison that a verb is like an atom attracting a certain number of arguments (cf. Tesnière 1965: 238). He also puts forward a description of a sentence such as The women danced on the table in terms of a play consisting of some event, some actors and, possibly, circumstances of the event. The actors (or arguments) are the entities that are involved in the event (cf. Tesnière 1965: 102). Additionally, so-called circonstants (or adjuncts) may describe the circumstances in which the event takes place (time, place, manner, etc.). In the above sentence, the verb dance describes the event, the women would be the actors (or the argument), and on the table would specify the circumstances of the event by specifying where the event takes place. To determine the valence of a given verb two points need to be considered: Firstly, arguments have to be distinguished from adjuncts. According to Tesnière, only arguments are governed by a verb. Thus only arguments such as the women, but not adjuncts such as on the table are relevant for the valence of the verb. This distinction, however, is not trivial, and was subject to a long debate in the literature following the publication of Tesnière's (1965) book (see Jacobs 1994 and 2003). Besides the identification of the arguments of a given verb, one has to consider that a verb can have more than one valence, in the sense that a verb does not always appear with the same number of arguments. For example, the verb eat has only one argument in (2a.), but two arguments in (2b.).

5 (2)

a. b.

John ate. John ate the soup.

We thus need to distinguish between the valence that a verb has in a given sentence (e.g. eat as a one-argument verb in (2a.)) and the valence potential of a verb, i.e. the sum of valences a verb can have (e.g. eat as a one-argument verb and a two-argument verb). A more detailed description of verbal valence which goes beyond the mere enumeration of the number of arguments can be given if the grammatical and semantic roles of the arguments are taken into account. Haspelmath & Müller-Bardey (2004: 1130) use the term valency pattern for this more detailed description of verbal valence. For example, the valence pattern of the sentence (2b.) is as illustrated in (3): (3)

John ate subject V agent V

the soup. direct object theme

grammatical roles semantic roles

Even more fundamental than the question of the valence of a given verb is the question at which level of representation this valence-related information is stored. The projectionist, the constructionist and the neo-constructionist approach that constitute three current theories on verbal semantics and syntax, provide different answers to this question. Generally speaking, their answers differ as to how they divide the labour between different levels of representation. The projectionist approach attributes the information that is relevant for valence to the lexical entry of the verb. Although there is a variety of different projectionist approaches, Levin & Rappaport Hovav (2005: 186) note that [...] all [projectionist approaches] share the fundamental assumption that a verb's lexical entry registers some kind of semantically anchored argument structure, which in turn determines the morphosyntactic expression – or projection – of its arguments.

An example for such a lexical entry of a verb is given below for the verb break. (4)

break: [[x do-something] cause [y become BROKEN]] (Levin & Rappaport Hovav 1995: 94)

The lexical semantic representation (LSR) for the verb break, as given above, contains information on both the number of arguments of the verb and their role in the event: the first argument x does something, while the second argument y undergoes a change. Note that such LSRs only explicitly spell out information that is relevant for argument realization. Consequently, verb-specific

6 information such as the difference in acoustic volume that distinguishes the speech act verbs say and scream would not be represented explicitly. As opposed to the projectionist approach, constructionists assume a less structured lexical entry for verbs, which thus contains only a core meaning. This core meaning nevertheless includes valence-related information. For example, the verb sneeze has a single "profiled participant role" (i.e. the "sneezer") (Goldberg 1995: 54), which suggests that the verb is a one-argument verb. However, under the constructionist view the valence of a given verb is not determined before the verb enters a given argument structure construction (the pairing of a meaning with a given syntactic frame). Crucially, this argument structure construction can offer further participant roles besides the profiled participant roles of the verb. Consider the following example: (5)

He sneezed the napkin off the table.

(Goldberg 1995: 55)

In the above example, the verb sneeze integrates itself in the "caused-motion construction" (Goldberg 1995: 54). The way in which the verb and the argument structure construction integrate is represented in figure 1. Sem

CAUSE-MOVE

Syn Figure 1:

< cause

SNEEZE

< sneezer

V

S

goal

theme > >

OBL

O

Sneeze in Caused-motion construction (Goldberg 1995: 54)

The sentence in (5) contains two arguments, the subject he and the direct object the napkin. However, only the argument in subject position (he) corresponds to an argument that is incorporated in the core meaning of the verb sneeze. The second argument (the napkin) is contributed by the argument structure construction. Thus, the actual valence that a verb has in a given sentence does not necessarily correspond to the number of participants in the lexical entry of the verb. Instead, both the verb itself and the argument structure construction contribute to the valence of the verb. The constructionist approach on valence is embedded in the framework of construction grammar. One of the basic tenets of this model is the negation of an autonomous syntax. As a consequence, constructions are considered to be pairings of form and meaning that are stored in the lexicon regardless of their "size". Argument structure constructions such as the caused-motion construction and verbs themselves have the same status; they are both linguistic units

7 with a form and a meaning and are stored in the lexicon. In that sense, the constructionist approach assumes all valence-related information to be stored in the lexicon, although this information is not necessarily stored together with the verb. The third and last approach presented here is the neo-constructionist approach (cf. e.g. Borer 1994, 2003). With the traditional constructionist approach the neo-constructionist approach shares the assumption that not all valence-related information is stored together with the verb (or, to be more precise, with the category-neutral root). Unlike traditional constructionists, neo-constructionists assume that a root enters a compositionally derived syntactic structure, where it is semantically specified by the semantics of the syntactic structure. The distribution of roots in syntactic constructions (be it words or phrases or sentences) and thus also the valence of the root depends on the semantic compatibility of the root and the respective syntactic structure. In section 2.4 I will present Alexiadou et al.'s (2006) and Schäfer's (2008) neoconstructionist treatment of anticausatives. 2.1.2

Valence change

The notion valence change refers to at least the following two phenomena. In its first use it refers to the diachronic change of the valence potential of a verb. For example, the French verb basculer 'tip (over)' started out as a strictly intransitive verb, but from the middle of the 19th century on it could also be used transitively (cf. Robert Historique s.v. basculer). The verb has thus changed or, more precisely, enhanced its valence potential. In its second use, the term valence change refers to a synchronic alternation between two valence patterns. Since the present book deals with both phenomena, I will reserve the term valence change for the diachronic change of the valence potential of the verb and the term valence alternation for synchronic alternations between valence patterns (cf. section 2.1.3 for a discussion of the latter). Different descriptions of valence change (as a diachronic phenomenon) are provided in Goyens (2001), Korhonen (2006) or Koch (1991, 2004). While the first two authors focus more on the syntactic aspect of valence change, Koch (1991, 2004) pays more attention to the relation between the valence change and the change of the semantics of the affected verb. 2.1.3

Valence alternations

Valence alternations are a verb's synchronic alternations between valence patterns. On a purely descriptive level, they can be classified according to several parameters, for example, (i) according to the argument affected by the alterna-

8 tion, (ii) according to whether arguments are added, removed or simply reorganized, and (iii) according to whether the alternation is formally marked on the verb. Typological overviews on valence alternations and the devices that languages use to formally mark these alternations are given in Haspelmath & Müller-Bardey (2004) and Dixon & Aikhenvald (2000). In (6), for example, an object-removing valence alternation is shown. The basic valence pattern is a transitive structure (cf. (6a.)) and the derived pattern is an intransitive structure (cf. (6b.)). (6)

Deobjective in Ainu (Japan) a. Sake a-ku. sake 1SG.TR-drink 'I drink sake.' b. I-ku-an. DEOBJ-drink-1SG.INTR 'I drink.'

(Haspelmath & Müller-Bardey 2004: 1131f.)

In (7), an example for an alternation where the number of arguments increases is given. A further object is added to a basic transitive valence pattern. The added argument bears the semantic role of benefactive, thus, the derived valence pattern is called benefactive applicative. (7)

Benefactive applicative in Indonesian a. Orang itu me-masak ikan. man ART TR-cook fish 'The man cooked fish.' b. Orang itu me-masak-kan perempuanitu itu ikan. man ART TR-cook-APPL woman ART fish 'The man cooked fish for the woman.' (Haspelmath & Müller-Bardey 2004: 1134)

In the above cases the valence alternations are formally marked on the verb, i. e. on the object-removing affix i in (6) and the object-adding affix kan in (7). The formal change on the verb allows us to determine the basic and the derived valence pattern within the alternation. If there is no formal marking on the verb, there is a priori no reason to assume any direction of derivation. To illustrate this point let us consider the English equivalents of the above examples. The following examples show the same alternation between the two valence patterns illustrated in (6) and (7), but with the important difference that only in (6) and (7) the alternation is formally marked on the verb. (8)

a. b.

I drink sake. I drink.

9 (9)

a. b.

The man cooked fish. The man cooked fish for the woman.

As mentionned above, these are purely descriptive and pretheoretical observations. Naturally, assumptions about what "really happens" when a verb alternates (formally marked or not) heavily depend on one's view on valence itself. For example, in the neo-constructionist approach valence alternations cannot involve modifications of the number of arguments represented in a lexical entry. Recall that in this approach no lexical entries for verbs, but only for category-neutral roots exist. The lexical entries of these roots, however, do not include information on the number of arguments either. In section 2.2 anticausatives will be defined as one part of the causativeanticausative alternation. As a consequence, the question of whether anticausatives are a derived valence pattern arises, and if so, the question of how they are derived arises. The various answers that have been proposed in the literature will be reviewed in section 2.4, and the position taken in this book will be introduced on the basis of this overview. Upon the presentation of these views on anticausatives, I will constantly refer back to the concepts of valence that these proposals endorse.

2.2

Definition(s) of the notion anticausative

Utterances such as the one in (10) have been given many different names in the literature, with anticausative being only one of them (Nedyalkov & Silnitsky 1973, Siewierska 1984, Haspelmath 1987, 1990, 1993, Cennamo 1993, 1998, Michaelis 1998, Alexiadou et al. 2006, Schäfer 2008, KoontzGarboden 2007, 2009). Other terms for the same construction are inchoative (Levin 1993, Folli 2002), decausative (Geniušienơ 1987, Mel'þuk 1993), spontaneous (Shibatani 1985, Kemmer 1993), or ergative (Zribi-Hertz 1987). (10) La branche s'est cassée. 'The branch broke.'

Despite its frequent use in the literature and the consensus that the notion anticausative applies to cases such as the one given in (10), there is no general consensus on its delimitation. The original definition of anticausative was "[...] the non-causative member of an opposition which is formally marked by means of an anticausative affix [...]" (Nedyalkov & Silnitsky 1973: 7), which is illustrated in the following example:

10 (11) Russian lomat'-sja 'become broken' (derived from lomat' 'make broken') (cf. Nedyalkov & Silnitsky 1973: 2)

Anticausatives would thus be the result of a morphological operation (anticausativization) whereby the external argument of a transitive-causative verb is deleted from the argument structure. A similar use of the term anticausative has also been adopted by Haspelmath (1987, 1993), among others. But unlike Nedyalkov & Silnitsky (1973), Haspelmath (1993) does not only include morphologically derived anticausatives, but uses the term for all cases with a formal derivation between the causative and the non-causative member of the alternation (e.g. when the non-causative member is formed with an auxiliary). Conversely, Alexiadou et al. (2006) and Schäfer (2008) consider anticausatives simply as the non-causative part of the causative alternation without imposing constraints on the direction of derivation or the formal marking of the verb. For both views it is necessary to further specify the non-causative part of the causative alternation in order to delimit the scope of the term anticausative. Nedyalkov & Silnitsky (1973: 1) define the term anticausative as a verb designating "some state". Their examples, which include verbs like break and boil, but also return and leave, show that non-causative verbs are very heterogeneous with respect to their semantics. In their definition of the causative alternation, Alexiadou et al. (2006) refer to Levin (1993: 26-32), who distinguishes between different types of causative alternations: (i) causative-inchoative, (ii) induced action alternation, and (iii) other instances of the causative alternation. Levin's use of the notion inchoative for the non-causative part of the first type of alternation is of limited value, since the term as such has no consistent definition in the literature. Levin (1993: 30) notes that her use of inchoative corresponds to what others have called anticausative, but Haspelmath (1993: 91), for example, uses the term anticausative only for a subset of inchoative verbs, namely for those that can be formally derived from a transitive-causative verb. Haspelmath (1993: 90) defines the semantics of inchoative verbs as describing a change of state or a going-on (i.e., a non-agentive activity) where "[...] the verb meaning excludes a causing agent and presents the situation as occurring spontaneously." This quote, which suggests that no cause is present in anticausatives, needs to be further specified: First, no cause is expressed in subject position, i.e. the position that it is normally attributed by the linking hierarchy. Second, no cause is semantically present unless it is overtly expressed. These specifications include cases where a cause is expressed in a PP (12a.), and predicts the semantic absence of a cause for cases such as (12b.) where no cause is overtly expressed. In chapter 5, however, I will take a closer look at the causal structure of anticausatives. Although the above insights remain valid, more fine-grained distinctions will be drawn.

11 (12) a. b.

The window broke through the storm. The window broke.

With respect to the type of event described by the anticausative verb, some authors, in accordance with Haspelmath's (1993) description of inchoatives, note that anticausatives are "mainly denoting a change of state" (cf. Cennamo 1998: 80), while others, like Nedyalkov & Silnitsky (1973), do not provide any further specifications. Compared to the original definition by Nedyalkov & Silnitsky (1973), the definition of anticausative applied in this book is less rigid on the formal side, but more rigid on the semantic/functional side. I will adopt the view expressed by Alexiadou et al. (2006) and Schäfer (2008) that anticausative verbs must participate in the causative alternation, but that the anticausative use of the verb does not have to be overtly marked. I take anticausatives to be sentences with only one argument that bears the grammatical role subject and the semantic role theme. Furthermore, they are formed with aspectually dynamic verbs (break as opposed to, for example, know), and the relevant event is described as coming about without the implication of a cause triggering the respective event. Finally, Nedyalkov & Silnitsky's (1973) criterion of the presence of an anticausative-affix does not figure in my definition.

2.3

Restrictions on anticausative formation

The size of the set of verbs that can form anticausatives in a given language naturally depends on the respective definition of anticausative. For example, the set of verbs would be larger if a purely semantic definition were applied than with a definition imposing further constraints, such as, for example, that the verb in question has to be formally derived from a transitive-causative verb. In the preceding section it was shown that despite the considerable variety among the characterizations of anticausatives (and their quasi-equivalent categories) most authors agree on the absence of a cause that brings about the event as a defining property of anticausatives. In the literature, restrictions on anticausative formation have mainly been discussed in the context of the causative-anticausative alternation and have been considered equivalent to the question which transitive verbs can form anticausatives and which verbs cannot (cf. Smith 1970, Fillmore 1970, Levin & Rappaport Hovav 1995). The most basic assertion one can make in this respect is that anticausative formation presupposes the anticausative conceptualization of the event described by the verb. A natural implication of this assertion is that transitive verbs de-

12 scribing an event that cannot be conceptualized without an external cause cannot form anticausatives. Thus, causation must play a crucial role in accounts which address the question of which verbs can form anticausatives. I will now present three analyses that have been put forward to determine the semantic features which allow or block anticausative conceptualization. The first analysis is proposed in Smith (1970). In her article, Smith (1970: 101) refers to alternating verbs as "verbs of the 'change' class" (exemplified with the verb break below). (13) a. b.

The window broke. John broke the window.

Smith (1970: 101f.) assumes that two binary semantic features are decisive for the participation of the verb in the alternation. The first one is the relative independence of the activity (from outside control), and the second one is the possibility of an external agent controlling the activity. (14) semantic features a. [+/- independent activity] b. [+/- external control]

In the case of alternating verbs such as break in (13) both features are marked as positive. The events denoted by such verbs may thus be described as independent activities which can be externally controlled. To prove the relevance of these features, Smith (1970) compares alternating verbs to verbs that do not alternate. Non-alternating verbs fall into two groups: verbs that can only be used transitively and verbs that can only be used intransitively (cf. (15)). (15) 2 types of non-alternating verbs a. strictly transitive: destroy, build, cut, slice, draw, etc. b. strictly intransitive: shudder, laugh, tremble, hesitate, etc.

I will first take a look at the verbs that can only be used transitively, as illustrated in (16). (16) a. b.

The storm destroyed half of the city. *Half of the city destroyed.

Smith argues that verbs like destroy are similar to alternating verbs like break in that the activity can be externally controlled, but they differ from them in that the activity is not independent. These strictly transitive verbs thus bear the features [+ external control] and [- independent activity]. In the case of verbs

13 that can only be used intransitively as shudder in (17) the feature settings are different. (17) a. b.

John shuddered. *The green monster shuddered John.

Verbs like shudder are similar to alternating verbs like break in that they describe an independent activity, but differ from them in that the activity cannot be externally controlled. Strictly intransitive verbs like shudder thus bear the features [- external control] and [+ independent activity]. Smith's (1970) answer to the initial question of this section, namely the question of which transitive verbs can form anticausatives, is that anticausative formation is restricted to those transitive verbs that bear the semantic feature [+ independent activity]. To sum up, Smith (1970) tries to account for the different valences of verbs by means of the binary features [+/- external control] and [+/- independent activity]. The relation between the manifestations of these features and verbal valence is represented in table 1. Table 1:

Valence and verb semantics (Smith 1970)

valence transitive transitive, intransitive intransitive

verb semantics external control + + -

independent activity + +

The next proposal that I will discuss, Levin & Rappaport Hovav's (1995) analysis, builds on and modifies Smith's (1970) account. Levin & Rappaport Hovav (1995) first introduce a distinction between internally and externally caused verbs. Crucially, only the latter – which are semantically characterized as described in the following quote – can be used transitively. [E]xternally caused verbs by their very nature imply the existence of an "external cause" with immediate control over bringing about the eventuality described by the verb: an agent, an instrument, a natural force, or a circumstance. (Levin & Rappaport Hovav 1995: 92)

An example for an externally caused verb is, again, break. The lexical semantic representation Levin & Rappaport Hovav (1995: 83) propose for this verb is given in (18). (18) break: [[x do-something] cause [y become BROKEN]]

14 Levin & Rappaport Hovav (1995) assume that all externally caused verbs are basically dyadic (involving two sub-events) and therefore allow for the transitive use. But break can also be used intransitively without the external cause (cf. (13a.) above). As there is only one lexical semantic representation, which includes the external cause, the intransitive use is derived via the deletion of the external argument (cf. Levin & Rappaport Hovav (1995: 108) and the description of this derivation in section 2.4 below). However, only a subset of the externally caused verbs allows for this operation and consequently only this subset can form anticausatives. Levin & Rappaport Hovav (1995) characterize the subset of detransitivizing externally caused verbs as describing an event that "[...] can come about spontaneously without the volitional intervention of an external agent." (Levin & Rappaport Hovav 1995: 102) For the moment, I will refer to this semantic feature as [+ spontaneous]. In both proposals laid out so far, anticausative formation depends on the lexical semantics of the respective verb. It is limited to verbs that describe an "independent activity" in Smith's (1970) terms or a "spontaneous event" in Levin & Rappaport Hovav's (1995) terminology. Although the view that verbal valence is determined by verbal semantics is widespread, the falsifiability and the epistemological status of explanations as the ones given in Smith (1970) and Levin & Rappaport Hovav (1995) are seldom considered. To illustrate the problem, we shall again take the verb break as an example. The chain of arguments leading to the explanation of the restriction on anticausative formation in Levin & Rappaport Hovav (1995) is as follows: The transitive verb break can form anticausatives because (i) it describes a spontaneous event and (ii) there is a generalization that verbs of this type form anticausatives (cf. (19)). (19) a. b. c.

spontaneous verbs form anticausatives break is a spontaneous verb ---------------------------------------------break forms anticausatives

(law) (case) (result)

The falsifiability of the explanation (and with it the value of the explanation) crucially relies on the possibility of categorizing verbs with the semantic feature [+/- spontaneous]. One would thus expect an applicable definition of the term spontaneous from the literature that actually employs this feature. From Levin & Rappaport Hovav's (1995) definition, as cited above, it is not entirely clear how a decision can be made with respect to whether or not break (or any other verb for that matter) can be considered to be spontaneous. Moreover, it is not possible to establish a reasonable delineation between externally caused verbs that are [+ spontaneous] and those that are [- spontaneous] on semantic grounds.

15 Levin & Rappaport Hovav (1995), however, propose that there is a grammatical reflex which is linked to the semantic parameter of spontaneity. According to Levin & Rappaport Hovav (1995: 103), the restriction on anticausative formation correlates with a restriction on the semantic role of the subject of the verb in the transitive use of the verb. Externally caused verbs that allow causes and natural forces as subjects are able to form anticausatives, while externally caused verbs which restrict their subject position to agents and instruments cannot form anticausatives. This generalization is further illustrated in (20) and (21). The examples in (20) show that the verb break, which allows for causes as subjects in its transitive use (cf. (20a.)), can form anticausatives (cf. (20b.)). Conversely, the verb cut, which only licenses agents and instruments, but not causes as transitive subjects (cf. (21a.)), cannot form anticausatives (cf. the ungrammaticality of (21b.)). (20) a. b.

The vandals / The rocks / The storm broke the windows. The window broke.

(21) a. b.

John / That knife / *The lightning cut the clothesline. *The clothesline cut.

The restriction on possible subjects in the transitive use and the restriction on anticausative formation are linked by the presence (or absence) of agentoriented meaning components in the semantics of the respective verb.1 Verbs describing an event that can only be brought about by an agent do not allow for any other semantic role in subject position than the roles of agent and instrument. The feature that decides whether an externally caused verb can form anticausatives thus seems to encode agent-oriented meaning components. For this reason, it seems possible to eliminate causes, but we can only do without agents if the subject position of the verb is not specified for them, i.e. if the semantics of the verb does not involve agent-oriented components. The selection restrictions on the subject of the transitive use of a given verb thus provide an independent diagnostic as to whether its semantics includes such meaning components or not. Levin & Rappaport Hovav's (1995) account relies on a threefold semantic distinction between (i) internally caused verbs, (ii) externally caused verbs with agent-oriented meaning components and (iii) externally caused verbs without agent-oriented meaning components. The relation between these three semantic verb types and their valence is illustrated in table 2 below.

_________ 1

Cf. Pinker (1989: 223f.) on the presence or absence of manner components in the lexical semantic representation of verbs of change of state.

16 Table 2:

Valence and verb semantics (Levin & Rappaport Hovav 1995)

valence transitive transitive, intransitive intransitive

verb semantics externally caused with agent-oriented meaning components externally caused without agent-oriented meaning components internally caused

In keeping with Smith's (1970) and Levin & Rappaport Hovav's (1995) accounts, Alexiadou et al. (2006) and Schäfer (2008) assume that the valence of the verb is determined by semantics. However, given that the latter approaches are implemented within the neo-constructionist framework of Distributed Morphology they do not consider semantic features of verbs, but rather semantic features of category-neutral roots to be responsible for the determination of valence. A second difference lies in the number of semantic types that are assumed. While Smith (1970) and Levin & Rappaport Hovav (1995) assume only three semantic verb types, Alexiadou et al. (2006) and Schäfer (2008) differentiate between four semantic types of roots. Table 3:

Valence and root semantics (Alexiadou et al. 2006, Schäfer 2008)

valence transitive transitive, intransitive intransitive

root semantics —agentive —externally caused —cause unspecified —internally caused

Besides the terminological differences, the substantial difference between Alexiadou et al. (2006) and Schäfer (2008) on the one hand and Smith (1970) and Levin & Rappaport Hovav (1995) on the other hand is that only the first two divide strictly transitive roots into two sub-types: agentive roots and externally caused roots. This sub-division is motivated by the empirical observation that not all strictly transitive verbs/roots obey Levin & Rappaport Hovav's (1995) generalization concerning the semantic role of the subject. Instead there are strictly transitive verbs/roots which only allow agents as subjects and strictly transitive roots which also allow causes as subjects, as illustrated in (22) and (23) below: (22) a. b.

The man / *the earthquake murdered the child. *The child murdered. (anticausative reading is impossible)

17 (23) a. b.

The man / the earthquake destroyed the city. *The city destroyed.

The existence of verbs such as destroy which, on the one hand, allow causes as subjects in their transitive use, but, on the other hand, are ungrammatical as anticausatives, falsifies Levin & Rappaport Hovav's (1995) generalization concerning the relation between agent-oriented meaning components and anticausative formation. Rather, the correct generalization seems to be that transitive verbs that do not allow for causes as subjects cannot form anticausatives. Therefore, the lack of agent-oriented meaning components is only a necessary but not a sufficient condition for anticausative formation. So far, there is no satisfactory description of further conditions that come into play, more specifically, the semantic differences that distinguish alternating verbs that allow for causes from non-alternating verbs that allow for causes are still unknown. At the present state of research, the answer to the initial question, namely, what the conditions for anticausative formation are, cannot be fully answered. The lack of agent-oriented meaning components is a necessary, but not a sufficient condition. The quantitative relation between externally caused roots and cause-unspecified roots nevertheless allows a statistical prediction. As there is a considerably higher number of instances of the first type of roots, we may at least conclude from the above-mentioned literature that roots/verbs that can form transitive sentences and do not have agent-oriented meaning components are very likely to form anticausatives.

2.4

Anticausative formation and anticausativization

A first important aspect of anticausative formation is related to our definition of anticausatives. Recall that in section 2.2 above, anticausatives are defined as constituting one part of the causative-anticausative alternation. Thus the anticausative use of a verb is always opposed by a transitive-causative counterpart. This fact leads to the question of whether there is a direction of derivation within the alternation. There are three logically possible answers to this question: (i) there is no direction of derivation, (ii), the causative is derived from the anticausative or (iii) the anticausative is derived from the causative. I use the term anticausativization for the last case where the anticausative member is derived from the causative member of the alternation (with or without a formal reflex). In the literature, each of the three possibilities mentioned above can be found. Furthermore, the proposals cover a wide range with respect to their overall scope. We will see that there are proposals which deal only with one

18 specific morphological operation within one language, as well as proposals that claim to have universal validity. In the following I will present current views on anticausative formation focussing on two aspects: (i) whether anticausative formation involves anticausativization and (ii) the theory of verbal valence in which the respective proposal is embedded (cf. section 2.1.1). At the end of the section I will present my own view on anticausative formation, which departs from the literature discussed.2 1) The projectionist view with anticausativization I begin my survey with proposals assuming a direction within the derivation, more specifically, with those assuming a derived anticausative. Levin & Rappaport Hovav (1995) are proponents of this view. As we have already seen in section 2.3, Levin & Rappaport Hovav (1995) assume that the lexical entry of alternating verbs includes two arguments, illustrated for the verb break: [[x do-something] cause [y become BROKEN]]. In the anticausative use of the verb, however, only one of the arguments is expressed, namely the argument of the become-predicate. Levin & Rappaport Hovav (1995: 84) suggest that the anticausative use of break "[...] must arise from an operation that prevents the causer argument from being projected to the lexical syntactic representation [...]". This operation is subsequently specified as lexical binding (Levin & Rappaport Hovav 1995: 108). Crucially this binding operation happens during the mapping from the lexical-semantic representation (LSR) to the lexical-syntactic representation (i.e. to argument structure).3 The absence of the external argument in the lexical-syntactic structure is the main difference between passives and anticausatives, a difference which manifests itself in the different behaviour of the constructions, concerning, for example, by-phrases and purpose clauses (cf. Levin & Rappaport Hovav 1995: 109). Levin & Rappaport Hovav (1995) thus propose that anticausatives are formed via the application of lexical binding on the causer argument of a causative lexical-semantic representation, as illustrated for the verb break in the example in (24) below. The result is a lexical semantic representation where only the theme argument is left. As the basic lexical-semantic representation of anticausatives is a causative representation, this proposal can be interpreted as involving the derivation of the anticausative from the causative. The proposal counts as a projectionist approach because the information on the

_________ 2

3

A recent overview of theories on anticausative formation is provided in Schäfer (2009). Levin & Rappaport Hovav (1995) use the terms lexical syntactic representation and argument structure interchangeably.

19 number of arguments is part of the lexical semantics of the verb (the fact that the number of arguments is manipulated during the formation of anticausatives does not change the locus of information about these arguments). (24) a. transitive break LSR [[x do-something] cause [y become BROKEN]] Linking rules Ļ Ļ Arg. structure x y b. intransitive break LSR [[x do-something] cause [y become BROKEN]] Ļ Lexical binding Ø Linking rules Ļ Argument structure y (Levin & Rappaport Hovav 1995: 108)

A similar proposal can be found in Reinhart (2002). Like Levin & Rappaport Hovav (1995), she assumes that alternating verbs are basically dyadic and that anticausatives are derived via a lexical operation, with the difference being that in Reinhart's (2002: 241) account, this operation is conceived of as expletivization. Both proposals share the view that the derivation does not necessarily have to be expressed formally. The English verb break, for example, is used both as causative and as anticausative verb without any formal difference. Furthermore, both Levin & Rappaport Hovav's (1995) and Reinhart's (2002) proposals consider the derivation of the anticausative alternate to be universal. 2) The neo-constructionist view without anticausativization Contrary to the proposals by Levin & Rappaport Hovav (1995) and Reinhart (2002), Alexiadou et al. (2006) and Schäfer (2008) argue that there is no direction of derivation in the causative-anticausative alternation at all. Instead, both alternants are derived with the same category-neutral roots; as illustrated in the following syntactic structure for causatives and anticausatives. (25) a. b.

causatives [agent [Voice [vCAUS [¥Root + theme]]]] anticausatives [vCAUS [¥Root + theme]] (Schäfer 2008: 221)

The following example shows the application of these structures to the root ¥open. (26) a. b.

causatives [He [Voice [vCAUS [¥open + window]]]] anticausatives [vCAUS [¥open + window]]

20 The arguments of Alexiadou et al. (2006) and Schäfer (2008) against a derivational account, which are also discussed in detail in Schäfer (2008: 119-131), are based on the following observations: (i) cross-linguistic variation in the formal encoding of the alternation, (ii) restrictions on anticausative formation and selectional restrictions, (iii) PPs expressing external arguments in the context of anticausatives. Their arguments address mainly the analyses which claim that there is one universal direction in the alternation. I agree with Alexiadou et al. (2006) and Schäfer (2008) that such a view is not tenable. Piñón (2001: 3) has already concluded that, in view of the different directions of formal derivation presented in Haspelmath (1993), "[...] any general account of the causativeinchoative alternation in which causative-inchoative verbs are derived from inchoative verbs or vice versa is misguided." The variation in the crosslinguistic encoding of the causative-inchoative alternation is illustrated in table 4 below, taken from Haspelmath (1993), who distinguishes five formal types of encoding the alternation: types 1 and 2 are directed alternations, and 3 to 5 are non-directed alternations. Table 4:

1 2 3 4 5

Encodings of the causative-anticausative alternation (Haspelmath 1993: 9092)

type causative alternation anticausative alternation

Georgian: duȖ-s 'cook INTR' a-duȖ-ebs 'cook TR'

labile alternation

English: break 'break INTR' break 'break TR'

equipollent alternation suppletive alternation

example

Polish: záamaü-siĊ 'break INTR ' záamaü 'break TR'

Japanese: atum-aru 'gather INTR' atum-eru 'gather TR' Russian: goret' 'burn INTR' žeþ 'burn TR'

However, the conclusion put forward in Alexiadou et al. (2006) and Schäfer (2008), i.e. the rejection of any kind of derivation within the alternation, is only one possible conclusion that can be drawn from the cross-linguistic variation outlined in table 4 above. More specifically, the possibility that languages may have different directions of derivation and that even within one language different types of anticausatives may be formed via different processes is neither mentioned nor considered in the analyses proposed in Alexiadou et al. (2006) and Schäfer (2008). Still, a priori there is no reason to disprefer this second possibility over the one proposed by Alexiadou et al. (2006) and Schäfer (2008).

21 In typological work, the direction of formal derivation is given a more prominent role. Haspelmath & Müller-Bardey (2004: 1132) assume that the "[...] anticausative affix eliminates the agent argument completely [...]". The affixation would thus spell out a derivational process. The authors further point out that such valence morphology has a more derivational than inflectional character (cf. Haspelmath & Müller-Bardey 2004: 1139). Anticausatives would thus be derived from a transitive-causative base and then stored in the lexicon, and there would not be a new derivation every time they are used. The acceptance of this view is, however, highly dependent on the choice of theoretical framework in which the assumptions on the formation of anticausatives are embedded. For example, the proposal summarized above is diametrically opposed to the analyses put forward by Alexiadou et al. (2006) and Schäfer (2008) who follow a theoretical framework which assumes a very slim lexicon and attributes most of the labour to the syntax. 3) The projectionist view with and without anticausativization I now present my own view on the formation of anticausatives. But before I do this a general remark concerning its scope is necessary. I do not assume that there must be a universal direction of derivation within the alternation. Hence, the lack of a universal formal direction of derivation – as shown by Haspelmath's (1993) data – does not imply that there is no direction of derivation at all. Instead I assume that languages might differ in the ways they form anticausatives. As a matter of fact, formal asymmetries between the causative and anticausative can be interpreted as showing the direction of derivation, regardless of the fact that these asymmetries suggest cross-linguistic variation in the direction of derivation within the alternation. The assumptions made below are thus limited to French because they are motivated by the facts of French. (This does not exclude the possibility that the situation may be similar in other languages.) For the case of labile verbs in the causative-anticausative alternation, Löbner (2002) assumes, like the neo-constructionists, that there is no direction of derivation involved. The motives and the view on the partition of labour between the syntax and the lexicon are, however, entirely different. More specifically, Löbner (2002: 105) maintains that the causative and the anticausative of a verb like open are formed by two different verbs. He assumes a case of polysemy with open1 and open2 as two separate lexical entries, the first being a transitive verb and the second being an intransitive one. (27) a. b.

John opened the door. The door opened.

His arguments for this view are expressed in the following quote:

22 These [i.e. open in a causative and an anticausative use] must be considered to be two different verbs as they belong to different word classes (intransitive vs. transitive verbs). Intransitive open predicates of its subject argument a certain change of state. Transitive open predicates of its subject argument an action which leads to a corresponding change of state of the direct object argument. Thus intransitive and transitive open express different predications about their respective subject arguments. Their meanings are clearly different. (Löbner 2002: 105)

In the case of polysemy, no use is synchronically derived from the other. Hence, no direction of derivation exists within the alternation. As indicated by the above quote, Löbner (2002) motivates the assumption of two different lexemes by the fact that the verbs in a. and b. belong to two different grammatical categories (transitive vs. intransitive) and have different predications. As Löbner's view entails that information on the number of arguments is part of the lexical entry, his view qualifies as a projectionist approach. As far as French is concerned, I take a projectionist stance that allows for both polysemy and valence operations. I assume that information about the number of arguments is part of the lexical entry of a verb and that the number of arguments can be manipulated by valence changing operations. For the case of reflexive anticausatives I assume a deletion operation that applies on the lexical entry of a dyadic verb (cf. Wehrli (1986), Mutz (2005) on French). The reflexive morpheme is the formal marking of this operation. (28) Formation of French reflexive anticausative input: VERB[cause, theme] reflexive: deletes cause output: se VERB[theme]

As far as the formation of unmarked anticausatives is concerned I extend Löbner's (2002) view on labile verbs to French and assume that they are formed on the basis of an intransitive lexical entry. (29) Formation of French unmarked anticausative input/output: VERB[theme]

The above assumptions on anticausative formation in French are motivated by the linguistic facts of French. In principle they rely on the semantic relation that exists between the reflexive anticausative and the unmarked anticausative use of a given verb. The distinction between the formation of French RAC and UAC described above is motivated by the linguistic facts of French. In principle, the distinction relies on the semantic relation between the reflexive anticausative use and unmarked anticausative use of a given verb. As will become apparent in chapter 5, there is a semantic difference between French RAC and UAC as far as

23 the focus on the resultant state is concerned. Based on a case study, I will argue in chapter 5 that the RAC but not the UAC emphasizes the resultant state of the event. Crucially, I will attribute this difference not to the verbs that appear in the two constructions (because they are identical), but to the constructions themselves. If the unmarked anticausative were to be formed in the same way as reflexive anticausatives (but without an overt sign of the valence operation), a semantic difference between the two constructions, such as the one discussed in chapter 5, would be highly unlikely.

2.5

Summary

In this chapter the properties and the formation of anticausatives have been described. I take anticausatives to be sentences formed by aspectually dynamic verbs which participate in the causative alternation, but the anticausative use of the verb does not have to be overtly marked. Anticausatives are sentences with only one argument that bears the grammatical role subject and the semantic role theme. Crucially, the event is described as coming about without the implication of a cause (unless it is overtly expressed in a PP). This semantic property excludes verbs with agent-oriented meaning components from anticausative formation, as a causal component is always present in these verbs. Concerning the formation (or derivation) of anticausatives, different proposals from the literature have been discussed. The proposals differ with respect to (i) whether valence related information is stored in the lexicon and (ii) whether the valence of a verb is manipulated in the course of anticausative formation. I have argued that the number of arguments is part of the lexical entry of a verb and that French anticausatives are formed in the following two ways: unmarked anticausatives are formed on the basis of an intransitive lexical entry, reflexive anticausatives are formed through a valence operation on a transitive lexical entry where the external argument is deleted from the argument grid.

3

The Emergence of the French reflexive anticausative

3.1

Introduction

Valence alternations can be formally encoded in various ways (cf. Haspelmath 1993, Dixon & Aikhenvald 2000, Nichols et al. 2004, Haspelmath & MüllerBardey 2004). A basic distinction in this respect is whether an alternation is formally marked or not. If the alternation is not marked and the same verb is used without any formal difference for both members of the alternation the terms labile encoding or labile verb are used. If a valence alternation is formally marked, then the type of marking can be further described. In section 2.4 we saw some of the various ways in which languages formally encode the causative-anticausative alternation (cf. especially Haspelmath 1993). There may not only exist cross-linguistic differences in the encoding of valence alternations, there may also be differences within a single language. In the latter case, a language possesses more than one way to formally encode a given valence alternation. The encoding of a valence alternation within a given language can also be subject to language change. The present chapter is concerned with the emergence of reflexive anticausatives, which constitutes the most substantial diachronic change in the encoding of the causative-anticausative alternation in French. In chapter 4 the spread of this new form will be discussed in detail. In this chapter I pursue two goals. The first goal is to delimit the exact point in time in the history of French which marks the emergence of reflexive anticausatives. Many claims regarding the time frame of this development have been suggested in the literature. Unfortunately, as we will see in what follows, these proposals often suffer from weaknesses in their methodology. In particular, since assertions concerning the emergence of a given linguistic form are quantitative in nature they should be explicitly stated as such. Moreover, the instantiations of the new form must be quantified and set in relation to the size of the corpus from which they are retrieved. However, a major part of the literature (which will be discussed in section 3.2) does not meet these requirements, and many authors do not even specify the empirical basis of their statements.1 Thus, in order to delimit the point in time when the French RAC emerged, I have conducted a corpus study. In section 3.2, I describe how the morphological form of anticausative verbs develops from Late Latin to

_________ 1

Cf. Stéfanini (1962: 597) for similar concerns regarding the chronology of the emergence of the French reflexive passive.

25 Modern French. As we will see, Late Latin had three morphological types of anticausative verbs (verbs with the suffix -r, reflexive verbs, and unmarked verbs). Out of these three types, only unmarked verbs survived the development from Late Latin to Old French. But unlike anticausatives with the suffix -r, reflexive anticausatives reemerge in French. The data from my corpus study suggests that this happened towards the end of the 12th century. While the first goal of the chapter is concerned with the chronology of this language change, the second one focuses on the manner of this change. I will first give a description of the relevant circumstances of the emergence, i.e. describe anticausatives and reflexive constructions in 12th century Old French (cf. section 3.3). Building on these findings I will discuss the mechanism of language change whereby the French RAC has emerged in section 3.4. I will argue that the emergence of the French anticausative can be modelled with both reanalysis and analogical extension and that the data does not permit a decision between the two mechanisms. I will further argue that with both mechanisms the existence of an intermediate reflexive construction (intermediate between the true reflexive and the reflexive anticausative) is a necessary condition for the emergence of the anticausative, and that not only the emergence via analogical extension, but also the emergence via reanalysis would be an extension of an existing pattern.

3.2

Chronology

This section gives an overview of how the form of anticausative verbs develops from Late Latin to Modern French. French follows Latin on the Italic branch of the Indo-European language family. Vulgar Latin is the language spoken by ordinary people and constitutes the origin of the Romance languages. Empirical sources of Vulgar Latin are, among others, notes in prescriptive grammars, inscriptions, and technical texts (cf. Kiesler 2006: 33-40) and further literary sources such as the comedies by Plautus (~254-184 BC) and Terence (195/85-159 BC) or the picaresque novel Satiricon by Petronius (~27-66 AD). For the present case, only Vulgar Latin in its stage immediately before the end of Latin as a language is relevant. I will use the term Late Latin for this chronological subsection of Vulgar Latin. The split into the Romance languages, one of which is French, marks the end of the Late Latin period. The first preserved texts written in French are the Serments de Strasbourg (842) and the Cantilène de Sainte Eulalie (soon after 878). Within the 1200 years of the diachronic development of the French language the following subdivisions can be made: Old French (9th - 13th century), Middle French (14th - 15th century), Pre-classical and Classical French

26 (16th - 18th century), and Modern French (19th century onwards). The following chronology gives the subdivisions from Late Latin to Modern French. (unattested) (30) Late Latin > Proto-Romance > Old French > Middle Fr. > 600 AD 842 1330 1500 (Pre)classical Fr. > Modern Fr. 1789

In Late Latin three morphological types of anticausative verbs existed: (i) verbs with the suffix -r, (ii) reflexive verbs and (iii) unmarked verbs. The verbs with suffix -r, as shown in (31), were used to express anticausative semantics from the earliest stages of Latin. At the onset of the Late Latin period, however, they were already in decline. (31) Late Latin anticausative verb with suffix -r tempora eius colligantur temporal:NOM.PL DEM.GEN.SG knot:PST:3PL.R 'His temporals knotted.' (4th AD, Mulomedicina Chironis; (Michaelis 1998: 89))

On the other hand, reflexive anticausatives (RAC) as in (32) increased in Late Latin (cf. Kemmer 1993: 153, Selig 1998: 27) and took over the ground previously covered by the verbs with suffix -r. Cennamo (1998: 88) notes that "[b]y the end of the 4th century, practically all verbs can mark anticausatives by means of the reflexive pattern [...]." (32) Late Latin reflexive anticausative a. diffundit se pupilla dilate:PRS.3SG REFL pupil:NOM.SG 'The pupil dilates.' b. horum statim nervi se contrahent DEM.GEN.PL immediately nerv:NOM.PL REFL contract:FUT.3PL 'Their nerves immediately will contract.' c. ubi iam se morbus ostendit where already REFL illness:NOM.SG show:PRS.3SG '...where the illness already shows itself.' (4th AD, Mulomedicina Chironis; (Kriegel 1999: 154f.))

In Late Latin the reflexive morpheme was inflected for case with se as the accusative/ablative form and sibi as the dative form. Cennamo (2000: 45) argues that in Late Latin the distribution of the two forms is not related to case, but

27 depends on the semantics of the verb it combines with and, more specifically, on the position of the verb on the unaccusative-unergative scale. Anticausatives, however, could be formed with both se and sibi (cf. Cennamo 2000: 46). Unmarked verbs as in (33), finally, form the third type of anticausative verbs in Late Latin. (33) Late Latin unmarked anticausatives a. postea rumpunt dentes later break:PRS.3PL teeth:NOM.PL 'Later the teeth will break.' (4th AD, Mulomedicina Chironis; (Feltenius 1977: 121)) b. donec cicatrices cludant until wound:NOM.PL close:PRS.IRR.3PL '...until the wounds would close.' (4th AD, Mulomedicina Chironis; (Feltenius 1977: 78))

None of the three types of Late Latin anticausative verbs was functionally restricted to the anticausative. The reflexive morpheme was used to indicate true reflexives and reciprocals, the suffix -r also marked the passive, and the unmarked intransitive was used to express a wide range of non-anticausative semantics, such as, for example, agentive activities. As already mentioned above, anticausative verbs marked with the suffix -r were in decline in Late Latin. It is commonly assumed that the suffix disappeared before the end of the Latin language period, as it is attested in none of the Romance languages. The first important change between Late Latin and Old French in the domain of anticausatives is thus the loss of the suffix -r as an anticausative marker. The second important change is that the reflexive anticausative was also lost. Hatcher (1942: 127) notes that reflexive constructions that fall under our definition of anticausative did not exist in French between at least 880 and the second half of 12th century. But unlike verbs with suffix -r, reflexive anticausatives re-emerged towards the end of the Old French period. In general, French reflexive verbs are one of the most studied topics in traditional Romance linguistics (for references see the extensive bibliography already in Stéfanini 1962). The emergence of reflexive anticausatives towards the end of Old French is one important change identified in the literature. More precisely, however, the literature did not identify the emergence of reflexive anticausatives as such, since no such term existed in traditional Romance linguistics. Instead, in the traditional literature, sentences that can be captured with the definition of anticausative presented in section 2.2 have been called

28 middle voice (cf. Gamillscheg 1957, Babin 1937, Lerch 1939), or passive voice (Nyrop 1930).2 However, these terms do not only employ different terms for the same category anticausative, but instead cover different domains which only partially intersect with the anticausative domain as defined in section 2.2. As for the chronological development of the distinct (but related) category of reflexive anticausatives, different views can be found in the literature. While Gamillscheg (1957: 330) assumes that the reflexive marking of the middle voice emerged in the 13th century, Babin (1937: 22), Lerch (1939: 347) and Nyrop (1930: 204) claim that it emerged in the 14th century. Herslund (2001: 41), in turn, does not propose a specific time frame for the emergence, but states that reflexive anticausatives did not exist in Old French. Since reflexive anticausatives are a subset of the middle voice which is reflexively marked only from the 13th century on, we can assume that reflexive anticausatives do not exist prior to the 13th century. The overall conclusion to be drawn from the literature mentioned above would thus be that French reflexive anticausatives did not emerge before the 13th century. To delimit the point in time where the reflexive anticausative re-emerged in French, and to learn more about the subsequent development of the new pattern, I conducted a corpus study in the Nouveau Corpus d'Amsterdam (NCA), the Dictionnaire du Moyen Français (DMF1), and Frantext. As a starting point, I chose the Chanson de Roland text (around 1100), as it allows the verification of both Hatcher's (1942) statement about the non-existence of reflexive anticausatives before the end of the 12th century, and the proposals of Gamillscheg (1957), Babin (1937), Lerch (1939) and Nyrop (1930) concerning the emergence of reflexive anticausatives only in the 13th and 14th century. The results of the study are given in figure 2 and table 5 below.3

_________ 2

3

Note that also the term middle voice (German Medium, French voix moyenne) is not used uniformly in the literature. Stéfanini (1962), for example, uses the term voix moyenne for all uses of third person SE, including true reflexives and reciprocals. Gamillscheg (1957), Babin (1937) and Lerch (1939), on the other hand, use a definition which is functionally more restricted and excludes true reflexives and reciprocals. See chapter 4.2 for a detailed description of the corpora and the method of data extraction.

RF = y x 5.04x10^-5

29

18 15 12 9 6 3 0 1100

11701179

12401269

13301359

14801520

16101615

1745

1990

Figure 2:

Relative frequency (RF) of RAC from Old to Modern French (value 1 assigned to RF of 1170-1179)

Table 5:

Absolute frequency of RAC from Old to Modern French

1100 1170-1179 1240-1269 1330-1359 1480-1520 1610-1615 1745 1990

corpus size (words) 29924 138770 248320 506019 629923 316898 343307 239269

abs. frequency 0 7 17 34 62 65 214 227

Table 5 gives the size of the respective corpora and the absolute frequencies of reflexive anticausatives for eight spot checks from Old to Modern French. The relative frequency (figure 2) establishes the relation between the absolute frequency and the corpus size. The value 1 was assigned to the relative frequency of reflexive anticausatives in the second spot check (1170-1179) and the remaining values were calculated accordingly. Figure 2 illustrates the development of the relative frequency of reflexive anticausatives from Old French to Modern French. As table 5 shows, no reflexive anticausatives were found in the oldest text of the corpus, the Chanson de Roland (around 1100). The first reflexive anticausatives are attested in the second spot check (1170-1179). From then onwards, reflexive anticausatives are attested in all spot checks. Generally speaking, the relative frequency strongly increases from Old to Modern French. However, as figure 2 illustrates, the degree of increase is not stable; the increase starts rather slowly, takes off in the 17th century and then continues at a slower pace. I will come back to the shape of this graph in chapter 4.

30 The following examples are some of the earliest instances of RACs that are attested in my corpus. (34) a.

b.

c.

d.

l eue en la terre s abaisse 'the water sinks into the earth' (Chronique des ducs de Normandie par Benoit, 1175; NCA) la vie terrienne se change sovant, 'the earthly life changes often' (L'histoire de Barlaam et Josaphat, 1250; NCA) quant la plaie s est refreidie 'when the wound cooled' (Miracles de Notre Dame de Chartres de Jean le Marchant, 1262; NCA) ou la plume en l air se souslieve 'where the feather rises in the air' (Le bestiaire d'amour rimé, 1250; NCA)

Below I give a series of examples of reflexive anticausatives from Middle to Modern French. In chapter 4, where I describe the lexical diffusion of French reflexive anticausatives, I will provide a number of lists of verbs forming reflexive anticausatives at different stages of French. (35) a. b.

c.

d.

et se doubla le nombre du peuple [...] 'and the size of the people doubled' (Bersuire, 1354-1359; DMF1) La terre s'ouvry et grans flammes En issy, 'the earth opened and big flames came out of it' (de Pizan, 1400-1403; DMF1) [...] le coeur fust une roche dure, Qui ne s'amoliroit [...] 'the heart was a hard rock that did not became soft' (Montchrestien, 1601; Frantext) [...] les bajoues et le ventre qui commençaient à s'alourdir. 'the cheek and the belly which began to become heavy' (Kristeva, 1990; Frantext)

Together with Hatcher's (1942) findings about the lack of reflexive anticausatives in the earliest French texts, the results of the corpus study suggest that reflexive anticausatives did not exist before the 12th century and thus did not exist for most parts of Old French. As a consequence, I assume that reflexive anticausatives were lost, just like anticausatives formed with r-verbs, before the onset of Old French. But unlike the suffix -r, the reflexive anticausative reemerges in French.4 An important difference in this respect seems to be that

_________ 4

I am aware that this is not the only possible interpretation of these facts. Two other possible interpretations are (i) that RACs were not part of the variety of Late Latin

31 the suffix -r was lost in all its uses, while the reflexive morpheme was lost as a marker of the anticausative, but persisted in other functions. In section 3.4, I will discuss the emergence of French reflexive anticausatives and attribute a crucial role to reflexive constructions existing prior to the emergence of reflexive anticausatives in this language change. The third morphological type of anticausative verbs attested in Late Latin were unmarked verbs. In line with Hatcher (1942), I assume unmarked verbs to be continuously used from Latin over Old French to Modern French. The following examples of unmarked anticausatives are taken from the Chanson de Roland (for which no reflexive anticausatives are attested). (36) a. b. c.

L'espee cruist, ne fruisset ne ne brise 'the sword creaks, but does not burst nor break' Qu'envers le cel en volent les escicles. 'that the splinters fly into the air' Dunc ad tel doel pur poi d'ire ne fent, 'he [=Roland] suffers such great pain that he nearly bursts with anger'

Further examples, illustrating the use of unmarked verbs in anticausatives from Middle to Modern French, are given below. (37) a. b.

c.

son corps assechera 'his body will dry up' (Les quinze joies de mariage, 1390; DMF1) et elle dit que sa fievre avoit augmenté depuis le souper à faute de dormir, 'and she says that her fever had increased since dinner due tolack of sleep' (Audiguier, 1624; Frantext) Soudain, près du plafond, une vitre cassa, et les éclats plurent sur le tapis. 'Suddenly, a pane cracked near the ceiling and the splinters fell on the carpet' (Jarry, 1902; Frantext)

The following table summarizes the development of the morphological form of anticausative verbs from Late Latin to Modern French, as described above.

that is the basis of Old French and (ii) that RACs continuously existed from Late Latin to Old French but are not attested in the written sources.

32 Table 6:

Morphological form of anticausative verbs from Late Latin to Modern French

Late Latin

Old French (before ~1150)

Old French (after ~1150)

Middle French

Modern French

Suffix -r

Suffix-r

Suffix -r

Suffix -r

Suffix -r

Reflexive Unmarked

Reflexive Unmarked

Reflexive Unmarked

Reflexive Unmarked

Reflexive Unmarked

In Late Latin, anticausative verbs belonged to one of the following three morphological types: verbs with the suffix -r, reflexive verbs, and unmarked verbs. Out of these three types, only unmarked verbs survived the development from Late Latin to Old French, leaving the French language with an impoverished system in comparison to Late Latin. Towards the end of the 12th century reflexive anticausatives re-emerged and are continuously attested since then. Following the discussion of the time of emergence of French RACs, I now turn to the question of how they emerged.

3.3

Zooming in: Anticausatives and reflexive constructions in 12th century Old French

The results of the corpus study presented in the previous section suggest that the RAC emerges in the course of the 12th century. The present section focuses on this period and pursues two goals: (i) to describe the first reflexive anticausatives, and (ii) to describe the relevant linguistic circumstances of these first reflexive anticausatives. Due to their small number, the description of the first reflexive anticausatives will be relatively short. More space will be given to the description of the linguistic circumstances of these first examples, i.e. anticausatives other than reflexive anticausatives and reflexive constructions other than reflexive anticausatives. The description given in this section will provide the empirical basis for a number of issues discussed in the further course of this book such as the development of the reflexive morpheme from true reflexive to anticausative, the mechanism of language change whereby the RAC emerges, and the spread of the RAC after its emergence. While data on the reflexive construction (including reflexive anticausatives) can be easily extracted from digitalized text corpora through searches for the reflexive (se, s', ...), no such searches are possible to extract anticausative verbs and other constructions expressing anticausative semantics. Hence, the

33 relevant data had to be searched "manually" in texts. This was done in five Old French texts from the first and the second half of the 12th century. These texts, which are presented in the following table, constitute the corpus of the present description. Table 7:

Corpus for 12th century Old French

author Chanson de Roland (Rol) Le Voyage de Saint Brendan (Bre) Roman de Thèbes (Thè) Lancelot (Lanc) Guillaume d'Angleterre (Gui)

date of creation

unknown

1100

Benedeit

1112

unknown

1160

Chrestien de Troyes Chrestien (de Troyes?)

1171 1175

edition Steinsieck (1999) Ruhe (1977) Olef-Krafft (2002) Jauss-Meyer (1974) Klüppelholz (1987)

verses 4002 1840 10562 (1-4000 considered)

7134 3308

The first question that shall be answered on the basis of this corpus is how anticausative semantics could be expressed in 12th century Old French. Expressions of anticausative semantics do not necessarily fall under the definition of anticausatives given in chapter 2. Instead, anticausatives are a subset of the expressions of anticausative semantics. While the first is limited to verbs, the latter also includes periphrastic constructions such as become STATE as in The men became drunk. The following expressions of anticausative semantics were found in the corpus: unmarked anticausative verbs, reflexive anticausative verbs, periphrastic constructions with devenir 'become'. Consider the following examples. (38) unmarked anticausatives a. Cum apresmout, la mer ne mot. 'while he was approaching, the sea did not move' (Bre: verse 1259) b. Que sa lance a estros peçoie. 'that his lance breaks into pieces' (Lanc: verse 859) c. Et por ce que trop vos grevastes Vos plaies sanz dote escrevastes 'and because you struggled too hard, your wounds burst open' (Lanc: verse 4899f.)

34 d.

La voile ront et li maz froisse. 'the canvas breaks and the mast breaks into pieces'

(39) reflexive anticausatives a. la ou l'espaule se devise. 'where the shoulder intersects' b. S'est antre eus la tançons montee 'the tensions between them increased'

(Gui: verse 2329)

(Thè: verse 2654) (Gui: verse 1061)

(40) devenir + state a. Quant li rois sot que voirs ce fu coulor mua si devint noirs. 'when the king realized that this was true, he changed colour and turned black' (Thè: verse 2509f.) b. si se mervoille, Honte an ot, si devint vermoille 'she was so astonished, that she was very ashamed and turned dark red' (Lanc: verse 4797f.)

Further, certain constructions with avoir 'have' might also be interpretated as expressing anticausative semantics. An example is given below. (41) avoir + state Pour l'amour de son compaingnon en metra son cors a bandon, dont il avra le cors percié et son escu tout despecié. 'For the love of his companion he will risk his life and, doing this, his body will be wounded, and his shield will be shattered.' (Thè: verse 1213-1216)

(41) is a possessive construction where the possessor il 'he' is in subject position, the possessum le cors 'the body' is a direct object and modified by a past participle percié 'pierced'. For the anticausative interpretation the modifying past participle is most important, since it expresses the event that the possessum undergoes. On this view a structure such as the one in a. may be considered as a near-equivalent to b. (42) a. b.

X have Y STATE X's Y become STATE

Whether such an interpretation is correct seems to depend on two variables which, however, are hard to determine. The first one is whether the past participle is interpreted as eventive at all (instead of resultative or stative). The second variable is the consequence of the participle being interpreted as eventive:

35 the event could be understood as being brought about by a cause or an agent. This would yield a passive instead of an anticausative interpretation. Turning to the two types of anticausative verbs that have been found in the corpus, namely unmarked and reflexive anticausative verbs, the first question is how often they appear in the corpus. 134 anticausatives (= tokens) were found in the five texts, 129 are unmarked and five are reflexive anticausatives. Unmarked anticausatives appear much more frequently than reflexive anticausatives in the texts (96.3% vs. 3.7%). Table 8:

UAC and RAC in a corpus of 12th century Old French

UAC

RAC

total

abs. freq.

129

5

134

rel. freq.

96.3

3.7

100%

The quantitative dominance of UAC over RAC holds for all five texts of the corpus. In none of the five texts the RAC exceeds 10% of the total of anticausatives. In the Chanson de Roland, the oldest text of the corpus, 37 anticausative verbs were identified and all of them are unmarked. In Le Voyage de Saint Brendan (beginning of the 12th century) 18 anticausative verbs were identified and again all of them are unmarked. In Lancelot (second half of 12th century) 23 anticausative verbs were found, 22 of them are unmarked and one is a reflexive verb. 32 anticausative verbs were found in the Roman de Thèbes, 30 of them are unmarked, and two are reflexive. In Guillaume d'Angleterre 24 anticausative verbs were found: 22 unmarked and two reflexive anticausatives. The following table and figure show the frequency of unmarked and reflexive anticausatives in the texts of the corpus. Table 9:

UAC and RAC in a corpus of 12th century Old French

date of creation

UAC

RAC

total

Chanson de Roland Le Voyage de Saint Brendan Roman de Thèbes

1100

37

0

37

1112

18

0

18

1160

30

2

32

Lancelot

1171

22

1

23

Guillaume d'Angleterre

1175

22

2

24

129

5

134

total

36 100% 75% 50% 25% 0% Rol

Bre

Thè UAC

Figure 3:

Lanc

Gui

RAC

Unmarked and reflexive anticausatives in 12th century Old French

Unmarked anticausatives clearly dominate reflexive anticausatives in all five texts, and thus in both the first and the second half of the 12th century. The figure further shows that the few attested reflexive anticausatives were found in texts from the second half of the 12th century. This corroborates the assumption that the reflexive anticausative emerges in the course of the 12th century (cf. section 3.2). I will now describe the unmarked anticausatives in more detail. The identification of unmarked anticausatives in Old French texts is not trivial and complicated by two grammatical features of Old French. Because of the relatively free word order and the possibility to leave the subject unexpressed, sentences including (i) a verb that can form both causatives and anticausatives, and (ii) only one argument bearing the semantic role theme (and thus no argument bearing the role agent or cause) can be ambiguous between a causative and an anticausative interpretation. In the case of a causative interpretation, the subject would be unexpressed and the theme-argument would be the object. In the case of an anticausative interpretation, the theme-argument would be the subject and no direct object would be involved (neither expressed, nor unexpressed). (43) 2 interpretations for sentences with TH and V(anti)causative a. causative: Ø S, Vcausative, THdO b. anticausative: Vanticausative, THS

Although the lack of a second overt argument and the position of that argument are not sufficient to identify the above type of sentence unambiguously as anticausative, other cues may disambiguate such sentences. Morphological cues would be number inflection on verb and argument and case inflection

37 within the NP (nouns, pronouns, articles). With respect to the first, a mismatch in number inflection between the verb and the argument rules out that the argument is the subject and leads to a causative interpretation with an unexpressed subject. With respect to case, the identification of the argument as cas sujet or cas régime would disambiguate the sentence as well. Unfortunately, it is not always possible to identify the case of the respective noun (or pronoun or article) via its morphological form. E.g. for a large number of feminine nouns cas sujet and cas régime are identical (cf. Buridant (2007) and Schøsler (1984) on the defectiveness of the Old French case system). Anticausatives can be described according to the semantic verb class that the verb in the sentence belongs to. By definition, anticausatives are a very specific phenomenon with respect to the number of arguments, the semantic role of the argument, and the aspectual properties of the verb. The verbs that form unmarked anticausatives nevertheless come from different semantic verb classes. I will now describe which semantic verb classes the verbs belong to and how often they are used. The verb classes that have been distinguished are verbs of change of state (COS), aspectual verbs (ASP), verbs of appearance (APP), and verbs of change of position (COP) (this distinction is inspired by Levin (1993)). How often are unmarked anticausatives formed with these four semantic verb classes respectively? The following table shows that there is a very uneven distribution of the classes. In addition, it shows that there exists a strong preference for verbs of change of state. Table 10: UAC in different semantic verb classes

COS COP APP ASP

total

abs. freq.

110

rel. freq.

85.3 12.4

16

1

2

129

0.8

1.5

100%

110 out of the total of 129 unmarked anticausatives are formed by verbs of change of state (e.g. break). Verbs of change of state are by far the most frequent semantic verb class. (44) unmarked anticausative formed with verb of COS Tuz lur escuz i fruissent e esquassent, 'their shields burst and shattered'

(Rol: verse 3879)

Verbs of change of position (e.g. turn) are the second most frequent verb class; 16 out of the 129 unmarked anticausatives are formed with verbs of this semantic class.

38 (45) unmarked anticausative formed with verb of COP Et la nez muet, qui ront et fant A force les ondes devant 'the ship starts to move, and it breaks and divides the waves'

(Gui: verse 2061f.)

The other two semantic verb classes, namely verbs of appearance and aspectual verbs, only play a minor role and are attested once and twice, respectively. (46) unmarked anticausative formed with ASP verb Des or comance sa raison. 'Now his story begins.'

(Lanc: verse 30)

(47) unmarked anticausative formed with verb of APP Quant alat en tapinage Apparut lui Deu message; 'When he secretely left, a messenger of God appeared to him.' (Bre: verse 725f.)

Different types of entities can undergo anticausative events. In a textbook example like the window broke the entity undergoing the event is inanimate. However, the subject position in anticausatives is not restricted to inanimate entities. Instead, despite their potential for volitional actions and agentivity, non-human and human animate entities can also undergo anticausative events. Both types of entities appear in the subject position of the unmarked anticausatives found in the corpus. Below examples for four types of subjects are given: inanimate concrete, inanimate abstract, animate human, and animate non-human. (48) inanimate (concrete) Tranche li fuz et ront li fers, 'the wood splinters and the iron breaks' (49) inanimate (abstract) Idunc agreget le doel e la pitet. 'Then pain and grief increase.'

(Lanc: verse 2702)

(Rol: verse 2206)

(50) animate (human) Puis enfle fort, e li quirs tent; Anguisus sui, pur poi ne fent. 'Then I swell, my skin is stretched and I am afraid because it almost cracks.' (Bre: verse 1429f.; mod. StH) (51) animate (non-human) Draguns i at qui la guardent; Si cume fus trestut ardent. 'There are dragons guarding him whose whole bodies burn like fire' (Bre: verse 1711f.; mod. StH)

39 Animate and inanimate entities subjects are very unevenly distributed in the 129 instances of unmarked anticausatives. Inanimate entities appear much more frequently (86.8%) than animate entities (13.2%) (cf. table 11 below). Table 11: Types of subjects in UAC

inanimate animate total abs. freq.

112

17

129

rel. freq.

86.8

13.2

100%

Following the description of the verb classes and classes of entities in subject position, the next step consists in verifying whether certain preferred combinations between verb classes and classes of entities in unmarked anticausatives exist. The four semantic verb classes and the two types of entities give rise to eight possible combinations. Out of these eight combinations six are actually attested in the corpus. Between them strong differences with respect to the frequency of occurrence exist, as shown by the following table. Table 12: UAC: Combinations of verb class and type of subject (abs. freq.)

COS

COP

ASP

APP

inanimate

95

animate

15

15

2

0

112

1

0

1

17

110

16

2

1

129

An inanimate subject (e.g. la hanste 'the lance') with a verb of change of state (e.g. rompre 'break') is by far the most frequent combination appearing in the unmarked anticausatives of the corpus. Out of the 129 unmarked anticausatives 95 instances (= 96.3 %) display this particular combination of verb class and class of entity. Only two more combinations appear more than ten times: inanimate & verb of change of state (15 instances), and animate & verb of change of state (15 instances). Despite this variety, it has to be stressed that the great majority of the unmarked anticausatives found in the corpus resemble the textbook examples of anticausatives: an inanimate concrete entity undergoes a change of state (e.g. the window broke). I end this description of unmarked anticausatives in the corpus with a short remark on the expression of causes in the context of unmarked anticausatives. Recall from section 2.2 that the expression of causes is possible also in the context of anticausatives. To illustrate this point consider the following example.

40 (52) My heart broke because of all you've done.

In this case, the cause of the anticausative event is expressed in the "larger context" of the anticausative, namely in a subordinate sentence. Similar examples were also found in the corpus. (53) a.

b.

Et por ce que trop vos grevastes Vos plaies sanz dote escrevastes. 'and because you struggled too hard, undoubtedly, your wounds burst open' (Lanc: verse 4899f.) Si grant doel ad que par mi quiet fendre; 'he was in such pain that he thought he would break apart' (Rol: verse 1633)

However, there is also the possibility that a cause is expressed in a prepositional phrase in the anticausative sentence itself. (54) The window broke [through the wind]PP

Again, similar examples are found in the corpus. In (55) an agent is expressed in a PP introduced by par, in (56) a cause is expressed in a PP introduced by de. The expression of causes in the context of anticausatives and whether unmarked and reflexive differ in this respect will be a main topic of chapter 5. For the moment we shall only note that Old French unmarked anticausatives allow for PPs expressing causes. (55) par-agent Par qui [=queen] creistrat lei de terre 'because of whom the earthly laws increase'

(Bre: verse 3)

(56) de-cause De la joie que il an orent, Lor croist force et s'an esvertüent 'their strength grows from the joy they felt' (Lanc: verse 2440f.)

It should also be mentioned that some non-anticausative intransitive verbs were found in the corpus which in present-day French are necessarily reflexively marked. Both of the following examples involve the verb lever 'rise/get up'. (57) a. b.

Par main lievent li soudoier 'in the early morning the soldiers get up' Li empereres est par matin levet; 'the emperor got up early in the morning'

(Thè: verse 1015) (Rol: verse 163)

41 I now turn to the reflexive construction and describe both the reflexive anticausatives and other reflexive constructions found in the corpus. As mentioned above, only five instances of reflexive anticausatives were identified in the corpus. Two of them were already cited above and are repeated below as a. and b.; the remaining three instances are given in c., d. and e. (58) reflexive anticausatives a. la ou l'espaule se devise. 'where the shoulder intersects' (Thè: verse 2654) b. S'est antre eus la tançons montee 'the tensions between increased' (Gui: verse 1061) c. A po qu'ele ne s'est ocise Maintenant que de Lancelot La mançonge et la novele ot; 'she nearly died when she heard the false rumour about Lancelot' (Lanc: verse 4178-4180) d. S'estaint de soif et de fain muert, 'he is faint with thirst and dies of hunger' (Gui: verse 913) e. de s'espee tel coup li donne que Thideüs tou s'en estonne; 'with his sword he gives him such a strike that Thideüs is numb' (Thè: verse 1801-1802)

A first interesting property of these data is that in three out of five examples the subject undergoing the anticausative event is human (cf. c., d., e.). Human subjects in reflexive constructions can give rise to ambiguities between different interpretations. In d., for example, a true reflexive interpretation is possible ('X kills herself'), although the context rather favors a non-agentive interpretation of the subject ('X faints'). For the other two examples with human subjects no true reflexive interpretation is possible. In both d. and e. the subject can only be interpreted as non-agentive. Concerning example e., however, it may appear that s'en estonne is ambiguous between the anticausative reading 'become dizzy' and an interpretation as a psych verb 'be astonished'. Again the context favors the anticausative interpretation, as being hit by a sword rather causes dizziness than astonishment. However, the possibility of having more than one possible interpretation is not limited to the case of human subjects. The homonymy of the form s' as (i) an allomorph of the reflexive se and (ii) an allomorph of the conjunction si (meaning 'so', 'and', or 'if') also gives rise to more than one possible interpretation. In b., for example, no ambiguity exists between the two reflexive constructions. Rather, the example could be interpreted either as a reflexive anticausative (s' = allomorphe of se) or as an unmarked anticausative (s' = allomorph of the conjunction si). With regard to the semantic verb classes of the verbs forming the five reflexive anticausatives,

42 the class of verbs of change of state is again the dominant class. Four out of the five reflexive anticausatives are formed by this class. Only the verb éteindre 'go out' (in the sense of 'dying') in d. is a verb of appearance and thus belongs to another class. A qualitative comparison (semantic verb classes, types of subjects) between the reflexive and the unmarked anticausatives in the corpus is not warranted because of the small number of instances of reflexive anticausatives (for a quantitative comparison see above). Instead, I will describe the circumstances of these first reflexive anticausatives on the basis of the different functions of the reflexive constructions in the corpus. Leaving aside true reflexives and reciprocals (which are attested in the corpus but are of minor interest), I concentrate on reflexive constructions which display similarities to the reflexive anticausative. These reflexive constructions occupy an intermediate position between true reflexives and reciprocals on the one hand and reflexive anticausatives on the other hand. Similarities between the reflexive constructions and the reflexive anticausative exist with respect to several properties: the number of arguments, the agentivity of the subject, and the interpretation of the reflexive. Recall from chapter 2 that anticausatives only involve one argument, which is non-agentive. Hence, in reflexive anticausatives, the subject must be interpreted as non-agentive, and the reflexive morpheme cannot be an argument. One type of reflexive construction with such similarities to reflexive anticausatives is the type involving reflexive psych verbs. As illustrated by the examples below, the subject of a reflexive psych verb can be non-agentive. A further similarity to anticausatives is that a transitive verb is turned into an intransitive verb via the use of the reflexive. (59) a.

b.

c. d.

N'i ait Franceis ki tot ne s'en merveilt. 'That there was no Frank that was not astonished because of that' (Rol: verse 571) Einsi tote jor se contient, Ne la nuit pas ne se rapeise 'so he behaves the whole day and does not calm down at night' (Gui: verse 948f.) Asez oez que Rollant se dementet! 'You hear that Roland is desperate!' (Rol: verse 1795) De ce gueires ne s'esmaioit 'he doesn't get scared because of this' (Lanc: verse 3119)

Another interesting reflexive construction is formed by verbs which cannot be used transitively. Although these verbs differ from verbs forming reflexive anticausatives (and anticausatives more generally) in that the latter can form transitives, these strictly intransitive verbs also show similarities to reflexive anticausatives (cf. the examples given below). First, the subject bears the semantic role theme which is also the role taken on by the subject of anticausa-

43 tives. Second, the verbs describe a change of state or position and thus belong to semantic verb classes that also form anticausatives. Finally, no cause of the event is expressed as an argument in the sentence. (60) a. b. c.

Sur lui se pasmet, tant par est anguissus. 'he is so shaken that he slumps down on him' (Rol: verse 2880) Mais lur isle mult tost s'en fuit: 'their island quickly moved away' (Bre: verse 464) A val s'en torne et l'enfant let qui mout s'angoisse et forment bret 'then he turns away and leaves the child that suffers and cries loudly' (Thè: verse 2377f.; mod. StH)

These data show that there are reflexive constructions which occupy an intermediate position between true reflexives and reflexive anticausatives. Crucially, intermediate has to be understood both in terms of the properties/functions, but also with respect to the chronology. At least for reflexive psych verbs, there are good reasons to assume that they exist prior to the reflexive anticausative. First, they are attested in a much greater number than reflexive anticausatives (although I have not collected them exhaustively and systematically in the corpus, more than 40 instances were found). Second, these reflexive constructions are also attested in the two texts from the first half of the 12th century in which no reflexive anticausatives were found.

3.4

The Course of the emergence

3.4.1

Reflexive constructions

It is well known that reflexive constructions frequently have more than one function. Kazenin (2001: 920) notes that "[p]olysemous verbal reflexive markers are more frequent than verbal markers denoting only [true] reflexives",5 and Klaiman (1992: 39) proposes the cover term middle reflexive for all nontrue reflexive uses of reflexive constructions. The use of the same marker for (at least) true reflexives and anticausatives is especially widespread in the

_________ 5

Kazenin's (2001: 916) definitions are the following: (i) verbal reflexives: "one of the coreferential NPs is deleted, and the verb receives special marking which signals coreferentiality of the two participants", and (ii) reflexive pronouns: "one of the coreferential NPs is replaced by a reflexive pronoun".

44 Indo-European languages. 24 out of the 25 Indo-European languages in Geniušienơ's (1987: 258) sample show this kind of polysemy. These languages are listed in (61a.). Examples of non-Indo-European languages with the same polysemy are given in (61b.). (61) Languages using the same morpheme for true reflexives and anticausatives a. Indo-European Lithuanian, Latvian, Russian, Ukrainian, Belorussian, Bulgarian, Macedonian, Serbo-Croatian, Czech, Slovak, Polish, German, Swedish, Norwegian, Icelandic, Gothic, Latin, French, Spanish, Italian, Romanian, Armenian, Ancient Greek, Sanskrit (cf. Geniušienơ 1987: 258) b. Non-Indo-European Hungarian, Quechua, Georgian, Hixkarayana, Uzbek, Ponapean, Arabic, Tigre, Worora, Udmurt (cf. Haspelmath 1987: 24f. and Haspelmath 1990: 36)

The use of the term polysemy (as in Kazenin's (2001) quote) requires some clarification. The first clarification is of terminological nature. In the present context the term poly-functional will be used instead of polysemous. The second point that needs to be clarified is whether the polysemy or poly-functionality is a property of the reflexive construction, the reflexive morpheme or both. For present-day French, it is uncontested that the reflexive construction can have different functions (passive, true reflexive, etc.), and different inventories of reflexive constructions have been proposed (cf. Burston 1979, Lyons 1982, Melis 1990, Ruwet 1972, Wehrli 1986, Zribi-Hertz 1987, among others).6 What is contested, however, is whether the reflexive morpheme itself is also poly-functional or whether it has an invariant function in all its uses. In the latter case, the poly-functionality of the reflexive construction would result from the interplay of this single function with different verbs and verb types. According to Kazenin (2001: 920), the argument that authors that adhere to this view often adduce is the observation that the different functions of reflexive morphology are lexically distributed, i.e. the combination of the reflexive with a given verb can have only one function (or, at most, two functions). A mono-functional analysis of the reflexive morpheme for French (and to some extent, also for other Romance languages) has been proposed in Burston (1979), Wehrli (1986), Hummel (2004), and de Alencar & Kelling (2005). Although these authors have in common that they assume a unitary function of the reflexive, they do not propose the same unitary function. Their accounts

_________ 6

For Italian, Wehr (1995: 49-56) describes the reflexive construction as a unitarian third voice (besides the active and the passive) with several syntactic and semantic variants.

45 differ as to whether the unitary function of the reflexive is linked to valence changing operations or not. Wehrli (1986: 264), for example, argues that the unitary function of French se is the absorption (also referred to as suppression) of an argument. The different functions of the reflexive construction – Wehrli (1986: 265-267) distinguishes between reflexive/reciprocal, inherent, neuter and middle – follow from two factors: (a) [...] the nature of the argument which has been absorbed (i.e., internal versus external argument), and (b) whether or not the complex se + verb has been lexicalized.

Hummel (2004: 213), on the other hand, gives a semantic description of the unitary function of the reflexive: it expresses reflexivity, i.e. the fact that the Ereignisträger (~actor) of the event is also the Betroffener (~undergoer) of the event.7 Koontz-Garboden (2007, 2009) suggests that at least in the true reflexive and the reflexive anticausative, the reflexive morpheme is a reflexivization operator and thus has the same function: Reflexivization, quite simply, is an operation that takes a relation as an argument, setting both arguments of the relation to be the same (Chierchia 2004: 29). In settheoretic terms, if a relation is conceived of as a set of pairs, what reflexivization does is to restrict the denotation of the relation to those sets of pairs each of whose members is identical to the other. (Koontz-Garboden 2009: 83)

As anticausativization is also a reflexivization operation, this view implies that no kind of argument deletion is involved in the formation of reflexive anticausatives. "Rather the relation denoted by the causative verb is simply reflexivized so that the participants in both the causing and caused eventualities are specified to be identical." (Koontz-Garboden 2009: 86) The view that not only the reflexive construction, but also the reflexive morpheme is poly-functional can be found for example in Labelle (2008). According to the author, the reflexive in true reflexives and reciprocals is a Voice head introducing the external argument (in Spec,Voice) on the basis of which the unsaturated internal argument of the verb is identified (cf. Labelle 2008: 835). In contrast to true reflexive and reciprocals, where the reflexive

_________ 7

"Se ist ein Reflexivmorphem, das Rückbezüglichkeit anzeigt, ohne die spezifische Art der Rückbezüglichkeit festzulegen. Es signalisiert einfach, dass der Ereignisträger zugleich ein Betroffener des Ereignisses ist, d.h. es differenziert nicht nach den Betroffenheitstypen, die mit dem direkten oder indirekten Objekt ausgedrückt werden. Welche Art von Betroffenheit vorliegt, ergibt sich aus der semantischen Interpretation des Ereignisses im jeweiligen Kontext." (Hummel 2004: 213)

46 introduces the external argument, the reflexive can also have another "flavour" (Labelle 2008). In the case of anticausatives (and also middles) the reflexive does not introduce an external argument, but instead allows the internal argument to appear as the subject of the sentence (cf. Labelle 2008: 871). The view one develops of the emergence of the reflexive anticausative naturally depends on one's view of the construction itself and other reflexive constructions which might exist prior to the reflexive anticausative. In order to establish the basis for this discussion, I will describe the reflexive constructions involved in the emergence of the French RAC in the next section. 3.4.2

True reflexives, reflexive psych verbs and reflexive anticausatives

Concerning the number of different reflexive constructions prior to the emergence of the French reflexive anticausatives, the most important finding of the corpus study presented in section 3.3 is that not only true reflexive and reciprocals existed prior to reflexive anticausatives, but also reflexive psych-verbs. Relevant properties of the latter are (i) the non-agentivity of the subject and (ii) the lack of a semantic role of the reflexive morpheme. I will first describe RPSY with respect to these two properties and then show how they differ from true reflexives and reflexive anticausatives and what they have in common with them. In the corpus of 12th century Old French (cf. table 7), I have found more than 40 examples of reflexive constructions such as the example given in (62), where the subject is non-agentive (cf. section 3.3 for more examples of this type). (62) A moi s'est Deus coreciez, 'God became angry over me'

(Gui: verse 890)

(63) provides a (non-exhaustive) list of verbs that are used in the same way, i.e. with a non-agentive subject and a non-referential reflexive morpheme, in the Chanson de Roland. (63) criembre 'fear', esbaudir 'become happy', esclargir 'become happy', dementer 'moan', doloser 'grieve', esragier 'become angry', merveiller 'be surprised', repentir 'repent' or 'feel remorse', rapaiser 'calm down'

Hatcher (1942: 92f.) notes with respect to this semantic verb class that verbs of agony, madness, and grief are generally reflexively marked in Old French. For her, verbs like se desver 'get angry', se esragier 'get angry', se esfreer 'be alarmed', or se doloir 'grieve' describe "[...] for the most part reactions that are purely involuntary and spontaneous [...]". (Hatcher 1942: 95)

47 The non-agentivity of the subject in these cases seems to contradict one important property of the Old French reflexive construction that has been mentioned in the literature: the reflexive morpheme expresses the intensive participation (Lerch 1939: 344) of the subject in the event.8 Given the existence of non-agentive reflexive psych verbs (RPSY) in 12th century Old French, intensive participation either does not imply agentivity or is no longer a general property of the subject in reflexive constructions at this point in time. A second important property of reflexive constructions, besides the semantic role (or agentivity/non-agentivity) of the subject, is the possibility that the reflexive can be interpreted as expressing a participant of the event or not, i.e. the possibility of receiving a semantic role or not. I assume that in the case of non-agentive reflexive psych verbs, the reflexive does not bear a semantic role: (64) a. b.

S(EXP) se(-) VERB A moi s(-) 'est Deus(EXP) coreciez, 'God became angry over me'

(Gui: verse 890)

If we now compare reflexive anticausatives with reflexive psych verbs on the one hand and with true reflexives on the other hand, it becomes clear that the reflexive anticausative has more in common with non-agentive reflexive psych verbs than with true reflexives. The difference between true reflexives and reflexive anticausatives with respect to the two parameters we are interested in is illustrated in (65). (65) a.

b.

true reflexive Jeani(AG) sei(TH) frappe avec un fléau. (S = AG, REFL = TH) 'Jean hits himself with a flagellum.' reflexive anticausative Le vase(TH) s'(-) est brisé. (S = TH, REFL = - ) 'The vase broke.'

In the case of the true reflexive, both the subject and the reflexive bear semantic roles. The subject bears the semantic role of an agent (AG), the reflexive morpheme is a theme (TH). In the case of the reflexive anticausative, only the subject, but not the reflexive morpheme, bears a semantic role; the subject is a theme. The change from true reflexive to reflexive anticausative involves thus (i) the loss of a semantic role and (ii) the change of the semantic role of the subject from agent to theme.

_________ 8

"Das Reflexiv drückte im Altfrz. die intensive Beteiligung des Subjekts an der Handlung aus; daher il se disne 'er isst für sich', il s'en vait, il se monte, il se rit, il se plore, il se pense, il se pent usw." (Lerch 1939: 344)

48 Note that in this representation a semantic role is attributed to the reflexive morpheme in its true reflexive use. Another way of representing this situation would be to state that the two semantic roles are attributed to the subject. Building on Kayne's (1975) insights on the behaviour of reflexive verbs in causative constructions, Grimshaw (1982), Wehrli (1986) and Reinhart & Siloni (2005: 392f.) (amongst others) argue that for present-day French, we must assume that the reflexive morpheme does not bear a semantic role in any of its uses and is always a valence operator that manipulates the argument structure in the lexicon. Recently, Labelle (2008) and Doron & Rappaport Hovav (2007) have argued against this view. Labelle's (2008) view on the role of the reflexive has already been presented above. Doron & Rappaport Hovav (2007: 3) assume that the reflexive in true reflexives and reciprocals can be an anaphor and accordingly, French true reflexives as in the example above would be syntactically transitive. For earlier stages of the language, especially for those stages at which we expect the emergence of the reflexive anticausative, a representation where the reflexive morpheme still bears a semantic role seems more adequate. If we now take a look at the three reflexive constructions together, the following picture emerges. Table 13: Comparison of true reflexive, reflexive psych verb and reflexive anticausative

true reflexive

reflexive psych verb

reflexive anticausative

animacy of S

animate

animate

inanimate

agency of S

agentive

non-agentive

non-agentive

macro role of S (semantic role)

actor (agent)

undergoer (experiencer)

undergoer (theme)

macro role of se

undergoer (theme)

-

-

The RPSY shares more common features with the RAC than the true reflexive does. The subject is non-agentive in the case of RPSY and RAC, but not in the case of the TR. Using the macro roles actor vs. undergoer (Van Valin & LaPolla 1997), the subject is an undergoer in the case of RPSY and RAC, while it is an actor in the case of TR. Finally, the reflexive does not bear a macro role or semantic role in the case of RPSY and RAC, while it does in the

49 case of TR. The only property that groups RPSY with TR is the animacy of the subject (subjects of RAC are predominantly inanimate).9 10 After comparing the three reflexive constructions with respect to their semantic properties we can now compare their formation. The different proposals on the formation of RAC have been presented in section 2.4. My own view on this topic is that the formation is achieved by means of a valence operation where the external (or actor) argument is deleted from the argument structure in the lexicon. In the following representations only the macro roles are given. (66) Derivation of reflexive anticausative input: VERB[actor, undergoer] the reflexive: deletes actor output: se VERB[undergoer]

For RPSY with non-agentive experiencers as subjects I assume the same operation. The difference in output – the subject is the experiencer and not the theme – results from a difference in the input. The lexical entry contains an experiencer and not a theme. But most importantly, RPSY and RAC have in common that they are formed with the same valence operation. (67) Derivation of non-agentive reflexive psych verbs input: VERB[actor, undergoer] the reflexive: deletes actor output: se VERB[undergoer]

In the case of true reflexives no such valence operation is assumed, given that the verb still has its two arguments. Note that in representations where no semantic role is attributed to the reflexive, a merging operation is assumed for the formation of true reflexives, where the cause and the theme merge in the argument in subject position (cf. Mutz 2005, Koontz-Garboden 2007, 2009). But this is not the representation I assume for Old French true reflexives as explained above.

_________ 9

10

Selig (1998: 31) makes the same observation with respect to animacy for the diachronic development of the Italian reflexive construction. Note that also in true reflexives there is not always identity between the referent of the subject and the referent of the reflexive (cf. Waltereit (1998) and Zribi-Hertz (1978)).

50 (68) Derivation of true reflexives (with merging of actor and undergoer in S) input: VERB[actor, undergoer] the reflexive: merges actor and undergoer output: se VERB[actor-undergoer]

To sum up, the chain TR > RPSY > RAC is both a chronological chain and a chain with the smallest possible differences between the constructions (cf. table 13). We may thus consider RPSY as an intermediate reflexive construction between TR and RAC both in terms of its properties and in terms of time.11 In the above, it became apparent that the difference between non-agentive reflexive psych verbs (with a transitive use) and reflexive anticausatives is only a comparatively small one, which is most obvious in the case of reflexive anticausatives with human subjects. This is also reflected in the literature, since, for instance, the difference between reflexive anticausatives with human subjects and non-agentive reflexive psych verbs cannot be captured by Reinhart's (2002) theta system. Reinhart would have to assume in both cases that there is a deletion operation which removes the external argument from the transitive theta grid (expletivization in her terms). In both cases, a [-c,+m] argument remains (cf. Reinhart 2002: 241f.). In Reinhart & Siloni (2005: 418), the same operation (decausativization) is proposed for the derivation of reflexive anticausatives and non-agentive reflexive psych verbs. To take this literature as a hint for the small difference between reflexive anticausatives and non-agentive reflexive psych verbs is only one possible interpretation. Another possible interpretation would be to consider non-agentive reflexive psych verbs as anticausatives, too. But even if one does so, the chronological precedence of one type of reflexive anticausatives, i.e. non-agentive reflexive psych verbs, with regard to another type of reflexive anticausatives, namely those which are not formed with psych verbs, remains. Depending on the delimitation of the term anticausative one chooses, the ideas laid out in this chapter are either about the emergence of reflexive anticausatives (from the point of view that psych verbs do not form anticausatives) or about the emergence of a new type of reflexive anticausatives (from the point of view that psych verbs can form anticausatives). A comprehensive description of French psych verbs is given in Kailuweit (2005). In this work, the author also discusses the question

_________ 11

Recall from sections 3.2 and 3.3 that I have interpreted the quantitative difference between reflexive anticausatives in the first and second half of the 12th century (none vs. very few) as indicating the emergence of the RAC in French in the course of the 12th century. But even if one does not follow this interpretation and assumes that the RAC emerged earlier in Old French, there is no doubt about a succession of RPSY > RAC given the quantitative difference between the two constructions.

51 of whether reflexive psych verbs are anticausatives. Kailuweit (2005: 129) points out that it is heavily debated in the literature whether reflexive psych verbs should be considered as lexical or grammatical pseudo-reflexives.12 Only the latter view would group them together with anticausatives. Before we can go on to the next section and discuss the mechanism of language change, one last point needs to be stressed which has already be mentioned before. The situation after the emergence of the reflexive anticausative differs from the one before in at least one obvious way: there is another reflexive construction with a non-agentive subject and a reflexive that does not bear a semantic role. One issue that is often mentioned in descriptions of the development of the French reflexive construction is that it does not acquire new functions that are added to the existing ones (A > A, B > A, B, C), but that instead it changes from a marker of the intensive participation of the subject to a valence operator (A > B) (cf. Lerch 1939, Mutz 2005). The question thus arises what role the emergence of the reflexive anticausative plays in this change. I assume that the reflexive in reflexive anticausatives deletes the external argument of a two-place predicate, so that the internal (or undergoer) argument surfaces in subject position and the reflexive is thus a valence operator. Accordingly, the existence of reflexive anticausatives implies the existence of a reflexive that is a valence operator (additionally it might also have other functions). However, I do not assume that the function of argument deletion is introduced by the emergence of the reflexive anticausative. Nonagentive reflexive psych verbs that have a transitive-causative counterpart are also formed via the deletion of the external argument. The function of argument deletion is thus not an innovation provided by the reflexive anticausative. Finally, it should be noted that the change of the reflexive construction is accompagnied by a change of its syntactic properties. Although the reflexive is a clitic since the beginning of Old French, in the course of Old French it becomes more and more attached to the (finite) verb of the sentence (a development that is not limited to the reflexive, but affects clitic object pronouns in general). Kaiser (1992: 148) (following Wanner 1987) notes that from the 11th century onwards a clear tendency of a verb based clitic placement can be found in the texts. The reflexive anticausatives that I have found for 12th century Old French are positioned next to the finite verb (sometimes another clitic element gets between the reflexive and the verb).

_________ 12

The term pseudo-reflexive is often used in Romance linguistics for reflexive constructions other than true reflexives and reciprocals (cf. Oesterreicher 1992).

52 3.4.3

The Mechanism of language change

Reanalysis is commonly considered to be an important mechanism for syntactic change (cf. De Smet 2009, Harris & Campbell 1995, Hopper & Traugott 1993). Reanalysis is the assignment of a new interpretation (syntactic or semantic) to an existing surface sequence without any change in the surface sequence itself (cf. De Smet 2009 1728f., Harris & Campbell 1995: 50f.). This is illustrated in figure 4: At t2 a new interpretation Y is assigned to a surface sequence A. As the interpretation Y did not exist for the surface sequence A at t1 Y is an innovation – A has thus been reanalyzed. t1 surface sequence interpretation

Figure 4:

A | X

t2 : reanalysis A / \ X Y

Reanalysis

A textbook example of reanalysis is the change of English going to from a verb of motion to a future auxiliary: (69) a. b.

Hermione is going to marry Ron. Structure: Hermione is goingverb of motion to marry Ron. Hermione is going to marry Ron. Structure: Hermione is goingfuture auxiliary to marry Ron. (Campbell 2004: 284)

This change can also be represented in a schema such as the one given in figure 5: At t1 the surface sequence going to only has the interpretation of a verb of motion. At t2 the surface sequence has been reanalyzed and can also have the new interpretation of a future auxiliary.

53 t1 surface sequence interpretation

Hermione is going to marry Ron. | verb of motion t2 : reanalysis

surface sequence interpretation

Figure 5:

Hermione is going to marry Ron. / \ verb of motion future auxiliary

Reanalysis of going to as a future auxiliary

The result of the reanalysis is the ambiguity of going to between an interpretation as a verb of motion and a future auxiliary. However, there are uses of going to in English where no such ambiguity exists, i.e. they can only receive an auxiliary interpretation and do not allow for the motion-interpretation. Consider the following examples from Campbell (2004: 285): (70) a. b. c.

It is going to rain on the muggles. Ron is going to like Hermione. Hermione is going to go to Hogwarts.

In none of these examples, going to is ambiguous, only the interpretation as a future auxiliary is available. As a consequence, these uses of going to as a future auxiliary could not emerge through the reanalysis of going to as a motion verb (hence the lack of a line relating these surface sequences with the interpretation verb of motion in figure 6). These uses emerge through a different mechanism: extension.

54 t1 Hermione is going to marry Ron. | verb of motion

surface sequence interpretation

t2 : reanalysis Hermione is going to marry Ron. / \ verb of motion future auxiliary

surface sequence interpretation

t3 : extension surface sequence

H. is going to marry R. |

interpretation

Figure 6:

verb of motion

\

H. is going to like R. /

future auxiliary

Reanalysis and extension of going to as a future auxiliary

Unlike reanalysis, extension creates new surface sequences and thus does not involve changes in the interpretation or underlying structure of an existing surface sequence, as proposed in Harris & Campbell (1995: 51). Similarly, De Smet (2009: 1729) includes two steps in his description of reanalysis: [...] the first step – the reanalysis proper – establishes a new category on the basis of ambiguous instances still licensed by the old source construction. The second step – the actualisation stage – creates new instances no longer licensed by the source construction. (De Smet 2009: 1729)

A representation of the succession of reanalysis and extension with variables is given in the following figure.

55

surface sequence Interpretation

Figure 7:

t1

t2 : reanalysis

t3 : extension

A | X

A / \ X Y

A B / \ / X Y

Reanalysis and extension

Turning to the reflexive anticausative, the mechanism of language change due to which it emerges is not entirely clear. Reanalysis would be one possible mechanism of language change to account for the emergence. Haspelmath (1990), for example, argues that it emerges through the reanalysis of true reflexives. He notes that "[d]ropping the agency restriction on [true] reflexives automatically leads to anticausative uses." (Haspelmath 1990: 44) This idea is illustrated with the following example. (71) Jean s' est blessé. Jean REFL is hurt

t1 surface sequence

t2 : reanalysis

Jean s'est blessé

Jean s'est blessé

| interpretation

Figure 8:

true reflexive

/ true reflexive

\ anticausative

Reanalysis of se blesser as anticausative

As long as the subject is interpreted as an agent, only the true reflexive reading where the subject willingly performs some action on himself is possible: Jeani(AG) s'i(TH) est blessé, i.e. that Jean hurt himself on purpose. At a given point in time, which would mark the emergence of the reflexive anticausative in the respective language, a speaker interprets the subject not as an agent but as a theme. Once the subject is interpreted as a theme, the reflexive morpheme can no longer be interpreted as bearing a semantic role and the anticausative reading naturally follows: Jean(TH) s'(-) est blessé. In the case of French, the new

56 and the old reading co-exist after the reanalysis, as strings like (71) are ambiguous between a true reflexive and an anticausative reading. Of course, not all true reflexives may be reanalyzed as reflexive anticausatives, as many of them are formed with verbs describing events that are not open to an anticausative conceptualization. The true reflexive use of se frapper 'hit oneself' is such a case. Presumably, as the lexical semantics of the verb frapper includes the existence of an external causer, it is this meaning component which blocks an anticausative interpretation of a sentence like (72). Due to this obligatory external causer, the interpretation of the subject as a theme would lead to a passive interpretation of the sentence. (72) Jean se frappe avec un fléau. 'Jean hits himself with a flagellum.'

Thus only a subset of the verbs that form true reflexives are potential candidates for the reanalysis as reflexive anticausatives.13 But there are not only examples of true reflexives such as Jean se frappe avec un fléau that cannot be reanalyzed as reflexive anticausatives, there are also reflexive anticausatives that are not ambiguous between a true reflexive and an anticausative interpretation and thus may not have emerged through reanalysis. An example of such a reflexive anticausative is given below. (73) Le vase s'est brisé 'The vase broke.'

Such reflexive anticausatives would thus emerge through extension, i.e. the creation of new surface sequences that have the interpretation introduced by the reanalysis of existing surface sequences. The following schema represents the succession of reanalysis and extension and involves three points in time: the time prior to the reanalysis (t1), the reanalysis (t2), and the extension (t3).

_________ 13

See chapter 2.3 for a discussion of the semantic restrictions on anticausative formation.

57 t1

t2 : reanalysis

t3 : extension

J. s'est blessé | | | TR

J. s'est blessé /\ / \ / \ TR AC

J. s'est blessé | \ Le vase s'est brisé | \ / | \ / TR AC

Figure 9:

Reanalysis and extension of se VERB as anticausative

Note that in the above succession the linguistic innovation proper, i.e. the first reflexive anticausative, emerges through the mechanism of reanalysis. In the literature this partition of labour (or chronology) between the two mechanisms is sometimes considered to hold generally for all language changes. For Campbell (2004: 285), extension is the process whereby new linguistic patterns emerging through reanalysis are extended to, for instance, new verbs. Hopper & Traugott (1993: 56) express the same view and argue for a similar partition of labour between reanalysis and analogy (or extension): "[R]eanalysis refers to the development of new forms out of old forms. [...] Analogy, by contrast, refers to the attraction of extant forms to already existing constructions [...]." However, in contrast to Campbell (2004) and Hopper & Traugott (1993), I follow Fischer (2007: 123) who rejects the above view on the partition of labour between reanalysis and extension and the primacy of reanalysis over analogical extension this view entails. I assume that linguistic innovations such as the ones that we are concerned with here may not only emerge through reanalysis. It might as well be the case that they emerge through extension. In this case the reflexive would be used in a new context (creating a new surface sequence with a new verb type or a new type of subject) and thereby receive an anticausative interpretation. I assume this process to be analogical. Analogy in language change means that "[...] one piece of the language changes to become more like another pattern in the language where speakers perceive the changing part as similar to the pattern that it changes to be more like." (Campbell 2004: 104) In the case of the emergence of reflexive anticausatives through analogical extension, first a similarity would be perceived between an existing use of the reflexive morpheme and the anticausative on the level of function (or semantics), and, as a consequence, a similarity in form would be established between the two by the use of the reflexive morpheme with a previously unmarked anticausative verb.

58 (74) Emergence of reflexive anticausatives through analogical extension (i) level of function: new similarity is perceived between unmarked intransitive anticausative and (non-anticausative) reflexive construction (ii) level of form: similarity of form is established, i.e. use of reflexive with previously strictly unmarked intransitive anticausative verbs

Sufficient functional-semantic similarities between an existing use of the reflexive morpheme and the anticausative are thus a necessary condition for this mechanism to apply. Given the considerable differences between true reflexives and anticausatives, e.g. with respect to the number and kind of semantic roles (cf. the discussion in the previous section), it does not seem plausible that sufficient similarities exist between the two constructions. The similarities between true reflexives and anticausatives are insufficient for speakers to treat the constructions as formally alike. Recall that after Late Latin the reflexive anticausative, but not the reflexive construction in general, was lost. In the description of the reflexive construction(s) prior to the emergence of reflexive anticausatives (based on the corpus for 12th century Old French), one reflexive construction with many similarities to reflexive anticausatives has been attested: reflexive psych verbs. Already before the emergence of reflexive anticausatives, non-referential uses of the reflexive morpheme and uses with non-agentive subjects existed. These uses could be possible anchors for the analogical extension of the reflexive morpheme to anticausative contexts. The following schema represents the emergence of reflexive anticausatives through analogical extension. There is a similarity between reflexive psych verbs on the one hand, and anticausatives on the other hand, with regard to the non-agentive subject, the number of arguments, or the lack of an external cause.14

_________ 14

I do not assume analogical extension to apply in a rule-like fashion to all sentences that have these three properties, i.e. there is no rule that could be paraphrased as "Reflexively mark all verbs in sentences with non-agentive subjects, with only one argument and without an external cause." The reason for rejecting this particular assumption lies in the fact that such a rule would make the wrong predictions. I will come back to this point in chapter 4, where I discuss the spread of RAC.

59 established similarity on the level of form: reflexively marked verb

source: A moi s'est Deus coreciez (75a.)

target: Que sa lance a estros ___ peçoie. (75b.)

similarity on the level of function: non-agentive subject, only one argument, no external cause Figure 10: Emergence of the French RAC through analogical extension (75) a. b.

A moi s'est Deus coreciez, 'God became angry over me' Que sa lance a estros peçoie. 'that his lance breaks into pieces'

(Gui: verse 890) (Lanc: verse 859)

In the above schema, the extension is represented as a surface-based change, i.e. the use of the reflexive morpheme creates a new surface sequence. However, what lies behind this creation of a new surface sequence is that the valence operation that I assume to apply in non-agentive reflexive psych verbs is extended to new verbs. This follows from the fact that the reflexive morpheme in non-agentive reflexive psych verbs and reflexive anticausatives is a valence operator and has the same function: it deletes the external argument from the argument structure. As the analogical extension is, in fact, the extension of a valence operation that does not exist in the language prior to nonagentive reflexive psych verbs, the existence of the non-agentive reflexive psych verbs is a necessary condition for the mechanism to apply. The idea that there must be an intermediate reflexive construction in the development of a marker from true reflexive to reflexive anticausative is not new and can also be found in Haspelmath (1987) (see also Geniušienơ 1987 and Kemmer 1993). In his monograph on anticausatives, he discusses the origin and further uses of anticausative morphology and consequently comes across the case where anticausative markers originally started out as markers of the true reflexive. For such cases he proposes the chain in (76) "[...] in which only minimal changes are allowed between two adjacent members [...]" (Haspelmath 1987: 29). (76) [true] reflexive > endoreflexive > anticausative

(Haspelmath 1987: 29)

60 Haspelmath's examples for endoreflexives are German sich hinsetzen 'sit down', sich anlehnen 'lean on', sich umdrehen 'turn', sich organisieren 'organize oneself', among others. He describes endoreflexives as reflexive constructions where the only participant of the event bears the semantic roles of actor and undergoer. The term endo- is chosen because the action described by the verb never "leaves" the participant and is only inside of him (which is not the case in true reflexives as e.g. hit oneself) (cf. Haspelmath 1987: 29). Unfortunately, Haspelmath (1987) does not specify what the relation ">", and, for that matter, the chain as a whole, is intended to signify. Some possibilities of interpretation could be chronological precedence, a synchronic implicational relation, or the statement that one member of the chain constitutes a necessary condition for the next member of the chain. However, since he does not provide any further specifications other than the ones indicated in the quote above we may in fact assume that this chain is the arrangement of the three reflexive constructions according to the smallest degree of semantic differences existing between them. In Haspelmath (1990), where he discusses the emergence of passive morphology, he proposes a different chain, illustrated in (77) below, to describe the development of reflexive morphology from the true reflexive to the passive via the anticausative. (77) [true] reflexive > anticausative > passive

One question that comes to mind if one compares the two chains is why the endoreflexive construction figures in the chain in (76) but is missing in the chain in (77). At first sight, this difference may be attributed to the fact that the two papers cover a different range of phenomena, and differ in the degree of detail that they provide. The chain in (77), from Haspelmath (1990), presumably does not include endoreflexives because the analysis put forward in this paper has a wider scope and does not describe the development from true reflexive to anticausative in the same detail as Haspelmath (1987). However, Haspelmath's description of the change from true reflexive to reflexive anticausative suggests that there is indeed a deeper difference between the chain in (76) and (77), irrespective of the overall aim of the respective paper. As a matter of fact, the difference lies in the observation that no intermediate construction is necessary: The main difference between [true] reflexive and anticausative is that the latter is mostly non-agentive. Dropping the agency restriction on reflexives automatically leads to anticausative uses. (Haspelmath 1990: 44f.)

The mechanism of language change that is responsible for the emergence of RAC directly applies to the true reflexive. To illustrate this point Haspelmath (1990) cites an example from Lakoff (1971: 158), which is given in (78).

61 (78) John hurt himself.

Depending on whether agency is attributed to the subject or not, the sentence in (21) either receives a true reflexive or an anticausative interpretation. From this perspective, it seems thus reasonable to assume that Haspelmath (1990) does not attribute any important role – which, for instance, could consist in being a necessary condition for RAC – to the intermediate endoreflexive in the chain in (76). However, the chain in (77), as proposed in Haspelmath (1990), has one important disadvantage: the author does not provide any empirical support for this hypothesis. Although Haspelmath (1990) states very clearly that reflexive anticausatives emerge via the reanalysis of an existing true reflexive string, not a single language is given for which this scenario has been empirically proven. Recall that in Old French there already existed an intermediate reflexive construction before the emergence of the reflexive anticausative. Thus the question arises whether the existence of an intermediate reflexive construction between true reflexives and reflexive anticausatives would also speak against reanalysis and for extension as the mechanism whereby the reflexive anticausative emerges. What implications does this fact have for the mechanism of language change whereby the French reflexive anticausatives might have emerged? Does it rule out reanalysis as the mechanism because the emergence is much simpler through the extension of an existing valence operation? In the following I will argue that a priori there is no reason to reject reanalysis (despite the existence of the intermediate construction). However, some properties commonly attributed to reanalysis must be relativized and reanalysis (at least in this case) appears more like a special case of extension. The first fact about reanalysis is that it creates a linguistic innovation, i.e. a pattern that has not existed in the language before. Thus the question arises where this innovation comes from ("the ontological issue" according to De Smet (2009: 1729)). One answer would be that certain surface sequences are structurally ambiguous between different interpretations or underlying structures and that reanalysis instantiates an interpretation that has not yet been in use. But if the ambiguity of a surface sequence is the result of the reanalysis, it cannot be the source or cause of the reanalysis (cf. De Smet 2009: 1729). Within theories of language assuming an innate grammar – e.g. generative grammar – this ontological problem can be solved relatively easily: the innovation is part of the innate grammar and simply has not yet been instantiated in the actual grammar of a given language. Before the reanalysis, one of the two interpretations would only exist in the innate grammar while the other interpretation would exist in both the innate and the actual grammar. After the reanalysis, both interpretations would be part of the actual grammar of the language (cf. also De Smet (2009: 1730) for a discussion of this issue). It is however questionable whether the term ambiguity should be used for the

62 situation prior the reanalysis at all. De Smet (2009) does not follow the assumption of an innate grammar and thus seeks a different solution to the "ontological issue", i.e. the question of where the innovations brought about by reanalysis come from. He rejects both the assumption of an innate grammar as the source of the new interpretation and the assumption that surface sequences are ambiguous prior to the reanalysis. He argues that [...] there is much more that goes into the process of reanalysis than just the two syntactic representations involved and the syntactic ambiguities that (supposedly) arise in particular surface sequences; and, […] much of the outcome of reanalysis is already available to the language user prior to reanalysis. (De Smet 2009: 1730)

He empirically motivates this view using the diachronic development of English worth and worthwhile + Gerund and for...to+infinitive as examples. The reanalysis of French true reflexives as reflexive anticausatives would also be a case where "much of the outcome" of the reanalysis existed in the language already before the reanalysis. Again, I am referring to reflexive psych verbs, which are attested prior to the emergence of reflexive anticausatives and which show strong similarities to them: the subject is non-agentive, the reflexive does not bear a semantic role, and they involve the deletion of the actor-argument from the argument grid and the linking of the undergoerargument in subject position. These properties are part of the outcome. I thus assume that, in a way, the reanalysis makes use of the ingredients that the intermediate reflexive construction offers, i.e. the possibility to interpret the reflexive morpheme as non-referential and the subject as non-agentive. In the figure below the idea of an extension being involved in the process of reanalysis is represented by the dashed line. t2 : reanalysis

t1 Jean s'est blessé | TR

Jean s'est blessé / TR

\

extension of existing valence operation: deletion of actor and linking of undergoer in subject position

AC

Figure 11: Reanalysis of se blesser 'hurt oneself/get hurt' as anticausative as a process of extension

The application of an existing valence operation is in fact the extension of an existing pattern in a new context. As a matter of fact, the emergence of reflexive anticausatives through reanalysis (as described above) and the emergence through extension proper (as described in figure 11 above) reveal themselves

63 to be very much alike. Reanalysis differs from extension insofar as in the latter case a new surface sequence is created, which is not the case with reanalysis. So far I have shown that the emergence of the French reflexive anticausative can be modelled with both reanalysis and analogical extension as the relevant mechanism of language change. I assume that in both cases the RPSY as an intermediate reflexive construction plays a crucial role for the change. As the emergence relies in both cases on the existence of this intermediate construction we may assume that it is a necessary condition for both mechanisms to apply and thus for the reflexive anticausative to emerge. As mentioned above there is no a priori reason to assume that the reflexive anticausative only emerges through reanalysis or through analogical extension. Hence, the determination of the relevant mechanism is an empirical question. Obviously, the data that needs to be investigated are the first occurrences of the reflexive anticausative in the language under discussion. A strong argument in favour of reanalysis as the relevant mechanism would be the observation that these first reflexive anticausatives have human subjects and are formed with verbs that can also appear in a true reflexive construction. Conversely, if the first attested reflexive anticausatives have inanimate subjects and are formed with verbs that cannot appear in any other reflexive construction, we would have a strong argument for analogical extension. The actual determination of the mechanism for a particular language on an empirical basis is, however, impeded by at least two methodological problems. The first problem is related to the nature of reanalysis itself. The first reflexive anticausatives must be ambiguous if they are supposed to have emerged through reanalysis. Just as it would be impossible to clearly determine the point in time from which on a given potentially ambiguous string with a reflexively marked verb can also receive an anticausative interpretation, it would equally be impossible to determine the first reflexive anticausatives if they had emerged through reanalysis. One solution to this problem would be to consider only those cases that can safely be categorized as anticausatives. But this leaves us with the possibility that the first reflexive anticausatives in the language are missing from the empirical base. Still, the data might speak in favour of one of the two mechanisms because there are also unambiguous examples of reflexive anticausatives with human subjects and verbs that can also appear in other reflexive constructions (e.g. the use of se blesser in the context of a fight). At this point, however, the second methodological problem comes into play. In order to decide between reanalysis and analogical extension as the relevant mechanism, the data must be conclusive in the sense that a significant quantitative difference exists between the instances of reflexive anticausatives pointing towards reanalysis on the one hand, and the instances pointing towards analogical extension on the other hand. However, as shown in section 3.2 and 3.3, the reflexive anticausative starts out with a very low level of frequency that increases only very slowly. As a consequence, the instances which

64 may be reasonably considered as the first reflexive anticausatives attested in the language form a very small set. For example, of the five reflexive anticausatives found in my corpus for 12th century Old French, three have an animate subject and two have an inanimate subject. But even if all five examples were to have the same type of subject with respect to animacy, the data would not be conclusive because of the small number of examples. It is unclear to me whether (and even if possible, how) these two methodological problems might be solved at all. Crucially, there is, to my knowledge, not a single empirical study which proves for any language that either reanalysis or analogical extension is the mechanism whereby reflexive anticausatives emerge. From the above considerations I must thus conclude that any statements with respect to the relevant mechanism remain speculative.15 Finally, with respect to the spread of the new reflexive anticausative the assumption is that both mechanisms apply. If the hypothesis that reflexive anticausatives emerged through the reanalysis of true reflexives is valid, there would still be cases such as Le vase s'est brisé 'The vase broke' which could not have emerged through reanalysis, given the lack of the possibility of a different reading. If analogical extension is the relevant mechanism, then we would still assume reanalysis to be responsible for the emergence of cases such as Jean s'est blessé 'John hurt himself' or 'John got hurt', i.e. a surface sequence that can also have a true reflexive interpretation. The fact that the data does not allow us to decide between reanalysis and analogical extension as the relevant mechanism is not satisfying. Still there are important lessons to be learned from this study of the emergence of French reflexive anticausatives. The first one is that extension – in this case the application of an existing valence operation in new contexts – is part of the emergence, regardless of whether the emergence is extension proper or reanalysis. The case studied here provide a further argument for the view that the difference between reanalysis and extension is rather small and that reanalysis does not necessarily create innovations "out of the blue" (cf. De Smet 2009). The second insight is that an intermediate reflexive construction is a necessary condition for the emergence of reflexive anticaustaives in French. This is another fact that reanalysis and extension have in common.

_________ 15

It should also be mentioned that Lerch (1939: 347) proposes yet another mechanism for the emergence of reflexive anticausatives, namely the personification of inanimate subjects. He argues that in the first reflexive anticausatives of the type La branche s'est cassée 'The branch broke.' the subject is understood as being the entity that brings about and undergoes the event. This personification of the inanimate subject would fade in the course of time leaving the language with an inanimate and non-agentive subject in a reflexive anticausative. However, the first examples of reflexive anticausatives do not provide any support for this assumption.

65

3.5

Summary

In this chapter I described the emergence of the French reflexive anticausative. Based on a corpus study I have argued that the reflexive anticausative emerges in the course of the 12th century. In a second corpus study I took a closer look at the situation in 12th century Old French in order to analyse the linguistic circumstances of the emergence. Two important facts have been observed. First, unmarked anticausatives are much more frequent in the 12th century than reflexive anticausatives (abs. freq. 129 vs. 5). Second, reflexive psych verbs with non-agentive subjects existed already before the emergence of the reflexive anticausative. Building on this latter finding I have argued that this intermediate reflexive construction (between true reflexive and reflexive anticausative) is a necessary condition for the reflexive anticausative to emerge. Crucially, both mechanisms of language change (reanalysis and extension) rely on this intermediate construction since in both cases an existing valence operation (applied in the formation of reflexive psych verbs) is extended. The difference between analogical extension proper and reanalysis is thus reduced to whether a new surface sequence is created in the course of the emergence of the reflexive anticausative. Finally, I have argued that the question of whether the first reflexive anticausatives emerged through reanalysis or analogical extension cannot be answered on empirical grounds nor on theoretical grounds (as the emergence can be modelled with both mechanisms). (79)

a.

b. true reflexive

intermediate reflexive construction

reflexive anticausative analogical extension

true reflexive

intermediate reflexive construction

reflexive anticausative reanalysis (with extension)

4

The Spread of the French reflexive anticausative

4.1

Introduction

In the previous chapter I described the emergence of the reflexive anticausative (as a new reflexive construction) in French. I argued that the reflexive anticausative did not directly emerge from the true reflexive, given that there is no direct link between the two constructions. Rather, the existence of an intermediate reflexive construction is assumed to be a necessary condition for the emergence of the RAC. In this sense, the previous chapter thus described the continual spread of the reflexive morpheme until it reached the domain of anticausatives. I will now continue my investigation of the historical development, with special emphasis on the spread of the reflexive morpheme within the domain of anticausatives. First of all, the term spread needs to be clarified and properly defined. It is important to bear in mind that a general characteristic feature, or even the very nature, of the spread of linguistic innovations is the fact that the spread occurs at different levels. These levels, in turn, may vary depending on the type of linguistic innovation. For the present case of reflexive anticausatives, for instance, the relevant levels would be (i) the speech community, (ii) the verbal lexicon and (iii) the anticausative use of individual verbs (cf. (80) below). (80) a. b. c.

Speech community: the share of speakers using RAC increases Verbal lexicon: the share of verbs forming RAC increases Anticausative use of individual verbs: the share of RAC (as opposed to UAC) increases

Concerning the speech community, the term spread denotes an increase in the share of speakers using RACs. With respect to the verbal lexicon, the notion means that the share of verbs that form RACs increases. Finally, regarding individual verbs, the term spread indicates that the share of reflexive anticausative uses of a given verb, based on the totality of anticausative uses of this verb, increases. We can now properly distinguish the notions emergence of the reflexive anticausative and spread of the reflexive anticausative. Both are phases of the same language change. However, the term emergence of the reflexive anticausative is reserved for the appearance of the first reflexive anticausatives in the language (late 12th century Old French; cf. chapter 3), the spread is the phase that follows this emergence. As new anticausative verbs may emerge in the

67 spread, the notion of emergence of new anticausative verbs must not be mistaken for the emergence of the reflexive anticausative as a new reflexive construction in the language. Thus the two phases of the language change in question are the emergence of the reflexive anticausative as the first and short phase and the spread of the reflexive anticausative as the second and – as will become apparent in the course of this chapter – long phase. The most complete and thus most satisfactory description of the development that reflexive anticausatives underwent after their emergence should include all three levels in (80). However, in the case under discussion, there are methodological constraints because of which the three levels cannot be considered to the same extent. Linguistic innovations start with one speaker or a small group of speakers and then spread through the population, much like a virus would. Following a given linguistic innovation to its source is, however, a very difficult task. Given the amount of possibly relevant data, one would have to have the good fortune to grasp the change at a very early stage, so that the community of speakers using the linguistic innovation can still be delimited in a straightforward manner. Labov's (1972) description of a sound change on the island of Martha's Vineyard is such a case. He was able to trace back the innovation (which was, in fact, the reactivation of an old existing sound pattern) to a small group of fishermen on the West coast of the island. From there the change spread and was taken over by non-fishermen who were affiliated to the group of fishermen in the sense that they represented the values that they also supported, i.e. the traditional, maritime-influenced way of life on the West side of the island, as opposed to the more touristy East coast. For the investigation of the spread of French RAC, there are no available data that would yield an investigation on a comparable micro-level. The empirical coverage of the French language for the period during which we might expect the emergence of RAC does not allow the tracing back of reflexive anticausatives as successfully as in the sound change case. That is, it is not possible to attribute the source of the change to a small group of speakers as potential sources/innovators, which was possible for Labov (1972) for the sound change case described above. Thus, since at present, there is no reliable method for measuring the spread of RAC in the speech community, I have to leave this aspect of the spread out of the discussion. However, as Aitchison (1991: 76) notes, "[l]anguage change spreads in two ways: outwardly through a community, and inwardly through a language". Given the methodological constraints mentioned above, I will not consider the level of the speech community but focus on the "inward spread through the language" of French reflexive anticausatives. This inward spread comprises the level of the verbal lexicon, i.e. the increase of the set of verbs forming reflexive anticausatives, and the level of the anticausative uses of individual verbs,

68 i.e. the increase of the reflexive anticausative use of a given verb at the cost of the unmarked intransitive anticausative use of the same verb. Unfortunately the corpus of examples which is the basis of this chapter's discussion does not yield a satisfactory analysis of the development of individual verbs, since their frequency of occurrence is not high enough. Level (iii), that is, the anticausative use of individual verbs, will be discussed in section 5.4, where I will present the results of case studies conducted on six anticausative verbs. There remain, however, enough aspects of the spread of RAC that can be described and that give us important insights on the manner and motivation of the spread of linguistic innovations. I will begin with a description of the development of the relative frequency and the relative lexical diffusion of RAC from Old to Modern French. Then I will discuss whether the (qualitative) properties of RAC change during the spread and whether these changes allow the identification of regularities in the spread. The properties of RAC that I will analyze are (i) semantic verb classes, (ii) the type of subject and (iii) the question of whether the respective verb can also be used as an unmarked anticausative. (The first two properties have also been analyzed for the unmarked anticausatives in 12th century Old French; cf. section 3.3.) At the end of this chapter I will turn to two more specific issues related to the spread of French RAC, namely (i) whether the course of the spread can be accounted for using statistical models of language change and (ii) the consequences that the spread of RAC had for UAC (which existed already prior to the emergence of the RAC).

4.2

Data basis and data extraction

- Old French Table 14: Corpus of Old French (source: NCA)

corpus (words)

1100

1170-1179

1240-1269

1270-1299

29924

138770

248320

354166

The data for Old French were extracted from the Nouveau Corpus d'Amsterdam (NCA).1 For each spot check the texts were searched for the

_________ 1

Cf. Kunstmann & Stein (2007) for a detailed description of the corpus.

69 forms se, s and soi. The results were then analyzed via a manual search for reflexive anticausatives. Since all instances of reflexive anticausatives needed to be extracted, this objective method of extraction was used. It would have been insufficient to limit the study to data extracted with searches for particular verbs that we expect to form anticausatives with the reflexive. - Middle French The data from Middle French were extracted from the electronic text corpus Dictionnaire du Moyen Français (DMF1). The DMF1 dictionary is, primarily, an electronic dictionary of Middle French. It does not only allow for various types of searches within the dictionary, but also for queries in the text corpus that is at its basis. This text corpus, called henceforth DMF1 corpus, is composed of 216 texts, dating from 1330 to 1520. The size of these texts varies from 317 to 216.062 words. The total number of words in the DMF1 corpus is 5.541.985. Again, all instances of the reflexive were extracted and were manually analysed. The first step in this procedure, the extraction of all instances of reflexive verbs in the DMF1 corpus, proved comparatively simple. To extract the relevant data, the atilf web interface for searches in the DMF1 corpus was used. In two queries every occurrence of the strings se and s, respectively, were extracted from the corpus, and specified with a context of two positions preceding and four positions following the reflexive. Note here that in Middle French the strings se and s did not only spell out the reflexive. For instance, se may also spell out demonstrative (cf. ce) or a conjunction (cf. si). The same holds for the string s. Therefore the results of the queries do not only include instances of reflexive verbs, but also examples that do not contain reflexive verbs. For the identification of reflexive anticausatives in this list of examples, in a second step the list was alphabetically ordered with respect to the position that followed se/s. This was achieved using UNIX tools. The ensuing analysis was, for the most part, done manually. Only the elimination of some nonreflexive occurrences of se/s, such as the deletion of instances where se/s is followed by a personal pronoun (il, ils, elle, elles), could be done automatically with UNIX tools. The Middle French period was divided into six sub-periods: Table 15: Corpus of Middle French (source: DMF1)

corpus (words)

13301359

13601389

13901419

14201449

14501479

14801520

506019

1359015

1547850

549735

949441

629923

70 The values for the respective periods are rounded values. More specifically, in all six sub-periods we find decimals rather than whole numbers to indicate the corpus size. This is due to the fact that some of the texts (28 out of 216) are counted into more than one sub-period.2 For every such text I thus calculated the average number of words per year in a first step, and then verified the number of years for the different sub-periods. The number of words per given sub-period is then calculated via the multiplication of the number of years by the number of words per year. Obviously, this approximation relies on the assumption that the total number of words of the respective text is evenly distributed over the years.3 The following example illustrates how the examples drawn from such texts are weighted. The Registre criminel de Châtelet is one of the texts that stretch over the boundaries of sub-periods; one year belongs to sub-period 2 (1360-1389) and three years belong to sub-period 3 (1390-1419). This gives us a ratio of 1:3 in terms of years. Thus, an example taken from this text would be counted as 0.25 for period 2 and 0.75 for period 3. As a consequence the absolute frequencies may include decimal numbers. In the presentation of the results these decimal numbers were, however, rounded to the next whole number. - Post-Middle French Table 16: Corpus of post-Middle French (source: Frantext)

1610-1615

1745

1990

29924

138770

248320

corpus (words)

As in the other two corpora, for each period, all occurrences of the strings se and s- were extracted from the corpus Frantext in a separate query. Unlike in Old and Middle French, however, the string se is only a reflexive and no longer the demonstrative ce or the conjunction si. S, on the other hand, can still be an instantiation of the conjunction si. As a consequence, the results of the queries do not only include instances of reflexive verbs, but also examples that do not contain reflexive verbs.

_________ 2

3

This can have different reasons: (i) the exact chronology of the text is unkown and the bibliographical information indicates a time span in which the text was produced, or (ii) the text is composed of a number of sub-texts written over a time span that stretches over the boundaries of the sub-periods. While this assumption is presumably a gross simplification, there seems to be no other way to manage the slight imbalance caused by these texts in the overall peryear evaluation.

71 By definition, verbs forming anticausatives must also have a transitive use (cf. section 2.2). For all relevant examples found in the corpora, I have thus verified whether the verb figuring in the example has a transitive use. Depending on the date of the example, the transitive use was checked in one of the following dictionaries: Altfranzösisches Wörterbuch, Dictionnaire du Moyen Français (DMF1), Trésor de la langue française informatisé (TLFi), Petit Robert. All findings on the emergence and spread of French reflexive anticausatives reported in chapter 4 rest upon the empirical basis described above. In line with Barra Jover (2007), I am aware of the limits of representativity that text corpora can have and acknowledge that expressions such as "the reflexive anticausative in Middle French" or "the emergence of the reflexive anticausative" should be regarded as short forms of "the reflexive anticausative in Middle French as it presents itself in my data" or "the first attestation of the reflexive anticausative in my corpus". Nevertheless I assume that corpora of historical texts offer a satisfying empirical basis for the study of a phenomenon like reflexive anticausatives. One huge advantage of reflexive anticausatives in this respect is that they are not determined by lexical material, but are instead a combination of three grammatical elements (subject, verb, and reflexive morpheme). One question that is given much space in Barra Jover (2007) is how one can determine for a language A that is no longer spoken (e.g. Middle French) whether the sequence X belongs to that language. Recall that I have argued in section 3.2 (based on data from the Nouveau Corpus d'Amsterdam) that the French reflexive anticausative emerges in the 12th century. This, however, must be understood as an existential and not an universal assertion (in the sense of Barra Jover (2007: 97)): The data does not suggest that every speaker of Old French towards the end of the 12th century uses reflexive anticausatives, but instead should be understood such that there existed speakers of Old French towards the end of the 12th century who use reflexive anticausatives. As a consequence, I assume that reflexive anticausatives belong to late 12th century Old French insofar as there are speakers using them, but I do not make any claims as to whether they belong to the idiolect of all speakers of late 12th century Old French. Variation among speakers and its role in the spread of the reflexive anticausative has not been taken into account in this study (cf. the remarks in section 4.1 and 4.5 on that matter).

72

4.3

Quantitative observations

4.3.1

The Increase of the relative frequency of RAC from Old to Modern French

As already mentioned in the introduction to chapter 4, the relative frequency of RAC is the first parameter with which I tried to capture the spread of RAC. The data pertaining to the present discussion were introduced in section 3.2. However, while section 3.2 focused on the point in time where RACs emerged, here the data is analysed as to the information they provide with respect to the spread of RAC. Before I turn to the discussion of the data one final remark on the presentation of the data is in order. In the previous section I outlined the empirical basis of the study, that is, that three spot checks were considered for Old French, six for Middle French and again three for the time after Middle French. For the sake of clarity I will only present data from two of the six spot checks that I have done for Middle French (1330-1359 and 1480-1520), since the four remaining spot checks for Middle French do not change the general picture concerning the development of the relative frequency of RACs from Old to Modern French. Table 17 gives the size of the respective corpora and the absolute frequency of reflexive anticausatives for eight spot checks from Old to Modern French. The relative frequency, which is given in figure 12, establishes the relation between the absolute frequency and the corpus size. The value 1 was assigned to the relative frequency of reflexive anticausatives in the second spot check (1170-1179) and the remaining values were calculated accordingly. Table 17: Absolute frequency of RAC from Old to Modern French

1100 1170-1179 1240-1269 1330-1359 1480-1520 1610-1615 1745 1990

corpus size (words) 29924 138770 248320 506019 629923 316898 343307 239269

abs. frequency 0 7 17 34 62 65 214 227

RF = y x 5.04x10^-5

73

18 15 12 9 6 3 0 1100

11701179

12401269

13301359

14801520

16101615

1745

1990

Figure 12: Relative frequency (RF) of RAC from Old to Modern French (value 1 assigned to RF of 1170-1179)

As table 17 shows, no reflexive anticausatives were found in the oldest text of the corpus, the Chanson de Roland (around 1100). The first reflexive anticausatives are attested in the second spot check (1170-1179). From then onwards, reflexive anticausatives are attested in all spot checks. Generally speaking, the relative frequency strongly increases from Old to Modern French. However, as figure 12 illustrates, the degree of increase is not stable; the increase starts rather slowly during Old and Middle French, takes off in the 17th century and then continues at a slower pace.4 The succession of slow, fast and slow increases results in a sigmoid graph. This is an interesting result, as it is often assumed that the succession of slow, fast and slow increase is not only one possible course of increase among many others, but that this particular succession of increases is instead the course of a "typical change" (cf. Aitchison 1991). In section 4.5 I will provide a detailed discussion on the shape of the graph and possible explanations for it. Apart from the temporal dimension of the increase, I will also address the question of the source of this increase. As described in the introduction to this chapter, linguistic innovations spread on different levels. The speech community, the verbal lexicon and the anticausative use of individual verbs are the relevant levels for the spread of reflexive anticausatives. With respect to the

_________ 4

As said, the increase during Old and Middle French is very slow. Thus the question arises whether the increase is statistically significant at all. It is difficult to answer this question properly since the total number of anticausatives for (the corpora of) the respective periods is unknown (which would be a necessary condition to run a chi-square-test). However, if we assume the relative frequency of anticausatives (RAC and UAC together) to be twice as high as in present-day French (spotcheck 1990), then the increase from 1170-1179 to 1480-1520 would be significant according to the chi-square-test (chi-square=11.2).

74 development of the relative frequency of RAC, it is important to note that all of the above levels can have an impact. The relative frequency of RAC is thus a parameter which is determined based on multiple causes. The most exhaustive way of spread would involve all three levels. Conversely, however, an increase of the relative frequency of RAC does not imply a spread of RAC on all these levels. In principle, it is even possible that the relative frequency of RAC increases while RAC actually decreases on two of the three levels. In order to find out more about the source for the increase of the relative frequency of RAC, as shown in figure 12, I will take a closer look at its development on the level of the verbal lexicon in the next section. 4.3.2

Lexical diffusion of RAC from Old to Modern French

Lexical diffusion is the spread of a given linguistic innovation on the level of the lexicon. In the present case this means that the share of verbs forming reflexive anticausatives increases. Although I will focus on the level of the lexicon, it has to be noted that linguistic innovations also spread on other levels (cf. section 4.1). To capture the quantitative aspect of the lexical diffusion, I use a very simple index, the relative lexical diffusion (RLD). This index relates the number of verbs that form reflexive anticausatives in a corpus to the size of the respective corpus and thus allows the comparison of the lexical diffusion of reflexive anticausatives (in the respective periods) with the different corpus sizes.5 verbs forming RAC (= types) (81) relative lexical diffusion (RLD) = -------------------------------corpus size

The development of the RLD from Old French to Modern French is illustrated in figure 13 (cf. table 18 for the number of verbs (verb types) and the respective corpus size). The overall development constitutes a strong increase from Old French to Modern French. However, there is no continuous increase over the whole time span. Instead, the RLD remains stable until 1520 and strongly increases only afterwards.

_________ 5

With respect to the parameter of relative lexical diffusion it should be noted that beyond a certain corpus size the verbs (as types) would repeat themselves.

RLD = y x 5.04x10^-5

75 12 10 8 6 4 2 0 11701179

12401269

13301359

14801520

1610-10

1745

1990

Figure 13: Relative lexical diffusion of RAC from Old to Modern French (value 1 assigned to RLD of 1170-1179) Table 18: Lexical diffusion of RAC from Old to Modern French

verbs corpus (words)

11701179 7

12401269 15

13301359 29

14801520 38

16101615 48

138770

248320

506019

629923

316898

1745

1990

92

130

343307

239269

RLD & RF = y x 5.04x10^-5

The following figure combines the developments of relative lexical diffusion and relative frequency of RAC.

20 15 10 5 0 11701179

12401269

13301359

14801520

RF of RAC

16101615

1745

1990

RLD of RAC

Figure 14: Relative lexical diffusion and relative frequency of RAC from Old to Modern French (value 1 assigned to RLD of 1170-1179)

76 Table 19: Relative lexical diffusion and relative frequency of RAC from Old to Modern French (value 1 assigned to RLD of 1170-1179)

11701179

12401269

13301359

14801520

16101615

1745

1990

RF

1

1.4

1.3

2.0

4

12.0

18.8

RLD

1

1.2

1.1

1.2

3

5.3

10.8

The increase of both parameters has already been described above. With respect to the relation between the two parameters it should be noted first of all that the RLD never exceeds the relative frequency. However, this in itself is not surprising since the number of types can never exceed the number of tokens. A second observation is that the distance between RLD and RF increases. The distance between the two graphs, in relation with the base line, indicates the average frequency of use of different verb types in RAC. If the two graphs are distant at a given point in time, then the average frequency of use of the verbs is higher, and lower if the graphs were closer together. Table 20 provides the average frequency of use for the verbs in RACs. Table 20: Average frequency of use of verbs in RAC from Old to Modern French

11701179

12401269

13301359

14801520

16101615

1745

1990

1

1.2

1.2

1.7

1.3

2.3

1.7

The diachronic comparison of the two parameters illustrates how the increase of lexical diffusion and the increase of frequency of use contribute to the relative frequency of RAC. For instance, the fact that the RF increases more strongly compared to the RLD after the Middle French period shows that both lexical diffusion and the increase of frequency of use contribute to the strong increase of the relative frequency of RAC.

4.4

Qualitative observations

4.4.1

Regularity

Both the relative frequency of RAC and the number of verbs that form RACs strongly increase from Old to Modern French. Thus, the data shows that the emergence of the RAC is followed by a spread of the RAC. Having discussed

77 these basic quantitative results of the spread, I now turn to the qualitative aspect of the spread which is closely related to the question of whether the spread of RACs is regular or not. A first observation in this respect is that the spread of French RACs is a slow process which began in the 12th century and presumably still continues. It is thus precisely not the case that the older pattern UAC is substituted with the emergence of the RAC nor that the RAC quickly becomes the only type of expression for verbal anticausatives in French.6 The spread of RAC is thus not rule-based in the sense that from a certain point in the history of French a given rule 1 "Form anticausatives with unmarked intransitive verbs" would be substituted by a different rule 2 "Form anticausatives with reflexively marked verbs". On the other hand, the slowness of the spread itself does not yet imply that the spread is irregular. Two relations are important when considering the regularity of the spread of RAC. The first one is the relation between RACs (or the properties of RACs) at different points in time. The second one is the relation between RAC and UAC (which can also be looked at at different points in time). These two relations are represented below as horizontal and vertical lines, respectively, in the following figure. t1

t2

t3

RAC

RAC

RAC

UAC

UAC

UAC

Figure 15: Relations and regularity

Both relations may be characterized by regularity or irregularity. I begin with the relation between RAC at different points in time (indicated by the dashed horizontal line). At this level the spread of RAC could be considered regular if the RAC spreads from one type of anticausative to another. Two possibilities need to be distinguished: A type enhancing spread where the reflexive spreads such that it enhances the types of anticausatives it appears with (as in (a.) below) and a type changing spread where the reflexive changes the type of anticausatives it appears with (as in (b.)). Obviously, a change of type constitutes a spread only if the new type is more frequent than the old one. In both cases the concept of regularity relies on which subtypes of anticausatives can

_________ 6

Cf. chapter 3.2 and, more generally, chapter 4.6 for a discussion of the persistence of UAC following the emergence of the RAC.

78 be expressed by RAC at different points in time and the spread would be considered regular if a succession of subtypes can be identified in the spread of RAC. (82) Regular spread of the reflexive in the domain of anticausatives a. type1 > type1,2 > type1,2,3 > type1-n type enhancing spread b. type1 > type2 > type3 > typen type changing spread

The idea that the spread of RAC can be regular although not all anticausatives are affected by it shows similarities to the Neogrammarian view on sound change. The Neogrammarians have postulated the lack of exceptions in sound changes. However, it is important to note that these sound laws did not necessarily apply on sounds as such, but could also include further conditions. A sound change could be limited to the use of a given sound in the onset, or, to use another example, be limited to stressed syllables. In this sense, there are many sound changes which do not affect a sound as such but only do so if the sound is specified for some further property (such as being in the onset, being in a stressed syllable, etc.).7 In a similar vein, if a sound does not have to be affected by a sound change in all its occurrences, not all unmarked anticausative verbs were necessarily affected by the emergence of RAC such that they were replaced by reflexively marked verbs. Similar to the creation of subtypes of a given sound via the postulation of further conditions for a sound change, there may exist subtypes within the category anticausative for which a regular spread is possible. The question is thus whether the spread of RAC is regular in the above sense, and which parameters create the subtypes of anticausatives that the regularity of the spread is based on.8 Semantic verb classes, the animacy of the subject and the distribution of the verb in UAC are the three parameters that will be presented as potential anchors of regularity.

_________ 7

8

Cf. Vennemann & Wilbur (1972), Labov (1994), and, more recently, Phillips (2006) for a discussion of the controversy on the nature of sound changes between the Neogrammarian view and the views involving lexical diffusion (as in the seminal work by Schuchardt (1885)). Von Heusinger (2008) shows that the spread of the differential object marker a in Spanish is an example of what has been called type enhancing spread above. The different types of direct objects that are marked with a (or not – depending on the period in the history of Spanish) are defined by (i) semantic and pragmatic properties of the object and (ii) lexical semantic properties of the verb with which the object co-occurs.

79 Table 21: Subtypes of anticausatives

specific category semantic verb class animacy of subject distribution in UAC

subtypes of anticausatives

verbs of change of state (COS) verbs of change of position (COP) aspectual verbs (ASP) verbs of appearence (APP)

AC formed with COS AC formed with COP AC formed with ASP AC formed with COP

animate inanimate

AC formed with animate S AC formed with inanimate S

yes no

AC-verb forms RAC and UAC AC-verb forms only UAC

The second level at which the regularity or irregularity of the spread of RAC may be analyzed is the relation between RAC and UAC. Concerning this relation, the spread would be regular if the relation between the two anticausative constructions is predictable or rule-based during the spread. To verify whether the spread fulfills this condition, one would have to compare reflexive anticausatives and unmarked anticausatives at various synchronic stages. Due to the lack of data on the unmarked anticausatives, I will focus on the relation between RAC at different points in time and consider the relation between RAC and UAC only insofar as I will compare the properties of RAC to the properties of UAC as in the corpus of 12th century anticausatives.9 4.4.2

Semantic verb classes

4.4.2.1

General points

Our first candidate for a parameter that may create relevant subtypes of anticausatives is the semantic verb class. In this case, a regular spread would imply that the RAC spreads from one semantic verb class to another (verb class1 > verb classes1,2 > verb classes1-n or verb class1 > verb class2 > verb classn). The following four semantic verb classes have been distinguished: verbs of change of state, aspectual verbs, verbs of appearance, and verbs of change of position.

_________ 9

Although the extraction of reflexive anticausatives from the text corpora was facilitated by the fact that the search could be restricted to the reflexive morpheme, for unmarked anticausatives no such restriction is possible. While a lot of "noise" had to be dealt with in the extraction of the reflexive anticausatives despite the restriction of the search, one would have to go through disproportionally more noise for the extraction of all unmarked anticausatives of large text corpora.

80 This distinction is inspired by Levin's (1993) verb classes for English. Examples for the use of the four semantic verb classes in reflexive anticausatives are given below. (83) a.

b.

c.

d.

Verbs of change of state (COS), e.g. se fondre 'melt' ils [=les cœurs] se fondent aux rayons du soleil [...] 'they [=hearts] melt in the rays of the sun' (Camus, 1615; Frantext) Aspectual verbs (ASP), e.g. se commencer 'begin' Lors se commancent [...] vendanges [...] 'then began the grape harvest' (La Vigne, 1496; DMF1) Verbs of appearance (APP), e.g. se manifester 'manifest itself' Car justice divine se manifeste en extermination des reprouvés 'Because divine justice manifests itself in the extermination of the reprobates' (Chartier, 1429; DMF1) Verbs of change of position (COP), e.g. se tourner 'turn' Le vent tellement se tourna [...] 'the wind turned in such a way' (Miracle de Saint Alexis, 1382; DMF1)

Before I go on to show how RACs are distributed in these semantic verb classes at different points in time, one short remark on the formation of semantic verb classes is in order. On the level of semantics the number of possible semantic verb classes is unlimited. For example, one could further divide the class of aspectual verbs in (83b.) above into (i) verbs of beginning, (ii) verbs of ending, and (iii) verbs of continuation. Thus the question arises of how a given inventory of semantic verb classes is motivated and which inventories are useful. The distinction of four semantic verb classes in (83) above is inspired by Levin's (1993) work on verb classes in English. However, the verb classes suggested in Levin (1993) are not semantic, since they are based on the distribution of verbs participating in valence alternations. The semantic labels of the verb classes are a result of the fact that the verb classes which result from the distribution in valence alternations possess, on the intuitive level, a semantic counterpart. Conversely, the classes proposed in Levin (1993) also cut through existing semantic verb classes (cf. the subdivisions within the class of aspectual verbs mentioned above). While Levin (1993) gives semantic names for classes which are motivated by the morpho-syntax, I apply the verb classes described above to the French RAC data, in the hope that these verb classes shed light on the diachronic developments under discussion.

81 4.4.2.2

Diachronic course

The diachronic development of reflexive anticausatives formed by verbs of change of state, aspectual verbs, verbs of appearance, and verbs of change of position is illustrated in figure 16 and table 22.

RF = y x 1.44x10^-5

50 40 30 20 10 0 11701179

12401269

13301359

COS

14801520 ASP

16101615 APP

1745

1990

COP

Figure 16: Relative frequency (RF) of RAC with 4 verb classes (value 1 assigned to RF of COS/1170-1179) Table 22: Relative frequency of RAC with 4 verb classes (value 1 assigned to RF of COS/1170-1179)

11701179

12401269

13301359

14801520

16101615

1745

1990

COS

1

2.8

2.3

2.7

8.5

26.5

44.1

ASP

0.5

0.6

0.4

2.3

0.7

2.4

6.7

APP

1

0.3

1.4

1.9

2.6

6.7

9.6

COP

1

1.1

0.7

0.4

2.4

7.7

5.5

As the above table shows, all four verb classes are continuously attested from the first period onwards, up to Modern French. There is thus no spread from one verb class to another in the sense discussed above. As a consequence, there is also no regularity in the spread which would be based on semantic verb classes. During the Old and Middle French period, when the overall relative frequency of reflexive anticausatives was still relatively low, the relative fre-

82 quency of the four verb classes changes only on a small scale. Substantial changes occur only after the Middle French period. As figure 16 illustrates, verbs of change of state constitute the verb class with the strongest increase. Crucially, from the 17th century onwards they outnumber the other three verb classes by far. The following figure shows the share of the different verb classes with respect to the totality of RACs, from the end of Middle French to present-day French.

100% 80% 60% 40% 20% 0% 1480-1520

1610-1615 COS

ASP

1745 APP

1990

COP

Figure 17: RACs in four semantic verb classes from Middle to Modern French

Figure 17 clearly indicates that during the strong increase of the relative frequency of RAC after Middle French there was also a strong increase in the class of verbs of change of state. The increase of the overall relative frequency is thus driven by the strong increase of verbs of change of state. It is only after the Middle French period that the RAC spreads to the semantic core class of anticausative verbs. To consider the class of verbs of change of state as the core of anticausative verbs is justified by two empirical observations: the frequency of the four classes with RAC in present-day French and their frequency with UAC in 12th century Old French. The distribution of the four semantic verb classes in 12th century unmarked anticausatives has already been described in section 3.3. Recall that verbs of change of state are the most frequent verb class (85.3%), followed by verbs of change of position with 12.4%. Only very rarely unmarked anticausatives are formed by verbs of appearance (0.8%) or aspectual verbs (1.5%). The dominance of verbs of change of state is an important commonality between 12th century unmarked anticausatives and post-middle French reflexive anticausatives. This comparison shows that the increase of reflexive anticausatives with verbs of change of state that can be observed after Middle French in fact constitutes the spread of RAC in the lexical core domain of anticausative verbs. The development of the reflexive anti-

83 causative is a development towards the situation of unmarked anticausatives in 12th century French. Despite this important commonality, 12th century UAC and post-Middle French RAC also show differences. Most importantly, the share of anticausatives formed with verbs of appearance or aspectual verbs is much higher with reflexive anticausatives than with unmarked anticausatives. The assumption that verbs of change of state are the core class of anticausatives is simply another expression of the fact that most verbs that can form anticausatives come from this class. Although the respective size of the four semantic verb classes is unknown for French, Levin's (1993) work on English still gives an idea of the dimension of the difference. In English, the number of verbs in the class of verbs of change of state amounts to more than 400, compared to, for example, only 13 aspectual verbs (cf. Levin 1993: 274). I assume that this difference is largely caused by extra-linguistic reasons. The number of aspectual verbs is naturally limited by the number of existing temporal relations that need to be expressed. Verbs of change of state, on the other hand, are a much larger class of verbs because there is a considerably greater number of different (types of) changes of state we perceive in the world around us. Therefore there is still good reason to believe that French aspectual verbs and verbs of change of state substantially differ with respect to class size and thus differ in their potential for lexical diffusion, even if we take into account the possibility of cross-linguistic and diachronic variation of class size. The comparison of the diachronic development of the relative frequency of the two verb classes with their potential for lexical diffusion shows that the two parameters correlate. The verb class with the higher potential for lexical diffusion, namely, the class of verbs of change of state, exhibits a higher degree of increase than the verb class with the much lower potential for lexical diffusion, that is, the class of aspectual verbs. The conclusion to be drawn with respect to the relevance of semantic verb class is thus twofold. On the one hand, the spread of the RAC in the verbal lexicon is not the spread from one semantic verb class to another. In this sense, the semantic verb classes have proved irrelevant for the identification of regularity in the spread of RAC. On the other hand, as shown above, not all four verb classes contribute to an equal degree to the strong increase of RAC after Middle French, in fact, the contribution of verbs of change of state is considerably higher than the contribution of the other three verb classes. The following tables list verbs forming reflexive anticausatives at different stages of French. For each stage a separate table sorted by semantic verb classes is given. Note that these lists are not exhaustive and thus do not represent the quantitative differences between the verb classes shown in figure 16.

84 Table 23: Examples of verbs forming reflexive anticausatives in Old French

semantic verb classes verbs of change of state aspectual verbs verbs of appearance verbs of change of position

verbs amollir 'soften', effondre 'cave in', croître 'grow', enfaiblir 'weaken', remuer 'change', refroidir 'cool out' arrêter 'stop', commencer 'begin', parfaire 'end' escorlogier 'disappear', éteindre 'cease', découvrir 'become visible', déserrer 'cease' bestourner 'turn', esparpiellier 'disperse', soulever 'rise', épandre 'spread', étendre 'spread', issir 'issue'

Table 24: Examples of verbs forming reflexive anticausatives in Middle French

semantic verb classes

verbs

verbs of change of state

aviver 'become lively', changer 'change', mouiller 'become wet', multiplier 'multiply', doubler 'double', amollir 'soften', augmenter 'increase', rompre 'break'

aspectual verbs

finir 'end', commencer 'begin', parfaire 'end', définer 'end', cesser 'end', terminer 'end', continuer 'continue'

verbs of appearance

évanouir 'disappear', former 'emerge', engendrer 'emerge', apparaître 'appear', demontrer 'show oneself'

verbs of change of position

étendre 'spread', élever 'rise', tourner 'turn', lever 'rise', arrêter 'stop', dilater 'spread'

85 Table 25: Examples of verbs forming reflexive anticausatives in 17th and 18th century French

semantic verb classes

verbs

verbs of change of state

durcir 'harden', transformer 'change', rallumer 'start to burn', enlaidir 'become ugly', dessécher 'dry out', renouveler 'renew', ouvrir 'open', embellir 'embellish'

aspectual verbs

achever 'end', parfaire 'end', arrêter 'stop', terminer 'end', éteindre 'end'

verbs of appearance

introduire 'come to existence', former 'emerge', effacer 'disappear', manifester 'manifest oneself', montrer 'appear'

verbs of change of position

glisser 'slide', tourner 'turn', arrêter 'stop', soulever 'rise', répandre 'spread'

Table 26: Examples of verbs forming reflexive anticausatives in present-day French

semantic verb classes verbs of change of state aspectual verbs verbs of appearance verbs of change of position

verbs alourdir 'become heavier', ouvrir 'open', élargir 'grow bigger', arrondir 'become rounder', obscurcir 'darken', gonfler 'swell' arrêter 'stop', achever 'end', accomplir 'end', terminer 'end' former 'emerge', effacer 'cease', montrer 'appear', éteindre 'disappear' lever 'rise', arrêter 'stop', glisser 'slide', dilater 'spread', rapprocher 'approach', abattre 'fall down'

These lists illustrate once again that there is no categorical difference between the reflexive anticausatives at different stages of French with respect to the semantic verb classes. The new construction does not spread from one semantic verb class to another, but instead it spreads in parallel fashion in all four semantic verb classes. In addition to the above lists, the following four sets provide for each semantic verb class a number of examples from Old to Modern French.

86 (84) verbs of change of state a. [...] la vie terrienne se change sovant, 'the earthly life changes often' (L'histoire de Barlaam et Josaphat, 1250; NCA) b. et les anciens mensonges se transforment en de belles petites veritez, 'and the old lies turn into beautiful little truths' (Béroalde de Verville, 1610; Frantext) c. [...] et son corps se desséchait comme celui d'une anorexique. 'and his/her body was drying up like the one of an anorexic' (Kristeva, 1990; Frantext) (85) verbs of appearance a. quant li segrei de mon cour s ovre et mis corage se descovre 'when the secret of my heart opens and my courage reveals itself' (Le livre des manieres de Etienne de Fougeres, 1176; NCA; mod. StH) b. de la fleur se forme le fruit, 'from the flower springs the fruit' (de Pizan, 1404; DMF1) c. [...] la reconnoissance des peuples, qui se manifeste toujours par de nouveaux services. 'the gratitude of the people which manifests itself always through new favours' (Duclos, 1745; Frantext) d. La foule se dissipa avec le départ du corps. 'the mob dispersed once the body was gone' (Guibert, 1990; Frantext) (86) aspectual verbs a. e se comensse le seruise 'and the service begins' (La vie du pape Saint Gregoire, 1175; NCA) b. et ainsi se fine le jeu. 'and so ends the play' (Miracle de sainte Bautheuch, 1376; DMF1) c. une tréve de douze jours, qui se termina par une paix [...] 'a ceasefire of twelve days that ended in a peace' (Duclos, 1745; Frantext) d. Et si la Révolution de 1789 s'était achevée seulement hier, en cette fin de Mai 68? 'And what if the Revolution of 1789 had ended only yesterday, in this end of may 1968? (Kristeva, 1990; Frantext) (87) verbs of change of position a. ou la plume en l air se souslieve 'where the feather rises in the air' (Le bestiaire d'amour rimé, 1250; NCA) b. le soleil se leve sans apporter le jour 'the sun rises without bringing the day' (Deimier, 1610; Frantext) c. Les jets de pierres s'abattent soudain sur les flics. 'the stones thrown come down on the cops' (Kristeva, 1990; Frantext)

87 4.4.3

Animacy of the subject

4.4.3.1

General points

Animacy is known to have various grammatical reflexes in the languages of the world. Examples of phenomena that can be controlled by animacy are case marking, number distinctions or verb agreement (cf. Comrie 1989: chapter 9). In the domain of anticausatives, animacy (as a binary distinction between animate and inanimate) is relevant in at least two respects. First, there is a relation between the animacy of an argument and the semantic roles it can fulfil. Inanimate entities, by definition, cannot bear the semantic role of experiencer and thus cannot figure in the subject position of a verb like fear which only allows for experiencers in this position. On the other hand, inanimate entities are ideal candidates for the subject position of anticausatives because such entities are non-agentive – a defining property of the subject of anticausatives. The subject position of anticausatives is, however, not limited to inanimate entities because animate entities may as well be non-agentive (cf. the discussion of reflexive psych verbs in chapter 3). Still, the typical example of anticausatives involves an inanimate subject that undergoes a change of state. (88) a.

b.

inanimate subject le vaisseau se brise; 'the boat breaks apart' (Baculard d'Arnaud, 1745; Frantext) animate subject Cette princesse [...] tomba de cheval, se blessa [...] 'this princess fell off the horse and hurt herself' (Duclos, 1745; Frantext)

The assumption that subjects in anticausatives are typically inanimate is supported by the data on 12th century Old French anticausatives (cf. section 3.3). 86.8% of the unmarked anticausatives had inanimate subjects. Animacy is thus a relevant category for anticausatives because they show a strong preference for inanimate subjects. The second point where one might expect animacy to play a role in anticausatives is more specific and closely related to the emergence of the reflexive anticausative. One could assume that if a marker X with the function A enhances its functions to B and C, it will first take over the function that is closer to the original function A. Similarly one could assume that if there are subtypes within the domain of anticausatives that differ with respect to their proximity to already existing reflexive constructions, then the reflexive will first spread to those subtypes that are closer to the already existing reflexive constructions. One of the main findings of chapter 3 was that reflexive psych verbs play an important role in the change in that they are an intermediate reflexive con-

88 struction between true reflexives and reflexive anticausatives. As both mechanisms of language change whereby the reflexive anticausative might have emerged crucially rely on non-agentive reflexive psych verbs I assume that this reflexive construction is a necessary condition for the emergence of the reflexive anticausative construction. Reflexive anticausatives can have both animate and inanimate subjects. Reflexive psych verbs, by their very nature, have animate subjects. Thus, the animacy of the subject creates subtypes of anticausatives that do not have the same proximity (or distance) to already existing reflexive constructions. Reflexive anticausatives with animate subjects are thus closer to existing reflexive constructions than reflexive anticausatives with inanimate subjects. If the above assumption about the relevance of proximity holds, we expect the reflexive to take over anticausatives with animate subjects first. (89) hypothesized spread reflexive psych verbs (S=anim.) > RAC with animate S > RAC with inanimate S

4.4.3.2

Diachronic course

The development of the relative frequency of reflexive anticausatives with animate and inanimate subjects is given in the table and figure below. As we can see, the two types develop very differently. While the relative frequency of reflexive anticausatives with animate subjects remains relatively stable over the whole time span, the relative frequency of reflexive anticausative with inanimate subjects strongly increases towards Modern French. In all periods, inanimate subjects are more frequent in reflexive anticausatives than animate ones. The development of the reflexive anticausatives with inanimate subjects resembles the one of the overall relative frequency (cf. section 4.3) in that the increase starts slowly, then accelerates after Middle French and slows down towards present-day French.

89

RF = y x 5.04x10^-5

20 15 10 5 0 11701179

12401269

13301359

14801520

animate

16101615

1745

1990

inanimate

Figure 18: Relative frequency of RACs with two types of subjects (value 1 assigned to RF of inanimate/1170-1179) Table 27: Relative frequency of RACs with two types of subjects (value 1: inanimate, 1170-1179)

animate inanimate

11701179 0 1

12401269 0.4 1

13301359 0.1 1.2

14801520 0.08 1.9

16101615 0.1 3.9

1745

1990

0.2 12.2

0.4 18.4

100% 80% 60% 40% 20% 0% 11701179

12401269

13301359 animate

14801520

16101615

1745

inanimate

Figure 19: Animacy of the subject in RACs from Old to Modern French

1990

90 The diachronic development does not show the succession predicted in the previous section. The reflexive anticausatives with animate subjects are not attested before the reflexive anticausatives with inanimate subjects (this has already been stated in chapter 3). It is not the case that there was first a period with only reflexive anticausatives with animate subjects, and then a period with reflexive anticausatives with animate and inanimate subjects. The analysis of the data with respect to animacy nevertheless reveals an interesting point, namely the diachronic stability of the frequency of reflexive anticausatives with inanimate subjects. The first conclusion to be drawn is relatively safe and straightforward: RACs with animate subjects reach their maximal frequency soon after the emergence of the RAC in French and long before the RAC with inanimate subjects. This fact, namely that the spread of the reflexive in the domain of anticausatives with animate subjects ends before the spread of the reflexive in the domain of anticausatives with inanimate subjects, might be interpreted as a hint that it also started first in this domain. Thus the diachronic development of the two types of reflexive anticausatives may be seen as indirect evidence for the fact that the first reflexive anticausatives predominantly had animate subjects (as predicted in the discussion in the previous section). The problems of finding direct evidence (or counter evidence) have already been discussed in chapter 3. 4.4.4

Possibility of use in UAC

In languages with two formal types of anticausatives (e.g. unmarked vs. reflexive anticausatives), anticausative verbs often fall into three classes, depending on their distribution with the two formal types: (i) verbs forming only type A, (ii) verbs forming both types and (iii) verbs forming only type B. For example, in present-day French, the verb durcir 'harden' can only form unmarked anticausatives, the verb casser 'break' can form both reflexive and unmarked anticausatives, and the verb briser 'break' can only form reflexive anticausatives. More French examples, together with examples from Italian and German, are given below.

91 Table 28: Distribution of verbs in RAC and UAC

Class I: UAC/*RAC

Class II: RAC/UAC

Class III: RAC/*UAC

cuire 'cook' durcir 'harden' sécher 'dry' pourrir 'rot'

casser 'break' gonfler 'inflate' refroidir 'cool out' ramollir 'soften'

briser 'break' améliorer 'improve' assécher 'dry' agrandir 'become bigger'

Italian (Folli 2002)

diminuire 'decrease' aumentare 'increase' invecchiare 'age' cambiare 'change'

fondere 'melt' cuocere 'cook' raffreddare 'cool' riscaldare 'warm up'

rompere 'break' alterare 'alter' svegliare 'wake up' aprire 'open'

German (Schäfer 2008)

schmelzen 'melt' kochen 'cook' zerbrechen 'break' umstürzen 'overturn'

abkühlen 'cool' abflachen 'flatten' kumulieren 'cumulate' ausdünnen 'thin out'

vergrößern 'enlarge' ausdehnen 'extend' verbreiten 'spread' verbessern 'improve'

French (Labelle 1992)

In the following, I will show that the distinction between the three classes of anticausative verbs given in the above table is relevant in the spread of the French reflexive anticausative. Obviously, before the emergence of the French reflexive anticausative, all anticausative verbs of the language belonged to the first of the above classes, i.e. they were all verbs forming only unmarked anticausatives. It is only after the emergence of the RAC that the second and the third class come into play. One might assume that classes two and three were successively filled with members during the spread of the reflexive anticausative (presumably at the expense of the first class). With respect to the relation between the classes two and three, one would expect that verbs originally forming only unmarked anticausatives do not directly pass from class one to class three. Instead, during a certain period, they were used in both constructions and only after that period, they were only used as a reflexive anticausative ((90a.) below). Besides this development there are two more possibilities for the rise of verbs forming only RAC. Verbs that came into the language after the emergence of the reflexive anticausative might start out either forming both RAC and UAC and then lose the latter option ((90b.) below) or forming only RAC right from the start ((90c.) below). (90) a. b. c.

UAC/*RAC > UAC/RAC > RAC/*UAC UAC/RAC > RAC/*UAC RAC/*UAC

Depending on how often these three scenarios occur, respectively, there will be a first phase in the spread of RAC where verbs forming both RAC and UAC

92 dominate verbs forming only RAC, or there will be no such phase. This is an empirical question and there are no a priori preferences for either of the two possibilities. If the reflexive anticausative first only applies to verbs that already form unmarked anticausatives, reflexive anticausatives would gain lexical ground from the UAC. If the reflexive anticausative first applies to verbs that have never formed unmarked anticausatives, reflexive anticausatives would conquer new lexical ground for the anticausative construction. In order to answer these questions, the diachronic development of the relative frequency of two types of reflexive anticausatives has been analyzed: (i) reflexive anticausatives formed with verbs that can also form unmarked anticausatives, and (ii) reflexive anticausatives formed with verbs that cannot form unmarked anticausatives. (91) a.

b.

RAC type 1: verb can be used in UAC La branche s'est cassée. (La branche a cassé.) 'The branch broke.' RAC type 2: verb cannot be used in UAC Le vase s'est brisé. (*Le vase a brisé.) 'The vase broke'

To obtain the relevant data, the examples for reflexive anticausatives have been coded as to whether the anticausative verb in the example can also form unmarked anticausatives. The possibility of an unmarked anticausative use was checked specifically for the period that the respective reflexive anticausative stems from.10 The development of the relative frequency of the two types of reflexive anticausatives is given in the table and figure below. Table 29: Relative frequency of two types of reflexive anticausatives (value 1 assigned to RF of type 1/1170-1179)

RAC type 1: UAC possible RAC type 2: UAC not possible

11701179

12401269

13301359

14801520

16101615

1.0

1.4

1.0

1.6

0.2

0.2

0.3

0.7

1745

1990

1.8

1.5

4.6

2.9

12.9

17.7

_________ 10

Sources: Altfranzösisches Wörterbuch for Old French; DMF2009 for Middle French; Furetière (1970) and Dubois & Lagane (1960) for 17th century French; Furetière (1970) for the 18th century French; Petit Robert for present-day French.

93

RF = y x 4.3x10^-5

20 15 10 5 0 11701179

12401269

13301359

RAC type 1 (UAC possible)

14801520

16101615

1745

1990

RAC type 2 (UAC not possible)

Figure 20: Relative frequency (RF) of two types of reflexive anticausatives (value 1 assigned to RF of type 1/1170-1179)

The frequency of use and the quantitative relation between the two types of reflexive anticausatives change drastically from Old to Modern French. The relative frequency of reflexive anticausatives of type 1, i.e. reflexive anticausatives formed by verbs which can also be used as unmarked anticausatives, remains relatively stable from Old French to the 18th century and only then increases towards present-day French. From Old French until the end of the Middle French period it is also the dominant type of reflexive anticausatives. For the development of the relative frequency of type 2, i.e. reflexive anticausatives formed with verbs that cannot be used as unmarked anticausatives, two phases can be distinguished: A phase of absence and very slow increase during Old and Middle French and a phase of strong increase of the construction after Middle French towards present-day French. In the course of this strong increase, the second type overtakes the first type and becomes the dominant type of reflexive anticausative. The current situation and thus the last observable stage of the development is such that the relative frequency of reflexive anticausatives of type 2 is four times higher than the relative frequency of reflexive anticausatives of type 1. The following table gives examples of the verbs that form the reflexive anticausatives in my corpus of examples. The verbs are grouped vertically according to period and horizontally according to whether they can form unmarked anticausatives or not.

94 Table 30: Verbs forming RAC and UAC and verbs forming only RAC from Old to Modern French

RAC/UAC

RAC/*UAC

Old French

abaisser 'go down/lower', changer 'change', guérir 'heal', muer 'move/change', ouvrir 'open', refroidir 'cool', tordre 'bend', tourner 'turn'

parfaire 'complete'

Middle French

accroître 'grow/increase', amollir 'soften', convertir 'convert/change', décroître 'decrease', ouvrir 'open', redoubler 'double', renforcer 'enforce/strengthen', terminer 'end'

dégâter 'take damage', démontrer 'show itself', manifester 'manifest itself', parfaire 'complete', prouver 'prouve itself', remplir 'fill'

16101615

augmenter 'increase', brûler 'burn', changer 'change', éclater 'burst', enlaidir 'become ugly', fondre 'melt', glisser 'glide', ouvrir 'open', tourner 'turn'

agiter 'move', corrompre 'become corrupt', dessécher 'dry out', dissiper 'desperse', fortifier 'become fortified', parfaire 'complete', perfectionner 'complete', putréfier 'rot', éteindre 'go out/fade', transformer 'change', vider 'empty'

1745

augmenter 'increase', briser (rare) 'break', changer 'change', embellir 'embellish', enfler 'swell', fermer 'close', fondre 'melt', multiplier 'multiply', ouvrir 'open', rompre 'break'

accroître 'grow/increase', affaiblir 'become weaker', blesser 'get hurt', convertir 'convert/change into', dessécher 'dry out', fortifier 'become fortified', effacer 'disappear', enflammer 'inflame', irriter 'become irritated', terminer 'end'

1990

allonger 'become longer', arrêter 'stop', casser 'break', fermer 'close', muer 'move/change', multiplier 'multiply', ouvrir 'open', plier 'bend', rompre 'break'

accroître 'grow/increase', amplifier 'increase/intenify', aplatir 'become flat', arrondir 'become round', briser 'break', éteindre 'go out/fade', réduire 'become less', remplir 'fill', transformer 'change', vider 'become empty'

Taking the developments of the two types of reflexive anticausatives together, it becomes evident that the role of the reflexive anticausative in the encoding of anticausatives changes from Old to Modern French. In Old and Middle French the reflexive anticausative was one of two possible anticausative uses for a given verb forming RAC. After the Middle French period, it often is the only possible anticausative use for a given anticausative verb forming RAC. During Old and Middle French, in the case of most verbs forming reflexive

95 anticausatives, the reflexive anticausative was only one option of two options and verbs forming only reflexive anticausatives were very rare. After the Middle French period, the frequency of reflexive anticausatives formed by verbs that can only form reflexive anticausatives strongly increases. Thus, the number of verbs where the reflexive anticausative is not just one of two possible uses but instead the only possible use for the anticausative increases. This is a first important conclusion. The second important conclusion concerns the diachronic course of the overall relative frequency of reflexive anticausatives (cf. section 4.3). We have seen that the increase of the relative frequency of RAC begins very slowly and accelerates not before the end of the Middle French period. If we compare the development of the relative frequency of RAC to the development of the two types of reflexive anticausatives, it becomes evident that the two developments are related. In fact, the strong increase of the frequency of RAC correlates with the strong increase of reflexive anticausatives formed with verbs which do not form unmarked anticausatives. This is an important finding because it increases our understanding of the specific temporal course of the change: The relative frequency of RAC increases once the reflexive anticausative ceases to be only one of two possible anticausative uses. The remaining question is whether the distinction between these two types of reflexive anticausatives provides us with types that reveal a regularity in the spread of the reflexive anticausatives. As the above data show, there is no succession of the two types – neither as type changing (type 1 > type 2), nor as type enhancing (type 1 > type 1, 2). Yet, if we compare the properties of reflexive anticausatives at different points in time, the distinction between the two types proves to be important.

100% 80% 60% 40% 20% 0% 11701179

12401269

13301359

RAC type 1 (UAC possible)

14801520

16101615

1745

1990

RAC type 2 (UAC not possible)

Figure 21: Share of two types of RAC from Old to Modern French

96 Although we cannot observe a succession of the two types, the comparison of the different synchronic cuts (or periods) shows that there is a succession with respect to the dominant type of reflexive anticausatives (cf. the table above). Until the end of Middle French, reflexive anticausatives formed with verbs that also form UAC dominate while after the Middle French period, it is just the other way around and the majority of reflexive anticausatives are formed by verbs that do not form UACs. Thus, there is a succession with respect to the dominant type of reflexive anticausative. This is also a difference with respect to the distinction based on semantic verb classes. Although the semantic verb classes also proved to be relevant for the understanding of the strong increase of RAC after Middle French, they did not reveal a change with respect to the dominant type of reflexive anticausatives – verbs of change of state were always the most frequent type. With respect to the question of regularity, I conclude that the modeling of the spread based on these two types of reflexive anticausatives comes very close to the hypothesized regular spread described in section 4.4.1.

4.5

The Temporal course of the increase of RAC

In the previous sections I pointed out that the emergence of the French RAC in the 12th century was not followed by an abrupt substitution of the older UAC by the new RAC, but rather by a slow spread of the RAC. The analysis of the parameters of relative lexical diffusion and relative frequency of RAC revealed that the spread is a continuous but not a constant process. There is a succession of slow, fast and slow increases of the relative frequency of RAC, resulting in an s-shaped (or sigmoid) graph. This form is often considered for the spread of linguistic innovations (cf. Aitchison 1991, Chen 1972, Crystal 1994). In the present section I will discuss possible explanations for the shape of this graph. A first explanation might lie in the succession of different types of anticausatives in the course of the spread of RAC. I illustrate this with the following hypothetical example. Let us assume that there are three types of anticausatives which are subsequently conquered by the reflexive and which do not have the same quantity.

97 type 1,2,3 type 1,2

type 1

t0

t1

t2

Figure 22: Succession of types of anticausatives in the spread of the RAC

At point t1, the RAC spreads from type 1 to type 2, and both types can subsequently be expressed by RACs. At t2, the RAC spreads from type 1 and 2 on the one hand to type 3 on the other hand, and all three types can subsequently be expressed by RACs. The characteristic shape of the graph, that is, the succession of slow, fast and slow increases, results from the quantitative differences between the different types of anticausatives. The anticausatives of type 2 constitute the largest group, and the conquest of this group by the reflexive would lead to a strong increase of RACs. The anticausatives of type 1 and 3, on the other hand, are quantitatively smaller groups, which is why their substitution by the RAC would not cause such a strong increase of RACs in comparison to type 2. The question is thus which parameters are responsible for the creation of such types. In section 4.4.2 we have seen that there is no spread of the RAC from one verb class to another, and that semantic verb classes are thus not as such a parameter that could provide an explanation for the course of the increase of the relative frequency of RACs. The similarity between the examples of RAC of Old French and those of present-day French, with respect to semantic verb classes, suggests that there is no succession of types of anticausatives involved in the spread of the construction. The spread seems to be rather a matter of quantity and not of quality. Instead, from its emergence on the RAC covers all types of anticausatives. Despite the lack of a succession of semantic verb classes, the consideration of semantic verb classes nevertheless contributes to a better understanding of the

98 temporal course of the increase of RAC. As shown in the previous section, the class of verbs of change of state shows the strongest increase of all four semantic verb classes after Middle French. Hence, the strong increase of the overall relative frequency may be attributed to changes of verbs of this semantic verb class in RACs (although not as in the idealized schema above). A second explanation for the particular development may lie in the change of the status of the reflexive anticausative, a fact which has been discussed in detail in section 4.4.4. I refer here to the observation that until the end of Middle French, reflexive anticausatives were predominantly formed by verbs which could also form unmarked anticausatives. For most anticausative verbs, the use in the reflexive anticausative was thus only one of two possible options. This changes drastically after the Middle French period, when the number of reflexive anticausatives formed by verbs which do not form unmarked anticausatives strongly increases. Until the end of the Middle French period, most verbs forming reflexive anticausatives where unspecified for the anticausative construction (i.e. allowing both anticausative constructions). It is only after the Middle French period that the language shows reflexive anticausative verbs – in the sense of verbs which can only be used in one anticausative construction, namely the reflexive anticausative construction. It is not surprising that this change in the status of the reflexive anticausative construction results in a change of its frequency of use. Historically, the change in the status and the increase of the relative frequency of the reflexive construction coincide, providing us thus with an explanation for the increase of the relative frequency after Middle French. The question of why the status of the RAC changed is yet another issue. In chapter 6 I will present a hypothesis on what made speakers use the RAC more frequently. The third possible explanation for the characteristic course of the increase of relative frequency of RAC does not presuppose a spread of RAC from one type of anticausative to another or a change in the status but instead relates the relative frequency of the innovation to the spread of the innovation in the speech community. Altmann (1983) proposes a mathematical model for the spread of linguistic innovations ("Piotrowski's law") which also models sigmoid increases. Such a sigmoid increase can be split into two phases, an exponential increase and a fade out (cf. (92)).

99 (92)

frequency of use of innovation

(i) exponential increase

(ii) fade out

t Osgood & Sebeok (1954-1965), in the quote given below, but also Altmann (1983), Aitchison (1991), Chen (1972), and Crystal (1994) attribute the reason for the sigmoid graph to the level of the speech community. Language change in a community will be gradual and cumulative, representing a continuous changing proportion of individuals who do or do not hear and produce a particular feature or set of features. The process of change in the community would most probably be represented by an S-curve. (Osgood & Sebeok 1954-1965: 155)

The sigmoid graph on the level of frequency of use of an innovation would thus reproduce the spread of this innovation in the speech community. One question that immediately arises is why the spread through the speech community would take the form of a sigmoid graph. Leopold (2005: 631) proposes that a "branching chain" might provide a suitable model to make the relation between the number of speakers using the innovation (infected speakers) and the degree of increase of the innovation explicit. If we assume that there is one speaker in t=1, then the number of infected speakers increases from 1 to 27 in the period t=3.11 The principal assumptions behind the exponential increase are thus that the innovation spreads through communication between speakers, and that – metaphorically speaking – with every newly infected speaker a new "diffuser" emerges. Therefore, the overall rate of diffusion increases, and the spread

_________ 11

Note that in this simplified illustration no infected speakers are "passed on" from one point in time to another. However, it is possible that at t2 there is the first infected speaker of t1 in addition to the three newly infected speakers of t2. The speed with which the increase proceeds could thus be even faster.

100 through the speech community accelerates even if the rate of diffusion of every individual speaker remains stable. I will now turn to the second phase of the spread, where the increase slows down until it comes to an end. In a finite world true exponential growth is impossible. As the growth of a population is limited by the finiteness of resources such as space, food, water etc., so is the spread of a linguistic innovation within a speaker community limited by the fact that the number of members of the community is finite. Under this premise we expect that the spread through the speech community comes to an end. However, the course of this second phase, i.e. the fade out of the increase, is not predicted by this hypothesis. It might be equally plausible to assume that the innovation exponentially spreads to the last speaker of the community and reaches an abrupt end. The question thus arises why spreads do not end abruptly but fade out instead. One explanation follows naturally from the above branching chain hypothesis. While the number of diffusers increases with every newly infected speaker, the number of non-infected speakers decreases. The consequence of this decrease of non-infected speakers is that at a certain point in time, not every infected speaker may make full use of his/her rate of diffusion because there is no sufficient number of non-infected speakers available anymore. At this point, the increase begins to slow down and eventually fades out. We can now turn to the second important issue, namely, the relation between the developments on the level of the speech community and the level of the frequency of the linguistic innovation in texts. As we have seen, the sigmoid increase of linguistic innovations is taken to be caused by the way linguistic innovations spread on the level of the speech community. Unfortunately, the spread of linguistic innovations on the level of the speech community is very difficult to investigate. For instance, suppose that, for a period of 100 years, only data from 8 authors is available. Then, obviously, no reasonable statements can be made with respect to the spread of a linguistic innovation in the speech community. Studies such as the one presented in Labov (1972) which do have the necessary depth of description are rare, especially for language changes that took place a long time ago. As a consequence, we have to rely on the assumption that changes in the speech community reproduce themselves on some other, more easily observable level, without having empirical evidence. There is, however, little evidence supporting this assumption. The assumption made above constitutes a projection of the observed change from the level of linguistic forms onto the level of the speech community, and not the reverse. Altmann's (1983) formulation of Piotrowski's law is strongly influenced by the epidemiological model of population growth, which he transfers to the level of the speech community. However, the law itself does not contain the parameter "speech community". It does not contain a parameter such as "the share of speakers that use an innovation vs. the share of speakers

101 that does not" in any conceivable form. Instead, Piotrowski's law makes a prediction about the development of the share of old as opposed to new forms. Leopold (2005: 631) has criticized Altmann's (1983) analysis for not explicitly stating the relation between the two parameters. Crucially, the relation between the spread in the speech community and the spread on other levels of language is purely stipulative, and nothing is known about its exact nature.12 The conclusion to be drawn is thus that the increase of the relative frequency of RAC obeys Piotrowski's law. However, given the current state of Piotrowski's law with respect to the modeling of the contribution of different levels to sigmoid increases, this generalization can be seen as providing an explanation for the course of the increase. Comparing now the three approaches to the problem (succession of types, change of status, level of speech community), it becomes evident that the change of the status of the reflexive anticausative contributes most to our understanding of the course of the increase of RAC. At the end of this section I will turn to one aspect of the spread that is also linked to the temporal dimension of the spread, but is less specific than the sigmoid shape of the graph: The slowness of the spread. The facts that I will consider are the following. Firstly, the spread of RAC started 800 years ago. Secondly, UAC still exist in present-day French. Thirdly, in present-day French the verbs forming RACs outnumber the verbs forming UACs by far (cf. section 4.6). RACs must be considered to be the preferred pattern to encode anticausatives in present-day French. This, however, implies that in the encoding of the causative-anticausative alternation the encoding with a formally marked anticausative is preferred over the labile type of encoding where the two members of the alternation are formally identical.13 However, prior to the emergence of French RACs, and most probably also for some time after the emergence of RACs, the labile type must have been the predominant type of encoding.14 The development of the relative frequency of RAC suggests that this situation lasted until the end of the Middle French period (about 1500). Abstracting away from the exact chronology it is obvious that the labile encoding was replaced by the encoding with a formally marked anticausative as

_________ 12

13

14

The present case under discussion, the spread of RAC, is further complicated by the fact that we are dealing with a linguistic innovation that also spreads through the lexicon. In fact, Piotrowski's law would predict a sigmoid spread through the speech community for every new verb that is conquered in the course of the lexical diffusion of RAC. Note that Haspelmath's (1993) anticausative type of encoding corresponds to my type of encoding with a formally marked anticausative. Cf. Kulikov (1999) and especially Kulikov (2003) and Lavidas (forthcoming) for a discussion of diachronic aspects of labile verbs.

102 the predominant pattern. It is also obvious that this change in the encoding of the French causative-anticausative alternation was a rather slow process. The question is thus whether this is a particularity of the change described in this book or a more general feature of changes in this domain of grammar. In a recent article Comrie (2006) addressed the diachronic stability of how languages formally encode the causative-anticausative alternation. He argues that great diachronic stability can be found in this domain and that substantial changes in this area take a long time to be fully established. The first set of empirical arguments that Comrie provides in support of this view is drawn from a comparison of Standard Arabic and Maltese, two languages that have developed separately for more than a millennium. Comrie (2006: 314) considers the fact that, despite this long time of separation, the two languages still encode 23 of 31 verb meanings with the same type, to suggest that there is diachronic stability in this domain. A second argument that he adduces is based on the observation that, in spite of the strong influence that Sicilian and, later on, Standard Italian have had on Maltese, the language has not adopted the predominant type of encoding of the two Romance languages that have been in close contact with it (cf. Comrie 2006: 315). The predominant type of encoding in Sicilian and Standard Italian is the reflexive anticausative. Comrie's third argument also goes back to the persistence of a certain type of encoding under (and despite) the strong influence of languages with a different type of encoding. The concrete case Comrie refers to are the Uralic languages. These languages have been in contact with the Indo-European languages of Europe for a long time, but did not adopt their preference for the formal marking of the anticausatives (cf. Comrie 2006: 316f.). Comrie (2006: 316f.) illustrates his claim with Finnish, as being one example among others. Finnish has kept its preference for the formal marking of the causative although it has been heavily influenced by neighboring languages in other domains of grammar: Given the heavy Indo-European influence on Finnish especially, with its adoption of subject - verb - object word order and of finite subordination, this would suggest that the transitivity profile of a language is indeed relatively stable diachronically. (Comrie 2006: 317)

Implicitly, Comrie's (2006) final argument also includes the change which has been central in this dissertation, the emergence and spread of French reflexive anticausatives. In doing so, Comrie (2006) refers to the time frame that the Indo-European languages of Central and Western Europe (aside from English) required to develop their preference for the reflexive anticausative following the loss of the productive causative morphology and the middle voice (which was also used to mark the anticausative) they had inherited from Proto-Indo-

103 European. Comrie (2006: 316) proposes two millennia as a plausible time frame for this process. In addition, he notes that the actual time span may have varied from language to language. At first sight, the results from my study on the spread of French reflexive anticausatives suggest a somewhat shorter time span for the development of the reflexive anticausative as a predominant pattern, for instance, 800 years since the emergence of RAC. However, if one adds the period in which the labile encoding was predominant, a situation that must be considered as a first important step in the restructuring of the encoding system, the actual time span covered is much closer to the two millennia that were proposed by Comrie (2006).

4.6

Consequences for UAC

I now turn to the consequences that the emergence and spread of the new type of anticausative had for the unmarked intransitive anticausative, that is, for the older type of verbal anticausatives in French. A very basic yet important observation is that the spread of the reflexive in the domain of anticausatives has not yet come to an end, in the sense that unmarked anticausatives would be fully replaced by reflexive anticausatives, and, hence, in present-day French, reflexive verbs would be the only verb type available to form anticausatives. As the examples in (93) show, unmarked intransitive anticausatives still exist in present-day French. (93) a.

b.

c.

le bois du lierre n'est pas solide, il casse facilement. 'the wood of the ivy is not hard, it breaks easily' (Le Monde 2002; TWIC; mod. StH) [...] le marché du charbon ne cesse de rétrécir en Grande-Bretagne. 'the coal market does not stop diminishing in Great Britain' (Le Monde 2002; TWIC; mod. StH) Le réalisateur aime les plans longs filmés de loin et à l'intérieur desquels la violence enfle et se déploie. 'the director likes long takes from a distance and inside of which violence swells and expands' (Le Monde 2002; TWIC; mod. StH)

In present-day French, besides verbs forming only reflexive anticausatives such as briser break', there are two other types of verbs, (i) verbs that can only form unmarked anticausatives and (ii) verbs that can form reflexive and unmarked anticausatives. The first case is illustrated with the verb commencer 'begin', the latter case is illustrated with the verb casser 'break'.

104 (94) RAC/*UAC a. Le verre s'est brisé. b. *Le verre a brisé. 'The glass broke.' (95) UAC/*RAC a. Les matchs ont commencé à 15 heures. (France-Info, 25.09.94 (Bassac 1995: 265)) b. *Les matchs se sont commencés à 15 heures. (Bassac 1995: 265) 'The matches began at 3 pm.' (96) RAC/UAC a. La branche s'est cassée. b. La branche a cassé. 'The branch broke.'

But how did the emergence and spread of the RAC change the status of UAC as a means to express anticausative semantics? The picture that emerges when considering the relevant evidence is twofold: on the one hand there is evidence that suggests that the spread of RAC resulted in massive substitution and loss for the UAC; on the other hand, UAC shows a lot of persistence during this process of loss and substitution. In the following I will present evidence for this view, beginning with the aspect of loss and substitution. The first evidence that shows how the role of UAC in the encoding of anticausatives must have changed since the emergence of RAC is the lexical diffusion of the reflexive and the unmarked anticausative in present-day French. As to the number of verbs that can form unmarked anticausatives in present-day French (that is, counting both the types of commencer and of casser), diverging statements can be found in the literature. On the basis of a spot check, Lagane (1967) estimates their number to be about 200. Rothemberg (1974), in turn, counts 311 such verbs in a corpus of 7.080 verbs. The only statement with respect to the number of verbs that can form reflexive anticausatives has been made by Zribi-Hertz (1987), who found 1.700 reflexive ergative verbs in a corpus of 5.400 verbs (these 1.700 verbs do not yet include verbs with the prefixes dé- and ré-). As Zribi-Hertz's definition of reflexive ergative is very similar to my definition of reflexive anticausative, we may assume that the numbers given above are fairly representative for the number of verbs that can form reflexive anticausatives in present-day French. The comparison of the numbers for RAC and UAC, on the other hand, indicates that in present-day French, the verbs that can form reflexive anticausatives largely outnumber the verbs that can form unmarked anticausatives (1700 reflexive anticausatives (cf. Zribi-Hertz 1987) vs. 311 unmarked anticausatives (cf. Rothemberg 1974)). These numbers refer to the situation in the verbal

105 lexicon. If this proportion reproduced itself on the level of frequency of use (which seems a reasonable assumption), then reflexive anticausatives outnumber unmarked anticausatives by far. A second empirical observation showing the loss of unmarked anticausatives comes from Heidinger (2005), a study on the diachronic development of twenty anticausative verbs with the prefix a- (cf. (97) below). (97) 20 anticausative verbs with prefix aabaisser 'make/become less', adoucir 'make/become softer', affadir 'make/become tasteless, dull', affaiblir 'make/become weak', affoler 'make/become crazy', aggraver 'make/become worse', agrandir 'make/become larger', alentir 'make/become slow', amaigrir 'make/become slim', amatir 'make/become dim', amenuiser 'make/become smaller', amoindrir 'reduce/become less', amollir 'make/become soft', anoblir 'make/become noble', apetisser 'make/become smaller', aplanir 'make/become even', appauvrir 'make/become poor', assécher 'make/become dry', attendrir 'make/become tender', aviver 'make/become lively'

All twenty verbs could form unmarked anticausatives in late Old French (around 1300). To find out how these verbs developed, their use was checked in various sources (Altfranzösisches Wörterbuch, Base de Lexiques du Moyen Français (BLMF), Base textuelle de Moyen Français (BTMF), Frantext and TLFi; the use as a reflexive anticausative was verified in TLFi). The unmarked anticausative use was checked for four points in time: 1300, 1500, 1700 and 2000 (representing present-day French). The reflexive anticausative use was only checked for the year 2000. The results are given in table 31 below. For each verb, the table shows whether it could be used in the unmarked or the reflexive anticausative at a given point in time. As already mentioned all twenty verbs allowed for the unmarked anticausative in 1300 and were thus labile at this point in time. Towards present-day French, however, all of them lost their ability to form unmarked anticausative. According to the TLFi, none of the verbs can be used in the unmarked anticausative in present-day French, but all verbs except one can be used in the reflexive anticausative.

106 Table 31: Diachronic development of anticausative use of 20 verbs with prefix a-

1300 abaisser adoucir affadir affaiblir affoler aggraver agrandir alentir amaigrir amatir amenuiser amoindrir amollir anoblir apetisser aplanir appauvrir assécher attendrir aviver

yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes

unmarked anticausative 1500 1700 2000 yes no no yes yes no no no yes no no yes yes no yes no yes yes yes no

no no no no no no no no yes no no no no no no no no no no no

no no no no no no no no no no no no no no no no no restricted no no

reflexive anticausative 2000 yes yes yes yes yes yes yes yes yes no yes yes yes yes yes yes yes yes yes yes

The results of this case study suggest that the consequences of the emergence and spread of the reflexive anticausative were indeed negative for the older pattern. The unmarked anticausative was frequently replaced by the reflexive anticausative. To sum up, the evidence shows that the emergence of the reflexive anticausative corresponded to the emergence of a concurring pattern for the unmarked anticausative. In the course of time, the reflexive anticausative conquered ground from the unmarked anticausative, gradually substituting the older pattern. Turning now to the second aspect, namely the persistence of the unmarked anticausative, I will discuss three observations as evidence: the lexical diffusion of UAC in present-day French, the emergence of UAC verbs during the spread of RAC, and finally, the temporal dimension of the spread.

107 The first piece of evidence is a statement that has already been made above, in the context of the number of verbs forming unmarked anticausatives in present-day French. According to Rothemberg (1974), their number amounts to more than 300, Lagane (1967) estimates their number to be about 200. In the above, these numbers were contrasted to the numbers of verbs forming reflexive anticausatives (about 1.700 according to Zribi-Hertz 1987). Although the number of 300 is less than 1.700, 300 verbs forming unmarked anticausatives is still a considerable number. Despite the strong increase of reflexive anticausatives, there are still a large number of verbs forming unmarked anticausatives in present-day French. One possible explanation for this would be that the unmarked anticausatives are remnants of an older stage of the language. While this might be the case for a certain number of verbs forming unmarked anticausatives, this does not hold in general. In fact, there are verbs where the unmarked form gained ground from the reflexive anticausative, contrary to the general trend described in section 3. Hence unmarked anticausatives (and with them the labile encoding of the alternation) are not just remnants of an older stage of the language. Fournier (1998: 89) gives the following examples of anticausative verbs which only marginally have the reflexive use in present-day French, but more freely allowed for both the unmarked anticausative and the reflexive anticausative in Classical French (17th century): commencer 'begin', changer 'change', empirer 'worsen', fondre 'melt', redoubler 'double'. In the case studies in chapter 5, I will present further verbs where the unmarked use increased with regard to the reflexive. The developments of the verbs augmenter 'increase', grossir 'grow/gain weight' and empirer 'worsen' have in common that the use in the reflexive anticausative decreases towards present-day French and the use in the unmarked anticausative increases. To summarize, unmarked anticausatives are not just remnants from an older stage of the language, and although the replacement of the unmarked anticausative by the reflexive anticausative may be the general trend, opposite developments can be observed on the level of individual verbs. The third and last observation concerning the persistence of the unmarked anticausative is the temporal dimension of the spread of the reflexive anticausative. The increase of the relative frequency, the increase of the relative lexical diffusion and the substitution of the unmarked anticausative by the reflexive anticausative in the case of the 20 anticausative verbs in Heidinger (2005) all are slow and not abrupt processes. Recall that during the Old and Middle French period the relative frequency and the lexical diffusion of the reflexive anticausative increased very slowly. Even today, 800 years after the emergence of RAC in French, the spread is still ongoing, in the sense that unmarked anticausatives still exist in the language. (Cf. also Comrie (2006) on the diachronic stability of the encoding of transitivity alternations as discussed in section 4.5.)

108

4.7

Summary

In this chapter I described the development of French RAC following its emergence in the 12th century. Both quantitative and qualitative aspects of this change have been taken into account. The parameters which have been investigated to capture the quantitative side spread of RAC are (i) the relative frequency of RAC and (ii) its relative lexical diffusion. The analysis of these two parameters has shown that the relative frequency as well as the relative lexical diffusion strongly increased from Old French to Modern French. The discrepancy between the increases of the two parameters shows that – unsurprisingly – the increase of the relative frequency is not only caused by a spread at the lexical level, but also by an increase of the frequency of use of the verbs forming RACs. With respect to the qualitative side of the spread, the main question was whether there are properties that distinguish Old from Modern French anticausatives. Three parameters have been considered: (i) semantic verb class, (ii) animacy of the subject and (iii) possibility of use in UAC. Beginning with semantic verb classes, one conclusion to be drawn is that the spread of RACs in the lexicon is not a spread from one verb class to another. Already in the first period when RACs were attested (1170-1179) all four semantic verbs classes that have been distinguished are attested (verbs of change of state, aspectual verbs, verb of appearance, verbs of change of position). Rather than there being a spread from one verb class to another, the new pattern spread in parallel fashion in all four semantic verb classes. A further result of the study is that verbs of change of state are the most frequent semantic verb class in RAC during the whole period. Hence there is no change as far as the semantic verb classes that form RAC (verbs of change of state, verbs of change of position, aspectual verbs, verbs of appearance) and the most frequent verb class (verbs of change of state) are concerned. What changes is the proportion between the four classes. Crucially, the dominance of the verbs of change of state over the other three classes increases after the Middle French period which is also the period when the overall relative frequency increases. With respect to the animacy of subject the situation is similar. Old and Modern French do not differ in an absolute way, but only the proportion between the different types changes. Already in Old French both animate and inanimate subjects are attested. But while the relative frequency with RAC with animate subjects remains stable until present-day French, the frequency of RAC with inanimate subjects strongly increases. Again, what sets apart Modern from Old French reflexive anticausatives with respect to animacy is the proportion between animate and inanimate subjects: the dominance of RACs with inanimate subjects increases.

109 The possibility of use in UAC proved to be the most interesting parameter because it shows the most drastic change from Old and Middle French to Modern French and it provides us with an explanation for the strong increase of the reflexive anticausative after Middle French. The difference between Old and Middle French and Modern French reflexive anticausatives is that until the end of Middle French, reflexive anticausatives were predominately formed by verbs which can also form unmarked anticausatives. After Middle French it is just the other way around and the huge majority of verbs in RACs cannot be used in UAC. With respect to the development of the relative frequency of RAC the study has revealed continuous but not constant increases from Old French to Modern French. The increase begins slowly, then accelerates after the end of the Middle French period before it slows down again after the middle of the 18th century. In a graphical representation, this succession can be represented by a sigmoid curve. This result is interesting because in the literature on language change, this particular shape is considered to be typical for the spread of linguistic innovations ("Piotrowski's law"). The change of the status of the RAC from one of two options for an anticausative verb (in Old and Middle French) to being the only option for most verbs forming reflexive anticausative verbs coincides with the strong increase of the relative frequency after Middle French. In chapter 6 I will present a hypothesis that relates the increase of the RAC to a change in the auxiliary selection of intransitive verbs in French. This might as well provide an explanation for the change of the status of the RAC.

5

The Semantic relation between French reflexive and unmarked anticausatives

5.1

Introduction

There is not only cross-linguistic variation in the encoding of anticausatives. Even within a single language, different formal types of anticausatives may exist. For the majority of languages for which this applies, the two types of anticausatives stand to each other in a marked-unmarked relation as far as the level of form is concerned (e.g. Greek, Albanian, German, Spanish, Italian). With the emergence of the reflexive anticausative in the 12th century, French also became such a language. There is a general consensus that in the lexicon, cases where a difference in form does not correspond to a difference in meaning (i.e. synonymy) are very rare if not non-existent. In most cases a difference in form does correspond to a difference in meaning. Similarly, on the level of grammatical constructions, Bolinger (1977) observes that different grammatical constructions always spell out different functions, a claim that effectively denies the possibility of synonymy in grammar. In this chapter, I will consider both the aspectual and the causal structure of French anticausatives, in order to provide insights as to whether the two formal types of anticausatives differ with respect to these two semantic parameters.1 The chapter is structured as follows. In section 5.2, I review the literature on French anticausatives, which suggests that RAC and UAC differ with respect to their aspectual structure.2 In section 5.3 I provide a review of the literature in which a difference in the causal structure of the two constructions is assumed. We will see that these proposals, while based on interesting empirical observations, sometimes require further empirical support. In section 5.4 I will provide a detailed description and discussion of a corpus study I conducted to test the alleged semantic differences between RAC and UAC with empirical data. In section 5.5 I discuss several aspects of these findings in more detail.

_________ 1

2

For recent overviews on possible semantic differences between French RAC and UAC cf. Larjavaara (2000), Ben Salah-Tlili (2007) and Kailuweit (to appear) (with the latter also considering other Romance languages besides French). Folli (2002) has argued for an aspectual difference between Italian reflexive and unmarked anticausatives.

111

5.2

Aspectual structure

5.2.1

Zribi-Hertz (1987)

Zribi-Hertz (1987) argues that reflexive and unmarked anticausatives have different aspectual properties. She bases her discussion on the following working hypothesis: While French reflexive anticausatives are perfective, French unmarked anticausatives tend to be imperfective (cf. Zribi-Hertz 1987: 41). Zribi-Hertz presents two arguments to empirically support this hypothesis. First, she argues that the verbs that appear in RACs and UACs have different aspectual properties. Second, she shows that aspectual operators such as perfective prepositional phrases influence the choice between RAC and UAC.3 I begin by reviewing the first argument, which crucially relies on the distinction between perfective and imperfective verbs. According to Zribi-Hertz's (1987: 41) definition, a verb is perfective if it can receive a resultative reading in a short passive in present tense. All other verbs are imperfective. Thus classified, the French verbs attendre and manger fall into two different aspectual classes. (98) a.

b.

Alfred est attendu Alfred is awaited (resultative reading impossible, attendre 'await' is an imperfective verb) Neuf pommes sont mangées nine apples are eaten (resultative reading possible, manger 'eat' is a perfective verb) (Zribi-Hertz 1987: 41)

Verbs that can occur in such a short passive are aspectually characterized as leading to a term (Zribi-Hertz 1987: 41). This characterization, together with the fact that the above test refers to a property of verbs and not of the use of verbs as such, suggests that Zribi-Hertz's (1987) notion of perfectivity corresponds to what others refer to as telicity, i.e. the level of aktionsart or lexical aspect of verbs. In the course of her argumentation, Zribi-Hertz extends her definition of perfective verb. According to the modified definition, a perfective verb is a verb with a morphologically related form which can receive a resultative reading in a short passive or a copula construction.

_________ 3

Zribi-Hertz (1987: 47) uses the French term complément prépositionnel which I replace by the more general term prepositional phrase (PP) in the English translation.

112 (99) a.

b.

c.

resultative passive Une situation de guerre civile est (*très) installée dans le pays. a situation of civil war is (very) set.in in the country adjectival passive Le gâteau est (très) caramélisé. 'The cake is (very) caramelised.' copula with related adjective4 Le temps est meilleur. 'The weather is better.' (Zribi-Hertz 1987: 42f.)

Despite this extension, the term perfectivity is still defined as being a property of verbs, while the working hypothesis refers to the aspectual properties of the two anticausative constructions themselves. The verbs and the constructions they occur in are linked by the assumption that the aspectual properties of the verbs that appear in RACs and UACs reflect the aspectual properties of the RAC and UAC. The aspectual restrictions that the constructions impose on the verbs are thus taken to be a reflex of the aspectual properties of the constructions themselves. The most clear-cut distribution of verbs that appear in the two constructions would be a complementary distribution of perfective and imperfective verbs, as illustrated in (100). (100) a. b.

reflexive anticausative unmarked anticausative (to be revised)

ļ ļ

perfective verb imperfective verb

Perfective verbs would then only be licit in the reflexive anticausative construction, and the reflexive construction would only allow for perfective verbs, while imperfective verbs would only be acceptable in the unmarked anticausative construction, and the unmarked construction would only permit imperfective verbs. The empirical facts of French are, however, not as straightforward. The verb casser 'break', for instance, clearly shows that the generalization in (100) does not hold, since casser is perfective according to Zribi-Hertz's definition, but can form both reflexive and unmarked anticausatives (cf. below). (101) a. b.

La branche s'est cassée. La branche a cassé. 'The branch broke.'

_________ 4

Zribi-Hertz (1987) needs to include this type of construction because of cases where the participle construction is blocked by an adjectival copula (cf. also Lagae (2005: 134f.) for a discussion).

113 Zribi-Hertz acknowledges these cases and develops a generalization which is able to capture these examples. In her generalization, which I will summarize here, Zribi-Hertz (1987: 45) relates the distribution of verbs to three morphosyntactic contexts, defined as follows: (i) a transitive-causative use of the verb, where the subject can bear the semantic role cause and the verb has a translative meaning, (ii) a construction denoting a resultant state, and (iii) the reflexive anticausative construction (RAC). She relates the specific transitive use and the existence of a resultant state on the one hand to the reflexive anticausative construction on the other hand. Unfortunately, the implicational relation that exists between these two poles is not clearly stated. In a different context, Zribi-Hertz (1987: 46) states that (i) a transitive-causative use of the verb, where the subject can bear the semantic role cause and the verb has a translative meaning, (ii) a construction denoting a resultant state, are both prerequisites for the existence of the RAC. This strongly suggests that she assumes that only perfective verbs can form reflexive anticausatives, as schematized in (102) below. (102) RAC ĺ perfective verb

The above generalization further implies that imperfective verbs cannot form reflexive anticausatives, as illustrated in (103) below. (103) imperfective verb ĺ *RAC

Note further that (102) is a generalization on the reflexive anticausative construction and not on perfective verbs. It does not make a statement about the distribution of perfective verbs, besides the observation that some of these verbs can form reflexive anticausatives. The case of casser already illustrated that perfective verbs can indeed form unmarked anticausatives. Being defined as a perfective verb thus does not imply that a particular verb that is covered by this definition can only form RACs. (104) perfective verb: RAC, UAC

A working generalization thus must pertain to the reflexive construction on the one hand and to imperfective verbs on the other hand. It is only the distribution of imperfective verbs that can be predicted by this generalization, and only the RAC is aspectually specified: D'une manière générale, la CRE [=RAC] est ouverte à des verbes perfectifs, et les verbes à renversements imperfectifs n'ont qu'une forme ergative non réflexive. (Zribi-Hertz 1987: 47)

114 The generalizations on the distribution of perfective and imperfective verbs in RAC and UAC that can be drawn from the above discussion are summarized in (105) below. (105) a. b. c. d.

RAC ĺ perfective verbs imperfective verbs ĺ *RAC perfective verbs: RAC, UAC UAC: perfective verbs, imperfective verbs

Lagae (1990) provides further empirical data in order to verify the empirical validity of Zribi-Hertz's (1987) generalization. She considers two series of data that, according to her, falsify the generalization. The first set of data consists of verbs with a transitive-causative use that can form RACs but do not occur in short passives with a resultative reading. (106) First type of alleged counter examples to Zribi-Hertz (1987) transitive-causative, RAC, *short passive with resultative reading (cf. Lagae 1990: 36)

Lagae gives gâter '(make) go bad', abêtir 'make/become stupid', enlaidir 'make/become ugly', épaissir 'thichen' as examples for verbs with the above properties. However, these properties actually differ from the ones given in Zribi-Hertz's final generalization (Zribi-Hertz 1987: 45), since only the first, very narrow definition of perfective verb is applied. In Zribi-Hertz's (1987) final generalization, perfective verbs are instead defined as verbs that have a morphologically related form that designates a resultant state. As a consequence, Lagae's (1990) data could only falsify the preliminary version of ZribiHertz's (1987) generalization. In a way, though, Lagae (1990) reproduces a dialectic step of Zribi-Hertz's (1987) argumentation that led Zribi-Hertz to broaden her definition of perfective verbs. Thus, despite the fact that Lagae (1990) refers to a pre-version of Zribi-Hertz's (1987) generalization, in the presentation of her alleged counterexamples, Lagae actually analyzes both (i) whether a verb has a short passive with a resultative reading, and (ii) whether the verb has some other morphologically related form designating a resultant state, the latter in fact applying Zribi-Hertz's (1987) final generalization. Her data is thus enabling us to verify the empirical validity of the generalization that only perfective verbs can form RACs. From the four verbs that do not have a short passive with a resultative reading, only gâter does not have a morphologically related form designating a resultant state (cf. below).

115 (107) a. b. c. d.

Cet événement gâte nos affaires 'This event ruins our matters.' Cet événement fait se gâter nos affaires 'This event makes our matters go bad.' Nos affaires se gâtent 'Our matters go bad.' *Nos affaires sont gâtées our matters are gone.bad

The data presented in Lagae (1990) thus suggest that the generalization that only perfective verbs form reflexive anticausatives holds, even if only as a very strong tendency and not categorically. The second type of alleged counterexamples provided in Lagae (1990) is a group of verbs with a transitive-causative use that can form a short passive with a resultative reading, but are ungrammatical in the RAC construction (cf. below). (108) Second type of alleged counter examples to Zribi-Hertz (1987) transitive-causative, short passive with resultative reading, *RAC (cf. Lagae 1990: 35)

According to Lagae (1990), verbs with these properties are, for instance, boucler 'curl', bouffir 'swell', brûler 'swell', carboniser 'carbonize', chauffer 'heat', cuire 'cook', frire 'fry', friser 'curl', rôtir 'roast', roussir 'make/go brown', sécher 'dry', tiédir 'cool'.5 However, again, these verbs do not falsify ZribiHertz's (1987) generalization since the generalization itself does not rule them out as such; rather, these verbs show that perfectivity and a transitive-causative use are not sufficient conditions to be grammatical in the RAC construction (cf. the example of brûler 'burn' from Lagae (1990: 35)). (109) a. b. c.

Les flammes brûlent le papier 'The flames burn the paper' Les flammes font brûler le papier 'The flames make the paper burn.' Le papier brûle 'The paper burns.'

_________ 5

It should be noted that there seems to be some variation among French native speakers concerning the acceptability of these verbs as RACs. Contrary to Lagae (1990), Anne Zribi-Hertz (p.c.) accepts most of them as RACs in the context of perfective PPs.

116 d. e.

*Le papier se brûle 'The paper burns.' Le papier est brûlé the paper is burned

Recall that so far, two subclasses of perfective verbs have been identified according to their distribution in anticausative constructions: verbs like casser that appear both in the RAC and the UAC construction, and verbs like briser that appear only in the RAC construction. The second type of the alleged counterexamples from Lagae (1990), which consists of verbs such as brûler, adds a third subclass to the set of perfective verbs, namely, perfective verbs that appear in the UAC, but are ungrammatical in the RAC. Thus, according to their distribution in RACs and UACs perfective verbs forming anticausatives fall into the three subclasses given below in (110). (110) Distribution of perfective verbs in anticausative constructions a. RAC, UAC (e.g. casser 'break') b. RAC, *UAC (e.g. briser 'break') c. *RAC, UAC (rare, e.g. brûler 'burn')

In her working hypothesis, Zribi-Hertz (1987) assumes an aspectual difference between RAC and UAC. Her first argument builds on the observation that verbs forming RACs have different aspectual properties than verbs forming UACs. However, Zribi-Hertz's (1987: 45) generalization shows that the aspectual properties of the two constructions are not complementary. As we have seen, it is not the case that only perfective verbs can form RACs and only imperfective verbs can form UACs. Instead, they are in a subset relation, where the RAC has only a subset of the aspectual properties of the UAC (cf. below). (111) RAC: perfective UAC: perfective, imperfective

Zribi-Hertz's (1987) second argument in favour of an aspectual difference between RAC and UAC is related more directly to the two anticausative constructions. The argument builds on the influence that perfective PPs can have on the grammaticality of a given verb in the two constructions. According to the grammaticality of the input and the output, these effects can be divided into the following three subtypes: (112) a. b. c.

UAC, RAC + UAC, *RAC + UAC, *RAC +

perfective PP = *UAC, RAC perfective PP = *UAC, RAC perfective PP = UAC, RAC

117 In the first type (cf. (112a.)), the verb is grammatical in both anticausative constructions, and the addition of the perfective PP renders the UAC ungrammatical. This is illustrated with the example of embellir 'embellish', friser 'curl' and blanchir 'make/become white' below. embellir 'embellish' (113) a. b. (114) a. b.

Le jardin a beaucoup embelli depuis l'hiver dernier. Le jardin s'est beaucoup embelli depuis l'hiver dernier. 'The garden has become much more beautiful since last winter.' *Le jardin a embelli d'une nouvelle pelouse. Le jardin s'est embelli d'une nouvelle pelouse. 'The garden has become more beautiful with a new lawn.' (Zribi-Hertz 1987: 47)

friser 'curl' (115) a. b.

(116) a. b.

Grâce à cette technique perfectionnée, les cheveux de Marie se sont frisés en très exactement deux minutes. Grâce à cette technique perfectionnée, les cheveux de Marie ont frisé en très exactement deux minutes. 'Thanks to this perfect technique, Marie's hair curled in exactly two minutes.' Sur cette toile de Delacroix, les cheveux du guerrier ?se frisent par endroits. Sur cette toile de Delacroix, les cheveux du guerrier frisent par endroits. 'On this painting by Delacroix, some of the hair of the warrior curls.' (Zribi-Hertz 1987: 49)

blanchir 'make/become white' (117) a. b. (118) a. b.

Les toits blanchissent de plus en plus. Les toits se blanchissent de plus en plus. 'The roofs become more and more white.' *Les toits blanchissent d'une épaisse couche de neige. Les toits se blanchissent d'une épaisse couche de neige. 'The roofs become white under a thick snow cover.' (Zribi-Hertz 1987: 47)

In the second type (cf. (112b.)), the verb is only grammatical in the UAC. Adding a perfective PP causes the RAC to be grammatical and the UAC to be ungrammatical, as illustrated with the verb muer 'change' below. muer 'change' (119) a. b.

L'oiseau a mué sous l'effet du virus. *L'oiseau s'est mué sous l'effet du virus. 'The bird changed under the influence of the virus.' (Zribi-Hertz 1986: 334)

118 (120) a. b.

*L'oiseau a mué en un monstre à cinq têtes. L'oiseau s'est mué en un monstre à cinq têtes. 'The bird turned into a monster with five heads.'

(Zribi-Hertz 1986: 334)

In the third type of example (cf. (112c.)), a verb that is ungrammatical in the RAC without a perfective PP becomes grammatical in the RAC in the context of a perfective PP, while the grammaticality of UAC remains unaffected by the perfective PP (cf. below). couler 'pour/flow' (121) a. b. c. d.

La cire coule *La cire se coule 'The wax flows.' La cire coule dans le moule La cire se coule dans le moule 'The wax flows into the mould.'

(Zribi-Hertz 1987: 47)

To sum up, perfective PPs can cause the RAC to become grammatical and perfective PPs can cause the UAC to become ungrammatical. The above examples show further that the UAC is not in general ungrammatical with perfective PPs. Despite this latter fact, the effects of the addition of a perfective PP still reveal an asymmetry between the RAC and the UAC. As already mentioned, the PPs can render the RAC grammatical and the UAC ungrammatical, but, crucially, they cannot render the RAC ungrammatical. This asymmetry can be interpreted as an argument for an aspectual difference between RACs and UACs. In conclusion, Zribi-Hertz (1987) provides two empirical arguments for an aspectual difference between RAC and UAC. Firstly, the RAC is only open to perfective verbs while the UAC allows both perfective and imperfective verbs. Secondly, perfective PPs seem to favour the use of RAC. 5.2.2

Labelle (1992)

Labelle (1992) discusses the distribution of verbs of change of state in the reflexive and the unmarked intransitive construction. Most of the cases she describes are anticausatives, but she also considers verbs that do not have a transitive use and thus are not covered by the definition of anticausative established in section 2.2. In her article, Labelle (1992) makes the following three claims: Firstly, she argues that there is a semantic difference between the two constructions with respect to the following two parameters: (i) the autonomy (vs. non-autonomy)

119 of the event, and (ii) the focus (vs. non-focus) on the resultant state of the event.6 7 Her second claim is that the distribution of verbs in UAC and RAC is predictable on the basis of the compatibility of the semantics of the verb and the semantics of the two constructions. Thirdly and finally, Labelle attributes the semantic differences between the two constructions to the fact that UAC and RAC have different syntactic structures; the UAC is unergative (the subject is generated in external position), while the RAC is unaccusative (the subject is generated in internal position and moves to the subject position). In the following table the three claims are summarized. Table 32: Differences between French UAC and RAC according to Labelle (1992)

UAC RAC

semantics (claim 1) autonomous event, no focus on resultant state non-autonomous event, focus on resultant state

distribution (claim 2) verbs compatible with semantics of UAC verbs compatible with semantics of RAC

syntax (claim 3) unergative unaccusative

In this section I will focus on the assumption that there is an aspectual difference between RAC and UAC in that the first but not the latter focuses on the resultant state of the event. Labelle offers a number of arguments that are intended to show the "aspectual flavour" (Labelle 1992: 395) of the difference between the two constructions. The first piece of evidence is the distribution of reflexive and unmarked intransitive verbs in the infinitival construction mettre quelque chose à 'to bring something to' (called henceforth the mettre-à construction). Labelle (1992: 396) characterizes the semantics of the construction as follows: "Mettre quelque chose à... describes the fact of creating the appropriate conditions for an autonomous process to take place." The distribution of verbs in the construction is such that only unmarked intransitives but not reflexives are grammatical (cf. below).

_________ 6

7

Labelle takes the aspectual difference between UAC and RAC to be secondary and a mere by-product of the semantic difference between RACs and UACs concerning the autonomy of the event: "The aspectual flavour of the distinction between the intransitive construction and the reflexive construction is an indirect consequence of the semantics of the construction." (Labelle 1992: 395) Note that focus in focus on the resultant state differs from the term focus as a category of the information structure of sentences.

120 (122) a. b.

Le sucre (se) caramélise. 'The sugar caramelises.' Le cuisinier a mis le sucre à (*se) caraméliser. 'The cook brought the sugar to caramelise.'

(Labelle 1992: 396)

Although the verb caraméliser 'caramelise' can form both intransitives and reflexives, only the intransitive use is grammatical in the mettre-à construction. According to Labelle (1992: 396), the ungrammaticality of the reflexive construction follows from the fact that the reflexive construction focuses on the resultant state. The semantics of the mettre-à construction and the semantics of the reflexive construction are thus incompatible. The second manifestation of the aspectual flavour of the distinction between RAC and UAC, as conceived by Labelle, is the influence that the passé composé can have on the grammaticality of verbs in RAC. She illustrates this effect by means of the verb empirer. The examples in (123) show that the unmarked intransitive use of empirer 'worsen' in a. is grammatical in the présent, the imparfait and the passé composé, while the reflexive use in b. is only grammatical in the passé composé. (123) a. b.

Son état {empire/empirait/a empiré} 'His/her state worsens/was worsening/worsened.' Son état s'{*?empire/*?empirait/est empiré} 'His/her state worsens/was worsening/worsened.'

(Labelle 1992: 397)

The argument that Labelle (1992) offers on the basis of this observation relies on the assumption that the French past tenses are aspectually sensitive: Simplifying somewhat, we might say that the Imparfait is aspectually imperfective, while the Passé Composé is perfective. (Labelle 1992: 397)

The fact that the reflexive is only possible with the perfective passé composé is thus considered to indicate the aspectual flavor of the reflexive construction. Labelle concludes that "[...] the reflexive construction, which focuses on the attainment of a final state, is somehow licensed by the Passé Composé, which asserts the end of the process." (Labelle 1992: 397) A further argument given in Labelle (1992: 397) builds on the generalization that "[p]erfective adverbials [...] appear to favour the use of a reflexive construction." The effect that the temporal adverbials with en 'in' and pendant 'for' can have on the choice between RAC and UAC is illustrated below. (124) a. b.

Le ciment {a/*?s'est} durci pendant 3 heures. 'The cement hardened for three hours.' Les joints de caoutchouc (se) durcissaient en quelques années. 'The joints of caoutchouc hardened in a few years.'

121 These examples show that the reflexive is ungrammatical in the context of the imperfective pendant adverbial, but grammatical in the context of the perfective en adverbial. Again, there is an asymmetry between the reflexive and the intransitive with respect to their respective aspectual preferences.8 It is important to note that these arguments, in fact, do not reveal a categorical difference between RAC and UAC. Neither tenses nor temporal adverbials are in complementary distribution with respect to the RAC and UAC. Both unmarked intransitives and reflexives are grammatical with the passé composé, and both are equally grammatical with en and pendant adverbials. The above observations should thus rather be interpreted as implications: If the reflexive use of a verb is only grammatical in one grammatical tense, this tense will be the passé composé. If the reflexive use of a verb is grammatical with a pendant adverbial, it will also be grammatical with an en adverbial. As already mentioned at the beginning of this section, Labelle's (1992) explanation of the aspectual difference between RAC and UAC builds on the assumption that the two constructions have different syntactic structures. She takes unmarked anticausatives to be unergative and reflexive anticausatives to be unaccusative; the difference being that the subject of the first construction is generated in external argument position, while the subject of the latter construction is generated in internal argument position and then moves to external argument position. In theories of (lexical) aspect and aktionsart it is often assumed that the object is crucial for the temporal delimitation of an event in that it "measures up" the event (cf. Labelle 1992: 394). Under the assumption that the subject of the RAC still has object-like properties (since it was generated in object position), the association of the RAC with a focus on the resultant state is not surprising. 5.2.3

Bassac (1995)

Bassac (1995) provides a contrastive study on French and English ergative verbs. A considerable part of his study is devoted to the difference between the reflexive and the unmarked intransitive use of ergative verbs in French. His main concern is the question of what the reflexive marks if it appears with verbs that can be used both as intransitives and as reflexives (e.g. casser vs. se casser 'break'). Bassac begins his discussion with a short reflection on the etymology of French se and English self. He notes that both forms go back to a common Indo-European root *s(e)we, which is composed of the two elements

_________ 8

Labelle also refers to the observations regarding the effect of perfective PPs, which were made by Zribi-Hertz (1987) (cf. the discussion in the previous section for details).

122 *s and *we. While *s presumably was a deictic expression, *we expressed 'separation' and 'distance' (cf. Bassac 1995: 247). Bassac maintains that deixis and, more importantly, 'separation' are still the "basic values" (valeur primitive, Bassac 1995: 248) of both the French and the English reflexive morpheme. He assumes that the uses of the reflexive morpheme with French ergative verbs that allow for both constructions can be motivated on the basis of these basic values. One parameter that he perceives to be relevant for the presence or absence of the reflexive morpheme is the parameter of actualization. This notion is defined by Bassac (1995: 256) as "[...] la détermination naissant de la prise en charge énonciative d'une relation prédicative permettant d'envisager une occurrence particulière de la notion." In the case of anticausatives, actualization is thus the determination of a predicative relation between a verb and its single argument. Bassac argues that the function of the reflexive morpheme, i.e. the actualization, follows from one of the basic values attributed to the reflexive morpheme. More specifically, he assumes actualization to be the process whereby a single occurrence is separated from the concept conveyed by the general notion. Based on this particular function of the reflexive morpheme, the prediction regarding the distribution of the reflexive is the following: Nous pouvons donc, selon notre hypothèse, nous attendre à trouver la forme ØV [=UAC] associée à des renvois à la notion, hors actualisation, et la forme se+V [=RAC] associée à l'actualisation. (Bassac 1995: 257)

In order to assess this hypothesis, it should be noted that its empirical falsification presupposes an applicable definition of the "actualized use" of a verb. Bassac (1995: 256) defines it as "[...] une occurrence particulière de la notion". I myself will refer to this parameter distinguishing notion from actualization as being the determination of the predicative relation. Bassac (1995: 284) further claims that "[...] il y a bien une corrélation entre le degré de détermination apporté à l'énoncé et l'alternance ØV / se+V." This suggests that Bassac himself does not assume a binary distinction between actualization vs. nonactualization, but rather a gradual distinction, depending on the degree of determination. This idea is also expressed by the following quote: La probabilité de la construction se+V croîtra avec le nombre de déterminations associées au procès: plus les déterminations seront nombreuses, plus se+V sera probable. Moins les déterminations seront nombreuses, plus la forme ØV sera probable. Et, en même temps, plus les déterminations seront nombreuses, plus on sera proche de l'actualisation; moins les déterminations seront nombreuses, plus on sera proche de la non-actualisation. (Bassac 1995: 285)

As already indicated above, this quote also suggests that the determination of the predicative relation is a gradual and not a binary parameter. Bassac, however, does not specify the degree of determination at which the boundary

123 between the reflexive and the unmarked intransitive use of the verb is located. Instead, he presents a number of empirical observations, where the opposition between RAC and UAC corresponds to different degrees of determination of the predicative relation to illustrate his claim. In what follows, I will summarize these observations. The first observation relates the choice between RAC and UAC to the concept of genericity. In the examples in (125a.) and (125b.), the predicative relation between the verb and the subject is not singled out ("singulariser"), as Bassac (1995: 257) puts it. Instead, the examples denote a predicative relation that generally holds for the referent of the subject. (125a.), thus, asserts a property of a certain type of chocolate, just as (126a.) asserts a property of books. In the examples in (125b.) and (126b.), on the other hand, the predicative relation between the subject and the verb is temporally anchored, and neither of the two examples is generic. Both the breaking of the ship and the opening of the door are described as singular occurrences of the predicative relation between the verb and the subject, which is therefore maximally determined. (125) a.

b.

(126) a.

b.

Un chocolat très noir doit provenir de fèves de très grandes qualités. Parfaitement uni, brillant, il casse net. 'A very dark chocolate must come from high quality beans. Perfectly plain, shiny, it breaks clean.' (Sud-Ouest Dimanche, 2.01.94 (Bassac 1995: 257)) Naufrage du pétrolier "Baer". L'épave se casse. 'Sinking of oil tanker "Baer". The wreck is breaking.' (Sud-Ouest, 9.01.93 (Bassac 1995: 258)) Les livres sont des portes qui ouvrent sur des mondes merveilleux. 'Books are doors that open towards marvellous worlds.' (Michel Lebris, Journal Télévisé A2, 11.05.92 (Bassac 1995: 257)) Je pressai le bouton. Comme d'habitude, la porte s'ouvrit automatiquement, je montai l'escalier. 'I pushed the button. As usual, the door opened automatically and I climbed the staircase.' (S. de Beauvoir, Mémoires d'une jeune fille rangée (Bassac 1995: 258))

Although Bassac does not refer to the notion of genericity, we may reasonably rephrase the first type of contrast, as described in the above, as a contrast between generic sentences where the predicative relation is not determined, and non-generic sentences where the predicative relation is determined. As predicted by Bassac, the unmarked construction is used with the generic interpretation, and the reflexive construction is used with the non-generic interpretation.

124 As already mentioned above, Bassac does not specify the degree of determination which, at a given point, decides between RAC and UAC. In the case of the opposition between generic and non-generic sentences, there is a maximal difference with respect to the determination of the predicative relation. However, within the domain of non-generic uses there is considerable variation with respect to the degree of determination. A second observation relates the choice between the two constructions to grammatical tense (cf. the previous section for a similar observation made by Labelle (1992)). For instance, Bassac (1995: 259) notes that the unmarked intransitive use of verbs like fermer and ouvrir is impossible with the passé simple (cf. (127)), and that speakers prefer the reflexive over the intransitive construction with the passé composé (cf. (128)). (127) a. b.

La porte *(s') ouvrit. 'The door opened.' La porte *(se) ferma. 'The door closed.'

(128) Le fil s'est cassé > Le fil a cassé 'The yarn breaks.'

Unfortunately, Bassac does not specify whether the relation between the grammatical tenses and the choice of construction is due to tense proper (that is, the relation between event time and speech time) or due to tense as the aspectual properties of grammatical tenses. However, Bassac's next point is clearly related to the notion of aspect. He argues that actualization implies the attainment of the final state of the event denoted by the verb and consequently assumes that "[...] la forme se+V tendra à être associée à la phase terminative d'un procès." (Bassac 1995: 264) Bassac provides three empirical arguments in favour of this assumption. First of all, the distribution of the reflexive morpheme with French aspectual verbs depends on whether the verb refers to the beginning or to the end of an event; terminative verbs are reflexively marked (se terminer, s'achever, se clôturer 'end'), while inchoative verbs are not (commencer, débuter 'begin'). Secondly, the choice of the construction does not only depend on whether there is a final state of the event, but also on its ultimacy. Bassac cites François' (1989: 92) observation that in pairs like durcir-endurcir 'harden' or sécher-assécher/déssécher 'dry' the verbs without prefix denote a change that is temporary, while the change denoted by prefixed verbs is permanent. Furthermore, these verbs differ with respect to their distribution with the reflexive: the latter are compatible with the reflexive, while the former "sont difficilement compatibles" with the reflexive (Bassac 1995: 264). Zribi-Hertz (1987: 49) made a similar observation:

125 Avec certains verbes à renversement, l'opposition aspectuelle entre les deux formes ergatives est corrélée à des effets de sens spécifiques. [...] [L]a [RAC] dénote un changement d'état durable, tandis que la forme ergative non réflexive peut aussi dénoter un changement d'état éphémère – conduisant à un "état final" non définitif [...].

Unlike François (1989), Zribi-Hertz (1987) refers to cases where the same verb can either denote a temporal or a permanent change, without the correlate of a present or absent prefix. (129) a. b. c.

(130) a. b. c.

*L'annonce de la catastrophe a pâli/rougi/blêmi le visage de Marie. 'The anouncement of the catastrophy made Marie's face turn pale/red/pale.' Le visage de Marie a pâli/rougi/blêmi à l'annonce de la catastrophe. 'Marie's face turned pale/red/pale at the announcement of the catastrophy.' *Le visage de Marie s'est pâli/rougi/blêmi sous l'effet de l'humidité. 'Marie's face turned pale/red/pale from humidity.' (Zribi-Hertz 1987: 49) L'humidité a pâli/rougi/blêmi la peau du cadavre. 'Humidity turned pale/red/pale the corps' skin.' La peau du cadavre a pâli/rougi/blêmi sous l'effet de l'humidité. 'The corps skin turned pale/red/pale from humidity.' La peau du cadavre s'est pâlie/rougie/blêmie sous l'effet de l'humidité. 'The corps skin turned pale/red/pale from humidity.' (Zribi-Hertz 1987: 49)

Bassac's third observation with respect to the impact of the resultant state on the choice of the construction also builds on an observation that has already been made by Zribi-Hertz (1987: 47-49). Bassac (1995: 283f.) notes that PPs denoting a resultant state may influence the choice of construction. This is illustrated by the following examples in (131) and (132): (131) shows the grammaticality of the reflexive use of a verb with a perfective PP, the sentences in (132a.) illustrate the ungrammaticality of the unmarked intransitive use of the same verb with the same perfective PP, and the sentences in (132b.), finally, show the grammaticality of the intransitive use without the perfective PP. (131) Dans un deuxième temps, cette indemnité transactionnelle s'augmente de frais de dossier. 'In a second time, this sum increases by the service fee.' (France Info, 5.04.94 (Bassac 1995: 283)) (132) a.

b.

*Dans un deuxième temps, cette indemnité transactionnelle augmente de frais de dossier. 'In a second time, this sum increases by the service fee.' Dans un deuxième temps, cette indemnité augmente. 'In a second time, this sum increases.' (Bassac 1995: 284)

126 5.2.4

Telicity, perfectivity and the resultant state

In sections 5.2.1 to 5.2.3 I have reported empirical observations from the literature that suggest a relation between the aspectual structure of an anticausative sentence on the one hand and the choice between RAC and UAC on the other hand. However, in my presentation of these data and observations I did not yet specify the empirical scope of the observations or the level of aspectual structure where the observed phenomena are supposed to be located. This was mainly due to the fact that the literature under discussion does not always provide these specifications. In the present section, in order to make the descriptions given in sections 5.2.1 to 5.2.3 as precise as possible, I will classify the observed phenomena in more common terms of aspectual theory and consider the scope of the observations in more detail. Before I begin with the classification proper, first a structured description of aspectual structure is necessary. Here I take the analysis in de Swart (1998) as a starting point, who proposes the following compositional model of aspect. (133) [Tense [Aspect [Eventuality Description]]]

(de Swart 1998: 348)

Eventuality descriptions (henceforth ED), which consist of a verb and its internal argument, constitute the input in de Swart's model.9 These eventuality descriptions fall into three types, according to their setting of the binary features of homogeneity and stativity. The homogeneity of an eventuality description (also referred to as (a)telicity) depends on the presence or absence of an inherent endpoint of the event. In the literature, a number of tests have been proposed in order to detect whether an ED possesses such an inherent endpoint or not. One of these tests is the distribution of temporal adverbials. Quantized (or telic) EDs pattern with in TIME, while homogeneous (or atelic) EDs pattern with for TIME (cf. below). (134) a. b.

John ate the apple in an hour. John worked for an hour.

A second point is the observation that the stativity of an ED depends on the stativity of the verb. Stative verbs form stative EDs whereas non-stative verbs form dynamic EDs. Again, there are tests that motivate this distinction. For instance, Vendler (1957: 144) notes that the possibility of providing an answer to the question "What are you doing?" sets apart dynamic from stative EDs.

_________ 9

This is also expressed in the following quote: "[...] aspectual class is determined at the level of predicate-argument structure, which I will identify as the level of the eventuality description." (de Swart 1998: 348).

127 (135) a. b.

What are you doing? - I'm working. What are you doing? *- I'm knowing the answer.

dynamic stative

As already implied in the above, the possible combinations of the binary features homogeneity and stativity result in the three types of EDs: states, processes and events (cf. de Swart 1998: 351). (136) a. b. c.

State (homogeneous, stative) Process (homogeneous, dynamic) Events (quantized, dynamic)

John is sick. John eats. John eats an apple.

Table 33: Types of eventuality descriptions in de Swart (1998)

homogeneous quantized stative dynamic State

Process

Event

Other authors have proposed inventories of EDs consisting of more than three types. Smith (1991), for example, who introduces the feature of durativity, distinguishes between five types of EDs: states, semelfactives, activities, achievements, and accomplishments. Table 34: Types of eventuality descriptions in Smith (1991)

stative durative States

dynamic atelic punctual

telic durative

punctual

durative

Semelfactives

Activities

Achievements

Accomplishments

The additional feature of durativity relies on the temporal extension of the respective ED. EDs with a temporal extension are durative, while those without this extension are punctual. Schaden (2007: 22f.) notes with respect to this distinction that [e]n réalité, probablement toutes les éventualités ont une extension temporelle, même si elle est minimale; mais dans le langage, il existe certains eventualités qui se comportent comme si elles n'occupaient qu'un seul instant.

In order to know whether the language treats an eventuality as punctual or durative, it is often tested whether the eventuality described can be interrupted or not.

128 (137) a. b. c. d.

John stopped eating the apple. *John stopped arriving. The glass stopped breaking. #John stopped jumping. (only iterative reading is possible)

On the next level of de Swart's (1998) model, aspectual operators come into play – which, as a matter of fact, they may not always do, given that aspectual operators are not always present (de Swart 1998: 348).10 If they are present, however, they "[...] impose a certain viewpoint on the eventuality introduced by the eventuality description." (de Swart 1998: 351). The relevant distinction in the present context is the distinction between perfective and non-perfective viewpoint; the first includes the perfect state, while the latter excludes the perfect state of the eventuality (where perfect state is the state that follows the end of the eventuality). The telic ED construire une maison is viewed as perfective (with a perfect state) and as non-perfective (without a perfect state). (138) a. b.

Jean a construit une maison. perfective, telic 'Jean built a house.' Jean est en train de construire une maison. non-perfective, telic 'John is building a house.'

While the above examples illustrate the possibility of both viewpoints for telic EDs, the following examples show that the same possibility holds for atelic eventualities. (139) a. b.

Jean a beaucoup travaillé. 'Jean worked a lot.' Jean est en train de travailler. 'John is working.'

perfective, atelic non-perfective, atelic

A perfective viewpoint includes the perfect state, i.e. the state that follows the end of the eventuality. In the literature, different types of perfect states are distinguished (see e.g. the overview in Nishiyama & Koenig 2004: 102 for details). The perfective sentences in (138a.) and (139a.) illustrate an important difference in this respect. Only the ED construire une maison in (138a.) lexically entails a perfect state that is also a resultant state. In the case of the ED travailler in (139a.), the perfect state is simply the state of not working anymore.

_________ 10

Evidently, the absence of aspectual operators makes it possible to determine the aspectual class of EDs in the first place.

129 Given that perfect states are introduced by a perfective viewpoint and that resultant states have been defined above as being one subtype of perfect states, it naturally follows that resultant states are only introduced under a perfective viewpoint. However, this restriction only holds if the resultant state is not overtly expressed. In the following example, the expression of a resultant state is combined with a progressive and thus non-perfective viewpoint. (140) Jean est en train de se changer en un vrai con. 'Jean is turning into a real jerk.'

The above example suggests that resultant states do not imply a perfective viewpoint. As indicated above, one constraint on the combination of resultant states with a non-perfective viewpoint is that the resultant state needs to be overtly expressed. Otherwise, as (141a.) shows, the resultant state interpretation is missing. Under a perfective viewpoint as in (141b.), however, the lexically entailed resultant state is introduced "automatically", i.e. without being overtly expressed. (141) a. b.

Jean est en train de changer. 'Jean is changing.' Jean a changé. 'Jean changed.'

The conclusion that can be drawn from these examples is the following: Under a non-perfective viewpoint, the resultant state is introduced and overtly expressed by the PP. Under a perfective viewpoint, the resultant state is introduced by the perfective viewpoint and overtly expressed by the PP. Only the latter type of resultant state is a subtype of a perfect state. In subsequent sections I will show that the lexical entailment of a resultant state and the overt expression of a resultant state actually are the two aspects of resultant states that are relevant for the choice between RAC and UAC. So far, I have introduced telicity, perfectivity and the resultant state as the three key components of aspectual structure. Using these components to build a classification, I will now reconsider the empirical observations on the choice between RAC and UAC reported in the previous sections. In addition to relating these observations to the three components described above, I will discuss the empirical scope of each observation as it is given in the respective literature.11 I begin my discussion with the concept of telicity.

_________ 11

Scope here is understood to be the range of contexts for which a given observation holds.

130 A first observation related to this component is found in Labelle (1992). Temporal adverbials like en TIME and pendant TIME are reported to have an impact on the choice between RAC and UAC. In Labelle's (1992) example, the use of the RAC is ungrammatical in the context of the adverbial en TIME. The distribution of these two types of adverbials is commonly assumed to be sensitive to telicity. Telic EDs are compatible with en TIME, atelic EDs are compatible with pendant TIME. Labelle's observation thus relates telicity and the use of RAC. With respect to the scope of this observation, Labelle (1992: 397) remains rather vague: "Perfective adverbials [...] appear to favour the use of a reflexive construction." She discusses durcir and cuire which are conceived to be two verbs where the distribution of en TIME and pendant TIME has an impact on the choice between the reflexive and unmarked intransitive anticausative use of the verb. (Labelle 1992: 398) Unfortunately, as Labelle (1992) does not provide any further examples or indicates that the generalization that she establishes goes beyond the cases discussed in her analysis, the scope of her observation remains limited to these two verbs. A second observation that relates telicity and the choice between RAC and UAC comes from Zribi-Hertz (1987). She gives examples where perfective PPs favour the use of RAC. The following list illustrates the variety of types of PPs that are possible in this context. (142) a. b. c. d.

La cire se coule dans le moule. 'The wax flows into the mould.' Le jardin s'est embelli d'une nouvelle pelouse. 'The garden has become more beautiful with a new lawn.' Le ballon se gonfle de gaz carbonique depuis cinq minutes. 'The ball is inflating with gas since five minutes.' La maison s'est empestée d'une intolérable odeur de moisi. 'The house became filled with an intolerable stench of mold.' (Zribi-Hertz 1987: 47f.)

In the first two examples, the PPs have an impact on the telicity of the ED in that they turn an atelic ED into a telic ED. Apart from this point that they have in common, the PPs in exhibit differences that need to be addressed. To begin with, the PP dans le moule delimits the event in that its endpoint is clearly expressed. The change of position described by the verb couler may be regarded as a gradable change, and the PP specifying the endpoint of this change of position quantizes the degree of change. The verb embellir also describes a gradable change. As in the case of dans le moule, the PP d'une nouvelle pelouse quantizes the degree of change. It does so by specifying the component that differentiates the subject before and after the change. In the same vein, ZribiHertz (1987: 47) notes that the PP in (142b.) denotes "[...] une composante de l'état final de la translation."

131 In neither of the two examples, the PP explicitly expresses the resultant state of the change. Instead, in both cases it has to be inferred to "be in PLACE" and to be "initial state plus une belle pelouse", respectively. Nevertheless, we may assume that the resultant state is more salient if such PPs are present than if they were absent. This seems to be an important point that these PPs share with the PPs in c. and d. Yet, there is a very important difference between the PPs in a. and b. on the one hand and the ones in c. and d. on the other hand: only the PPs in a. and b. turn atelic EDs into telic EDs. The observation that the PPs in c. and d. also have an impact on the choice between RAC and UAC raises some doubts as to whether telicity is a relevant parameter at all. I will come back to this issue. Concerning the scope of her observation, Zribi-Hertz (1987) gives examples for the verbs couler 'flow', embellir 'embellish', blanchir 'make/become white', gonfler 'swell/inflate' and empester 'stink', where the presence of a perfective PP influences the choice between RAC and UAC. Again, as no further comment is made as to whether these examples are representative for the anticausative use in general or whether they are only anecdotal instances that have no universal impact, the observation does not go beyond these examples which thus constitute the scope of the observation. The heterogeneity of the examples in (142) also raises more general questions concerning the definition of the term perfective PP. As only some but not all of Zribi-Hertz's (1987) perfective PPs imply telicity, perfective may not be taken to be synonymous to telic. Furthermore, these PPs do not relate to the distinction between perfective and imperfective viewpoint as they are compatible with both. Apart from the terminological issue about possible interpretations of the term perfective PP, it seems that these PPs draw attention to the resultant state – regardless of whether the resultant state is attained (perfective viewpoint) or not (imperfective viewpoint), and regardless of whether the resultant state follows an inherent endpoint (telic) or not (atelic). I will discuss this view in more detail below in the context of the observations related to the resultant state. The question whether French RAC and UAC generally differ with respect to telicity has been investigated by Lagae (1990). Using a slightly modified version of the well-known entailment test by Garey (1957: 105), Lagae tries to verify whether a part of an ED entails the full, given ED (atelic) or not (telic). Her working hypothesis is as follows: [...] les formes réflexives se trouvent en effet exclusivement dans des énoncés perfectifs (téliques) et les formes non réflexives dans des énoncés imperfectifs (atéliques). (Lagae 1990: 37)

Garey's (1957) test consists of a question that needs to be answered. If the answer is positive, then the tested verb is atelic. If it is negative, then the tested verb is telic.

132 (143) Si on verbait, mais a été interrompu tout en verbant, est-ce qu'on a verbé? (Garey 1957: 105)

Note that this test applies to individual verbs, while Lagae's (1990) working hypothesis is concerned with the telicity of sentences. Therefore, she reformulates the question in (143) as in (144b.), to test the telicity of the proposition in (144a.). (144) a. b.

Le prix du pain a augmenté à deux reprises. 'The bread price increased in two steps.' Si le processus était interrompu, est-ce que le prix du pain aurait augmenté à deux reprises? 'If the process had been stopped, would the bread price have increased in two steps?' (cf. Lagae 1990: 37)

The answer to the question is negative and the proposition in (144a.) would thus test positive for telicity. The example would contain a telic UAC and thus falsify her working hypothesis. However, this result proves to be of little use since there is one important complication in both the design and the application of the test. First of all, a possible reason for criticism would be that telicity is a property of EDs (that is, of verbs and their internal argument) and not a straightforward property of propositions. Therefore, one might argue that only a proposition similar to le prix du pain a augmenté should be tested. This, however, is not the main problem in the present case. In fact, it might indeed be interesting to see the influence that PPs such as à deux reprises have on the telicicty of EDs. The real problem is, however, that such PPs in and of themselves render the test invalid, as the resultant state is triggered by the PP regardless of the properties of the verb or the ED that originally should be tested. Crucially, à deux reprises divides the event into two subevents: a first increase and a second increase. In the question in (144b.), however, it is not specified when the process is interrupted, that is, during the first or the second increase. The following examples illustrate how a reformulation of the question leads to different answers. (145) a. b.

Le prix du pain a augmenté à deux reprises. 'The bread price increased in two steps.' Si le processus était interrompu à la deuxième reprise, est-ce que le prix du pain aurait augmenté à deux reprises? 'If the process had been stopped at the second step, would the bread price have increased in two steps?'

133 c.

Si le processus était interrompu à la première reprise, est-ce que le prix du pain aurait augmenté à deux reprises? 'If the process had been stopped at the first step, would the bread price increased in two steps?'

The answer to the question in (145b.) is positive and the answer to (145c.) is negative. Importantly, here the negative answer has nothing to do with telicity, but is simply due to a mismatch between the number of subevents (two) and the subevent when the process is interrupted (one). Similarly, the fact that the answer to (144b.) is negative is a result of an uncertainty concerning the question when, i.e. during which subevent, the process is interrupted. Unfortunately, many of Lagae's (1990) counterexamples to the working hypothesis are invalid as they suffer from the very same shortcoming: The event is split up into subevents by means of plural subjects and operators such as l'un après l'autre 'one after the other' or successivement 'successively' (cf. (146)). (146) a.

b.

L'un après l'autre, les sommets blanchissent complètement à cause de la neige 'One after another, the summits turned completely white from the snow.' Les trois pommiers ont verdi successivement 'The three apple trees turned green successively.' (Lagae 1990: 37)

Others of Lagae's (1990) examples prove to be more instructive. In the discussion of perfective PPs we have seen that EDs describing gradable changes become telic once the degree of change is quantized. In two of Lagae's (1990) seven examples of telic UACs, the degree of change is quantized by means of a measure phrase (147a.) or an adverb of completion (147b.). These sentences test positive for telicity. (147) a. b.

Les trois peupliers ont rapetissé de moitié à cause de la distance 'The three poplars decreased (only) half because of the distance.' L'organisation de l'entreprise a entièrement changé 'The organization of the enterprise changed completely.'

With respect to RACs, Lagae's (1990) working hypothesis predicts that they should be excluded from atelic contexts. The counterexamples she provides are given in (148). (148) a. b.

A cause de la chaleur, des morceaux de sucre se sont caramélisés 'Because of the heat, pieces of sugar caramelised.' Ce vin s'améliore sans cesse 'This wine becomes better all the time.'

134 c. d. e.

Tous les jours, des fleurs se fanent 'every day, flowers wilt' La source s'est tarie régulièrement 'The source dried up regularly.' Toutes les nuits, le parquet se bombe à cause de l'humidité 'Every night, the floorboards camber because of the humidity.' (Lagae 1990: 38)

Lagae's (1990) overall conclusion, which I follow despite the methodological problems discussed above, is that RACs and UACs are not in complementary distribution with respect to telicity: RACs are not limited to telic contexts and UACs are not limited to atelic contexts. Still, the fact that the counterexamples provided by Lagae (1990) seem to require a certain degree of complexity (adverbs, measure phrases, quantificational properties of NP) might indicate that the two constructions have different preferences with respect to telicity. I will now turn to the next component, namely, perfectivity as viewpoint. Clearly related to the concept of perfectivity, Labelle (1992) observed that grammatical tense can have an impact on the choice between RAC and UAC. She gives examples where the use of the RAC is licensed by the use of the passé composé. This observation is linked to the notion of viewpoint by her assumption that "[s]implifying somewhat, we might say that the Imparfait is aspectually imperfective, while the Passé Composé is perfective." (Labelle 1992: 397) Although it is generally admitted in the literature that the French passé composé can be perfective, it is also stressed in the literature that this is not always the case; Labelle's cautiousness in stating that the generalization pertaining to the use of perfective passé composé vs. imperfective imparfait is a simplification is thus warranted. The scope of the observation is once again stated very vaguely: "[S]ome verbs which are normally not constructed with the reflexive are accepted in a reflexive construction when they are in the Passé Composé." (Labelle 1992: 397) The only example Labelle gives to support her claim is the verb empirer. I will now consider the resultant state in more detail, which is the third and last component of aspectual structure, as motivated above. The following two observations that were reported in sections 5.2.1 to 5.2.3 are related to this component. First of all, Zribi-Hertz (1987: 45) argues that only verbs with a morphologically related form that denotes a resultant state can form RACs. The scope of the observation is thus clearly stated and very general: The existence of a resultant state form of any given verb is a necessary condition for the formation of RAC with that same verb. The second observation refers to the fact that "perfective PPs" (Zribi-Hertz 1987) and PPs expressing the "consequences or resultant state of the event" (Bassac 1995) have an impact on the choice between RAC and UAC. In the discussion above we already saw that some but not all of these PPs have an

135 impact on the telicity of an ED in that they turn atelic EDs into telic EDs (cf. (149)). (149) a.

b.

telic Le jardin s'est embelli d'une nouvelle pelouse. 'The garden has become more beautiful with a new lawn.' atelic Le ballon se gonfle de gaz carbonique depuis cinq minutes. 'The ball is inflating with gas since five minutes.' (Zribi-Hertz 1987: 47f., mod. StH)

The above observation –PPs that do not imply telicity also favour the use of the RAC – suggests that the effect of these PPs is not exclusively related to telicity. Rather, independently of their impact on the telicity of the ED, these PPs specify the resultant state, and in doing so, they draw attention to the resultant state. One might thus argue that the resultant state is more salient in the presence of such PPs. With respect to the scope of the observation, however, the statements put forward in Zribi-Hertz (1987) and Bassac (1995) differ considerably. Zribi-Hertz (1987) provides examples where the presence of such a perfective PP influences the choice between RAC and UAC, specifically, the verbs couler, embellir, blanchir, gonfler and empester. Unfortunately, though, she does not indicate whether these examples are representative for French anticausatives in general or whether the particular verbs she discusses are idiosyncratic in their behaviour. Bassac (1995: 284), on the other hand, explicitly claims that the presence of such PPs implies the use of RAC, which is a much stronger hypothesis. The following table summarizes the discussion provided in the preceding sections. It lists the relevant observations made in the literature, their scope as it is given in the respective literature, and the component of aspectual structure the observation is related to.

136 Table 35: Aspectual structure and the choice between French RAC and UAC

observation en 'in' vs. pendant 'for' adverbials

scope of observation Labelle (1992: 396): "Perfective adverbials also appear to favour the use of a reflexive."

preference for RAC in the context of PPs turning atelic EDs into telic EDs

Zribi-Hertz (1987: 47): examples for individual verbs Labelle (1992: 399f.) only repeats some of Zribi-Hertz's (1987) examples Bassac (1995: 284): "La forme se+V est utilisée quand mention est faite dans l'énoncé des conséquences du procès, de son résultat."

perfectivity

Passé composé can license RAC

Labelle (1992: 396): "Interestingly, some verbs which are not normally constructed with the reflexive are accepted in a reflexive construction when they are in the Passé Composé."

resultant state

RAC ĺ Resultant state form preference for RAC in the context of PPs specifying the resultant state

telicity

5.3

Causal structure

5.3.1

Causes in anticausatives

Zribi-Hertz (1987: 45): necessary condition on verbs forming RACs cf. above

The question whether causes are present in anticausatives is a core issue of the description of anticausatives. In chapter 2 (Anticausatives), the following two remarks have been made as to the presence of causes in anticausatives: First, no cause is expressed in subject position, i.e. the position that it is normally attributed by linking hierarchy. Second, no cause is semantically present unless it is overtly expressed. Hence the examples below can be commented as follows: In a. no cause is semantically present because no cause is overtly ex-

137 pressed. Despite the overt expression of a cause, example b. is an anticausative because the cause is not in subject position. (150) a. b.

The window broke. The window broke because of the storm.

Thus, when referring to the presence of a cause in anticausatives, the semantic and the formal level need to be distinguished. The implicit semantic presence of the cause is considered as the main difference between anticausatives and passives. The two constructions have in common that they can present an event without explicitly expressing the cause of the event, but they are assumed to differ in that only in the case of the passive an implicit cause is always present. (151) a. b.

The window broke. The window was broken.

Standard diagnostics to show this are the licensing of agent-oriented adverbs, control into purpose clauses, instrumental phrases or secondary predication (see Fagan (1992), Fellbaum & Zribi-Hertz (1989) or Lekakou (2005), among others, for a discussion of these diagnostics in French). The tests give negative results for anticausatives as opposed to passives. However, with respect to agent-oriented adverbs and purpose clauses Schäfer (2008) notes that their grammaticality only hints at the presence of agents and that their ungrammaticality does not rule out the presence of non-agentive external arguments. A similar remark can be found in Kallulli (2006: 211): [...] all that purpose clauses and so-called agent-oriented adverbs do is identify an intention-bearing (i.e., animate) event participant as the source or initiation of the event named by the verb.

The fact that the tests mentioned above give negative results for anticausatives does not imply that there are no causes in anticausatives. As a consequence, the tests cannot be used to detect a possible difference between RAC and UAC either. Härtl (2003) proposes a semantic test which supposedly detects the presence of a cause independently of agency. The test consists of a sequence of two sentences. The first sentence describes a change-of-state event, and the second sentence contrasts two possible causes for this event. The idea behind this test is that if a cause is semantically present in the first sentence it should be possible to refer to this cause in the second. More specifically, Härtl (2003: 897) maintains that the contrastive focus in the second sentence presupposes a specific referential status of the focussed entity. The fact that the second sentence is not acceptable shows that this specific referential status is not provided by the anticausative sentence. The conclusion is thus that no cause is semantically present in anticausatives such as (152a.) and (b.).

138 (152) a. b.

Der Teller zerbrach. ??Aber es war nicht Peter, sondern Maria. 'The plate broke, but it wasn't Peter but Maria.' Das Papier verbrannte. ??Aber es war nicht Peter, sondern Maria. 'The sheet burnt, but it wasn't Peter but Maria.' (Härtl 2003: 896f.)

The following examples show that the grammaticality does not change once inanimate entities are used as causes in the second sentence. (153) a. b.

Der Teller zerbrach. ??Aber es war nicht das Erdbeben, sondern der Wind. 'The plate broke, but it wasn't the earthquake but the wind.' Der Teller zerbrach. ??Aber es war nicht der Hammer, sondern die Zange. 'The plate broke, but it wasn’t the hammer but the pliers.' (Härtl 2003: 897)

To this, Schäfer (2008) adds reflexive anticausatives for which the same results as for the above unmarked anticausatives can be obtained. (154) a. b. c.

Das Benzin entzündete sich 'The petrol caught fire.' ??Aber es war nicht Peter, sondern Maria 'But it was not Peter but Mary.' ??Aber es war kein Blitz sondern die Sonneneinstrahlung 'But it was not a lightning but the sun.'

Passives behave differently from anticausatives in this test. It is possible to refer to the cause in the second sentence if the change-of-state event is described by a passive in the first sentence. Cf. the following examples: (155) a. b.

Der Teller wurde zerbrochen. Aber es war nicht Peter, sondern Maria. 'The plate was broken, but it wasn’t Peter but Maria.' (Härtl 2003: 898) Der Teller wurde zerbrochen. Aber es war nicht das Erdbeben, sondern der Wind. 'The plate was broken, but it was not the earthquake but the wind.' (Schäfer 2008: 56)

To sum it up, Härtl's (2003) test thus does not separate German reflexive from unmarked anticausatives but, more generally, distinguishes between anticausatives and passives. The same holds for the other phenomena mentioned above. The above differences between anticausatives and passives are confirmed for French by native speakers. Härtl (2003) concludes from the above test that no causal component is present in the structure of anticausatives and I follow this conclusion as far as French anticausatives are concerned. Although Schäfer (2008) draws a similar conclusion with respect to this test, he still assumes that a causal component is present in anticausatives (fol-

139 lowing Alexiadou et al. 2006). He motivates this assumption by the distribution of PPs expressing causes in the context of anticausatives. He argues that there are crosslinguistic restrictions as far as the expression of causes in the contexts of anticausatives is concerned. These restrictions involve both the semantic role of the cause (causers and causing events are allowed, agents are not allowed), and the preposition whereby the PP is introduced (cf. Schäfer 2008: 139f. and Alexiadou et al. 2006). To account for these restrictions he assumes a causal component in the structure of anticausatives that licences the causes. Note that the fact that causes can be expressed in the context of anticausatives could be easily accounted for by assuming that the PP introduces the cause and does not pick a causal component already in the structure of anticausatives. Schäfer (2008: 147) himself discusses this alternative explanation for the possibility of the PPs. However, he rejects this view because of the above-mentionned restrictions. One drawback for the account presented in Schäfer (2008) and Alexiadou et al. (2006) is that there is no direct semantic evidence for the presence of a causal component. One would expect that the presence of a causal component in the structure of anticausatives would also have a semantic correlate. Unlike Alexiadou et al. (2006) and Schäfer (2008), Koontz-Garboden (2007, 2009) does not assume a causal component in the syntax of anticausatives, but he assumes that at least in formally derived anticausatives "[...] the individual undergoing the change is also the EFFECTOR participant in the event that causes the change of state event." (Koontz-Garboden 2009: 36) He suggests that at least in the true reflexive and the reflexive anticausative, the reflexive morpheme is a reflexivization operator and thus has the same function: Reflexivization, quite simply, is an operation that takes a relation as an argument, setting both arguments of the relation to be the same (Chierchia 2004: 29). In settheoretic terms, if a relation is conceived of as a set of pairs, what reflexivization does is to restrict the denotation of the relation to those sets of pairs each of whose members is identical to the other. (Koontz-Garboden 2009: 83)

This assumption, however, is problematic. The semantics predicted by the reflexivization account do not capture the semantics of the anticausative construction. The subject is not at the same time the effector and the undergoer; it is only the undergoer. Crucially, it has been maintained that a cause that is external to the subject is more present in the case of RAC than UAC; in fact the opposite would be predicted under the reflexivization account. To conclude, I follow Härtl (2003) in that I do not assume that a cause is present as a constant in the semantic or syntactic structure of anticausatives. Contrary to Schäfer (2008) and Alexiadou et al. (2006) I assume that the PPs expressing causes do not pick up a causal component in anticausatives, but introduce a causal component. Contrary to Koontz-Garboden (2007, 2009) I do

140 not assume that the subject of marked anticausatives is at the same time the effector and the undergoer of the event, but only the undergoer. Given this conclusion, does it even make sense to further investigate the question of whether French reflexive and unmarked anticausatives may differ as far as causal structure is concerned? I still assume this question to be valid, because the lack of a cause as a constant in the structure of anticausatives does not imply that a difference may not emerge on some other level of representation. 5.3.2

Markedness theory and the spontaneity of events

In Haspelmath's (1993) overview of the different formal means to encode the causative-anticausative alternation (cf. the discussion in the previous section) he examined 31 causative-anticausative verb pairs, e.g. English break: X breaks Y and Y breaks, in a sample of 21 languages. The following table shows once again the five formal types of expressing the alternation, as distinguished by Haspelmath (1993: 91f.): 1) and 2) are directed and 3) - 5) are non-directed alternations. Table 36: Encodings of the causative-anticausative alternation (Haspelmath 1993: 9092)

1 2 3 4 5

type causative alternation anticausative alternation labile alternation equipollent alternation suppletive alternation

example Georgian: duȖ-s 'cook INTR' a-duȖ-ebs 'cook TR' Polish: záamaü-siĊ 'break INTR ' záamaü 'break TR' English: break 'break INTR' break 'break TR' Japanese: atum-aru 'gather INTR' atum-eru 'gather TR' Russian: goret' 'burn INTR' žeþ 'burn TR'

The data from Haspelmath's study reveals that there is considerable crosslinguistic variation in the distribution of these 5 types. For instance, there are languages with a strong preference for the labile type (e.g. English: 25 out of 31 verb pairs), for the equipollent type (e.g. Japanese: 20.5 out of 31) or the

141 causative type (e.g. Mongolian: 22 out of 31).12 Apart from these prototypical instantiations of one type, Haspelmath (1993) found many languages that do not display a special preference and use several types (e.g. Hindi-Urdu: 7.5 anticausative, 14 causative, 17 equipollent, 2 labile, 0 suppletive). One obvious conclusion to be drawn from this data is that there is no crosslinguistic stability in the encoding of the alternation. This fact causes serious problems for accounts proposing one universal direction of derivation within the alternation. But this cross-linguistic variation also poses problems for theories that try to account for morphosyntactic differences by a principle of iconicity of complexity. I will briefly outline these problems in what follows. In the case of the causative-anticausative alternation one would assume the causative part to be semantically more complex. If we abstract away from the different proposals that have been suggested for the semantic decompositions of the two alternants, one conclusion to be drawn is that the causative alternant is more complex in that it includes more predicates.13 (156) a. b.

[y become STATE] x cause [y become STATE]

anticausative causative

However, the cross-linguistic variation in the encoding of the causativeanticausative alternation shows that a relation in which the semantic complexity of the causative part is higher than the one of the anticausative part is not iconically represented by the actual formal encoding in the various languages discussed in the literature (cf. Haspelmath 1993, Nichols et al. 2004). Even if we assume the opposite hypothesis and take causatives to be semantically simpler than anticausatives, due to the cross-linguistic variation in this respect, there will always be counter examples, i.e. languages where the semantically simpler alternate is formally more complex than the other form. Haspelmath (1993) argues that although no iconic relation can be found between semantic complexity and formal complexity, there is an iconic relation between conceptual complexity and formal complexity. This is the point where the external cause comes into play. In the present case, conceptual complexity is the result of the interplay of the following two parameters: (i) the position of the event described by the verb on the spontaneity scale and (ii) the way the event is linguistically expressed. The position on the spontaneity scale is related to the probability of the event to come about without an external cause.

_________ 12

13

"When there are two synonymous verb pairs that show different expression types, each of them was counted as 0.5." (Haspelmath 1993: 100) Note that this asymmetry in complexity also exists under Alexiadou et al.'s (2006) and Schäfer's (2008) view that a CAUSE-component is present in the anticausative alternant, because causatives have a voice-head and a cause-head.

142 Haspelmath (1993) gives 'wash' and 'laugh' as examples to characterize the negative and the positive ends of the spontaneity scale. 'wash' 'close' 'melt' 'laugh'

- spontaneous + spontaneous Figure 23: Spontaneity scale (cf. Haspelmath 1993: 105)

With respect to the second parameter, the event can either be linguistically expressed with an intransitive sentence without the explicit expression of the cause or with a transitive sentence with the explicit expression of the cause. Instantiations of high conceptual complexity are characterized by either of the following: (i) if an event that is located close to the negative end of the spontaneity scale is expressed with an intransitive structure (without a cause) or (ii) if an event that is located close to the positive end of the spontaneity scale is expressed with a transitive structure (with a cause). Accordingly, low conceptual complexity is given if an event that is located at the positive end of the spontaneity scale is expressed with an intransitive structure (without a cause) or if an event that is located at the negative end of the spontaneity scale is expressed with a transitive structure (with a cause) (cf. Haspelmath 1993: 106f.). Table 37: Conceptual complexity in the causative-anticausative alternation

+ spontaneous

- spontaneous

1 argument

- conceptually complex

+ conceptually complex

2 arguments

+ conceptually complex

- conceptually complex

Anticausatives only involve one argument, and the cause is either not expressed at all or is not expressed as an argument. As a consequence, anticausatives that are located at the negative end of the spontaneity scale are conceptually complex, while those located at the positive end of the spontaneity scale are conceptually simple. On the assumption that conceptual complexity is formally marked, the above considerations lead to the following prediction regarding the formal marking of anticausative verbs: If two verbs occupy different positions on the spontaneity scale and if only one of them is formally marked, the marked verb will be the element that is located nearer the negative end of the scale. With respect to the central question of this section, namely, the question of whether the formal difference between RAC and UAC mirrors a difference in the causal structure, we thus need to change perspective. On the basis of what

143 has been said so far, we have to conclude that the formal difference between RAC and UAC expresses a semantic difference. Events expressed with RACs are located nearer the negative end of the spontaneity scale than events expressed with UACs. As the position of an event on the spontaneity scale was defined as a function of the probability of the event to occur without an external cause, it follows that the formal difference between RAC and UAC marks a semantic difference with respect to the probability of the respective event to occur with or without an external cause (cf. Haspelmath 1993: 103). The probability of the occurrence of a particular event without an explicit cause is higher with UAC than with RAC. On the other hand, this difference also suggests that the cause is semantically more present in the context of RAC than in the context of UAC (which was the assumption discussed at the beginning of this section). The difference between RAC and UAC is expected to hold not only between verbs forming RAC and verbs forming UAC, but also for verbs which can form both RAC and UAC (e.g. French casser). For such verbs, a semantic difference between the two uses would be predicted, to the effect that in the case of the reflexive use the cause is semantically more present than it is in the case of the unmarked intransitive use. The choice between RAC and UAC would thus be a means to either stress or neglect the cause of the event. The falsifiability and, as a consequence, the validity of the hypothesis, relies on the degree of objectivity according to which events can be located on the spontaneity scale. In table 38 below, based on Haspelmath's (1993) overview of the formal encoding of the causative-anticausative alternation, a selection of verb meanings is given along with the total number of languages that were investigated. This total (that is, the sample population) is then evaluated with respect to whether they formally derive the causative or the anticausative part of the alternation. The result of this comparison is captured in the anticausative-causative ratio. Evidently, there are considerable differences between the verb meanings. Haspelmath (1993: 105) takes these differences to reflect the position of the verb meanings on the spontaneity scale. According to him, these differences also illustrate the relevance of spontaneity (and, consequently, conceptual complexity) for the encoding of the alternation. However, in order to establish a clear link between the different values in table 10 and the concept of spontaneity as such, an objective way to locate verb meanings on the spontaneity scale is required.

144 Table 38: Formal encoding of selected verb meanings (adapted from Haspelmath 1993)

verb meaning

sample (languages)

'freeze' 'dry' 'melt' 'burn' 'open' 'break'

21 20 21 21 21 21

formally derived anticausative 2 3 5 7 13 12.5

formally derived causative 12 10 10.5 5 1.5 1

ratio 0.17 0.30 0.48 1.40 8.67 12.50

Schäfer (2008: 224) takes up Haspelmath's (1993) idea and proposes that roots forming anticausatives (¥cause-unspecified in his terminology) are located on the spontaneity scale according to the above hypothesis. In particular, roots forming marked anticausatives are located closer to the negative end of the scale than roots that form unmarked anticausatives. Schäfer further assumes that this principle holds cross-linguistically, although languages differ with respect to the point on the scale where the line is drawn between marked and unmarked anticausatives. - spontaneous + spontaneous

German: … … … … … … … … … … … … … … Italian: … … ... …… … … … … … … … … … … Greek: … … … … … … … … … … … … Figure 24: Spontaneity scale (for ¥cause-unspecified) (Schäfer 2008: 224, modified)

Although Schäfer (2008) does not specify the empirical implications of the above table, every attempt to do so would face the methodological problem mentioned above, that is, the fact that there are no objective criteria for the positioning of events (or roots, for that matter) on the spontaneity scale. 5.3.3

Proposals on French: RAC vs. UAC

In the following I will take a look at literature on French in which a semantic difference between RAC and UAC with respect to the presence of the cause has been proposed. Rothemberg (1974) has argued that the formal difference between French RACs and UACs corresponds to a semantic difference between external and internal causation. In her dissertation on French labile

145 verbs, Rothemberg needed to accommodate many cases in which the unmarked intransitive use of a labile verb has a reflexively marked counterpart (e.g. casser and se casser 'break'). She describes the difference between the unmarked intransitive and the reflexive construction as follows. Unmarked anticausatives: L'élément lexical assumant la fonction du sujet est le siège de l'action, du processus qui est vu comme se développant organiquement à partir de lui et rien qu'à partir de lui, grâce à ses qualités inhérentes, à ses propriétés intrinsèques, à des forces internes qui permettent la réalisation de l'action du processus. (Rothemberg 1974: 67)

Reflexive anticausatives: Le se [...] marque que les qualités inhérentes, les forces internes, les propriétés de l'élément lexical assumant la fonction du sujet sont insuffisantes à elles seules pour permettre la réalisation de l'action ou du processus. (Rothemberg 1974: 67)

This description of the semantics of the UAC and RAC resembles the descriptions of internally and externally caused verbs given by Levin & Rappaport Hovav (1995) (cf. section 2.3). An important difference between the proposals is, however, that Levin & Rappaport Hovav (1995) use the distinction to account for the fact that not all English intransitive verbs of change of state can be used transitively, while Rothemberg (1974) uses her definition to account for the formal asymmetry within the class of French anticausative verbs. Forest (1988) assumes a difference between French UAC and RAC, the description of which is similar to Rothemberg's (1974) definition and diverges from it only in using a different terminology. He refers to the Aristotelian term entéléchique for the semantics of the unmarked intransitive: "[...] Aristote, qui parle d'entéléchie pour désigner la réalisation de ce qui est en puissance dans quelque chose [...]" (Forest 1988: 152). He uses the verb casser as an example: (157) a. b.

La branche se casse. La branche casse. 'The branch breaks.'

He argues that the reflexive use expresses the "[...] changement d'état purement extérieur aux virtualités de l'entité considérée [...]", while the unmarked intransitive use "[...] prend en compte dans la description la conformité du changement à une virtualité intime." (Forest 1988: 152) Rothemberg (1974) and Forest (1988) also have in common that they consider this semantic difference to be a tendency rather than a categorical distinction. This sets them apart from the analysis proposed in Labelle (1992), who also tries to account for the distribution of verbs in the two constructions on the basis of the semantic difference. Another difference between Rothemberg (1974) and Forest (1988) on

146 the one hand and Labelle (1992) on the other hand is the fact that only Labelle proposes a syntactic explanation for the semantic difference between the two constructions. According to Labelle (1992), the semantic characteristics of the UAC are the following: (i) (ii) (iii)

denotes an event internal to the subject presents the change as an event unfolding naturally the subject is the main actor in the event and is responsible for the event

To illustrate these properties, Labelle (1992: 391) gives the examples in (158) and (159) (the sentences in a. and c. are taken from Rothemberg (1974)): (158) a. b. c. d.

(159) a. b. c. d.

Jeanne rougit. *Jeanne se rougit. 'Jeanne blushes.' Il vit le mouchoir se rougir soudain. *Il vit le mouchoir rougir soudain. 'He saw the handkerchief suddenly turn red.' Après l'extraction du nerf, les dents noircissent. *Après l'extraction du nerf, les dents se noircissent. 'After the extraction of the nerve, the teeth turn black.' Les murs près de la chimenée se noircissent. ?Les murs près de la chimenée noircissent. 'The walls close to the chimney turn black.'

Following Rothemberg (1974), Labelle argues that the examples in a. denote an event "internal to the subject", while the examples in c. denote an event "[...] that is not internal to the subject but that affects it [...]" (Labelle 1992: 391). Labelle, however, does not conceive this semantic difference to be a difference between internal and external causation. Rather, by choosing the unmarked intransitive construction [...] the speaker asserts the autonomy of the process and linguistically attributes the responsibility for the change to the entity undergoing it. The event is presented as being internally driven or as unfolding naturally without obvious external control. (Labelle 1992: 401)

She stresses that there is an important difference between "[...] entities causing a change of state and entities linguistically attributed responsibility for the event." (Labelle 1992: 401), and illustrates the necessity for this distinction with the following examples.

147 (160) a. b. c.

Marie a rougi de honte. 'Marie blushed from shame.' La neige a fondu sous l'effet de la chaleur. 'The snow melted from the heat.' L'oiseau a mué sous l'effet du virus. 'The bird changed under the influence of the virus.'

(Labelle 1992: 401)

In these examples, the unmarked intransitive construction is followed by a PP expressing the cause of the change of state (la honte, la chaleur, le virus). Labelle seems to conclude from this configuration that the event cannot be internally caused by the subject, since the cause of the event is expressed in the PP. Therefore, she only attributes the responsibility but not the causation of the event to the subject of unmarked intransitive anticausatives. Despite this terminological distinction between causation and responsibility, I assume the semantic difference between RACs and UACs described by Labelle to be related to causal structure. Turning to the evaluation of Labelle's (1992) analysis, I begin with a series of examples that can be interpreted in favour of Labelle's (1992) proposal. Consider the following examples. (161) a. b.

*Après son coup, mes lèvres se sont gonflées tout de suite. Après son coup, mes lèvres ont gonflé tout de suite. 'After his punch, my lips started to swell immediately.'

The event described in (161) is an autonomous change internal to the subject, in the sense that the lips start to inflate once the outside impetus applies. The hitting of the lips is only the impetus for the change of state, which itself happens autonomously. It is not the case that the hitting itself inflates the lips. A similar case is illustrated by the examples in (162). Although the verb gonfler 'swell' can be used in both RACs and UACs, the two constructions are in complementary distribution with respect to their preferred context. (162) a. b.

Le ballon s'est gonflé. Le ballon a gonflé. 'The ball inflated.'

A context for (162a.) would be a situation in which somebody inflates the ball with a pump, that is, a non-autonomous event. A context for (162b.) would be a situation in which the ball has already been inflated, left outside in the sun, and then, influenced by the heat of the sun, becomes even bigger. Only in the second case the event would be autonomous. One might object that in the example in (162b.) too the event is not autonomous, as it is caused by the sun. (Cf. McKoon & MacFarland (2000) on natural forces in the causal chain.)

148 The following examples illustrate the use of the verb épaissir 'thicken', a verb that appears both in the unmarked intransitive and the reflexive construction. In the context given in (163) the reflexive construction is preferred over the intransitive, whereas the opposite holds in (164). (163) a. b.

La couche de neige s'est épaissie depuis hier. *La couche de neige a épaissi depuis hier. 'The cover of snow has thickened since yesterday.'

(164) a. b.

*Pendant une heure la crème s'est épaissie. Pendant une heure la crème a épaissi. 'The cream thickened for an hour.'

Intuitively, the two events described in (163) and (164) seem to differ with respect to the autonomy of their external causes. The autonomy is greater in (164) than it is in (163). The choice of construction is in accordance with Labelle's (1992) prediction. Besides cases as in (163) and (164) which can be interpreted in favour of Labelle's (1992) analysis, there are also cases which pose severe problems for Labelle's (1992) proposal. These cases are instances in which the intuitions on the parameter of autonomy are quite clear, but where the choice of construction does not follow Labelle's generalization. In further cases the situation arises that we do not have clear intuitions on the parameter at all, which makes the choice of construction seem rather arbitrary with respect to the parameter. First, consider the example in (165). (165) a. b.

Avec l'âge, ses lèvres se sont épaissies de plus en plus. ?Avec l'âge, ses lèvres ont épaissi de plus en plus. 'With age, his lips became thicker and thicker.'

Here, the intuition on the parameter of autonomy is quite clear. The change that the lips undergo is internal to the subject and occurs without obvious external causes. Therefore, the conditions for the use of the unmarked intransitive construction seem to be satisfied. However, contrary to Labelle's (1992) predictions, the reflexive is not only grammatical but even preferred over the unmarked intransitive. A different example, which cannot be categorized as to the parameter of event autonomy, is the verb refroidir 'cool down', illustrated below. (166) a. b.

Le pain (*se) refroidit hors du four. 'The bread cools down outside of the oven.' Marie / la chambre *(se) refroidit. 'Marie / the room cools down.'

149 Based on our knowledge about physical processes such as cooling in the example above, both the events that are described in a. and b. are identical with respect to the parameter of autonomy. In both cases the change is internal to the subject, and the influence of the outside world is exactly the same in both cases. Thus, the choice of construction, which is different in the two cases, cannot rely on the parameter of autonomy. What sets apart the two examples is that in a. the event is a change towards a normal state, whereas in b. it is a change away from the normal state.14 The choice of the construction seems to depend rather on this difference than on the parameter of autonomy. A similar case is provided by the different distributions of the verbs sécher and assécher 'dry'. The example in (167) describes an event where the subject is put in a certain position, and then the change of state occurs autonomously. As predicted by Labelle, the unmarked intransitive construction is chosen. (167) a. b.

*Il a mis l'éponge dans l'eau et son volume s'est augmenté. Il a mis l'éponge dans l'eau et son le volume a augmenté. 'He put the sponge into water and its size increased.'

For the example in (168), where the intransitive construction is chosen, it is much less clear whether the event is autonomous or not. Is it the case that the managers make certain decisions and the profits rise autonomously? Can profits rise, just like shy girls blush, and a lip that has been hit swells? (168) Grâce aux bonnes décisions des dirigeants, 'Thanks to the operators' smart decisions,' a. *les bénéfices se sont augmentés. b. les bénéfices ont augmenté. 'the profits increased.'

These questions, however, inevitably lead to more fundamental considerations about the quality of the parameter that Labelle's (1992) proposal depends on. Recall that she claims that there is a general semantic difference between the unmarked intransitive and the reflexive construction in that only the intransitive describes an autonomous event. However, in her description of this semantic difference it remains rather unclear on the basis of which criteria one should categorize an event in terms of the parameter of autonomy. How can we know whether the subject is the main actor in the event, whether the change is presented as internal to the subject, or whether the event is unfolding naturally and autonomously? Relying only on our intuitions is quite problematic,

_________ 14

This observation is due to Florian Lionnet (p.c.).

150 since "the autonomy of an event" does not seem to be a parameter that we have clear intuitions on. The vagueness and gradual nature of this parameter may explain why the two constructions in general do not differ with respect to this parameter. Instead, we find examples that display that particular difference next to clear counterexamples and cases which cannot be categorized at all with respect to this parameter. Labelle (1992) seems to be aware of this problem related to her semantic distinction and admits herself that the autonomy of an event cannot be objectively determined. The way she tries to overcome this problem is illustrated in her discussion of the contrast between (169a.) and (169b.). (169) a. b.

*Il vit le mouchoir rougir soudain. 'He saw the handkerchief suddenly turn red.' ?Les murs près de la cheminée noircissent. 'The walls close to the chimney turn black.'

(Labelle 1992: 391)

Labelle (1992) notes that [t]he difference between [(169a.) and (b.)] appears to be that while a speaker may know that soot is an external factor, he or she may decide to overlook this knowledge and present the change as an event unfolding naturally since in everyday life the change in color of a wall adjoining a chimney is a phenomenon occurring gradually, independently of obvious external influences. (Labelle 1992: 391, mod. StH)

Thus, Labelle anchors her notion of autonomy on the level of how speakers conceptualize the world around them, and not on the level of how the world around them actually is. The crucial point here is that, under this view, the unmarked intransitive construction is not used to describe autonomous events – instead, it describes an event as being autonomous. The unmarked intransitive and the reflexive are thus seen as two different ways of conceptualizing events. As we will see in what follows, this move has negative consequences for the epistemological value of Labelle's (1992) analysis. Recall that Labelle claims that her analysis allows the prediction of the distribution of verbs in the two constructions on the basis of the semantics of the two constructions. The unmarked intransitive construction only allows for verbs that describe an event which can be conceptualized as occurring autonomously. This observation is, however, not falsifiable, as the only indicator of the possible conceptualizations of the event described by the verb is its distribution in the two constructions. The argumentation thus remains circular in nature, until an independent criterion for the evaluation of events with respect to the parameter of autonomy is found.

151 Labelle (1992), as well as Rothemberg (1974) and Forest (1988), is thus confronted with the problem that I already mentioned at the end of the first part of this section, in the discussion of Haspelmath's (1993) approach. Due to the lack of proper criteria, the categorization of events according to their autonomy from external causes proves to be as difficult as the location of events on the spontaneity scale. For the time being, the hypothesis according to which a semantic difference exists between RAC and UAC with respect to their causal structure thus remains in the unsatisfying state of unfalsiability. My own contribution to the question of whether French RACs and UACs differ with respect to the semantic presence of a cause follows a somewhat different approach. Given the problems outlined above, which inevitably arise if the issue is considered from the semantic side, I will put forward a proposal that does not rely on such problematic semantic notions as "spontaneity", "autonomy" or "internal vs. external causation". If RAC and UAC differ with respect to the semantic presence of a cause, then it is also expected that causes are more often overtly expressed in the context of RAC than in the context of UAC. This assumption is motivated by the idea that in the case of such a semantic difference the overt expression would be a factor favoring the use of RAC. The obvious advantage of looking at the overt expression of causes is that it can be objectively observed and quantified. Every hypothesis relying on this parameter would thus be falsifiable. In the next section I will describe how the above assumption concerning the expression of causes and the assumption of differences between RAC and UAC with respect to their aspectual structures (cf. section 5.2) can be empirically tested with a corpus study.

5.4

Case studies

5.4.1

Method

In the previous sections, I have motivated the view that systematic semantic differences exist between French RACs and UACs that involve both the aspectual and the causal structure of the constructions. Still, as the discussion of the relevant literature has shown, there is little empirical proof for this difference. In order to find out more about both the empirical validity and the quality of the difference I conducted a corpus study.15 From the text corpus Frantext, I

_________ 15

Note that this is the third corpus study after "Anticausatives and reflexive constructions in 12th century French" (chapter 3) and "Reflexive anticausatives from

152 have extracted 1584 anticausative examples containing the following six verbs: augmenter 'increase', durcir 'harden', empirer 'worsen', enfler 'swell', gonfler 'swell', grossir 'grow, gain weight'.16 All six verbs belong to the semantic class of verbs of change of state. This choice is motivated by the fact that this class must be considered as the core class of anticausatives. Further, these verbs are degree achievements, i.e. verbs which can show both telic and atelic behaviour (a fact that makes these verbs particularly interesting for the present issue). The time spans that were considered for the different verbs do not temporally coincide. For example, those for the verb enfler are 1600-1639, 17501769, 1830-1859, 1900-1919 and 1970-1989; those for the verb gonfler are 1650-1799, 1890-1914, and 1950-1969. These (admittedly undesirable) differences between the verbs under investigation are a consequence of the fact that the verbs have different text frequencies and that single verbs have different text frequencies at different times. It was thus not possible to provide an upfront definition of five spot checks with the same temporal extension and check the use of all six verbs in these identical spot checks. In the qualitative analysis, sometimes even examples outside the spot checks needed to be considered, because the text frequency in the spot checks was too low to yield statistically significant results (e.g. the low frequency of the reflexive anticausative use of augmenter in the 20th century). The examples of the anticausative uses of these verbs were coded for the presence of certain indicators which I take to represent different manifestations of aspectual and causal structure. The question that I intend to answer with the corpus study is whether the distribution of the indicators is different with the RAC than with the UAC. I will first describe the different indicators and then formulate a working hypothesis on their distribution (cf. (182) below). Beginning with the aspectual structure, a distinction can be made between indicators that are related to telicity and indicators that are related to the focus on the resultant state. Note that some indicators are relevant for both telicity and the focus on the resultant state. Indicators for telicity are overt linguistic material which imply the telicity of the ED, such as the degree phrase in the following example. (170) a. b.

The road is widening 5m. does not entail The road has widened 5m. The road is widening. entails The road has widened. (Hay et al. 1999: 133)

Only the sentence in a., i.e. the sentence with the degree phrase, tests positive for telicity in the entailment test. In the above example a gradable change is

16

Old to Modern French" (chapter 3 and 4). Only data from the third one are considered here, although the verbs from the case studies may also appear in the other two corpus studies. Note that the corpus of examples does not include infinitives.

153 described. The telicity of the ED depends on whether the degree of this change is quantized or not. In their paper on degree achievements, Hay et al. (1999) distinguish three ways in which the degree of change may be quantized; overt linguistic material being only one of them. The other two are lexical and contextual inference. If a verb is derived from a closed-range adjective (e.g. straight/straighten), the degree of change is quantized by the lexical semantics of the underlying adjective. By contrast, this does not hold for verbs based on open-range adjectives (cf. long/lengthen). (171) a. b.

They are straightening the rope. does not entail They have straightened the rope. They are lengthening the rope. entails They have lengthened the rope. (Hay et al. 1999: 134)

The second subtype where inference plays a role concerns sentences such as the one in (172). Here, the degree of change is neither quantized by overt linguistic material nor can it be lexically inferred, as the verb is based on an open-range adjective. (172) The tailor lengthened my pants.

According to Hay et al. (1999: 136), "[r]eal-world knowledge informs us that there is a conventional maximal length for pants [...]". This maximal length then quantizes the maximal degree of change. As expected, the sentence tests positive for telicity: (173) The tailor is lengthening my pants. does not entail The tailor has lengthened my pants.

As already mentioned, I will only consider the first of the three ways indicated above, overt linguistic material. The following examples show how the degree of change can be quantized by overt linguistic material in French: a PP quantizing the change (PQC), a degree phrase (DegP), a quantized degree adverb (QDA), or an expression of resultant state (ERS). (174) PPs quantizing the change a. [...] Orebro, dont la population s'est augmentée de 49000 à 75000 habitants entre 1940 et 1960, 'Orebo, of which the population increased from 49000 to 75000 inhabitants between 1940 and 1960' (Les Grands ensembles d'habitations, 1963; Frantext)

154 b.

l'outillage de la petite colonie s'était augmenté d'une machine soufflante, 'the equipment of the small colony increased by a blower' (Verne, 1874; Frantext)

(175) degree phrases a. elle [=cette envie] s'est augmentée à un point que, si elle n'est bientôt satisfaite, je crains qu'il ne m'arrive quelque disgrâce. 'it [=desire] increased to a point that if it were not soon be satisfied, I am afraid that some disgrace will happen to me.' (Galland, 1715; Frantext) b. [...] et son mal s'est tellement augmenté, qu'il est enfin réduit au pitoyable état que j'ai eu l'honneur de vous dire [...] 'and his trouble increased in such a way that he is finally reduced to the pitiful state that I have had the honour to tell you about' (Galland, 1715; Frantext; mod. StH) (176) quantized degree adverb a. [...] il faut que le métal se soit solidifié complètement; 'it is necessary that the metal has solidified completely' (Barneriasm 1934; Frantext) b. je crois qu'ils verraient le monde se transformer complètement, [...] 'I think that they will see the world change entirely' (Malraux, 1928; Frantext) (177) expression of resultant state a. le Gean mort se mua en Serpent, 'the dead giant turned into a snake' (Flore, 1537; Frantext) b. le moindre bruit suspect s'enfle dans son cerveau en tonnerres destructeurs, [...] 'the smallest noise grows into destructive thunders in his brain' (Pergaud, 1910; Frantext)

In the context of an explicitly expressed resultant state, some verbs seem to describe a transformation rather than a gradable change. By expressing the resultant state, the PPs in (177) indicate that the resultant state differs from the initial state such that the event has to be understood as a transformation. This, in turn, is a crucial factor for the telicity of the ED: If an ED describes a transformation from state A into state B, then the event automatically has an inherent endpoint: the achievement of state B. The effect on the telicity of the ED in (76) is, however, the same as in (177): the PPs always imply telicity. The overt linguistic material presented in examples (174) to (177) are the indicators for telicity that the examples were coded for: expressions of resultant state, PPs quantizing the change, degree phrases, or quantized degree adverbs.

155 Turning now to the indicators for the focus on the resultant state, it needs to be mentioned first that some indicators for telicity are also indicators for the focus on the resultant state. This is true for PPs quantizing the change and expressions of the resultant state. At first sight, PPs quantizing the change only specify the degree of change. However, as the resultant state may be inferred as being the initial state plus the degree of change, I assume that these PPs also specify the resultant state. In the very same way, the resultant state is specified by the PPs in the examples in (178). Recall that Zribi-Hertz (1947: 47) characterizes such PPs as denoting a component of the resultant state. (178) PPs specifying the change a. [...] une petite ferme qui s'était encore augmentée récemment de granges et de hangars, [...] 'a small farm which recently increased by barns and sheds' (Zola, 1887; Frantext) b. [...] sa haine [...] s'était-elle augmentée de quelques regrets [...] 'his hate increased by some regrets' (Balzac, 1844; Frantext)

The only difference between the PPs in (174) and (178) is that the PPs in (178) do not imply telicity. As the NPs in the PPs are not quantized, they cannot quantize the degree of change, and the ED thus remains atelic. This difference in telicity is, however, irrelevant with respect to whether the PPs specify the resultant state or not; in fact, both of them do. The indicators for the focus on the resultant state are thus PPs which directly specify the resultant state or specify the degree of change. The following table summarizes the various indicators for aspectual structure. Table 39: Indicators for focus on the resultant state and telicity

indicators

focus on resultant state

telicity

expression of resultant state (ERS) PP quantizing change (PQC) PP specifying change (PSC)

expression of resultant state PP quantizing change degree phrase (DegP) quantized degree adverb (QDA)

I now turn to the indicators marking the difference in the causal structure of anticausatives. Here I consider the overt expression of causes in the context of anticausatives. Again, there exists a considerable variety of means of expression. The following examples illustrate the three configurations that are used most frequently: PPs introduced by de, PPs introduced by par and PPs with à cause de. Cf. Heidinger & Schäfer (2008) for a description of the diachronic development of par-phrases in the context of French RAC and reflexive passives.

156 (179) a. b.

c.

[...] ma joye s'augmente de ce que j'entens [...] 'my joy increases because of what I hear' (Calvin, 1543; Frantext; mod. StH) [...] la chair se gonfle par l'affluence des humeurs, et du sang, [...] 'The skin swells because of the accumulation of aqueous fluids and blood.' (1684, Bernier; Frantext; mod. StH) je suis triste en songeant à ce que vous souffrez à cause de moi; 'I am sad thinking about what you suffer because of me' (Bloy, 1900; Frantext; mod. StH)

The literature that discusses PPs expressing causes has mainly focused on the following two questions: (i) which are verbs the different PPs appear with, and (ii) which types of NPs are allowed in the PP. For instance, with respect to the first question, Leeman (1991) hints at a difference between Modern French decause and par-cause. While the first expression appears with verbs that "désignent le témoinage involontaire [...] d'un sentiment ou d'une sensation [...]", the latter chooses verbs "[...] impliquant une activité élaborée [...]" (Leeman 1991: 81). The following examples illustrate this contrast. (180) a. b. c. d.

Luc *Luc Luc Luc *Luc Luc

rougit rougit blush travaille travaille work

de par P par de P

plaisir. plaisir. pleasure plaisir. plaisir. pleasure

(Leeman 1991: 80)

The distribution of such PPs with reflexive and unmarked anticausatives has not yet been investigated in detail. In particular, the effect that the presence of such PPs has on the choice between the two constructions is still an open question. In section 5.3 I have discussed the hypothesis that there is a difference between RAC and UAC with respect to the causal structure. One way to empirically test this hypothesis is to look at the distribution of overtly expressed causes in the context of RACs and UACs. These are the indicators that the examples for the six verbs have been coded for. One methodological difficulty in the actual coding of the examples arose from the ambiguity of certain PPs with de. As we have already seen in the above discussion, the preposition de can either introduce NPs that specify the degree of change or NPs that express the cause of the anticausative event. Moreover, the very same PP is sometimes ambiguous between a causal and a degree interpretation (Benjamin Massot, p.c.) (cf. the following example). (181) Ses yeux [...] se gonflèrent de larmes [...]. 'his/her eyes filled with tears'

(1963, Moinot; Frantext)

157 This sentence may either be interpreted such that his eyes become bigger because of the tears (causal reading) or that his eyes become filled with tears (degree reading).17 In the historical data, where such semantic differences are not easily detectable due to the lack of native speakers, such cases will be coded as ambiguous and counted as 0.5 for both possible functions. The question to be answered is thus whether the indicators are distributed differently with RAC than they are with UAC. Recall that RACs are assumed to have a preference for telicity (compared to UACs) and to focus on the resultant state of the event, and that the cause is semantically more present in the case of the RAC. With respect to the distribution of the indicators this insight can be reformulated, yielding the following hypothesis. (182) Hypothesis: The share of RAC with indicator A on the total of RACs (of verb X) is significantly higher than the share of UAC with indicator A on the total of UACs (of verb X).

For instance, if the share of reflexive anticausative uses of verb X with the indicator A is 20% of the total of reflexive anticausative uses of verb X, then the prediction is that the share of unmarked anticausative uses of verb X with the indicator A is significantly lower than 20% of the total of unmarked anticausative uses of verb X. 5.4.2

Results

5.4.2.1

Quantitative development of reflexive and unmarked anticausative uses

For the following six French anticausative verbs, the relation between the reflexive and unmarked anticausative use has been investigated: augmenter 'increase', durcir 'harden', empirer 'worsen', enfler 'swell', gonfler 'swell', grossir 'grow, gain weight'. The main goal of these case studies is to verify whether the reflexive and the unmarked anticausative use of these verbs differ from each other with respect to the aspectual and the causal structure (as described in the previous sections). Apart from these qualitative issues, the diachronic development of the quantitative relation between the reflexive and the unmarked anticausative use of these verbs is also of interest. While the first

_________ 17

The term degree reading is to be understood here in the sense that the PP expresses the degree of change by specifying what is added. I am aware of the fact that this is a rather informal use of the notion of degree.

158 issue will be dealt with in the next section, I will describe the latter in the following paragraphs. The first verb under consideration is augmenter. Table 40 and figure 25 show the quantitative relation between the reflexive and the unmarked anticausative uses of this verb for five spot checks from the 16th to the 20th century. Table 40: Augmenter in RAC and UAC (abs. freq.)

1530-1599

1630-1669

1670-1699

1700-1719

1950-1959

RAC

21

19

86

65

9

UAC

21

45

37

12

153

100% 75% 50% 25% 0% 1500-1599

1630-1669 RAC

1670-1699

1700-1719

1950-1959

UAC

Figure 25: Augmenter 'increase' in RAC and UAC

The distribution of the verb augmenter in RAC and UAC substantially changed from the 16th to the 20th century. The general development can be summarized as follows: the share of UACs on the total of anticausative uses of augmenter increases, while the share of RACs decreases. In the first period (1500-1599), the two constructions are equally frequent, while in the last period (1950-1959) the UAC largely outnumbers the RAC (153 UAC vs. 9 RAC). An exception to this general trend is the change from the first to the second period, when the share of RAC increases. The second verb that has been investigated in this context is enfler, which has been analyzed for four periods from the 16th to the 20th century. As shown in table 41 and figure 26, there is no continuous development but a succession of increase, decrease and increase of UAC, which results in the wave form in figure 26. In the first four spot checks the reflexive is the most frequent anticausative use of enfler, and it is only in the fifth and last spot check that the unmarked anticausatives are in the majority.

159 Table 41: Enfler in RAC and UAC (abs. freq.)

1600-1639

1750-1769

1830-1859

1900-1919

1970-1989

RAC

21

14

34

36

15

UAC

4

12

1

6

18

100% 75% 50% 25% 0% 1600-1639

1750-1769 RAC

1830-1859

1900-1919

1970-1989

UAC

Figure 26: Enfler 'swell' in RAC and UAC

Gonfler is the next verb that I investigated. For this verb, three spot checks were conducted. In the first spot check, for which quite a large time frame has been chosen due the low frequency of anticausative uses, the reflexive anticausative largely outnumbers the unmarked anticausative (48 vs. 1 uses). Towards present-day French, however, the share of unmarked anticausative uses increases, as the data for the second and third spot check illustrate. Nevertheless, this increase does not change the dominance of the reflexive anticausative use of gonfler over its unmarked anticausative use; in all three spot checks the reflexive largely outnumbers the unmarked anticausative. Table 42: Gonfler 'swell' in RAC and UAC (abs. freq.)

1650-1799

1890-1914

1950-1969

RAC

48

54

36

UAC

1

19

6

160

100% 75% 50% 25% 0% 1650-1799

1890-1914 RAC

1950-1969

UAC

Figure 27: Gonfler in RAC and UAC

For the verb grossir five periods from the 16th to the 20th century have been examined. Table 43: Grossir 'grow/gain weight' in RAC and UAC (abs. freq.)

1550-1619

1700-1749

1800-1819

1890-1899

1950-1959

RAC

0

7

4

2

1

UAC

16

28

30

13

37

100% 75% 50% 25% 0% 1550-1619

1700-1749 RAC

1800-1819

1890-1899

1950-1959

UAC

Figure 28: Grossir in RAC and UAC

The development that this verb undergoes is the mirror image of the development of the verb gonfler. In the case of grossir the UAC dominates over the RAC, with changes on a small scale, such as the increase of RAC from the first to the second period. From the second period onwards to present-

161 day French, the share of unmarked anticausative uses decreases again. Over the whole time span under consideration, the unmarked anticausative use of grossir outnumbers the reflexive anticausative use. The verb durcir, in turn, could form both unmarked and reflexive anticausatives during the whole time span considered. The quantitative relation between the two types of anticausatives is given in table 44 and figure 29. Table 44: Durcir 'harden' in RAC and UAC (abs. freq.)

1550-1699

1700-1799

1850-1899

1940-1949

1980-2001

RAC

11

31

8

32

7

UAC

6

5

11

20

9

100% 75% 50% 25% 0% 1550-1699 1700-1799 1850-1899 1940-1949 1980-2001 RAC

UAC

Figure 29: Durcir in RAC and UAC

The quantitative relation between the two constructions is quite unstable over the whole period of consideration. Durcir has a preference for the reflexive in the 16th and 17th century (11:6) that becomes even stronger in the 18th century when the reflexive largely outnumbers the intransitive (31:5). Then, the quantitative relation changes and becomes almost balanced in the second half of the 19th century (8 RAC vs. 11 UAC). From the 19th century on towards present-day French the quantitative relation stabilizes, with only minor variation. For empirer, the sixth and last verb, four spot checks from the 17th to the 20th century were considered. The development of the anticausative use of this particular verb during this period is characterized by the decrease and loss of the RAC. In the first spot check, the share of RACs is still above 25%, but decreases in the next spot checks and finally disappears altogether.

162 Table 45: Empirer 'worsen' in RAC and UAC (abs. freq.)

1620-1699

1750-1799

1840-1859

1950-2000

RAC

7

2

1

0

UAC

13

12

28

30

100% 75% 50% 25% 0% 1620-1699

1750-1799 RAC

1840-1859

1950-2000

UAC

Figure 30: Empirer in RAC and UAC

To summarize, even the small sample of six verbs shows a considerable variety of different developments. There are verbs where the quantitative relation gravitates in one direction (e.g. empirer), verbs where increase and decrease succeed in a way that no general trend can be detected (e.g. enfler), and, finally, verbs where the quantitative relation remains relatively stable (e.g. grossir). Moreover, the increase and decrease of both RAC and UAC are attested. In the case of augmenter and empirer, the RAC decreases (and even disappears in the latter case), and in the case of enfler the RAC increases. These case studies corroborate the view that the emergence and spread of reflexive anticausatives does not result in a unidirectionality of changes of the quantitative relation between RAC and UAC. 5.4.2.2

Qualitative differences between reflexive and unmarked anticausatives

A total of 1584 anticausative examples of these six verbs have been collected and coded for the indicators presented in section 5.4.1. The time span under consideration stretches from the 16th to the 20th century. As the main focus of this study is the diachronic development of French anticausatives, a natural step to take is to consider the difference between the reflexive and unmarked

163 anticausative uses of the six verbs from a diachronic perspective. Unfortunately, as the frequency of the indicators is quite low in the texts of the corpora, only an analysis of the whole time span will give sufficient data for statistically significant results, while the comparison of data for several synchronic cuts will not. As we will see below, even the consideration of the whole time span does not always yield significant results. Nevertheless, the data provides interesting insights into the difference between the two types of anticausative verbs as such, and, in addition, permits observations with respect to the quality of this difference. Furthermore, we will see that the quality of the semantic difference between RACs and UACs can be linked to the quality of the spread of French RACs, as described in chapter 4. The presentation of the data proceeds from the general to the particular. Thus, I will first describe the distribution of the indicators with respect to the totality of occurrences of the verbs under discussion. On the basis of this general picture, then, a detailed discussion is possible where we will investigate whether there are differences between the individual verbs. Concerning the expression of causes, the most general but still reasonable way to look at the data is to put all six verbs together and compare the expression of causes in the context of reflexive and unmarked anticausatives. The following table shows that there is a statistically significant difference between the two types of anticausatives at this level (a chi-square test was used to measure the statistical significance; only distributions with P>0.95 are considered significant (P=1-p)). Table 46: Expression of causes with RAC and UAC for all six verbs (P>0.995)

total

with expression of cause (%)

RAC

721

74.5 (10.3%)

UAC

863

30 (3.5%)

With 10.3 percent of the reflexive anticausatives, the cause is expressed in a prepositional phrase, while this is only the case with 3.5 percent of the unmarked anticausatives. Causes are thus expressed nearly three times as often in the context of reflexive anticausatives than they are in the context of unmarked anticausatives. Despite this considerable difference it should be noted at this point that the difference is statistical and not absolute. It is not the case that only RACs, but not UACs allow for the expression of causes. Instead, the expression of a cause seems to be a factor that favors the use of the reflexive anticausative. The following series of examples illustrates the expression of causes in the context of RAC and UAC.

164 (183) expression of causes in the context of RAC a. Le desir de souffrir s'augmente par ma peine, 'the desire to suffer increases with/because of my pain' (Mareschal, 1637; Frantext; mod. StH) b. Peu à peu les matières molles [...], se seront durcies par leur propre poids, 'little by little the soft material will have hardened because of its own weight' (Buffon, 1749; Frantext) (184) expression of causes in the context of UAC a. d'heure en heure augmente Par trop de lascheté la force violente. 'hour by hour, the violence increases because of too much slackness' (Garnier, 1585; Frantext; mod. StH) b. Les mortiers proprement dits sont des mélanges de pâte de chaux et de matières inertes qui durcissent par l'action de l'air ou de l'eau. 'the mortar itself is a mixture of lime and inert material that harden because of the air and the water' (Pérès, 1896; Frantext)

The impression that the expression of a cause is a factor that favors the use of the reflexive anticausative persists when we narrow our focus to the level of individual verbs. With two of the six verbs of the sample, there is a statistically significant difference between RACs and UACs with respect to the distribution of causes. In both cases, causes are more frequently expressed with the reflexive anticausative. On its own, this result presumably does not suffice to assume a difference between RAC and UAC in this respect. However, a second argument for this claim comes from the remaining four verbs, where the difference is further attested in the sense that in these cases as well, causes are more frequently expressed with RAC than with UAC. Although it is a critical step to rely on statistically non-significant data I consider them to be a further argument that points in the same direction as the statistically significant differences. The following table gives the percentages for the six verbs.

165 Table 47: Expression of causes with six anticausative verbs

verb augmenter (P>0.995) enfler (P>0.995) gonfler (P0.995)

PP specifying change (P>0.995)

total

abs.

%

abs.

%

abs.

%

RAC

721

6

0.8

39.5

5.5

32

4.4

UAC

863

0

0

24

2.8

2

0.2

Table 51: Indicators for telicity with RAC and UAC

total

abs.

%

PP quantizing change (P>0.995) abs. %

RAC

721

6

0.8

39.5

5.5

21

2.9

1

0.1

UAC

863

0

0

24

2.8

22

2.5

2

0.2

expression of resultant state (P>0.995)

degree phrase (P RAC

0

2

difference without instance 11

170 The most important result in table 52 is that in all four cases with a significant difference between RAC and UAC the indicator appears more frequently with RAC than with UAC. However, the fact that in 26 cases no such difference has been found might raise doubts as to whether this result suggests a difference between RAC and UAC with respect to aspectual structure. Although it is critical to rely on data that is not statistically significant, in the present case, the 15 combinations of non-significant differences between RAC and UAC are nevertheless quite interesting, since in 13 out of the 15 cases with a nonsignificant difference between RAC and UAC, the indicator appears more frequently with RAC than with UAC. I suggest that the considerable quantitative difference between the non-significant differences of type RAC>UAC on the one hand, and UAC>RAC on the other hand (13 vs. 2), together with the fact that this result points in the same direction as the four statistically significant differences, should be seen as a further piece of evidence for the difference between RAC and UAC with respect to aspectual structure. The following table is a more detailed version of table 52. In addition to the information in table 52, it also lists the verb-indicator combination that the values in table 52 represent, and gives the relative frequencies of the respective indicator. differences P>0.95

differences PUAC

grossir/PQC (18% vs. 1.7%) augmenter/PSC (4% vs. 0.3%) gonfler/PSC (7.5% vs. 0%) grossir/PSC (6% vs. 0.4%)

augmenter/PQC (9.8% vs. 5.7%) augmenter/DegP (6.4% vs. 3.2%) enfler/ERS (1.5% vs. 0%) enfler/PQC (4.4% vs. 0%) enfler/PSC (3.3% vs. 0%) enfler/DegP (2.9% vs. 0%) enfler/QDA (0.7% vs. 0%) gonfler/ERS (0.5% vs. 0%) gonfler/PQC (1.4% vs. 0%) gonfler/DegP (0.9% vs. 0%) grossir/ERS (4% vs. 0%)19 durcir/DegP (2.9% vs. 1.5%) empirer/DegP (4.3% vs. 2.8%)

UAC>RAC

-

grossir/DegP (2.9% vs. 0%) durcir/QDA (3% vs. 0%)

Table 53: Statistically significant and non-significant differences in distribution of indicators of aspectual structure

_________ 19

In this case the chi-square test leads to a significant difference, but the number of total instances is lower than 6.

171 The above table indicates four significant differences in the distribution of indicators of aspectual structure with RAC and UAC. In all four cases, the indicators appear more frequently with RAC than with UAC. However, the frequency of the indicators as well as the degree of the difference between the frequencies of the indicator with RAC and UAC (i.e. the difference between the percentage of the respective indicator with RAC and with UAC) vary. The indicators involved in the four significant differences are PPs quantizing the degree of change on the one hand, and PPs specifying the change on the other hand. In the instances with the non-significant differences, all five indicators of aspectual structure are involved. As with the significant differences, the frequencies of the indicators and the degree of the difference between RAC and UAC vary. In the description of the situation on the level of the totality of verbs, the indicators were grouped as to whether they put the focus on the resultant state or imply telicity (cf. tables 50 and 51 above). Looking at all six verbs together, it seems that telicity is not a relevant parameter for the choice between RAC and UAC. The main point was that the two indicators that are linked exclusively to telicity (degree phrases and quantized adverbs) show no statistically significant difference in their distribution with RAC and UAC. The impression gained from the above, namely, that the focus on the resultant state but not the implication of telicity is a relevant parameter for the choice between RAC and UAC, is corroborated when we look at the level of individual verbs. Indeed, neither for degree phrases nor for quantized degree adverbs a statistically significant difference between the reflexive and the unmarked anticausative use of any of the verbs under discussion has been found. Furthermore, the two non-significant differences, where UAC is more frequent than RAC, are found with these two indicators (cf. grossir (UAC+DegP: 2.9% vs. RAC+DegP: 0%) and durcir (UAC-QDA: 3% vs. RAC-QDA: 0%)). To summarize, the study has revealed a difference in the distribution of indicators for a focus on the resultant state in the context of anticausatives: these indicators appear more frequently with RAC than with UAC. This difference is, however, only a statistical difference, as the indicators appear with both reflexive and unmarked anticausatives. The use of a given verb in the reflexive anticausative is preferred over the use in the unmarked anticausative in contexts with a focus on the resultant state.

172

5.5

Discussion

5.5.1

Resultant state or telicity as the relevant feature?

Zribi-Hertz (1987) puts forward the generalization that only perfective verbs are allowed in the RAC and that perfective verbs are defined as having a resultant state as one part of their meaning. The diagnostics to detect whether a verb has such a resultant state is the distribution of the verb in certain types of copulas (short passive with a resultative reading, adjectival passive, être with morphologically related adjective). As far as the distribution in resultative passives and adjectival passives is concerned, the question arises whether this diagnostics is sensitive to a resultant state in general or, more specifically, to telicity. In order to answer this question, two types of resultant state need to be distinguished: the positive vs. the comparative resultant state. In a recent paper, Rappaport Hovav & Levin (in press) argue for a principled distinction in the verbal lexicon between manner and result verbs. Both verb types have in common that they are dynamic and therefore involve change. The criterion that sets the two types of verbs apart is that only result verbs involve a scalar change: Verbs denoting events of scalar change lexically specify a scale, where a scale is a set of degrees – points or intervals indicating measurement values – on a particular dimension (e.g., height, temperature, cost), with an associated ordering relation [...]. The dimension represents an attribute of an argument of the verb, with the degrees indicating the possible values of this attribute. A scalar change in an entity involves a change in value of this attribute in a particular direction along the scale, with the direction speci¿ed by the ordering relation. (Rappaport Hovav & Levin in press: 8)

Aspectually relevant features such as telicity and durativity follow from the properties of the scale. A two point scale describes a punctual event; a scale with more than two points describes a durative event. A scale where the value of the attribute of the changing entity is fixed would be telic, if the value is not fixed it would be atelic. The difference between a fixed and a non-fixed attribute value is also related to the two types of resultant states mentioned above. In the first case the resultant state would be positive, in the second case, the resultant state would be comparative (cf. Kennedy & Levin 2008). (192) two types of resultant state: positive vs. comparative a. become (x state) positive ĺ telic b. become (x more state) comparative ĺ atelic

Let us now turn back to Zribi-Hertz's (1987) aspectual restriction on verbs forming RAC. As far as their distribution in the resultative and adjectival pas-

173 sive is concerned, the two types of resultant states behave differently: a positive resultant state is a necessary condition. The restriction of the resultative and adjectival passive thus only identifies verbs with a positive resultant state, and not, more generally, with a resultant state as such (positive or comparative). Riegel et al. (1999: 437f.), Vikner (1985: 100f.), Lagae (2005: 129f.) state that the resultative passive is only possible with telic verbs (verbes perfectifs in the terminology of Riegel et al. (1999), and verbes d'événements in the terminology of Vikner (1985)). The conclusion is thus that only verbs which can be telic form reflexive anticausatives. After considering the aspectual restriction on verbs forming RAC, let us now turn to the RAC itself. One important outcome of the corpus study was the observation that indicators emphasizing the resultant state appear more often in the context of RAC than in the context of UAC (i.e. the share of examples with such indicators is higher for RAC than for UAC). Crucially, the indicators create a preference for the RAC indicating either a positive or a comparative resultant state. Further empirical arguments supporting the view that the resultant state as such and not the positive resultant state is the relevant feature for the RAC can be found in Manente (2008). A first set of relevant data consists of combinations of UAC and RAC with time span and time frame adverbials. Time frame adverbials (in x TIME) are assumed to be only possible in the context of telic predicates, whereas time span adverbials (for x TIME) only occur with atelic predicates. Manente (2008) analyses the distribution of the adverbials pendant x TIME (= atelic) and en x TIME (= telic) with French UAC and RAC. The empirical basis of her analysis are acceptability judgements of 10 native speakers of French on 52 anticausative sentences. For a total amount of 13 verbs she tested the acceptability of combinations of (i) RAC with pendant x TIME, (ii) RAC with en x TIME, (iii) UAC with pendant x TIME, and (iv) UAC with en x TIME. The main conclusions are the following (cf. Manente 2008: 200): Firstly, with verbs which allow for both constructions, RAC is preferred over UAC in the context of en x TIME, and UAC is preferred over RAC in the context of pendant x TIME. Secondly, verbs which allow for only one construction (UAC or RAC) are (with some exceptions) compatible with both en x TIME and pendant x TIME. Hence, the distribution of the two types of adverbials with RAC and UAC is not complementary, both constructions are compatible with telic and atelic adverbials. Unfortunately, the non-categorical preferences of UAC for pendant x TIME and RAC for en x TIME with verbs which allow for both RAC and UAC could not be verified in my corpus, as the token frequency of en x TIME and pendant x TIME in the context of anticausatives is too low to yield statistically significant results. Another interesting result of Manente's (2008) work is that there is considerable variation among native speakers with regard to the acceptability of the sentences described above. The subjects had to judge the sentences as acceptable (¥), not fully acceptable (?), or unacceptable (*). Crucially, 17 of

174 the 52 sentences that were tested received all three types of judgements, i.e. they have been judged acceptable, not fully acceptable and unacceptable by at least one subject; for the remaining 35 sentences, the judgements of the 10 subjects were consistent. A second set of relevant data in Manente (2008) consists of verb pairs like baisser - abaisser and their distribution in French RAC and UAC. Some properties of these verb pairs are the following: (i) both verbs describe changes of states, (ii) both verbs are derived from the same adjectives, and (iii) one verb is formally derived from the other via prefixation. Manente (2008: 223) argues that the prefixes are not related to telicity (i.e. to a positive resultant state), but, more generally, to a scalar change (i.e. a positive or a comparative resultant state). Building on Di Sciullo (1996), Manente (2008: 223) maintains that the prefixes express the scale. À partir de ce fait, nous défendrons l'hypothèse que dans ces couples de verbes le préfixe a une incidence sur l’aspect. Plus précisément, nous formulerons l'hypothèse que le préfixe est porteur d'un sens aspectuel scalaire qui est compatible avec l'interprétation d'état résultant. Nous montrerons donc que, dans ces couples de verbes, il y a des différences aspectuelles entre le verbe préfixé et le verbe non préfixé en ce qui concerne l'interprétation d’état résultant. (Manente 2008: 223)

The prefix thus shows semantic similarities to some of the indicators for a resultant state that appeared in the case studies: both emphasize the resultant state by emphasizing the scale.20 But not only is their function similar. Just like the indicators, these prefixes have an influence on the distribution of verbs in RAC and UAC: En outre, la forme préfixée ne peut pas être employée intransitivement, tandis que la forme non préfixée a un emploi intransitif. (Manente 2008: 224)

Based on my own observations and Manente's (2008) observations, I assume that RAC and UAC differ in that the RAC but not the UAC "profiles" (cf. also Cornips & Hulk 1996) the resultant state of the event (positive or comparative). (193) (se) casser 'break' UAC: profiles event RAC: profiles result of event

As far as the RAC is concerned, I thus assume that the presence of a resultant state in general and not the presence of a positive resultant state is the relevant feature. The fact that if a verb can form both RAC and UAC, the RAC is pre-

_________ 20

For an analysis of the prefixes in French deadjectival verbs in terms of transitivity see Junker (1987).

175 ferred in telic contexts and the UAC is preferred in atelic contexts, does not contradict this assumption. Presumably, there is a strong relation between telicity and the resultant state: telicity implies a positive resultant state and therefore also implies a resultant state as such. It is thus not surprising that RAC is preferred over UAC in telic contexts despite the fact that RAC is not limited to telic contexts. The only mismatch that remains is that the RAC is only open to verbs which can be telic while the RAC is not restricted to telic contexts. In the next section where I discuss the structural origin of the aspectual difference between RAC and UAC, I will try to provide an explanation for this mismatch. 5.5.2

The Resultant state in RAC

The RAC but not the UAC emphasizes the resultant state of the anticausative event. It may thus seem surprising that the UAC is open to verbs which have a resultant state in their lexical semantics and that the UAC also allows for PPs specifying the resultant state. In order to clarify this point, one must distinguish between resultant states on two levels: the level of the lexical semantics of verbs, and the level of the semantic structure of syntactic constructions. The RAC is restricted to verbs that have a resultant state in their structure because the RAC itself has a resultant state in its structure. The use of verbs without a resultant state would thus result in a semantic incompatibility. The UAC focuses on the event rather than on the resultant state of the event. This, however, does not create an incompatibility with verbs that have a resultant state in their semantics because these verbs also involve an event (that leads to the resultant state) which can be emphasized. Hence, the UAC is open to verbs with and without a resultant state. Note further that the resultant state is only focused with the RAC but not with an RS-verb in UAC. The above-mentioned relations between verbs and the two anticausative constructions are illustrated in the figure below.21

_________ 21

I prefer this twofold distinction over a threefold distinction between RAC and two UAC-constructions (one with an RS component in it and one without) where verbs with a resultant state would appear in RAC or the UAC with a resultant state, and verbs without a resultant state would appear in the second UAC. Such a model would predict the possibility of a focus on the resultant state also in UAC, a hypothesis that is not supported by the data.

176 verb

+ resultant state

- resultant state

construction

RAC focuses on resultant state

UAC focuses on event

focus on resultant state

resultant state, but no focus on resultant state

outcome

no resultant state

Figure 31: Resultant state, verbs and anticausative constructions

I will now turn to the question of why the two constructions show different aspectual behaviour. What is it that makes the RAC, but not the UAC, focus on the resultant state? In the case of semantically and formally distinct constructions, semantic differences are usually attributed to the item that formally distinguishes the two constructions.22 In the present case, the reflexive morphology would be assumed to be responsible for the focus on the resultant state. This is, however, only a stipulation relating formal and semantic differences without describing how the presence of the extra morphology is related to the semantic difference. In the literature, different accounts of the role of the reflexive have been proposed. A first account, given by Labelle (1992: 395), is that the reflexive morpheme "[...] 'discharges' the subject theta-role, allowing it not to be expressed." The subject position could be filled by the theme argument which moves from object to subject position. The RAC is thus an unaccusative construction. The subject of the UAC, on the other hand, is assumed to be generated in subject position, and the UAC would thus be an unergative construction. Recall Labelle's (1992) claim that RAC and UAC differ both on the aspectual and the causal level. Crucially, Labelle relates both these abovementioned differences to the fact that RAC and UAC have different syntactic structures. The inheritance of object-like properties of the subject of RAC,

_________ 22

Note that here, the term construction is used pretheoretically and not as in the framework of Construction Grammar. For a representation of Romance anticausatives (and the differences between marked and unmarked anticausatives) in a constructionist RRG approach see Kailuweit (to appear).

177 which does not hold for UAC, would not only explain Labelle's (1992) presumed semantic differences between RAC and UAC, but would also account for the results of my case studies. Labelle's (1992) arguments for a syntactic difference between RAC and UAC concern the following phenomena: (i) Perfect auxiliary selection (ii) Impersonal construction (iii) Embedded infinitival relative (iv) Object raising (v) Participial constructions Beginning her discussion with auxiliary selection, Labelle (1992: 380) notes the well-known fact that in present-day French, reflexive anticausative verbs select être 'be', whereas unmarked intransitive verbs select avoir 'have' (cf. (194)). (194) a. b.

La branche s'est cassée. La branche a cassé. 'The branch broke.'

RAC: être 'be' UAC: avoir 'have'

However, Labelle does not take the selection of avoir as a diagnostic for the unergativity of a given verb. The only conclusion she draws is that perfect auxiliary selection does not suggest that UACs are unaccusative (cf. Labelle 1992: 381).23 The second issue concerns the distribution of these verbs in the impersonal construction and with en clitization (cf. Labelle 1992: 381-383). Labelle first observes that "[u]nlike unergatives, typical unaccusative verbs [(195a.)] and passive verbs [(195b.)] freely enter into an impersonal construction." (Labelle 1992: 381) (195) Que s'est-il passé? 'What happened?' a. - Il est arrivé trois hommes. 'There arrived three men.' b. - Il a été vendu beaucoup de popcorn. 'There was sold a lot of popcorn.'

(cf. Labelle 1992: 381)

With respect to unergatives, Labelle (1992: 381) notes that their distribution in impersonal constructions is possible but more restricted than the occurrence of unaccusatives (only the presentational and the existential reading is possible).

_________ 23

Cf. Manente (2008) on the relation between perfect auxiliary selection and unaccusativity in French (and Italian).

178 The basic assumption behind the diagnostic, Labelle (1992: 381) suggests, is that impersonal il fills a subject position that is empty because the only referential argument remains in object position. Thus, if the use of unergatives in impersonal constructions is possible, they should not qualify as unergatives. Furthermore, Legendre & Sorace (2003) show that even the restriction on certain readings of unergative verbs in impersonal constructions, as noted in Labelle (1992: 381), does not hold. Some of Legendre & Sorace's (2003: 205) examples for "event-reporting" impersonal constructions with typical unergative verbs are given below. (196) a.

b.

Il a échappé une bourde au président. 'There escaped a blunder from the president.' (uncontrolled process, telic, unergative) D'ici peu il voyagera de nombreux millionnaires dans l'espace. 'There soon will travel numerous millionaires in space.' (controlled process, atelic, unergative)

Thus, according to these data, the impersonal construction does not set apart typical unaccusative from typical unergative verbs. Turning to verbs of change of state, from Labelle's (1992) perspective one would expect that reflexive verbs are grammatical in the impersonal construction, while this should not be the case for unmarked intransitive verbs. Labelle provides examples such as (197), which suggest that this is true. (197) Que s'est-il passé? 'What happened?' a. Il s'est cassé trois branches. b. *Il a cassé trois branches. 'There broke three branches.'

For Labelle (1992: 381f.), "[t]he difficulty of using the non-reflexive use of verbs of change of state in an impersonal construction groups this use with unergatives rather than with unaccusatives." However, we have already seen above that the distribution of typical unergative verbs in impersonal constructions is less restricted than Labelle assumes. This argument is further weakened by the fact that unmarked intransitive verbs of change of state are only ungrammatical in the impersonal construction if the unmarked intransitive is in direct competition with a reflexive form (cf. casser in (197)). Verbs of change of state that only allow for the unmarked intransitive but not for the reflexive, however, can be used in an impersonal construction (cf. sécher 'dry' (198)).

179 (198) a. b. c.

Trois chemises sèchent dehors. *Trois chemises se sèchent dehors. 'Three shirts dry outside.' Il sèche trois chemises dehors. 'There are three shirts drying outside.'

Although this fact has been noted by Labelle (1992: 382), she does not discuss the consequence of this distribution, which, in effect, groups unmarked intransitive verbs such as sécher together with reflexives like se casser. As a consequence, the distinction between unergative and unaccusative verbs would not correlate with the distinction between UAC and RAC: sécher and se casser would be unaccusative, while casser would be unergative. We may thus conclude from the discussion above that the distribution in the impersonal construction does not suggest a syntactic difference between the RAC and the UAC. In a next step, Labelle looks at impersonal constructions with encliticization of postverbal NPs. According to Labelle (1992: 383), there, the restrictions are even tighter, and unergative verbs that can be used in the impersonal construction lose this option once the postverbal NP is cliticized. (199) a. b.

Il s'en est cassé trois. *Il en a cassé trois. 'There of.them has broken three.'

Recall that in the impersonal construction the single argument appears in postverbal position, which, in SVO languages like French, seems to be reserved for objects. As we have seen, there are, however, instances in which the single argument can appear in post-verbal position, provided that the expletive il fills the subject position. The grammaticality of the single argument appearing post-verbally depends on the verb. Therefore it is assumed that in the case of verbs that allow the single argument to appear post-verbally, the single argument has a different status than it has in the case of verbs where the single argument can only appear pre-verbally. The question is then why encliticization would impose stronger restrictions than the impersonal construction with a full NP. As mentioned above, Legendre & Sorace (2003) argue that the impersonal construction does not distinguish unaccusative from unergative verbs and they extend this view to cases with en-clitization. As empirical evidence they show that verbs like voyager 'travel' – which are typically considered as unergative verbs – in the construction. (200) D'ici peu il en voyagera de nombreux dans l'espace. 'Soon there will of.them travel many in space.' (Legendre & Sorace 2003: 208)

180 The next parameter Labelle (1992: 383-385) discusses involves embedded infinitival relatives. This diagnostic, originally suggested for French by Pollock (1985), builds on grammaticality contrasts such as the one in (201). (201) a. b.

L'homme que je croyais être arrivé/entré/avoir disparu. 'The man who I thought to have arrived/come in/disappeared.' ?*L'homme que je croyais avoir téléphoné/toussé/plongé dans l'eau. 'The man who I thought to have telephoned/coughed/dived into the water.' (cf. Labelle 1992: 384)

Pollock assumes that only unaccusative verbs, passive participles and être can assign case to their postverbal argument: the trace of wh-movement would be case-marked in (a.) but not in (b.) (cf. Labelle 1992: 384). Concerning verbs of change of state, Labelle (1992: 384) argues that reflexives pattern with unaccusative verbs in that they allow for the embedded infinitival, which is not the case with unmarked intransitives. (202) a.

b.

Le vase que tu croyais avoir cassé est en parfait état. 'The vase that you thought to have broken is in perfect condition.' (only transitive-causative reading of casser possible) Le vase que tu croyais s'être cassé est en parfait état. 'The vase that you thought to have broken is in perfect condition.' (anticausative reading)

However, there are data suggesting that the impossibility of the anticausative reading in (202a.) is again due to a conflict between a transitive-causative reading and an anticausative reading, which leans toward the transitive-causative. In the following examples the strictly intransitive change of state verbs faiblir 'weaken' and croître 'increase' are used in an embedded infinitival construction. (203) Les cours boursiers que je croyais avoir faibli ont en réalité augmenté. 'The stock market prices that I thought to have decreased have in fact increased.' (204) Le peuple qu'on croyait avoir cru a en réalité disparu. 'The people that one thought to have grown has in fact disappeared.'

The decisive factor for the possibility of the embedded infinitival construction is thus not related to the initial syntactic position of the subject argument, but instead to the possibility of a transitive-causative reading. In cases such as (202a.), the anticausative reading is ruled out by the transitive-causative reading. In cases such as (202b.), (203) and (204), where the transitive-causative reading is ruled out, either by the reflexive or by the fact that the verb does not

181 have a transitive-causative use, the anticausative reading is available. We may thus conclude that the construction is not sensitive to unaccusativity. The next parameter Labelle (1992: 385-387) discusses is Object Raising (OR). However, as Labelle (1992) concludes that reflexive and unmarked intransitive verbs of change of state do not differ with respect to OR, and that OR is not a diagnostic for unaccusativity, we can leave aside this parameter and turn directly to the last parameter, namely, the adjectival passive and participial constructions (cf. Labelle 1992: 387-389). This last diagnostic is based on the assumption that the argument of perfect participles must be a deep structure object. Accordingly, if a verb cannot form a participial construction, one must assume that the verb does not have a deep object. Note that this test applies primarily to verbs and not to the reflexive or intransitive construction. Still, Labelle (1992) argues that the test reveals a difference in behaviour of RAC and UAC. Consider the following example from Labelle (1992: 388): (205) a. b. c. d. e.

Marie a grandi. 'Marie has grown (up).' *Marie semble grandie. 'Marie seems grown.' Cette expérience a grandi Marie. 'This experience has made Marie grow (in stature).' Marie semble grandie par cette expérience. 'Marie seems to have grown (in stature) as a result of this experience.' *Marie s'est grandie. 'Marie has grown (up).'

Labelle (1992) takes the ungrammaticality of the participial construction in (205b.), which is related to the intransitive structure in (205a.), to suggest that the intransitive structure in (205a.) does not have a deep object and is thus unergative. On the other hand, the participial construction in (205d.), which is related to the transitive structure with the deep object in (205c.), is grammatical. An argument for the lack of a deep object in the unmarked intransitive use of verbs of change of state could be found in the observation of verbs that cannot be used in the reflexive construction and thus cannot form a participial construction that is related to this intransitive use. Labelle (1992) assumes that participial constructions that are formed with verbs of change of state which have both an intransitive and a reflexive use are related to the reflexive use (cf. Labelle 1992: 389). The assumption that the unmarked intransitive use of verbs of change of state is unergative thus predicts a distribution as in table 54.

182 Table 54: Predicted distribution of verbs of change of state in participial constructions

finite construction verb type I

participial construction

unmarked

if available, not related to unmarked

verb type II

unmarked & reflexive

yes, related to reflexive

verb type III

reflexive

yes, related to reflexive

However, there are some indications that this prediction does not hold. Crucially, Legendre & Sorace's (2005: 38) English translation for the French phrase une fois le lait bouilli, 'once the milk (had) boiled', suggests that the participle construction can be related to an intransitive use of a verb of change of state. Note that bouillir 'boil' does not have a reflexive use.24 Labelle (1992: 389) concludes on the basis of the tests described above that there is no evidence that the unmarked intransitive use of verbs of change of state is unaccusative, which is why she assumes it to be unergative. The reflexive use of verbs of change of state, on the other hand, is assumed to be unaccusative. This conclusion is surprising given the similar behaviour of the unmarked intransitive and the reflexive use of verbs of change of state in the tests applied (summarized in table 55). Table 55: Summary of unaccusativity tests

unmarked verbs of change of state

reflexive verbs of change of state

avoir

être

impersonal construction

possible

possible

embedded infinitival relative

possible

possible

object raising

possible

possible

participial construction

possible

possible

perfect auxiliary selection

My conclusion is therefore that the above tests are insufficient to assume a syntactic difference between the unmarked intransitive and the reflexive use of verbs of change of state. As the set of verbs of change of state intersects with the set of anticausative verbs, this conclusion has to be extended to the difference between RACs and UACs. One consequence of this lack of a syntactic difference is that the preference of the aspectual and causal indicators for RAC cannot be explained on syntactic grounds.

_________ 24

Cf. Manente (2008) for a description of the aspectual properties of the verb bouillir.

183 Another type of explanation for the aspectual properties of the reflexive anticausative builds on the idea that the reflexive morpheme introduces a small clause. Folli & Harley (2005) argue that the Italian anticausative reflexive morpheme si is the realization of a light verb. This light verb si selects a complement that encodes the resultant state of the event. (206)

A similar idea can be found in Cornips & Hulk (1996) who maintain that French anticausative se is the head of an aspectual small clause (similar to certain particles or prefixes). (207)

The syntactic structure proposed by Folli & Harley (2005) for Italian and the structure proposed by Cornips & Hulk (1996) have in common that the theme argument (it. la casa 'the house' and fr. la branche 'the branch') form a small clause with a stative predicate designating the resultant state of the anticausative event. In Folli & Harley (2005) this stative predicate is the adjectival core of the verb describing the anticausative event, while in Cornips & Hulk (1996) the resultant state is expressed by the PP en deux 'in two pieces'. Although both structures seem able to explain the aspectual difference between RAC and UAC, they have one major drawback in that they are unaccusative structures (the theme argument is in a VP (or v) internal position). If the aspectual difference between RAC and UAC relies on this distinction, we must also assume that UAC is not unaccusative. As a consequence, we would thus expect a clear difference between RAC and UAC with respect to syntactic

184 tests for unaccusativity. However, this difference would predict different behaviours between RAC and UAC in a number of syntactic tests; a prediction which is not borne out. My conclusion as far as the structural explanation for the semantic difference between RAC and UAC is concerned is therefore rather modest. We observe a semantic difference between RAC and UAC as far as the focus on the resultant state is concerned. Since this difference also exists when the same verb is used in both constructions, the difference cannot be attributed to the verb itself, but must be attributed to the constructions. In the literature, this difference has been accounted for by proposing different syntactic structures for RAC and UAC, where only the RAC contains a small clause. 5.5.3

The Cause in French RAC

Until now, I have discussed the aspectual properties of French anticausatives. The results of the case studies, however, also suggest a difference between RAC and UAC as far as the semantic presence of a cause is concerned. In this section I will focus on this latter result and try to provide an explanation for this observation. In section 5.3 the question of whether a cause is part of the semantic or syntactic structure of anticausatives has been discussed. Unlike Schäfer (2008) and Alexiadou et al. (2006) I do not assume that the syntax of anticausatives includes a cause-head, and unlike Koontz-Garboden (2009) I do not assume that the macro roles actor and undergoer merge in the subject of reflexive anticausatives. My own view unites two facts that have already been mentioned several times in this book. Firstly, I assume that se is a valence operator. It manipulates the number of arguments by deleting the cause-argument. The reflexive morpheme is thus an overt sign of anticausativization. Crucially, there is no such overt sign in the case of unmarked anticausatives.25 The presence of a formal trace of the transitive-causative use in RAC but not in UAC leads us to better understand the fact that speakers assign a semantics in which the cause of the event is more salient to the RAC rather than to the UAC. This, however, does not imply a constant semantic difference between RAC and UAC as far as the presence of a cause is concerned.

_________ 25

Recall from chapter 2 that I assume unmarked anticausatives to be formed from a lexical entry with only one argument. Note that the difference between RAC and UAC with respect to an overt sign of anticausativization would also exist if one assumes that unmarked anticausatives are also formed by an operation of argument deletion.

185 Since Hopper & Thompson (1980), transitivity has been conceived as a multidimensional phenomenon in the typological literature. Transitivity is no longer reduced to the presence or absence of a direct object (i.e. the number of arguments) but also includes aspectual properties such as telicity. Further properties of transitivity are (i) affectedness of the object, (ii) agentivity of the subject, (iii) modality, and (iv) punctuality. For all these properties, there exist values with high and with low transitivity, as shown in the table below. Table 56: Features of transitivity (Hopper & Thompson 1980: 252)

participants

high transitivity

low transitivity

two or more

one

kinesis

action

non-action

aspect

telic

atelic

punctuality

punctual

non-punctual

volitionality

volitional

non-volitional

affirmation

affirmation

negative

mode

realis

irrealis

agency

A high in potency

A low in potency

affectedness of O

totally affected

not affected

individuation of O

highly individuated

non-individuated

Crucially, Hopper & Thompson (1980) assume that the values of these properties correlate. This assumption is formulated in their transitivity hypothesis: (208) If two clauses (a) and (b) in a language differ in that (a) is higher in Transitivity according to any of the features [in table 56 ], then, if a concomitant grammatical or semantic difference appears elsewhere in the clause, that difference will also show (a) to be higher in Transitivity.

Similar to the semantic differences between RAC and UAC, this hypothesis groups aspectual and causal properties. In Hopper & Thompson (1980), it is the presence of a cause (as an argument) and telicity that are grouped together, in the difference between RAC and UAC it is the semantic presence of a cause and the focus on the resultant state. As far as the aspectual difference between RAC and UAC is concerned I have assumed in the previous section that the reflexive morpheme does not only delete the cause-argument of a transitive verb but also creates a syntactic structure that profiles the resultant state of the event. The strong relation between aspectual and causal properties of anticausatives can thus be formulated as follows: Se is a valence operator that creates

186 a structure focussing on the resultant state. Additionally, se remains in the structure as a trace of the transitive-causative lexical entry of the verb (cf. also Kupferman 2008). Finally, it should be noted once again that unaccusativity might also explain differences between RAC and UAC with respect to the semantic presence of a cause and focus on the resultant state. According to Labelle (1992), the fact that French RAC is unaccusative and French UAC is unergative can explain both the aspectual and the causal difference between them (cf. sections 5.2.2, 5.3.3 and 5.5.2 for discussions of her view). The analysis of her empirical arguments has, however, shown that the evidence can also be interpreted differently, namely that there is no such syntactic difference between RAC and UAC.

6

Auxiliary selection in unmarked anticausatives and the spread of the reflexive anticausative

In this chapter, two findings on French anticausatives shall be related: the quality of the spread of the RAC on the one hand, and the aspectual properties of the RAC on the other hand. I will put forward the hypothesis that the strong increase of the RAC after the Middle French period is at least favoured by the decrease of être 'be' as a perfect auxiliary for unmarked anticausative verbs. The auxiliaries être 'be' and avoir 'have' with intransitive verbs are considered to have different aspectual values – with the first focussing on the resultant state of the event and the latter focussing on the event itself. With the loss of être as a perfect auxiliary, this aspectual distinction could no longer be overtly expressed by the choice of the auxiliary. But at this point the language already had another means to express this distinction, namely, the choice between RAC and UAC. The RAC, just like être+Vpp, focuses on the resultant state of the event. I thus assume that the loss of être+Vpp was compensated by a more frequent use of RAC. In other words, the strong increase of RAC is causally related to the decrease of être as a perfect auxiliary. To evaluate the plausibility of this hypothesis, more needs to be known about the diachronic development of the perfect auxiliary selection in French. Before I describe this development, some general remarks on why the RAC could increase so strongly are in order. It has been argued in chapter 5 that the RAC, but not the UAC is specified with regard to the presence of a resultant state in the construction. Because of this specific constraint the set of verbs that could form RAC is more restricted than the set of verbs forming UAC. Hence, the set of verbs that could form RAC is a subset of the verbs that could form UAC. Given this asymmetry with respect to the restrictions imposed by RAC and UAC, it might seem contradictory that the RAC and not the UAC (which imposes fewer restrictions) is the most frequent type of AC in present-day French. The contradiction disappears once the following two points are taken into account. Firstly, the fact that a verb meets the semantic requirements of a construction does not imply that the verb is used in this construction. The verbs forming anticausatives in French fall into three classes, depending on their distribution in RAC and UAC: verbs forming only RAC, verbs forming only UAC, and verbs forming both RAC and UAC. Thus the verbs forming UAC are not a superset of the verbs forming RAC. There are verbs that fulfil the requirements of the UAC, but are not used in UAC. (The fact that the distribution of the verbs cannot be predicted on the basis of semantic or formal criteria gives the distribution an idiosyncratic character.)

188 The second point is that verbs without a resultant state are a relatively small set among the verbs that can form anticausatives. Thus the restriction implied by the RAC does not rule out many anticausative verbs. Therefore, the RAC can be the most frequent type of anticausative despite its additional restriction because (i) the set of verbs that forms UAC is smaller than the set of verbs that could form UAC, and (ii) the set of verbs that is ruled out from the RAC via the resultant state-restriction is relatively small. As far as the increase of the RAC is concerned, the observations above only describe the condition for the possibility of the increase, but do not explain the increase itself. At this point the change in the auxiliary selection comes into play. I will first give a short description of changes in French perfect auxiliary selection, and then discuss how the increase of the RAC might be related to these changes. In present-day French, transitive verbs select avoir 'have' and reflexive verbs select être 'be' as a perfect auxiliary.1 Most intransitive verbs select the auxiliary avoir; only some intransitive verbs of motion and even fewer verbs of change of state select only être. Cf. the list below, taken from Grevisse & Goose (2008: 1032): (209) intransitive verbs selecting only être as a perfect auxiliary (present-day French): aller 'go', arriver 'arrive', décéder 'die', devenir 'become', entrer 'enter', mourir 'die', naître 'be born', partir 'leave', rester 'stay', retourner 'return', sortir 'go out', tomber 'fall', venir 'come'

Additionally, Grevisse & Goose (2008: 1033) give a longer list of intransitive verbs which can select both être and avoir and where the choice between the two auxiliaries expresses an aspectual difference. Certains verbes intransitifs ou pris intransitivement se conjuguent avec avoir quand ils expriment l'action – et avec être quand ils expriment l'état résultant de l'action accomplie. (Grevisse & Goose 2008: 1033)

Examples of verbs from their list that are relevant for the present case are given in (210). (210) intransitive verbs selecting both être and avoir as a perfect auxiliary (present-day French): accroître 'grow/increase', augmenter 'increase', baisser 'go down', changer 'change', commencer 'begin', diminuer 'decrease', éclater 'burst', embellir 'embellish', empirer 'worsen', enlaidir 'turn ugly', finir 'end', grandir 'grow in size', grossir 'grow/gain weight' (Grevisse & Goose 2008: 1033)

_________ 1

Note that in some dialects and varieties of French, especially those spoken in Northern America, avoir has been generalized even more and has become the only auxiliary (cf. Manente 2008: 42-47 for an overview).

189 In present-day Italian, some verbs of change of state display the behaviour that I assume to characterize earlier stages of French: they can form the perfect with both essere 'be' and avere 'have' and the choice between the two auxiliaries is driven by the aspectual difference. Manente (2008: 203) gives marcire 'rot', fiorire 'flower', impallidire 'turn pale' and arrossire 'blush' as examples of verbs which select both essere and avere. She maintains that the selection of essere favors a telic interpretation, while avere is more common in atelic contexts (Manente 2008: 204). This is illustrated by the following series of data where the combinations of essere 'be' and time span adverbials and avere 'have' with time frame adverbials are marked as questionable, and not as ungrammatical: (211) a. b.

(212) a. b.

Gianni è/ ?ha arrossito in un minuto. 'Gianni blushed in a minute.' Gianni ?è/ha arrossito per un minuto. 'Gianni blushed for a minute.'

(Manente 2008: 204)

Il melo è/ ?ha fiorito in due mesi. 'The apple tree flowered in two months.' Il melo ?è/ha fiorito per due mesi. 'The apple tree flowered for two months.'

(Manente 2008: 204)

As far as present-day French is concerned, Grevisse and Goose (2008: 1033) maintain that in most of the cases the possibility to choose between the auxiliaries is rather "théorique que pratique". The choice is restricted in two ways: (i) for most verbs of change of state the construction être+Vpp can no longer receive an eventive interpretation, but only a stative resultative interpretation (e.g. changer 'change', décroître 'decrease', grandir 'increase in size'), and (ii) many verbs of movement show a clear preference for être (e.g. descendre 'go down', passer 'pass', etc.). This is confirmed by other descriptions of auxiliary selection in present-day French and judgements of native speakers. For example, Mackenzie (2006: 104) lists only the following verbs selecting être in present-day French: aller 'go', apparaître 'appear', arriver 'arrive', descendre 'go down', devenir 'become', entrer 'enter', intervenir 'intervene', monter 'go up', mourir 'die', naître 'be born', partir 'leave', rentrer 'come back', rester 'remain', retourner 'return', revenir 'return', sortir 'go out', survenir 'arise', tomber 'fall', venir 'come'. One might thus wonder why Grevisse & Goose (2008: 1033) mention the possibility of the selection of both avoir and être and then dispense with this hypothesis in the next paragraph. Apparently, they try to capture uses of the auxiliaries that could be found in texts of older stages and that can still be found in literary texts displaying the situation of these older stages of the language. In Classical French, a stage of French that is still present in grammar writing and lexicography, the situation was indeed such that some

190 anticausative verbs could be used with both auxiliaries, as will become apparent in the following discussion. The development of the perfect auxiliary selection from Old French to Modern French is described in Mackenzie (2006: 129-144). His starting point is the idiosyncratic situation in present-day French (and other Romance languages) where auxiliary selection cannot be semantically determined. The question he poses is thus how French could arrive at such a situation. His description of the development consists of three stages. In the first stage, the sequence être+Vpp was not used to express the perfect, but was a resultative copula describing the resultant state after an event. Because of the semantics of the construction, only telic verbs could appear in this construction. In the second stage, the construction was interpreted as a perfect tense (without losing the possibility to be interpreted as a resultative copula) and être was grammaticalized as a perfect auxiliary. In the early period of being a perfect auxiliary only telic verbs selected être (a heritage of the aspectual restriction of the resultative copula). In the third stage, the auxiliary selection was conventionalized independently of the aspectual properties of verbs. According to Mackenzie (2006: 142), a "[...] dissociation of the 'resultant state' constraint from its original syntactic locus (viz. a copula-adjective construction) [...]" took place. In French this dissociation led to an increase of verbs selecting avoir, and to a decrease of verbs selecting être. (213) 3 stages for être + Vpp in French a. Stage 1: resultative copula; possible only with telic verbs b. Stage 2: grammaticalization as perfect, only telic verbs, unified semantics c. Stage 3: conventionalization of auxiliary assignment, loss of unified semantics

Since most verbs forming anticausatives could also figure in copula constructions, we must assume that they have also undergone a change from selecting only être to selecting only avoir with an intermediate stage where they could select both auxiliaries, and where the choice of the auxiliary could express aspectual distinctions. For Old French, Benzing (1931: 451) notes that être was the dominant auxiliary of intransitive verbs. Clédat (1903: 38f.) maintains that verbs of change frequently selected être as a perfect auxiliary in the middle ages ("l'ancienne langue"), but only select avoir in present-day French. Middle French is considered by Förster (1908: 103) as an intermediate period between Old French (dominance of être) and Modern French (dominance of avoir) as far as auxiliary selection is concerned. In his study on the development of auxiliary selection he distinguishes between verbs where the auxiliary selection remained the same since Old French, and verbs where the auxiliary selection changed. Within the latter class of verbs Förster distinguishes three subclasses:

191 (i) verbs which only take être in Modern French, (ii) verbs which only select avoir in Modern French, and (iii) verbs where the use of avoir increased since Old French without fully substituting être. Anticausative verbs only appear in the latter two classes, i.e. they have undergone a change in auxiliary selection where avoir expanded at the expense of être. Examples of anticausative verbs with avoir as perfect auxiliary in Modern French and with être and avoir at older stages of French are approcher 'approach', avancer 'move', disparaître 'disappear', lever 'rise', tourner 'turn', finir 'end', périr perish', crever 'burst', rompre 'break', diminuer 'decrease', and multiplier 'multiply' (Förster 1908: 69100). Examples of anticausative verbs with être and avoir as perfect auxiliaries in Modern French where avoir increases since Old French are apparaître 'appear', descendre 'go down', monter 'go up', cesser 'end', commencer 'begin', croître 'grow', accroître 'grow, increase', and changer 'change' (Förster 1908: 7-53). As far as the latter verbs are concerned I do not follow Förster's claim that they still select être in present-day French. Crucially, Förster's (1908) data only covers the period from the 14th to the 17th century. The following examples illustrate the possibility of using être as the perfect auxiliary of unmarked anticausative verbs during this period. (214) rompre 'break' a. [...] par ou il est rompu 'where it broke' (14th century, Miracles de Nostre Dame (Förster 1908: 93)) b. Mes biens mondains sont rompuz et cassez 'my earthly goods shattered and broke' (16th century, Œuvres complètes de Gringore (Förster 1908: 93)) (215) tourner 'turn' a. Més ma vie est tournée en desespoir 'but my life turned into despair' (14th century, Froissart (Förster 1908: 83)) b. Nostre ris est tourné en pleurs 'our laughter turned into tears' (16th century, Le Mystère du vieil testament (Förster 1908: 84))

For the 16th century, Gougenheim (1973: 119) notes that the possibility to choose between être and avoir existed for more verbs than in present-day French. He cites entrer, voler, couler, marcher, courir, périr, croître as examples (cf. Gougenheim 1973: 119f.). Spillebout (1985: 191) notes for the 17th century that the auxiliary selection was similar to present-day French. Relevant changes (and thus differences) he notes are the loss of être as a perfect auxiliary for the verbs cesser and croître. (cf. Spillebout 1985: 192) For 18th century French, Seguin (1972: 84) notes the following anticausative verbs that could select both être and avoir: cesser, croître, descendre, monter. In Wagner & Pinchon (1989: 283), a grammar for Classical and Modern French, the

192 possibility to form the perfect of a given verb with être or avoir is also discussed. Among their examples, we also find the anticausative verbs changer and embellir. They note that by choosing être the resultant state in the present is described, while choosing avoir yields a construction describing a process that began in the past (Wagner & Pinchon 1989: 283). However, their description of the perfect formed with être rather resembles a resultative construction and not a perfect tense. Moreover, the examples with être that could receive an eventive interpretation all stem from Classical French (from authors such as Racine, Corneille, or Molière). This survey of the existing literature shows that the change of perfect auxiliary selection from Old to Modern French strongly affected anticausative verbs: the possibility to form their perfect with être or avoir has been lost. The sequence être+Vpp could no longer receive an eventive interpretation (with a focus on the resultant state), but only a resultative (passive) and a (eventive) passive interpretation. With the loss of être as a perfect auxiliary, the semantic difference between focus on the event and focus on the resultant state could no longer be overtly differentiated by the choice between être and avoir. I assume thus that because of the loss of the aspectual distinction between être+Vpp and avoir+Vpp, a different formal means to express this semantic difference started to be used more frequently: RAC (focus on the resultant state) vs. UAC (focus on the event). Simplifying somewhat, we observed two changes and assume a causal relationship between them. Change 1 is the decrease of the perfect focusing on the resultant state, change 2 is the increase of the RAC (which focuses on the resultant state). The assumed causal relation is that the change in the domain of auxiliary selection accelerates the spread of the RAC. Given the empirical observations it seems safe to assume that the loss of être is a factor favoring the use of the RAC. It is much more difficult to determine the exact impact of the loss of être as a perfect auxiliary for unmarked anticausatives on the course of the increase of RAC. Recall from chapters 3 and 4 that the strong increase of the relative frequency of RAC began after 1500. In the literature, the decrease and loss of être as a perfect auxiliary with anticausative verbs is not described in such depth that would allow correlating the rising graph of the relative frequency of RAC with a falling graph of the use of être as a perfect auxiliary with anticausative verbs. Still, the existing information about the chronology of the two changes makes a causal relation between the increase of RAC and the decrease of être very plausible.2 Note

_________ 2

The empirical observations concerning the increase of RAC and the change of the auxiliary selection are open to two different interpretations: (i) the decrease of être pulls the increase of RAC, or (ii) the increase of RAC pushes the decrease of être. The fact that the change in the auxiliary selection is a process that goes beyond verbs forming anticausatives suggests that the assumption of a pull-relation is correct.

193 that this does not imply that the decrease of être is the only reason for the increase of RAC. The change in auxiliary selection is not only relevant for the course of the increase of the relative frequency but is also important for the quality of the spread of RAC. To be more precise, if we take the change in the auxiliary selection to be an important factor for the increase of the RAC we are able to understand the most important finding on the spread of RAC reported in chapter 4: the increase of the relative frequency of RAC is not accompanied by a spread from one type of anticausative to another or by the enhancement of types of anticausatives. As a consequence, the strong increase of the RAC cannot be linked to any internal changes of the RAC. Early on, the RAC expressed the same types of anticausatives that it expresses in present-day French. What has changed, however, is the frequency of different types. As far as semantic verb classes are concerned, a strong increase of verbs of change of state has been observed. Still, the data on the spread of RAC does not suggest that the RAC changed and it is because of its new properties that speakers began to use it more frequently. Instead, I assume that the RAC (with its constant properties) became to be used more frequently (at least partially) because of an independent change, namely the change in the auxiliary selection of UAC. Hence the consideration of the auxiliary selection helps us to understand why the frequency of the RAC could increase without spreading to new types of anticausatives.

7

Conclusion

The main goal of this book was to investigate the diachronic development of French anticausatives. I concentrated on the emergence and spread of reflexive anticausatives (RAC) and also discussed the consequences that these two processes had on the unmarked anticausative (UAC), which existed already before the emergence of RAC. This last chapter shall now bring together the insights and results that emerged in the analysis of these three central topics. The emergence of the French reflexive anticausative has been described in chapter 3 of this book. Based on a first corpus study I have argued that the reflexive anticausative emerges in the course of the 12th century. I then zoomed in on the situation in 12th century Old French in order to analyze the linguistic circumstances of the emergence of the French RAC. This second corpus study revealed three important facts. Firstly, unmarked anticausatives are much more frequent in the 12th century than reflexive anticausatives (abs. freq. in my corpus: 129 vs. 5). Secondly, reflexive anticausatives are only attested in texts from the second half of the 12th century (this is why I assume that the French RAC emerges in the second half of the 12th century). Thirdly, reflexive psych verbs with non-agentive subjects existed already before the emergence of the reflexive anticausative. I have argued that this intermediate reflexive construction (between true reflexive and reflexive anticausative) is a necessary condition for the reflexive anticausative to emerge. As far as the mechanism of language change that is responsible for the emergence of the French RAC is concerned I have made the following three points: Firstly, the mechanism of language change (more precisely: reanalysis or analogical extension) cannot be determined on an empirical or theoretical basis. Secondly, both mechanisms of language change (reanalysis and extension) rely on the above-mentioned intermediate construction, since in both cases an existing valence operation (applied already in the formation of reflexive psych verbs) is extended. Thirdly, the difference between analogical extension proper and reanalysis is smaller than commonly assumed. In fact, it is reduced to whether a new surface sequence is created or not created in the course of the emergence of the reflexive anticausative. In chapter 4 I described the development of French RAC following its emergence in the 12th century. Both quantitative and qualitative aspects of this change have been taken into account. To capture the quantitative side of the spread of RAC, (i) the relative frequency of RAC and (ii) its relative lexical diffusion have been analyzed. Both the relative frequency and the relative lexical diffusion strongly increased from Old French to Modern French. The discrepancy between the increases of the two parameters shows that – unsurpris-

195 ingly – the increase of the relative frequency is not only caused by a spread at the lexical level, but also by an increase of the frequency of use of the verbs forming RACs. The relative frequency of RAC increases continuously but not constantly from Old French to Modern French. It begins slowly and accelerates after the end of the Middle French period before it slows down again after the middle of the 18th century. With respect to the qualitative side of the spread, the main questions that I discussed were whether there are properties that distinguish Old from Modern French anticausatives and whether a spread of the RAC from one type of anticausative to another (or others) could explain the particular course of the increase of use of the RAC. (i) Semantic verb class and (ii) animacy of the subject have been considered. Beginning with semantic verb classes, one conclusion to be drawn is that the spread of the RAC in the lexicon is not a spread from one verb class to another. Already in the first period when RACs were attested (1170-1179) all four semantic verbs classes that have been distinguished are attested (verbs of change of state, aspectual verbs, verb of appearance, verbs of change of position); thus, there is no spread of the RAC from one verb class to another. Instead, the new pattern spreads in parallel fashion in all four semantic verb classes. A further result is that during the whole period verbs of change of state constitute the most frequent semantic verb class in RAC. Hence there is no change as far as the semantic verb classes that form RAC (verbs of change of state, verbs of change of position, aspectual verbs, verbs of appearance) and the most frequent verb class (verbs of change of state) are concerned. What changes, though, is the proportion between the four classes. Crucially, the dominance of the verbs of change of state over the other three classes increases after Middle French, which is also the point in time where the overall relative frequency increases. With respect to the animacy of the subject the situation is similar. Old and Modern French do not differ in an absolute way, only the proportion between the different types changes. Already in Old French both animate and inanimate subjects are attested. But while the relative frequency of the RAC with animate subjects remains stable until present-day French, the frequency of those with inanimate subjects strongly increases. Again, what sets apart Modern from Old French reflexive anticausatives with respect to animacy is the proportion between animate and inanimate subjects: the dominance of RACs with inanimate subjects increases. Furthermore, I have analyzed whether the verbs found in RAC could also be used in UAC. This parameter proved to be the most interesting one, not only because it shows the most drastic change from Old and Middle French to Modern French but also because it correlates with the strong increase of the reflexive anticausative after Middle French. The difference between Old and Middle French and Modern French reflexive anticausatives lies in the fact that until the end of Middle French, reflexive anticausatives were predominately

196 formed by verbs which can also form unmarked anticausatives. After Middle French the situation changed completely and the vast majority of verbs forming RAC could not be used in UAC. The change of the status of the RAC from being one of two options for an anticausative verb (in Old and Middle French) to being the only option for most verbs forming reflexive anticausative verbs coincides with the strong increase of the relative frequency after Middle French. However, this change is not a spread of the RAC from one type of anticausative to another. Since the status of the RAC changes, the question of why this change occurred naturally arises. The explanation that I proposed in chapter 6 of this book relies on an observation on the semantic differences between French RAC and UAC. My case studies on the distribution of certain aspectual indicators with six anticausative verbs (as well as other literature on the topic) suggest that the RAC and not the UAC focuses on the resultant state of the anticausative event. In chapter 6 I related this result to a change in the auxiliary selection of unmarked anticausatives. The auxiliaries être 'be' and avoir 'have' with intransitive verbs are considered to have different aspectual values – the first focusing on the resultant state of the event and the latter focusing on the event itself. With the loss of être as a perfect auxiliary, this aspectual distinction could no longer be overtly expressed by the choice of the auxiliary. But at this point, the language already possessed a different means to express the aspectual distinction, namely, the choice between RAC and UAC. It thus seems reasonable to assume that the decrease of être as a perfect auxiliary with French UAC was a factor favoring the use of the RAC and thus contributed to the increase of the relative frequency of the construction after Middle French. This observation also helps us to understand how the RAC could increase so strongly without changing its inherent properties. The consequences of the spread of the RAC have been discussed in section 4.6. There, I have argued that the UAC decreases from Old to Modern French and is often replaced by the RAC. The hypothesis put forth in chapter 6, namely, that the increase of the RAC is accelerated by the change of the auxiliary selection of UAC, also makes a prediction about the relation between the increase of the RAC and the decrease of the UAC the decrease of the UAC would pull the increase of the RAC (and the increase of the RAC would not push the decrease of the UAC).

References

Aitchison, Jean (1991): Language change: progress or decay? – Cambridge: Cambridge University Press. Alexiadou, Artemis, Elena Anagnostopoulou and Florian Schäfer (2006): "The properties of anticausatives crosslinguistically." – In: Mara Frascarelli (ed.): Phases of Interpretation, 175-199. Berlin: Mouton. Altfranzösisches Wörterbuch = Blumenthal, Peter and Achim Stein (2002): ToblerLommatzsch: Altfranzösisches Wörterbuch. – Stuttgart: Steiner. Altmann, Gabriel (1983): "Das Piotrowski-Gesetz und seine Verallgemeinerungen." – In: Karl-Heinz Best and Jörg Kohlhase (eds.): Exakte Sprachwandelforschung, 5490. Göttingen: Edition Herodot. Babin, Gerhard (1937): Das Medium im Altfranzösischen. – Jena, Leipzig: Gronau. Barra Jover, Mario (2007): "S'il ne restait que l'induction: corpus, hypothèses diachroniques et la nature de la description grammaticale." – In: Recherches linguistiques de Vincennes 36, 89-122. Bassac, Christian (1995): Le statut de verbe dit ergatif: Étude contrastive anglaisfrançais. – Doctoral dissertation: Université de Nancy II. Ben Salah-Tlili, Imen (2007): "Contribution à l’étude des 'verbes symétriques' en français contemporain." – In: Jacques François and Ahmed Brahim (eds.): Morphosyntaxe et sémantique du verbe. Cahier n°23 du CRISCO, 15-37. Caen: Université de Caen. Benzing, Josef (1931): "Zur Geschichte von ser als Hilfszeitwort im Spanischen." – In: Zeitschrift für Romanische Philologie LI, 385-460. BLMF = ATILF (CNRS, Université Nancy 2): Base de Lexiques du Moyen Français (BLMF). – Nancy. – www.atilf.fr [access: 2005]. Bolinger, Dwight (1977): Meaning and Form. – London, New York: Longman. Borer, Hagit (1994): "The Projection of Arguments." – In: University of Massachusetts Occasional Papers 17, 19-47. – (2003): "Exo-skeletal vs. Endo-skeletal Explanations: Syntactic Projections and the Lexicon." – In: John Moore and Maria Polinsky (eds.): The Nature of Explanation in Linguistic Theory, 31-68. Stanford: CSLI Publications. BTMF = ATILF (CNRS, Université Nancy 2): Base textuelle de Moyen Français (BTMF). – Nancy. – www.atilf.fr [access: 2005]. Buridant, Claude (2007): Grammaire nouvelle de l'ancien francais. – Paris: Sedes. Burston, Jack L. (1979): "The pronominal construction in French: An argument against the fortuitous homonymy hypothesis." – In: Lingua 48, 147-176. Campbell, Lyle (22004): Historical linguistics: An introduction. – Cambridge, Mass.: MIT Press. Cennamo, Michela (1993): The Reanalysis of Reflexives: a diachronic perspective. – Naples: Liguori. – (1998): "The loss of the voice dimension between Late Latin and Early Romance." – In: Monika Schmid, Jennifer Austin and Dieter Stein (eds.): Historical Linguistics 1997, 77-100. Amsterdam: Benjamins. – (2000): "Patterns of 'Active' Syntax in Late Latin Pleonastic Reflexives." – In: John Smith and Delia Bentley (eds.): Historical Linguistics 1995, Vol. I: General Issues and non-Germanic Languages, 33-55. Amsterdam: Benjamins.

198 Chen, Matthew (1972): "The Time Dimension: Contribution toward a Theory of Sound Change." – In: Foundations of Language 8, 457-498. Chierchia, Gennaro (2004): "A semantics for unaccusatives and its syntactic consequences." – In: Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert (eds.): The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface, 22-59. Oxford: Oxford University Press. Clédat, Léon (1903): "Le participe passé, le passé composé et les deux auxiliaires." – In: Revue de Philologie Française XVII, 19-62. Comrie, Bernard (21989): Language universals and linguistic typology. – Chicago: University of Chicago Press. – (2006): "Transitivity pairs, markedness, and diachronic stability." – In: Linguistics 44 2, 303-318. Cornips, Leonie and Aafke Hulk (1996): "Ergative reflexives in Heerlen Dutch and French." – In: Studia Linguistica 50 1, 1-21. Crystal, David (1994): The Cambridge encyclopedia of language. – Cambridge: Cambridge University Press. de Alencar, Leonel F. and Carmen Kelling (2005): "Are reflexive constructions transitive or intransitive? Evidence from German and Romance." – In: Miriam Butt and Tracy Holloway King (eds.): Proceedings of the LFG 05 Conference, 1-20. Standford: CSLI Publications. De Smet, Hendrik (2009): "Analysing reanalysis." – In: Lingua 119, 1728-1755. de Swart, Henriëtte (1998): "Aspect Shift and Coercion." – In: Natural Language and Linguistic Theory 16, 347-385. Di Sciullo, Anna-Maria (1996): "Prefixes and Suffixes." – In: Claudia Parodi, Carlos Quicoli, Mario Saltarelli and Maria-Luisa Zubizarreta (eds.): Aspects of Romance Linguistics, 177-194. Washington: Georgetown University Press. Dixon, R. M. W. and Alexandra Y. Aikhenvald (2000): "Introduction." – In: R. M. W. Dixon and Alexandra Y. Aikhenvald (eds.): Changing valency Case studies in transitivity, 1-29. Cambridge: Cambridge University Press. DMF 2009 = ATILF (CNRS, Université Nancy 2): Dictionnaire du Moyen Français (DMF 2009). – Nancy. – www.atilf.fr. DMF1 = ATILF (CNRS, Université Nancy 2): Base de Lexiques de Moyen Français (DMF1). – Nancy. – www.atilf.fr. Doron, Edit and Malka Rappaport Hovav (2007): "Towards a Uniform Theory of Valence-changing Operations." – In: Yehuda N. Falk (ed.): Proceedings of Israel Association for Theoretical Linguistics 23. Dubois, Jean and René Lagane (1960): Dictionnaire de la langue française classique. – Paris: Belin. Fagan, Sarah M. B. (1992): The syntax and semantics of middle constructions: A study with special reference to German. – Cambridge: Cambridge University Press. Fellbaum, Christiane and Anne Zribi-Hertz (1989): "La construction moyenne en français et en anglais: étude de syntaxe et de sémantique comparées." – In: Recherches Linguistiques de Vincennes 18, 19-57. Feltenius, Leif (1977): Intransitivizations in Latin. – Uppsala: Almkvist & Wiksell. Fillmore, Charles (1970): "The grammar of Hitting and Breaking." – In: Roderick A. Jacobs and Peter S. Rosenbaum (eds.): Readings in English Transformations Grammar, 120-133. Waltham, MA: Ginn. Fischer, Olga (2007): Morpho-syntactic Change: Functional and Formal Perspectives. – Oxford: Oxford University Press.

199 Folli, Raffaella (2002): Constructing telicity in English and Italian. – Doctoral dissertation: University of Oxford. – and Heidi Harley (2005): "Flavours of v: Consuming Results in Italian and English." – In: Paula Kempchinsky and Roumyana Slabakova (eds.): Aspectual Enquiries, 95120. Dordrecht: Springer. Forest, Robert (1988): "Sémantisme entéléchique et affinité descriptive: pour une réanalyse des verbes symétriques ou neutres du français." – In: Bulletin de la Société de Linguistique de Paris 83 1, 137-162. Förster, August (1908): Avoir und être als Hilfsverba beim intransitiven Zeitwort in ihrer Entwicklung vom Alt- zum Neufranzösischen. – Darmstadt: Otto. Fournier, Nathalie (1998): Grammaire du français classique. – Paris: Belin. François, Jacques (1989): Changement, causation, action: trois catégories sémantiques fondamentales du lexique verbal français et allemand. – Genève: Droz. Frantext = ATILF (CNRS, Université Nancy 2): Base textuelle de Frantext. – Nancy. – www.frantext.fr. Furetière, Antoine (1970 [1690]): Dictionnaire universel contenant généralement tous les mots français tant vieux que modernes et les termes de toutes les sciences des Arts, Tome 1-3. – Genève: Slatkine Reprints. Gamillscheg, Ernst (1957): Historische französische Syntax. – Tübingen: Niemeyer. Garey, Howard B. (1957): "Verbal aspect in French." – In: Language 33, 91-110. Geniušienơ, Emma S. (1987): The typology of reflexives. – Berlin, New York: Mouton. Goldberg, Adele E. (1995): Constructions: A Construction Grammar Approach to Argument Structure. – Chicago: University of Chicago Press. Gougenheim, Georges (1973): Grammaire de la langue française du seizième siècle. – Paris: Picard. Goyens, Michèle (2001): "L'origine des verbes français à construction dative." – In: Lene Schøsler (ed.): La valence, perspectives romanes et diachroniques: actes du colloque international tenu à l'Institut d'Études Romanes à Copenhague, du 19 au 20 mars 1999, 43-58. Stuttgart: Steiner. Grevisse, Maurice and André Goosse (142008): Le bon usage: grammaire française. – Bruxelles: De Boek & Larcier. Grimshaw, Jane (1982): "On the Lexical Representation of Romance Reflexive Clitics." – In: Joan Bresnan (ed.): The Mental Representation of Grammatical Relations, 87148. Cambridge, Massachusetts: MIT Press. Harris, Alice C. and Lyle Campbell (1995): Historical syntax in cross-linguistic perspective. – Cambridge: Cambridge University Press. Härtl, Holden (2003): "Conceptual and grammatical characteristics of argument alternations. The case of decausative verbs." – In: Linguistics 41 5, 883-916. Haspelmath, Martin (1987): Transitivity Alternations of the Anticausative Type. – Köln: Universität zu Köln. – (1990): "The grammaticalization of passive morphology." – In: Studies in Language 14 1, 25-72. – (1993): "More on the typology of inchoative/causative verb alternations." – In: Bernard Comrie and Maria Polinsky (eds.): Causatives and transitivity, 87-120. Amsterdam: Benjamins. – and Thomas Müller-Bardey (2004): "Valency change." – In: Geert Booij, Christian Lehmann and Joachim Mugdan (eds.): Morphologie: Ein internationales Handbuch zur Flexion und Wortbildung. Band 2, 1130-1145. Berlin: de Gruyter.

200 Hatcher, Anna (1942): Reflexive verbs: Latin, Old French, Modern French. – Baltimore: John Hopkins Press. Hay, Jennifer, Christopher Kennedy and Beth Levin (1999): "Scalar Structure underlies Telicity in "Degree Achievements"." – In: Tanya Mathews and Devon Strolovitch (eds.): Proceedings of SALT IX, 127-144. Ithaca: CLC Publications. Heidinger, Steffen (2005): "A diachronic perspective on synchronic valence mismatches." – In: Alexandru Mardale, Mélodie Soufflard and Géraldine Vercherand (eds.): Proceedings of 9ème Atelier des Doctorants en Linguistique, 135-141. Paris: Université de Paris 7. – and Florian Schäfer (2008): "The French reflexive passive and anticausative: a diachronic view from the par-phrase." – In: Bejamin Fagard, Sophie Prévost, Bernard Combettes and Olivier Bertrand (eds.): Evolutions en français: Etudes de linguistique diachronique, 135-152. Bern: Peter Lang. Herslund, Michael (2001): "L'actant fondamental et les verbes symétriques et réfléchis de l'ancien français." – In: Lene Schøsler (ed.): La valence: Perspectives romanes et diachroniques, 34-42. Stuttgart: Steiner. Hopper, Paul J. and Sandra A. Thompson (1980): "Transitivity in grammar and discourse." – In: Language 56, 251-299. – and Elisabeth Traugott (1993): Grammaticalization. – Cambridge: Cambridge University Press. Hummel, Martin (2004): "Semantische Rollen bei reflexiven Verben." – In: Rolf Kailuweit and Martin Hummel (eds.): Semantische Rollen, 206-227. Tübingen: Narr. Irimia, Monica-Alexandrina (2005): "Types of secondary predication." – In: Toronto Working Papers in Linguistics 25, 20-29. Jacobs, Joachim (1994): Kontra Valenz. – Trier: Wissenschaftlicher Verlag. – (2003): "Die Problematik der Valenzebenen." – In: Vilmos Àgel, Ludwig M. Eichinger and Hans-Werner Eroms (eds.): Dependenz und Valenz. Ein internationales Handbuch der zeitgenössischen Forschung. Band 1, 378-399. Berlin & New York: Gruyter. Jauss-Meyer, Helga (1974): Lancelot. – München: Fink. Junker, Marie-Odile (1987): "Transitive, Intransitive and Reflexive Uses of DeAdjectival Verbs in French." – In: Jean Pierre Montreuil and David Birdsong (eds.): Advances in Romance Linguistics, 189-199. Dordrecht: Foris. Kailuweit, Rolf (2005): Linking: Syntax und Semantik französischer und italienischer Gefühlsverben. – Tübingen: Niemeyer. – (to appear): "Romance Anticausatives: A Constructionist RRG Approach." – In: Wataru Nakamura (ed.): Proceedings RRG 2009 Conference Berkeley. Kaiser, Georg (1992): Die klitischen Personalpronomina im Französischen und Portugiesischen: Eine synchronische und diachronische Analyse. – Frankfurt am Main: Vervuert. Kallulli, Dalina (2006): "A unified analysis of passives, anticausatives and reflexives." – In: Olivier Bonami and Patricia Cabredo Hofherr (eds.): Empirical Issues in Formal Syntax and Semantics 6, 201-225. Kayne, Richard S. (1975): French Syntax: The transformational cycle. – Cambridge, Mass.: MIT Press. Kazenin, Konstantin I. (2001): "Verbal reflexives and the middle voice." – In: Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.): Language Typology and Language Universals, 916-927. Berlin, New York: Mouton de Gruyter.

201 Kemmer, Suzanne (1993): The Middle Voice. – Amsterdam: Benjamins. Kennedy, Christopher and Beth Levin (2008): "Measure of Change: The Adjectival Core of Degree Achievements." – In: Louise McNally and Christopher Kennedy (eds.): Adjectives and Adverbs: Syntax, Semantics and Discourse, 156-182. Oxford: Oxford University Press. Kiesler, Reinhard (2006): Einführung in die Problematik des Vulgärlateins. – Tübingen: Niemeyer. Klaiman, Miriam Holly (1992): "Middle verbs, reflexive middle constructions, and middle voice." – In: Studies in Language 16 1, 35-61. Klüppelholz, Heinz (1987): Guillaume d'Angleterre. – München: Fink. Koch, Peter (1991): "Semantische Valenz, Polysemie und Bedeutungswandel bei romanischen Verben." – In: Thomas Krefeld and Peter Koch (eds.): Connexiones Romanicae. Dependenz und Valenz in romanischen Sprachen, 279-306. Tübingen: Niemeyer. – (2004): "Rollensemantik - diachronische Aspekte." – In: Rolf Kailuweit and Martin Hummel (eds.): Semantische Rollen, 421-434. Tübingen: Narr. Koontz-Garboden, Andrew (2007): States, changes of state, and the Monotonicity Hypothesis. - Doctoral dissertation: Stanford University. – (2009): "Anticausativization." – In: Natural Language and Linguistic Theory 27, 77138. Korhonen, Jarmo (2006): "Valenzwandel am Beispiel des Deutschen." – In: Vilmos Ágel, Ludwig M. Eichinger and Hans-Werner Eroms (eds.): Dependenz und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung. Band 2, 1462-1474. Berlin: de Gruyter. Kriegel, Sibylle (1999): "Les constructions pronominales dans deux textes du latin tardif." – In: Cahier du C.I.E.L.1996-1997, 135-156. Kulikov, Leonid (1999): "May he prosper in offspring and wealth. A few jubilee remarks on the typology of labile verbs and Sanskrit pús̙yati 'prosers; makes prosper'." – In: Ekaterina V. Rakhilina and Yakov G. Testelets (eds.): Typology and Linguistic Theory. From Description to Explanation. For the 60th Birthday of Aleksandr E. Kibrik, 224-244. Moscow: Jazyki russkoj kul´tury. – (2003): "The labile syntactic type in a diachronic perspective: the case of Vedic." – In: SKY Journal of Linguistics 16, 93-112. Kunstmann, Pierre and Achim Stein (2007): "Le Nouveau Corpus d'Amsterdam." – In: Pierre Kunstmann and Achim Stein (eds.): Le Nouveau Corpus d'Amsterdam: Actes de l'atelier de Lauterbad, 23-26 février 2006, 9-28. Stuttgart: Franz Steiner. Kupferman, Lucien (2008): "Les deux anticausatifs du français: invariants lexicaux, configurations syntaxiques et inférences sémantiques." – In: Danblon, Emmanuelle, Mikhail Kissine, Fabienne Martin, Christine Michaux and Svetlana Vogeleer (eds.): Linguista sum: Mélanges offerts à Marc Dominicy, 235-250. Paris: Harmattan. Labelle, Marie (1992): "Change of state and valency." – In: Journal of Linguistics 28, 375-414. – (2008): "The French reflexive and reciprocal se." – In: Natural Language and Linguistic Theory 26, 833-876. Labov, William (1972): Sociolinguistic Patterns. – Philadelphia: University of Philadelphia Press. – (1994): Principles of linguistic change: Internal factors. – Oxford UK, Cambridge USA: Blackwell.

202 Lagae, Véronique (1990): "Les caractéristiques aspectuelles de la construction réflexive ergative." – In: Travaux de Linguistique 20, 23-42. – (2005): "Les formes en être + participe passé à valeur résultative dans le système verbal français." – In: Arie Molendijk and Co Vet (eds.): Temporalité et attitude. Structuration du discours et expression de la modalité, 125-142. Amsterdam: Rodopi. Lagane, René (1967): "Les verbes symétriques: économie morphosyntaxique et différenciation sémantique." – In: Cahiers de lexicologie 10 1, 21-30. Lakoff, Robin (1971): "Passive resistance." – In: Chicago Linguistic Society 7, 149163. Larjavaara, Meri (2000): Présence ou absence de l'objet: limites du possible en français contemporain. – Doctoral dissertation: University of Helsinki. Lavidas, Nikolaos (forthcoming): "The diachrony of labile verbs: evidence from the history of English and Greek." – In: Seppo Kittilä and Leonid Kulikov (eds.): Diachronic typology of voice and valency-changing categories, Amsterdam: Benjamins. Leeman, Danielle (1991): "Hurler de rage, rayonner de bonheur: Remarques sur une construction en de." – In: Langue française 91 1, 80-101. Legendre, Géraldine and Antonella Sorace (2003): "Auxiliaires et intransitivité en français et dans les langues romanes." – In: Danièle Godard (ed.): Les langues romanes: Problèmes de la phrase simple, 185-234. Paris: CNRS Éditions. – (1997): "Secondary Predication and Functional Projections in French." – In: Natural Language and Linguistic Theory 15 1, 1-45. – and Antonella Sorrace (2005): Split intransitivity in French: an optimality-theoretic perspective, Ms. available at http://web.jhu.edu/cogsci/people/faculty/Legendre/ [access: 2005]. Lekakou, Marika (2005): In the Middle, Somewhat elevated. The semantics of middles and its crosslinguistic realization. – Doctoral dissertation: University of London. Leopold, Edda (2005): "Das Piotrowski-Gesetz." – In: Reinhard Köhler, Gabriel Altmann and Rajmund G. Piotrowski (eds.): Quantitative Linguistik: Ein internationales Handbuch, 627-633. Berlin: de Gruyter. Lerch, Eugen (1939): "Gerhard Babin, Das Medium im Altfranzösischen." – In: Zeitschrift für Romanische Philologie LIX, 341-351. Levin, Beth (1993): English verb classes and alternations: a preliminary investigation. – Chicago: University of Chicago Press. – and Malka Rappaport Hovav (1995): Unaccusativity: At the Syntax-Lexical Semantics Interface. – Cambridge, Massachusetts: MIT Press. – and Malka Rappaport Hovav (2005): Argument realisation. – Cambridge: Cambridge University Press. Löbner, Sebastian (2002): Understanding semantics. – London: Arnold. Lyons, Christopher (1982): "Pronominal Voice in French." – In: Nigel Vincent and Martin Harris (eds.): Studies in the Romance verb, 161-204. London: Croom Helm. Mackenzie, Ian E. (2006): Unaccusative verbs in Romance languages. – Basingstoke: Palgrave Macmillan. Manente, Mara (2008): L'aspect, les auxiliaires 'être' et 'avoir' et l'hypothèse inaccusative dans une perspective comparative français/italien. – Doctoral Dissertation: Università Ca' Foscari Di Venezia & Université de Paris 8. McKoon, Gail and Talke Macfarland (2000): "Externally and internally caused change of state verbs." – In: Language 76 4, 833-858.

203 Mel'þuk, Igor A. (1993): "The inflectional category of voice: towards a more rigorous definition." – In: Bernard Comrie and Maria Polinsky (eds.): Causatives and transitivity, 1-46. Amsterdam: Benjamins. Melis, Ludo (1990): La voie pronominale. – Paris: Duculot. Michaelis, Susanne (1998): "Antikausativ als Brücke zum Passiv: fieri, venire und se im Vulgärlateinischen und Altitalienischen." – In: Wolfgang Dahmen, Günter Holtus, Johannes Kramer, Michael Metzeltin, Wolfgang Schweickard and Otto Winkelmann (eds.): Neuere Beschreibungsmethoden der romanischen Syntax, 6998. Tübingen: Narr. Mutz, Katrin (2005): "Reflexiva und Verwandtes im Französischen und französischbasierten Kreolsprachen. Ein Vergleich." – In: Carsten Sinner and Georgia Veldre (eds.): Diathesen im Französischen / Les diathèses en français, 115-136. Frankfurt: Peter Lang. Napoli, Donna Jo (1992): "Secondary Resultative Predicates in Italian." – In: Journal of Linguistics 28 1, 53-90. NCA = Stein, Achim, Pierre Kunstmann and Martin-D Gleßgen (2006): Nouveau Corpus d'Amsterdam. Corpus informatique de textes littéraires d'ancien français (ca. 1150-1350), établi par Anthonij Dees (Amsterdam 1987), remanié par Achim Stein, Pierre Kunstmann et Martin-D. Gleßgen. – Stuttgart: Institut für Linguistik/Romanistik. Nedyalkov, Vladimir P. and G. G. Silnitsky (1973): "The typology of morphological causatives." – In: Ferenc Kiefer (ed.): Trends in Soviet Theoretical Linguistics, 1-32. Dordrecht: D. Reidel Publishing. Nichols, Johanna, David A. Peterson and Jonathan Barnes (2004): "Transitivizing and detransitivizing languages." – In: Linguistic Typology 8, 149-211. Nishiyama, Atsuko and Jean-Pierre Koenig (2004): "What is a Perfect State?" – In: Benjamin Schmeiser, Vineeta Chand, Ann Kelleher and Angelo J. Rodriguez (eds.): WCCFL 23 Proceedings, 101-113. Somerville: Cascadilla Press. Nyrop, K. (1930): Grammaire historique de la langue française, tome 6, Syntaxe: particules et verbes. – Copenhague: Det Nordiske Forlag. Oesterreicher, Wulf (1992): "SE im Spanischen: Pseudoreflexivität, Diathese und Prototypikalität von semantischen Rollen." – In: Romanistisches Jahrbuch 43, 237260. Olef-Krafft, Felicitas (2002): Roman de Thèbes. – München: Fink. Osgood, Charles and Thomas Sebeok (1954-1965): Psycholinguistics: A Survey of Theory and Research. – Bloomington: Indiana University Press. Petit Robert = Rey-Debove, Josette and Alain Rey (1993): Le Nouveau Petit Robert. – Paris: Dictionnaires Le Robert. Phillips, Betty S. (2006): Word frequency and lexical diffusion. – Basingstoke: Palgrave Macmillan. Pinker, Steven (1989): Learnability and cognition: the acquisition of argument structure. – Cambridge, Mass.: MIT Press. Piñón, Christopher (2001): A finer look at the causative-inchoative alternation, Ms. available at http://www.phil-fak.uni-duesseldorf.de/_pinon/papers [access: 2006]. Pollock, Jean-Yves (1985): "On Case and the syntax of infinitives in French." – In: Jacqueline Guéron, Hans-Georg Obenauer and Jean-Yves Pollock (eds.): Grammatical representation, 293-326. Dordrecht: Foris.

204 Rappaport Hovav, Malka and Beth Levin (in press): "Reflections on Manner/Result Complementarity." – In: Edit Doron, Malka Rappaport Hovav and Ivy Sichel (eds.): Syntax, Lexical Semantics, and Event Structure, Oxford: Oxford University Press. Reinhart, Tanya (2002): "The Theta System: An Overview." – In: Theoretical Linguistics 28 3, 229-290. – and Tal Siloni (2005): "The Lexicon-Syntax Parameter: Reflexivization and other Arity Operations." – In: Linguistic Inquiry 36 3, 389-436. Riegel, Martin, Jean-Christophe Pellat and René Rioul (51999): Grammaire méthodique du français. – Paris: Presses Universitaires de France. Robert Historique = Rey, Alain (1993): Dictionnaire historique de la langue française. – Paris: Dictionnaires Le Robert. Rothemberg, Mira (1974): Les verbes à la fois transitifs et intransitifs en français moderne. – The Hague, Berlin: Mouton. Ruhe, Ernstpeter (1977): Le voyage de Saint Brendan. – München: Fink. Ruwet, Nicolas (1972): Théorie syntaxique et syntaxe du français. – Paris: Ed. du Seuil. Schaden, Gerhard (2007): La sémantique du Parfait: Étude des "temps composés" dans un choix de langues germaniques et romanes. – Doctoral dissertation: Université de Paris 8. Schäfer, Florian (2008): The syntax of (anti-)causatives: External arguments in changeof-state contexts. – Amsterdam: Benjamins. – (2009): "The Causative Alternation." – In: Language and Linguistics Compass 3 2, 641-681. Schøsler, Lene (1984): La déclinaison bicasuelle de l'ancien français son rôle dans la syntaxe de la phrase, les causes de sa disparation. – Odense: Odense Univ. Press. Schuchardt, Hugo (1885): Ueber die Lautgesetze. Gegen die Junggrammatiker. – Berlin: Oppenheim. [available at http://schuchardt.uni-graz.at, access: 2010] Seguin, Jean-Pierre (1972): La langue française au 18e siècle. – Paris: Bordas. Selig, Maria (1998): "Pseudoreflexivität im Altitalienischen: Voraussetzungen und Richtungen eines Grammatikalisierungsprozesses." – In: Hans Geisler and Daniel Jacob (eds.): Transitivität und Diathese in romanischen Sprachen, 21-42. Tübingen: Narr. Shibatani, Masayoshi (1985): "Passives and related constructions: a prototype analysis." – In: Language 61, 821-848. Siewierska, Anna (1984): The passive: A comparative linguistic analysis. – London: Croom Helm. Smith, Carlota S. (1970): "Jespersen's 'Move and Change' Class and Causative Verbs in English." – In: Mohammad A. Jazayery, Edgar C. Polomé and Werner Winter (eds.): Linguistic and Literary Studies in Honor of Archibald A. Hill Vol. 2: Descriptive Linguistics, 101-109. The Hague: Mouton de Gruyter. – (1991): The Parameter of Aspect. – Dordrecht: Kluwer. Spillebout, Gabriel (1985): Grammaire de la langue francaise du XVII siécle. – Paris: Picard. Stéfanini, Jean (1962): La voix pronominale en ancien et en moyen français. – Aix-enProvence: Ophrys. Steinsieck, Wolf (1999): Das altfranzösische Rolandslied. – Stuttgart: Reclam. Tesnière, Lucien (21965): Éléments de syntaxe structurale. – Paris: Klincksieck. TLFi = ATILF (CNRS, Université Nancy 2): Trésor de la langue française informatisé. – Nancy. – www.atilf.fr.

205 TWIC = Stein, Achim: Tagged Words In Context (TWIC). – Stuttgart: Institut für Linguistik/Romanistik. Van Valin, Robert D. and Randy J. LaPolla (1997): Syntax: structure, meaning and function. – Cambridge: Cambridge University Press. Vendler, Zeno (1957): "Verbs and Times." – In: The Philosophical Review 66 2, 143160. Vennemann, Theo and Terence H. Wilbur (eds.) (1972): Schuchardt, the Neogrammarians, and the Transformational Theory of Phonological Change. – Frankfurt/M.: Atenäum. Vikner, Carl (1985): "L'aspect comme modificateur du mode d'action: à propos de la construction être + participe passé." – In: Langue française 67, 95-113. von Heusinger, Klaus (2008): "Verbal Semantics and the Diachronic Development of Differential Object Marking in Spanish." – In: Probus 20, 1-33. Wagner, Robert Léon and Jacqueline Pinchon (1989): Grammaire du français classique et moderne. – Paris: Hachette. Waltereit, Richard (1998): Metonymie und Grammatik: Kontiguitätsphänomene in der französischen Satzsemantik. – Tübingen: Niemeyer. Wanner, Dieter (1987): The development of Romance clitic pronouns from Latin to old Romance. – Berlin: Mouton de Gruyter. Wehr, Barbara (1995): SE-Diathesen im Italienischen. – Tübingen: Narr. Wehrli, Eric (1986): "On some properties of French clitic se." – In: Hagit Borer (ed.): The Syntax of Pronominal Clitics, 263-283. New York: Academic Press. Zribi-Hertz, Anne (1978): " Économisons-nous: A propos d'une classe de formes métonymiques en français." – In: Langue Française 39, 104-128. – (1987): "L'Ergativité réflexive en français moderne." – In: Le français moderne 55, 23-54.