Printer Friendly

The distribution of French intransitive predicates.


This article investigates the distribution of intransitive predicates in French. It argues that within each individual construction, the constraints that govern predicate selection are determined by the necessary semantic compatibility between the participating predicates and the construction itself. This "local" analysis based on the lexical semantics of specific verbs and constructions contrasts with most previous accounts that follow the Unaccusatire Hypothesis' claim that predicate distribution follows the specific structure of two distinct classes of intransitives. The local analysis is shown to be preferable to its structural counterpart to describe the behavior of intransitives in three constructions, namely: active impersonals, object raising in the context of the faire 'make' construction, and croire union. Because object raising and croire union are traditionally considered unaccusative diagnostics, the results obtained in this article raise serious questions about the validity of the Unaccusative Hypothesis for French, and suggests a possible alternative methodology to investigate the syntactic behavior of intransitive predicates.

1. Introduction

Most current accounts of the syntactic behavior of intransitive predicates are based on the Unaccusative Hypothesis (UH), namely the proposal that "certain intransitive clauses have an initial 2 [object] but no initial 1 [subject]" (Perlmutter 1978:160 insertions into brackets mine), and that consequently, intransitives can be separated into 'unaccusative' and 'unergative' predicates depending on their initial syntactic structure. (1) The high level of generalization UH affords both language internally and crosslinguistically goes a long way towards explaining its lasting popularity.

Within a single language, the initial structure of a predicate determines its behavior in several different constructions. For example, in English, the unaccusative status of 'break' accounts for its felicity in the resultative construction in (lb) and the causative alternation in (2). Similarly, the unergative structure of 'shout' explains its impossibility in the same constructions. The example in (lb) is from Levin and Rappaport Hovav (1995: 35):

(1) a. The door broke open

b. * Dora shouted hoarse

(2) a. The door broke

b. Paul broke the door

(Paul made the door break)

(3) a. The students shouted

b. * The teacher shouted the students

(The teacher made the students shout)

The evocation of a single characteristic to explain the behavior of specific predicates across constructions is highly desirable because it represents a way of grouping together constructions that might otherwise appear unrelated.

Crosslinguistically, initial unaccusativity versus unergativity is predicted by universal principles, and therefore does not vary from language to language. This property particularly attracted researchers, who immediately set out to validate this structural universal by documenting its presence in multiple languages. This endeavor yield prolific results, and the unaccusative/unergative split was described in Dutch, (Perlmutter 1978; Rosen 1984), English (Levin and Rappaport Hovav 1995), French (O1ie 1984; Ruwet 1989; Legendre 1989; Legendre and Sorace 2003), German (Kaufmann 1995), Italian (Burzio 1986; Perlmutter 1989), Japanese (Kishimoto 1996) to name just a few.

This article argues that in the case of French, the claims about the relevance of UH to the distribution of intransitive predicates are largely overstated, and that the predicates that participate in three specific constructions, namely active impersonals, object raising within the faire 'make' construction, and croire 'believe' union are selected solely on the basis of their semantics rather than their argument structure (Rosen 1984), or a combination of their semantics and syntax (Levin and Rappaport Hovav 1995). (2) The evidence in favor of this position comes from corpus data. When the constructions are considered in their natural context, they emerge as meaningful structures that code highly specific situations. The predicates that felicitously occur with them are therefore selected because their own meaning is compatible with that of the constructions themselves. Importantly, because object raising and croire union are traditionally considered "unaccusative diagnostics" (Levin and Rappaport Hovav 1995: 3), that is structural contexts where the predicates' behavior reflects their membership in the unaccusative or unergative class, the results obtained in this article raise important questions concerning the range of application of the Unaccusative Hypothesis in French.

This article suggests a "local" approach to the problem of the distribution of French intransitives because it focuses on the lexical semantics of individual predicates and constructions in order to evaluate their compatibility. The examination of these match ups reveals tightly delineated semantic domains where verbs and constructions elegantly combine to express highly specialized meanings in restricted contexts. The analysis is structured in the following fashion. Section 2 introduces the basic tenets of the Cognitive Grammar (CG) framework (Langacker 1987, 1991) within which the data are analyzed. Section 3 investigates the distribution of intransitive predicates with active impersonal constructions. Section 4 proposes an in-depth semantic analysis of object raising and croire union. Section 5 evaluates the status of the Unaccusative Hypothesis in French in the light of the findings. Section 6 concludes the article.

2. Theoretical assumptions

CG's main concern is to describe the semiological function of language, that is the association of meaning and form into symbolic elements. All linguistic expressions are symbolic units that contain a phonological and a semantic pole. In that view, the grammar of a language is entirely describable as a structured inventory of symbolic units that cannot be clearly separated between the lexicon, morphology, and syntax. Because of their symbolic value, all units are meaningful, and the description of their meaning constitutes an essential aspect of linguistic investigation. In CG, meaning is equated with conceptualization, and thus anthropomorphic and subjective. It critically includes, beside the objective properties of the described object, the way in which the conceptualizer chooses to present it. A linguistic expression is characterized by the specific profile its presence imposes on a conceptual base, or in other words how it structures that base. Two competing expressions provide alternative construals of the same conceptual base by structuring it in different ways.

Although all linguistic expressions are similar in their symbolic nature, they obviously differ in their level of complexity and abstraction. In particular, grammatical structures such as causatives and impersonals can also be described as complex symbolic units. These constructions take the form of templatic schemas that generalize over overtly occurring expressions and sanction their felicity. Because these schemas are symbolic, they are meaningful. Their meaning, however, is more abstract than that of the attested expressions which instantiate them. An expression is felicitous when the meaning of the component parts matches up with that of the construction itself.

Cognitive Grammar subscribes to a Usage-Based conception of language (Langacker 1988, 2000; Barlow and Kemmer 2000; Bybee 2001). As such, the model is "maximalist", "nonreductive" and "bottom up" (Langacker 1988). Maximalist refers to the position that since the cognitive representation of language is massively redundant, the corresponding grammar that describes linguistic production should also incorporate all its facets, regardless of size considerations. The model's nonreductive aspect states that the specification of rules should not prevent the explicit representation of their specific instances. A generalization (the rule) and its instantiation (the attested expression) constitute different facets linguistic knowledge that speakers can access at different times for different purposes. Consequently, any attested expression should be represented in a CG grammar, regardless whether it can be incorporated into a more general statement (a rule). Finally, the top down directionality of Cognitive Grammar indicates that the most general statements, i.e., the rules can only arise as schematization of overtly occurring expressions.

The direct relation that exists between the meaning of an expression and its usage represents the most crucial characteristic of CG's usage-based organization for the purposes of this article. Because the schemas that constitute grammatical constructions "spring from the soil of actual usage" (Langacker 2000: 3), no semantic characterization of those constructions can abstain from careful consideration of their context of use. Sections 3 and 4 show that the consideration of large corpora yields surprising semantic consistency among the attested examples of active impersonal, object raising, and croire union constructions, and hence solid evidence for the characterization of these constructions' meaning.

3. The distribution of intransitive predicates with active impersonals

The majority of researchers basically agree on the type of verbs that predominantly occur with active impersonals (AI), but often analyze this distribution in radically divergent manners. This section argues that the binary nature of the unaccusative/unergative distinction, on which most analyses in the literature rely, has contributed to this state of affairs by forcing researchers to adopt more rigid stances than the flexibility of usage warrants. By contrast, a local analysis that evaluates the semantic fit between the predicate and the construction is particularly well suited to express this flexibility.

Unaccusative predicates are unanimously recognized as felicitous in the construction. Examples such as the ones in (4) illustrate this position:

(4) a. Il est sorti trois personnes de la banque (3)

'There exited three people from the bank'

b. Il est nO six enfants le soir du reveillon

'There was born six children on new years' eve'

c. Il est arrive la voiture du facteur

'There arrived the mailman's car'

At first sight, the potential presence of unergatives in the construction seems to elicit diametrically opposed responses. Some researchers (Herschensohn 1982, 1996; Postal 1984; Ruwet 1989; Labelle 1992) view the possible participation of a predicate in the construction as a test of its unaccusative structure.4 Others (Kayne 1979; Grimshaw 1980; Legendre 1989) claim that unergatives are "fully productive" (Legendre 1989: 155) with active impersonals. Legendre illustrates her statement with the examples in (5), taken from Legendre (1989: 155):

(5) a. Il lui telephonait de nombreuses personnes a cette Opoque

'There used to call him many people in those days'

b. Il a cede beaucoup de candidats gt ce genre de pression

'There gave in many candidates to this kind of pressure'

c. Il a eternue beaucoup d'enfants pendant le concert

'There sneezed many children during the concert'

d. Il a violemment reagi beaucoup de personnes a l'annonce de cette nomination

'There violently reacted many people to the announcement of this nomination'

e. Savez-vous qu'il mendie beaucoup de personnes dans les rues de la capitale?

'Do you know that there beg many people in the streets of the capital?'

f. Il a cogne plusieurs imbkciles a la porte

'There knocked many jerks at the door'

g. Il a souri beaucoup de personnes a l'annonce de cette nouvelle

'There smiled a lot of people at the announcement of this news'

Upon closer scrutiny, however, these seemingly irreconcilable positions are considerably tempered by the caveats their proponents introduce in their analyses. The advocates of AI as an unaccusative test realize that it sometimes tolerates unergatives. Their presence however is too heavily constrained to seriously challenge the construction's diagnostic status: "While unergatives may be found in impersonal constructions, the availability of this construction for unergatives is much more restricted than it is with unaccusatives" Labelle (1992: 381). Furthermore, the unergative predicates that appear in AI undergo a semantic shift which bleaches them of all semantic content beyond that of "appearance in the world of discourse" (Gurron 1980: 653-654, see also Herschensohn 1982:211 and Labelle 1992: 381).

Similarly, researchers who claim that unergatives are felicitous in AI recognize that they are subject to specific constraints that do not apply to unaccusatives. For example, Legendre (1989) acknowledges that if unaccusatives freely occur with definite subjects, as the example in (4c) shows, unergatives are only felicitous with indefinite subjects. This indefinite subject constraint is illustrated by the contrast between the examples in (6) and (7):

(6) a * II mange Paul dans ce restaurant

'There eats Paul in this restaurant'

b. * Il travaille Marie dans cette usine

'There works Mary in this factory'

c. "Il dormait Fido dans un coin de la piece

'There slept Fido in the corner of the room'

(7) a. II mange beaucoup de linguistes dans ce restaurant

'There eat many linguists in this restaurant' (Pollock 1978)

b. Il travaille des milliers d'ouvriers dans cette usine 'There work thousand of workers in this factory' (Postal 1982)

c. Il dormait un chat dans un coin de la piece

'There slept a cat in the corner of the room' (Martin 1970)

Moreover, her claim of the full productivity of unergatives in AI is restricted to specific dialects: "There are dialects of French in which EXI is totally productive with unergatives, though the existence of such dialects has never been documented in the literature" (Legendre 1990: 82).5

The way in which researchers from both camps hedge their analyses reveals the basic agreement that obtains about the facts of the construction. One could therefore wonder why such divergent positions might arise in the first place. I believe that the diametrically opposed statements concerning IC's status as an unaccusative diagnostic result in great part from the polarizing nature of the unaccusative/unergative split. The invocation of the structure of the predicates to explain their distribution has forced analysts into too rigid an assessment of the presence of unergatives in AI, because these verbs do not comfortably fit in either option UH provides. Full productivity is obviously too strong, and needs to be considerably nuanced. Infelicity is also too strong, and needs to be qualified by considerations of frequency and semantic reanalysis. In a nutshell, the structural distinction between unaccusative and unergative predicates is simply too coarse to capture the flexibility of usage representative of the active impersonal construction.

The remainder of this section shows that an analysis based on the semantic compatibility between the participating predicates and AI is better suited to explain their distribution. First, the examination of the frequency of the predicates reveals that a very limited number of verbs accounts for an overwhelming majority of the attested examples. The specific lexical semantic structure of these verbs is shown to make them most directly compatible with the meaning of AI. Secondly, among the other predicates that occur in the construction, some have been argued to be unaccusative and others unergative. Despite their different argument structure, however, these predicates can easily be grouped together into consistent semantic classes. Their presence in AI is therefore argued to reflect their meaning rather than their structure. The analysis presented in this section is largely compatible with the one proposed in Cummins (2000).

3.1. Distribution of intransitive predicates in AI

The analysis presented in this section is based on a corpus of journalistic prose and literary texts. The newspaper data are taken from a series of articles from the news agency Agence France Press (AFP) that totals 70,013,327 words. The literary corpus consists of 25 twentieth-century works selected from the internet-based database Project for American and French Research on the Treasury of the French Language (ARTFL).6 A total of 610 tokens of active impersonal constructions were analyzed, 307 from AFP, and 303 from the literary sources. The 307 journalistic tokens instantiate 16 different verbs, while 39 verbs are represented in the 303 literary examples. Despite the apparent disparity in number, the data from the two corpora are remarkably similar, because all but 2 of the predicates found in journalistic texts are also attested in the literary corpus, and the two remaining ones have very close synonyms among the attested verbs. Furthermore, the relative frequency of the most commonly found predicates is very similar in both corpora. The difference in total number of verb types might be attributed to the wider range of narrative situations covered in the literary texts.

Two aspects of the distribution of intransitive predicates immediately come into view. The first one is the extreme disparity in the relative frequency of the participating predicates. Out of the 41 attested verbs, exister 'exist', and rester 'stay' account for 363 tokens, that is over 50% of all instances. Furthermore, only nine predicates, namely exister 'exist', rester 'stay', manquer 'lack', arriver 'arrive', venir 'come', etre 'be', resulter 'result', passer 'pass', and couter 'cost' produce 10 or more tokens. On the other hand, fourteen verbs contribute between 2 and 6 tokens each, and eighteen only provide 1 token. Secondly, the 41 verbs of the corpus can easily be grouped into the five remarkably well defined semantic classes illustrated in (8)-(12). The number that follows each class indicates the number of tokens for the whole class, the one that follows each verb represents the number of tokens for that verb. Following each class, an example is provided as an illustration. (7)

The overwhelming majority of the attested predicates assess the presence or absence of the elements that compose the conceptualized scene. The verbs of this class are given in (8):

(8) Elements that compose a scene (452): exister 'exist' (195), etre 'be'(16), rester 'stay' (168), manquer 'lack' (61), subsister 'subsist'

(6), suffire 'be enough' (5), demeurer 'remain' (1), durer 'last (1)

(8') ... Et puis, j'incarne un module de femme qu'on voit rarement au petit ecran. Il existe de nombreux modules masculins (Dechavanne, Ardisson, Rapp), mais peu de femmes ... '.... Plus the fact that I represent a type of women you don't often see on TV. There exist many such males (Dechavanne, Ardisson, Rapp), but few women

The second most frequent class includes predicates that denote the appearance/disappearance of a participant on a scene. The attested predicates are illustrated in (9):

(9) Appearance/disappearance on/from a scene (83): arriver 'arrive' (40), venir 'come' (21), naitre 'be born' (3), sortir 'go out' (5), surgir 'rush' (1), entrer 'come in' (4), revenir 'come back' (3), apparaitre 'appear' (2), survenir 'occur' (2), paraitre 'appear' (1), fuser 'gush out' (1)

(9') ... il me semble que tous mes livres ont eu pour point de depart le traversin ou je pose ma tete, mais dire comment je vois m'est impossible. En tout cas. il arrive toujours une seconde ou ces images d'une nettete parfois hallucinante se decomposent tout 5 coup et glissent les unes dans les autres.

'It seems to me that all my books started with the pillow where I rest my head, but it is impossible for me to express the way I see.

In any case, there always comes one second when these sometimes incredibly sharp images suddenly decompose and fade into one another'

The third (related) class includes predicates that denote a logical result or consequence. The complement of the predicate is presented as an expected occurrence that enters the conceptualized scene.

(10) Expected consequence (21): resulter 'result' (15), s'ensuivre 'follow' (5) (8), ressortir 'come out' (1)

(10') Je erois que lorsque cette unite se rompt, il en resulte des crises qui sont ni plus ni moins que des crises de folie.

'I believe that when this unity is broken, there results crises that are no less than crises of dementia.'

The fourth class describes the motion of a participant within a scene. It includes the predicates in (11):

(11) Motion within a scene (29): tomber 'fall' (6), monter 'go up' (4), passer 'pass' (10), couler 'flow' (1), courir 'run' (1), remuer 'shake' (1), descendre 'go down' (1), voler 'fly' (1), circuler 'circulate' (1), glisser 'glide' (1), roder 'roam' (1), pousser 'grow' (1) (9)

(11') Entends-tu ce que je te dis? Joseph hocha la tete.--tu attendras dans les bois que la nuit tombe, ce moment, tu descendras dans le ravin et tu rejoindras la route. Tu attendras encore, une heure s'il le faut. I1 passera une voiture qui ralentira et s' arretera a la hauteur du ravin pour te laisser monter.

'Can you hear what I am telling you? Joseph nodded. You will wait in the woods until nightfall. At that time, you will go down the ravine toward the road. You will wait again, one hour if you have to. There will pass a car that will slow down and stop by the ravine to let you in.'

Finally, a fifth and related class is composed of verbs that describe the general sensory impression the conceptualized scene evokes. The relevant predicates are given in (12):

(12) Sensory impression that permeates a scene (13): regner 'rein' (6), flotter 'float' (3), emaner 'emanate' (1), filtrer 'filter (1), cuire 'cook' (2) (10)

(12') Sur sa table de travail un poudrier d'argent etait ouvert, laissant voir une houppe ronde et blanche, pareille a un petit nuage. I! flottait entre ces tours une odeur affreusement douce et grisante qu'il s'efforga de ne pas respirer, une odeur de lilas.

'On his desk, a silver powder box was open, revealing a puff round and white like a little cloud. 'There floated between these walls a horribly sweet and enticing odor that he tried not to breathe in, an odor of lilac.'

This brief overview of the corpus suffices to show that the unaccusative/unergative distinction offers little insight into the distribution of intransitives in AI. First, the attested unergatives do not appear to be bleached of their semantic content, as illustrated in (13) and 04). (11)

(13) le ciel etait gris de nuages il y volait des oies sauvages qui criaient la mort au passage au-dessus des maisons des quais 'The sky was gray with clouds there [in it] flew wild geese screaming death as they flew above the houses on the pier.'

(14) Quelquefois, derriere la barre de la lagune, un aviron par intervalles tdtait l'eau gluante, ou tout pros s'etranglait le cri falot et obscOne d'un rat ou de quelque byte menue comme il en rode aux abords des charniers.

'Sometimes, behind the reef, an oar intermittently probed the sticky water, near where a rat or some other little creature such as there [of these creatures] roams near mass graves choked its timid and obscene scream.'

Furthermore, these predicates are perfectly well integrated semantically among the other participating verbs despite their different structure. (12) Finally, structural considerations fail to address the discrepancy in the frequency of the different classes. In particular, it does not explain why such a reduced number of unaccusative predicates constitute the overwhelming majority of the total instances, while other unaccusatives (most notably internal change predicates) occur so seldom. The remainder of this section proposes an analysis that accounts at the same time for centrality of the most frequent predicates, and the specific conditions under which others (including unergatives) can be felicitous in the construction. Consistent with the claim made at the beginning of this article, the predicates most frequently attested in AI are those whose meaning is most straightforwardly compatible with that of the construction. I will first consider the meaning of the impersonal construction before investigating the lexical semantics of the predicates that participate in it.

3.2. Meaning of impersonal constructions

Space considerations do not allow a thorough exploration of semantic structure of impersonals, but this section focuses on the aspects of their meaning most directly related to the concerns of this article.

Different Idealized Cognitive Models (Lakoff 1987; Langacker 1991) that pertain to our conception of events constitute the base relative to which the various elements that get integrated into the concept of clause structure can be described (Langacker 1991: 282). Le Control Cycle model (Langacker 2002, 2004) describes the interaction between participants within a certain setting (the field). The cycle consists of four phases, each representing a separate stage of the interaction. Langacker's explanatory example is that of a cat resting in the shade, whose attention is aroused by a passing mouse. The cat instantly views the mouse as a potential prey, and pounces on the smaller animal, thereby establishing his control over it. The initial static situation (i.e., the cat at rest) is the baseline. The cat is the agent, and the area around it within which it can interact with other participants is the field. Once the target (the mouse) appears in the field, it creates the potential for interaction, or in other words, a state of tension. The action phase follows, where the agent exerts force upon the target in order to bring it under control. The result phase is the stable situation that obtains once the target is in the agent's control. For the purposes of this article, it is important to stress the difference between the participants in the interaction (the cat and the mouse in our example), and the field, namely the setting in which they interact. The interaction proper strictly occurs between the participants. The field represents the area where the interaction occurs (or the potential for interaction exists). This model is relevant to the concerns of this article for two reasons. First, it affords a definition of the meaning of impersonal constructions, and secondly, it provides the base with respect to which the meaning differences between different classes of intransitive verbs will be characterized.

Different clause types profile specific conventionalized construals of events. Grammatical relations (in particular subject and object selection) are predominantly a matter of focal prominence. The primary figure in the relation profiled by the main verb is coded as the subject. The secondary figure in the main relation is coded as the object. The most common clause types select participants as focal figures. This is the case in transitive constructions. In 'the cat ate the mouse' for example, the cat is the focal figure in the relation profiled by 'eat', and the mouse is the secondary figure. Consequently, these participants are respectively coded as the subject and the object of the clause. Less conventional clauses, however, select the setting as the main figure in the profiled relation. For example, in 'under the table is all dirty', it is the setting (i.e., the field) within which the predication 'be dirty' is observed (the area under the table) that is taken as the main figure and thus chosen as the clausal subject. In the recent CG literature, impersonals have been analyzed as abstract setting constructions. The abstract setting is selected as the focal figure of the profiled relation (Smith 1985; Langacker 1991, 2002, 2004; Achard 1998) and thus coded as clausal subject. The abstract setting can be identified as the field within which certain interactions potentially occur (Langacker 2004). In the case of French impersonals, that field is profiled by the pronoun il 'it'. It can be equated with the subsection of reality necessary to the identification of the event or proposition coded in the complement (Achard 1998). (13)

The primary importance of the field constitutes the most important characteristic of impersonal constructions for the purposes of this article. The next section shows that the relevant factor in the distribution of intransitive predicates with AI is the level of salience of the field in the lexical semantic structure of the predicate (its scope of predication in CG terminology).

3.3. Semantic distinction between intransitives: field and participants

Predicates are commonly evaluated along several semantic dimensions. The number of participants in the main relation separates transitives from intransitives. The degree of volition of their subjects as well as their inherent aspectual properties are also routinely invoked to differentiate predicates in numerous constructions. I would like to suggest that verbs can also be distinguished with respect to the relative salience of their participant and the field within which their interaction takes place in their scope of predication. This section argues that the most felicitous predicates in AI are precisely those that most relevantly and naturally include the field as part of their scope of predication (to be made specific throughout this section). This semantic distinction allows us to explain i) the overwhelming frequency of the stative and (to a lesser extent) motion verb classes in the construction, and ii) the possible occurrence of a very large quantity of predicates (including unergatives) within the other classes.

The previous section already mentioned that transitive verbs predominantly focus on the interaction between the participants in the main relation. For example, in Marie a mange la tarte 'Mary ate the pie', the verb manger 'eat' exclusively profiles the interaction between Mary and the pie. Even though the eating episode necessarily occurs at a given time and place, these circumstances are not necessarily invoked when the verb is conceptualized. The field is therefore not part of the verb's scope of scope of predication.

Intransitives present more diversity. Different classes of predicates wildly differ with respect to the respective salience of the participant and the field in their lexical semantic structure. The verbs of mental activity such as rever 'dream', or reflechir 'think' constitute one extreme where the communicative focus is turned exclusively toward the participant. Marie a reflechi 'Mary thought (about it)' for example profiles a process internal to the participant Mary, while the field has little recognized salience. (14) At the other extreme, stative predicates such as etre 'be', exister 'exist', rester 'stay', manquer 'lack' assess the current state of a specific scene by identifying the elements that compose it. Importantly, that scene can be described as the field. The latter constitutes a crucial aspect of the scope of predication of those verbs because it provides the search domain within which the elements described in the complement can be identified. For example, in il reste de la biere dans lefrigo 'there is some beer left in the fridge', rester 'remain' describes the current state of the inside of the refrigerator, such that it includes the left over beer. The field is a necessary component of the verb's scope of predication because it is within its confines that the presence of the beer can be assessed.

In between these polar opposites, a large number of verbs profile different degrees of respective salience between the participant and the field. For instance, the verbs of physical activity and internal change predominantly focus on their participant, but nonetheless include the field in their scope of predication. Let us take courir 'run' and grandir 'grow' as exampies. In both cases, the descriptive focus is predominantly placed on the participant to the detriment of the field. In Marie a couru 'Mary ran', courir most prominently profiles Mary's activity. Similarly, in Marie a grandi 'Mary grew', grandir describes the changes internal to Mary herself. However, the processes these verbs profile are not exclusively internal to the participant, as was the case with the mental activity verbs. Consequently, their meaning involves some interaction with the field, even though the latter is not always prominently displayed. Finally, the semantic import of some verbs precisely consists in profiling their participant's interaction with the field. For example, in Marie est arrivee 'Mary arrived', the meaning of arriver 'arrive' profiles Mary's appearance onto the conceptualized scene. The field, namely the scene that gets modified by Mary's appearance thus constitutes a necessary part of the verb's scope of predication.

3.4. Compatibility between intransitives and AI

The distribution of intransitive predicates in AI is almost straightforwardly compatible with the hypothesis presented in Section 3.1. The most frequently attested predicates necessarily include the field in their scope of predication. This semantic feature makes them an ideal match for a construction where the setting is profiled as the focal figure in the main relation.

This compatibility is particularly obvious for the stative predicates illustrated in (8') and repeated here for convenience:

(8') ... Et puis, j'incarne un modele de femme qu'on voit rarement au petit ecran. li existe de nombreux modeles masculins (Dechavanne, Ardisson, Rapp), mais peu de femmes. . .

'... Plus the fact that I representa type of women you don't often see on TV. There exist many such males (Dechavanne, Ardisson, Rapp), but few women'

In (8') il profiles the setting where the event described in the complement can be located, that is the relevant subsection of French TV within which the presence of male show hosts can be ascertained. Exister profiles the presence of the hosts within this search domain.

The verbs of sensation presented in (12) are quite similar to the stative predicates because they describe the overall atmosphere of the conceptualized scene, or, in other words, the presence of specific sensations within that scene. The field also constitutes a crucial part of their scope of predication because it defines the search domain within which these sensations can be experienced. In (12') for example repeated here for convenience, flotter 'float' profiles the olfactory sensation that permeates the room.

(12') Sur sa table de travail un poudrier d'argent etait ouvert, laissant voir une houppe ronde et blanche, pareille a un petit nuage. I!

flottait entre ces murs une odeur affreusement douce et grisante qu'il s'efforca de ne pas respirer, une odeur de lilas.

'On his desk, a silver powder box was open, revealing a puff round and white like a little cloud. 'There floated between tbese walls a horribly sweet and enticing odor that he tried not to breathe in, an odor of lilac.'

The verbs of appearance/disappearance from a scene are equally straighfforward. Here again, the field constitutes a necessary part of these verbs' scope of predication because it marks the limit at which the conceptualized entity comes into or disappears from view. Let us turn to (15) for illustration:

(15) En 1988, une equipe d'enseignants et de bibliothecaires de la maison d'arret de Grenoble-Varces (Isere) organisait ainsi une rencontre entre le dessinateur Didier Savard et un groupe de detenus, Il en etait ne une bande dessinee en six planches, Vae Victis, publiee dans le mensuel A suivre.

'In 1988, a team of teachers and librarians from the Grenoble-Varces (Isere) detention center organized a meeting between the cartoonist Denis Savard anda group of inmates. From that meeting [there] was bom a six-page cartoon, Vae Victis, published in the monthly A Suivre.'

In the example in (15), il profiles the field, that is to say the part of the world that includes the relevant parts of the meeting that witnessed the creation of the comic strip. Naftre profiles the way in which that setting was modified by the emergence of a new entity. This kind of construction enhances the collaborative nature of the project by presenting the interaction that led to the birth of the comic strip as an emergent property of the meeting itself (i.e., the field), rather than as the sum of the individual actions of the attendees (the participants).

The semantic class presented in (11) under the label "motion within a scene" deserves a little more attention because it contains inherently directed motion predicates (monter 'go up', passer 'pass', descendre 'go down), as well as physical activity predicates (courir 'run', roder 'roam', voler 'fly'), and an intemal change predicate (pousser 'grow'). The inherently directed motion predicates are unproblematic because they have already been argued to profile their participants' interaction with the field. These verbs are therefore perfectly compatible with AI because the field constitutes a necessary part of their scope of predication. This is illustrated in (11') repeated here:

(11') Entends-tu ce que je te dis? Joseph hocha la tete.--tu attendras dans les bois que la nuit tombe, ce moment, tu descendras dans le ravin et tu rejoindras la route. Tu attendras encore, une heure s'il le faut. Il passera une voiture qui ralentira et s'arretera a la hauteur du ravin pour te laisser monter.

'Can you hear what I am telling you? Joseph nodded. You will wait in the woods until nightfall. At that time, you will go down the ravine toward the road. You will wait again, one hour if you have to. There will passa car that will slow down and stop by the ravine to let you in.'

In (11'), the road is clearly established as the field within which the arrival of the car is expected to take place. Passer 'pass' profiles the participant's trajectory through that field.

The physical activity and internal predicates on the other hand representa potential difficulty because they were previously argued to place their descriptive focus on their participant, and thus not to prominently include the field in their scope of predication. This difficulty, however, disappears once the specific discourse conditions that favor their presence in AI can be brought to light.

So far, the compatibility between the intransitive predicates and AI has been treated as a matter of strict lexical semantics. Specific verbs were considered more directly compatible with the construction because of their lexical structure. However, in CG, grammatical relations are first and foremost a matter of construal, and conditions particular to the context of the utterance can easily override any verb's lexical characteristics. I would like to suggest that this is indeed the case with the attested examples of physical activity and internal change predicates.

It was noted in several instances that these predicates place the descriptive focus on their participant. It is useful to remember, however, that in each instance, the participant's activity or change necessarily involves some interaction with the field, even though the latter is not prominently displayed. For example, the activity predicate courir 'run' contains as part of its scope of predication the same motion component as an inherently directed motion verb such as passer 'pass'. Both involve their subject's trajectory within the field, but they diverge in the descriptive focus they each place on different aspects of the conceptualized scene. Activity predicates are more concerned with the participant's role in the process, while inherently directed motion verbs more specifically concentrate on the trajectory through the field.

The inherent salience of the participant represents the main reason why activity and internal change predicates are not always felicitous with AI. These predicates are too strongly centered on their participant to match up well with a construction that selects the field as clausal subject. However, if due to particular circumstances in the context of the conceptualized situation, the participants' salience and individual character decreases to the point that their activity can be construed as a mere feature of the field, the latter can easily be involved as the main figure in the profiled relation. In these conditions, the participation of the verbs of activity and internal change in AI is perfectly natural. Consider the example in (14) repeated here:

(14) Quelquefois, derriere la barre de la lagune, un aviron par intervalles tatait l' eau gluante, ou tout pres s' etranglait le cri falot et obscene d' un rat ou de quelque bete menue comine li en rode aux abords des charniers.

'Sometimes, behind the reef, an oar intermittently probed the sticky water, near where a rat or some other little creature such as there [of these creatures] roams near mass graves choked its timid and obscene scream.'

Despite the fact that roder 'roam' is an activity predicate, its use in (15) makes it quite similar to the sensation predicates presented in (12). In a way similar to odors and sounds, motion can also permeate the conceptualized scene if it is construed as one if the features that importantly contributes to its overall atmosphere. In this sense, the predicates that profile motion within a scene can be viewed as the visual instances of the sensation predicates, because both profile a specific facet of the setting's ambiance. In (14), the roaming of the rats is presented not as the activity specific rodents are involved in, but as a distinctive feature that strongly contributes to the unique ambiance of mass graves.

This analysis receives strong support from the fact that specific grammatical contexts consistently favor the selection of the field as clausal subject with verbs that usually favor their participants. The most commonly recognized of these contexts is the presence of an indefinite subject. The indefinite subject constraint presented in Section 3 can be explained by the fact that the presence of an indefinite nominal decreases the inherent salience of the participant in the scene, and therefore favors a construal where the setting holds focal prominence. (15) Let us illustrate this point with the contrast in (6a) and (Ta) repeated here:

(6) a. * II mange Paul dans ce restaurant 'There eats Paul in this restaurant'

(7) a. II mange beaucoup de linguistes dans ce restaurant (Pollock 1978) 'There eat many linguists in this restaurant'

In (6a), Paul is an object of current immediate perception, and the salience of that participant is extremely high. In (Ta), the object of perception can only be formed over time, and by the consideration of many instances. Furthermore, it is not achieved by observation alone, but with the help of other cognitive abilities such as comparison and analysis. The frequency of observation required, as well as the analysis of those observations are instrumental in considering the presence of linguists as a statement about the restaurant, or in other words, a specific feature of the field itself. This constraint is motivated by the strong natural emphasis placed on participants and their actions, and therefore occurs on both activity (unergative) and internal change (unaccusative) predicates. (16)

The felicity of the verbs of activity and internal change in IC is also greatly enhanced if explicit emphasis is placed on the field in the context. Consider for example the pairs in (16) and (17) where the locational y denotes specific focus on the field.

(16) a. ?Il a cuit beaucoup de gigots dans ce four 'There cooked a lot of roasts in this oven'

b. Ce four est genial, il y cuit un gigot en 45 minutes 'This oven is great, there cooks in it a roast in 45 minutes'

(17) a. ?II a marine beaucoup d'olives dans le pot bleu 'There marinated a lot of olives in the blue pot'

b. Ne prends pas le pot bleu, il y marine des olives 'Do not take the blue jar, there marinate olives in it'

In (16b) and (17b), the oven and the jar are strongly topical, and they clearly constitute the field within which the cooking and marinating respectively occur. The presence of the locational pronoun y 'there' reinforces the setting function of the two containers. As a result, these examples are much more natural than the ones in (16a) and (17a). The contrasts in (16) and (17) lend further support to the current analysis where the compatibility of the predicates with AI depends on the salience of the field in their scope of predication.

We are now in a position to reconsider the semantic "bleaching" the activity and internal change predicates have been argued to undergo with impersonals. While the position that reduces their meaning to "appearance in the world of discourse" (Gueron 1980: 653-654; Herschensohn 1982:211; Labelle 1992: 381) seems exaggerated, the meaning of these verbs in AI indeed differs from the one they have in personal situations. I believe this semantic variance directly results from the construal shift characteristic of their presence in AI. When the activity and internal change predicates are used in their personal sense, they profile the interaction between their participants. Consider for example the personal counterpart of (14)in (18):

(18) Des rats rodent aux abords des charniers 'Rats roam near mass graves'

In (18), the physical energy involved in the roaming act is predominantly anchored in the rats themselves. (17) The speaker's main role is mostly to observe and report that display of activity that comes from a well delineated concrete source. Her conceptualization of the roaming act can thus be called maximally objective (Langacker 1985, 1990). The same verb in (14) evokes a very different construal. The roaming is not contained in the highly individualized source of the rats themselves. It is more diffuse and identified as a part of the scene's atmosphere. Because the construal of the scene, and hence the selection of the movement of the rats as one of its distinguishing characteristics represents a mental act by the speaker, the roaming is essentially anchored in her conceptualization. The diffusion of the physical energy from a well defined source contained in the main figure of the main relation to a primarily mental force associated with the construal relation itself, and thus located in the speaker, reflects its subjectification (Langacker 1985, 1990). The speaker's (mental) energy to conceive the roaming represents the subjective counterpart of the (physical) energy that emanates from the animals. The specific meaning of the verbs of activity and internal change in AI can therefore be attributed to the subjectification of the main relation which becomes more diffuse and associated with the speaker's construal of the conceptualized scene. (18)

As a brief summary, this section showed that a solution based on the necessary compatibility between the meaning of AI and the verbs that participate in it provides a convincing explanation of the distribution of intransitives in active impersonal constructions. This local account presents several advantages over a structural analysis. First, it allows us to group together the verbs of activity and internal change, when they are structurally different. Secondly, it is capable of accounting for the overwhelming frequency of some classes of predicates in the construction, as well as for the possible presence of a very large number of verbs under the appropriate conditions. Finally, it sheds some light on the long lasting debate that surrounds the grammaticality of some activity predicates in AI. For example, the possible felicity of Legendre's example presented in (5a) II lui telephonait de nombreuses personnes a cette epoque 'There called him/her many people at that time' depends on the hearer's ability to construct a context where the calling--an activity usually centered on the participants--could legitimately be construed as a distinguishing characteristic of the described time frame. These kinds of examples are therefore not problematic because of the syntactic structure of the predicate, but because of the unconventional match up between the main verb and the impersonal construction.

The next section provides further evidence of the viability of local accounts by considering the two constructions of object raising within the context of the faire construction, and croire union (Legendre 1989). The investigation of these constructions in their natural context reveals that the predicates that participate in them are selected for their lexical semantic properties rather than their structure. Because both object raising and croire union are traditionally considered unaccusative diagnostics (Legendre 1989), the results obtained seriously decrease the evidence in favor of the relevance of the Unaccusative Hypothesis for French.

4. Two unaccusative diagnostics

The basic insight behind the idea of an unaccusative diagnostic is that the unaccusative or unergative structure of the participating predicate is solely responsible for its behavior in the test construction. The legitimacy of the concept relies on the fact that no other factor can be responsible for the distribution of intransitives. Syntactic diagnostics are the most compelling when they group together predicates so semantically diverse that their behavior could not possibly be attributed to their meaning. On the other hand, they lose considerable explanatory force when that behavior can be explained in purely semantic terms. This section shows that it is precisely the case for object raising and croire union. An investigation of their context of use reveals that the two constructions cannot be considered broad structural patterns inclusive of a large array of semantically diverse predicates whose similar behavior can only be explained in syntactic terms. Rather, each construction emerges as a narrowly defined island with a highly specialized semantic function. The embedded predicates are therefore selected for their semantic import, not their structural properties. (19) The data for the investigation of the usage of the constructions comes from the journalistic corpus of 70,013,327 words presented in Section 3.

4.1. Object raising

Legendre (1989:98 insertions into brackets mine) claims that "Transitive 2's [objects] freely undergo object raising whether or not the structure contains the causative predicate faire 'make'." With respect to intransitive verbs, unaccusatives are argued to occur freely in the construction, while unergatives are deemed impossible. This is illustrated in (19) and (20), respectively (11k) and (12a) in Legendre (1989):

(19) Les prix seront faciles a faire augmenter 'The prices will be easy to make increase'

(20) * Le president est difficile a faire agir 'The president is difficult to make act'

Because it contains an initial 2, the unaccusative augmenter 'increase' is felicitous in (19). Agir 'act' on the other hand is unergative. Its subject is never a 2, which makes the sentence in (20) infelicitous.

4.1.1. A usage-based analysis. The relevante of the unaccusative/ unergative distinction to the distribution of intransitive predicates in object raising is difficult to maintain once the construction is investigated in its natural context. A mere glance at the corpus reveals two striking characteristics of the construction. The first one is its extreme scarcity. Only 25 examples were attested in the corpus, 18 involving transitive verbs, and 7 involving intransitive verbs. The second characteristic of the collected data concerns the great semantic consistency of the 25 instances of the construction. The examples in (21) and (22) present transitive verbs; the ones in (23) and (24) illustrate intransitive verbs.

(21) L'avocat David Bruck, qui defend Susan Smith, compte plaider un coup de folie.... Cette strategie de defense risque toutefois d'etre difficile a faire admettre au jury d'assises. 'David Bruck, the lawyer who represents Susan Smith, intends to use the insanity plea.... This defense strategy might however be difficult to make admit to the jury [for the jury to admit].'

(22) Le secretaire general de I'ONU, Boutros Boutros-Ghali, a estime lundi qu'un embargo sur les armes a l'encontre du Nigeria serait difficile a faire respecter, quelle que soit la volonte des Nations unies ...

'The general secretary of the United Nations, M. Boutros BoutrosGhali, estimated on Monday that an embargo on the arms destined to Nigeria would be difficult to make respect [enforce] regardless of the United Nation's position ...'

(23) M. Kozyrev a estime que la guerre froide avait pour origine l'incompatibilite entre le systeme communiste et la demoeratie, et que la nouvelle Russie "est un allie naturel de la democratie dans le monde." Mais il y a des "residus" de la guerre froide dans les bureaucraties des deux pays et "cela n'est pas si facile a falte disparaitre", a-t-il.

'M. Kozyrev voiced the opinion that the origin of the cold war was the incompatibility between the communist system and democracy, and that new Russia is "a natural ally of democracy in the world." However, there is a residue of cold war in the bureaucracies of both countries, and "it is not that easy to make disappear," he said.'

(24) Au dela de la motion de censure, l'opposition veut tenter d'atteindre Boris Eltsine.... "Le vote ne recueillera que 210-215 voix (contre une majorite necessaire de 226 voix), mais nous lancerons egalement mercredi une procedure de destitution du president Eltsine" a affirme le depute. Cette procedure est cependant extremement difficile a faire aboutir, selon la Constitution.

'Beyond the censorship motion to censor, the opposition wants to try to reach Boris Eltsine.... The motion will only receive 210215 positive votes (against a required majority of 226), but on Wednesday "we will also start a procedure to impeach President Eltsine" the representative added. However, according to the constitution, this procedure is extremely hard to make succeed [see to its completion].'

The semantic consistency of the 25 examples is clearly visible in the nature of both raised nominals and embedded predicates. Raised nominais describe administrative objects that pertain to our political, social, or judicial organization. A representative list includes rapplication de cette mesure 'the application of this measure', cette position inflexible 'this inflexible position', cerre proposition 'this proposal', un embargo 'an embargo', un message 'a message', cette strategie de defense 'this defense strategy'. The 6 transitive predicates include respecter 'respect', admettre 'admit', comprendre 'understand' passer 'pass', adopter 'adopt', appliquer 'apply'. The 2 intransitive predicates are disparaftre 'disappear' and aboutir 'come through'.

The semantic consistency of the component elements of object raising constitutes a strong indication that the latter describes a highly specific kind of event. Consistent with the methodology presented in the preceding section for AI, the specific construal of that event represents the construction's semantic import. I suggest that the meaning of the object raising construction can be described in the following fashion: In the context of the faire construction, object raising codes the evaluation of the induced manipulation of a social (administrative) object; often the degree of difficulty of the induction. Given the construction's meaningful nature, it seems reasonable to suggest that the constraints imposed on the participating predicates are imputable to their lexical semantics, not their argument structure. Object raising straightforwardly favors the predicates that are compatible with its own semantic import regardless of their argument structure. The semantic fit between the embedded predicates and the construction is clear. Some administrative object is proposed to a given community for them to "manipulate", that is, to recognize, accept, reject, or utilize. The embedded predicate indicates the kind of manipulation the community is expected to perform. The 6 transitive verbs presented above only present two kinds of administrative manipulations. Respecter 'respect', admettre 'admit', and comprendre 'understand' pertain to the integration of the new administrative object into someone's system (or everyday life). Passer 'pass', adopter 'adopt', and appliquer 'apply' are more procedural because they involve the formal passage through an official mechanism. The intransitive verbs exhibit similar semantic consistency. Disparaitre 'disappear' describes the elimination of an administrative object, aboutir 'come through' simply indicates the natural path a procedure takes when it is successful.

The recognition that object raising is meaningful does not per se invalidate its function as an unaccusative diagnostic. The two intransitive predicates attested in the construction are indeed unaccusative, and the parallel behavior of transitive and unaccusative verbs seems to argue in favor of a hypothesis where the two classes are structurally related. However, the narrow semantic range object raising exhibits indicates that UH might not provide the relevant distinction to account for the distribution of predicates in the construction. A predicate's argument structure constitutes a very general structural property, which we would expect to cut across a variety of semantic environments. In the restricted semantic domain of the object raising construction, the most pertinent question to ask is not why transitive and unaccusative verbs occur together to the exclusion of unergative predicates, but why the distribution is restricted to such a small number of semantically related verbs. The answer clearly invokes the semantics of the attested predicates rather than their argument structure.

It is interesting to note that in the context of object raising, the contrast between transitives and intransitives is neutralized. This fact does not per se constitute a direct argument in favor of a semantic analysis, because structural accounts are precisely based on the similar behavior of transitive objects and intransitive subjects. However, in the case of object raising, it combines with an extremely narrow semantic range to paint the picture of a construction that places major emphasis on the semantics of the participating predicates and considers their argument structure largely irrelevant.

The main difference between the two classes of predicates concerns the presence of an additional participant (the causee) with transitive verbs. However, the object raising construction places a strong emphasis on the raised object to the detriment of the causee. Achard (2000) argues that the raised variant of a raising construction is predominantly chosen when the nominal to be raised is highly topical. This construction is no different. The raised object not only represents what the utterance is primarily about, but its intrinsic properties constitute the main factor that determines the outcome of the manipulation. On the other hand, the causee receives little emphasis. Most of the times, it is not even mentioned. In fact, the causee is only mentioned four times in the corpus. The first one has already been introduced in (21). The phrase au jury d'assises 'to the court' specifies the entity in charge of accepting the lawyer's strategy. Another example is given in (25) where the causee is underlined for convenience sake:

(25) L'administration Clinton est en faveur du renouvellement a la Chine de la clause de la nation la plus favorisee, mais raffaire sera difficile a faire adopter par le congres a majorite republicaine, a indique mardi a Hong Kong le secretaire americain au Tresor Robert Rubin. (20)

'The Clinton administration is in favor of the renewal of China's status of favored nation, but the issue will be difficult to make adopt by the Republican Congress the American Secretary of Treasury Robert Rubin said in Hong Kong on Tuesday.'

The specific mention of the causee is rendered necessary by its contrastive nature which directly bears on the outcome of the induced manipulation. We might expect a democratic Congress to be more sympathetic to a democratic president, and thus more likely to yield to his desires. Outside such cases, however, the lack of mention of the causee seriously minimizes the difference between transitive and intransitive verbs, and casts serious doubt about the relevance of the argument structure of the embedded predicate for the object raising construction.

A consequence of that irrelevance is that in the cases where a given predicate has a transitive and intransitive sense, it is often almost impossible to tell them apart. This is the case in the corpus with passer 'pass'. The verb has the transitive sense of passer une loi 'passa law' for example, but it also has another unaccusative (perhaps more colloquial) sense of something being hard to accept, as for example in le gateau est trop lourd, il ne passe pas 'the cake is too heavy, it won't pass'. These two senses are virtually undistinguishable in the context of the object raising construction, as illustrated in (26) and (27):

(26) Clause de la nation la plus favorisee: le renouvellement sera difficile a faire passer, selon M. Rubin.

'Status of most favored nation: The renewal will be difficult to make pass, according to M. Rubin.'

(27) M. Preval a precise que les deux premieres entreprises privatisees seraient Le Ciment et la Minoterie ... "Nous avons la conviction que le prive est meilleur gestionnaire que l'Etat," a-t-il ajoute, soulignant cependant que la privatisation etait "un concept difficile a faire passer en Hafti."

'M. Preval specified that the first two companies to be privatized would be Cement and Mills ... "We are convinced that the private sector is a better manager than the government," be added, while noting, however, that privatization was "a concept difficult to make pass in Haiti.'"

Outside of context, it is hard to tell if passer in (27) is transitive or intransitive. It is the context of other articles [see the text in (26) about the same topic] that allows us to interpret it as the transitive sense of the verb with the causee (most likely the United States Congress) left unspecified. However, it would also be perfectly plausible to interpret it as the intransitive passer with a meaning close to accepter 'accept', where it could easily refer to the difficulty to make the American public accept China's status. In a similar way, we can safely treat passer in (26) as an intransitive (unaccusative) predicate because we know concepts are not usually voted on. This information, however, is provided by the context and our world knowledge, not the construction itself.

The predicates attested in the data do not participate in the object raising construction because they are unaccusative, but because of their specific lexical semantic structure. The unaccusative/unergative distinction does not provide a satisfactory explanation of the distribution of intransitive predicates in the construction because it completely eschews its distinct semantic function. A more revealing solution analyses the selectional restrictions imposed on the participating predicates as the result of the necessary semantic fit between the different components of a meaningful construction.

4.2. Croire union

Croire 'believe' is analyzed as a raising verb in Ruwet (1972) and Kayne (1975), and as a Union predicate in Fauconnier 0983). When it is in main verb position, the object of an embedded transitive predicate can raise to the main clause. This is illustrated in (28) and (29). The example in (29) is (40a) in Legendre 0989):

(28) Il croyait que son adversaire etait elimine

'He believed that his opponent was eliminated'

(29) Il croyait son adversaire elimine

'He believed his opponent eliminated'

As far as intransitives are concerned, unaccusatives are felicitous in the construction, but unergatives are not. This contrast is illustrated in (30)-(33), respectively (47a) and (47b) and (46a) and (46b) in Legendre (1989):

(30) On croyait la neige fondue dans toutes les stations de ski

'We believed the snow (to have) melted in all ski resorts'

(31) On croyait le magasin ouvert

'We believed the store (to be) open'

(32) * On croyait le president agi

'We believed the president (to have) acted'

(33) * On croyait l'homme parle

'We believed the man (to have) spoken'

4.2.1. A usage-based analysis. Following the methodology introduced for the previous constructions, this section investigates the usage of croire union and its validity as an unaccusative diagnostic. Ina way strikingly similar to object raising, croire union is also shown to be meaningful and semantically restricted. The constraints that preside over the distribution of the embedded predicates are therefore best explained in terms of semantic compatibility between the construction and its component parts. The 114 examples of the construction considered included 39 tokens representing 30 transitive verbs, and 20 tokens representing 8 intransitive verbs. The remaining forms, namely 34 predicate adjectives and 21 predicate nominals will be briefly considered at the end of the section.

The primary semantic function of croire union is to present a belief about the raised nominal. This is indicated by the latter's high degree of topicality (Achard 2000). For example, out of the 114 raised nominals, 56 are relative pronouns that directly follow their antecedents, and are thus as strongly topical as possible. This situation is illustrated in (34). In (34), the past participle perdue 'lost' is predicated of the relative pronoun que, which directly follows its antecedent efficacite 'efficiency'.

(34) Le retour au premier plan de la Scuderia n'etait-il qu'un simple feu de paille ou, au contraire, les nouvelles mesures adoptees au nom de la securite permettaient-elles a Ferrari de retrouver une e fficacite que l'on croyait a jamais perdue?

'Was the return of the Scuderia merely a flash in the pan, or to the contrary, do the new measures adopted in the name of security allow Ferrari to recover an efficiency we thought [was] lost forever?'

Just as was the case with object raising, the semantic consistency of the predicates attested in croire union represents a strong indication that the construction has its own semantic import. More specifically, I suggest that croire union presents the belief held by some conceptualizer that the raised nominal has gone through some process before settling into a stable state. The only possible embedded predicates are therefore those capable of expressing that stable state. (21) The 30 transitive and 8 intransitive verbs profile different kinds of stable states. Both transitive and intransitive predicates can be grouped into consistent semantic groupings.

The 30 transitive verbs fall within 4 related semantic categories. The first one describes the stable situation that follows a traumatic event (often a battle). The raised object is viewed as a reward. The verbs include maitriser 'master', gagner 'win', recompenser 'reward', conquerir 'conquer', acquerir 'acquire', atteindre 'reach'. This semantic grouping is illustrated in (35):

(35) Des combats se deroulent aussi autour de la localite de Muaka, au centre-ouest, a l'ouest de Gitarama, que l'on croyait "conquise" la veille par le FPR, a-t-il declare.

'Fighting also went on around the town of Muaka, in the western center, west of Gitarama which we believed [was] "conquered" by the FPR the day before he declared.'

The second kind of predicates describes the result of a violent process. Some force was required in order to overcome the situation evoked by the raised nominal. This represents the largest category, and includes the following verbs. Abolir 'abolish', eradiquer 'eradicate', terminer 'finish', eteindre 'extinguish', dissiper 'dissipate', eliminer 'eliminate', detruire 'destroy', conjurer 'overcome', regler 'solve', rompre 'break', boucler 'close', calmer 'calm', perdre 'lose'. An example of this semantic category is given in (36):

(36) Des centaines de familles ont commence a affluer a Bombay (ouest), fuyant une epidemie de peste pneumonique....Un responsable des services de sante de Surat, une ville industrielle de 2 millions d'habitanta situee a 270 km au nord de Bombay, a indique qu'au moins 100 personnes avaient ete tuees par le fleau que l'on eroyait eradique depuis des annees.

'Hundreds of families began to rush to Bombay (west) to escape an epidemic of bubonic plague....A health official from Surat, an industrial city of 2 million people 270 kilometers, North of Bombay, indicated that at least 100 people had been killed by the disease which we believed [had been] eradicated years ago.'

The third kind of transitive predicates describe a situation that endures despite attacks against it. The following verbs are attested: preserver 'preserve', assurer 'assure', epargner 'spare', reserver 'reserve', cantonner 'restrict', oublier 'forget, promettre 'promise', menacer 'threaten'. This semantic grouping is illustrated in (37):

(37) Le plus celebre de ces singes, Kanzi,... a demontre ces dernieres semaines son aptitude a maitriser des techniques que l'on croyait jusqu'a present reservees aux hommes, notamment la fabrication d'outils en pierre.

'The most famous of these monkeys, Kanzi,... showed this past few weeks his capacity to master techniques that we so far believed [were] reserved to humans, particularly the fabrication of stone tools.'

Finally, a class of verbs composed of posseder 'possess', envouter 'curse', and ensorceler 'bewitch' describe a situation where the observed stability is the result of an occult power. This class is illustrated in (38):

(38) Selon la police, la mere et la grand-mere de l'enfant auraient organise dimanche avec deux amis une ceremonie destinee a exorciser la petite fille qu'elles croyaient possedee par le demon.

'According to the police, the child's mother and grandmother would have organized a ceremony on Sunday with two friends in order to exorcise the little girl whom they believed [was] possessed by the devil.'

The eight intransitive verbs that participate in the construction also present great semantic consistency. Three semantic categories have been isolated. The first one describes death, and includes the verbs deceder 'decease', and mourir 'die'. The second one evokes a situation that comes into view, with the verbs venir 'come', revenir 'come back', and arriver 'arrive'. The third one presents a situation that is fading from view, and includes disparaitre 'disappear', passer 'pass', and envoler 'fly'. These categories are respectively illustrated in (39)-(41):

(39) Ce premier roman etait rernarque d'emblee pour sa force de moraliste et son talent de conteur acerbe. Son heros, que l'on croyait mort au cours d'un bombardement, revient dans son douar et prend conscience que sa presence devient genante pour tous.

'This first novel was noticed right away for its moral strength and its talent of acerbic storyteller. Its hero whom one believed [was] dead in a bombing comes back to his douar, and realizes that his presence disturbs everybody.'

(40) Un homme d'affaires de Hong Kong, qui croyait sa derniere heure arrivee au moment du tremblement de terre a affirme qu'il ne savait plus quoi faire, ne parlant pas le japonais.

'A Hong Kong businessman who believed his last hour [had] arrived when the earthquake hit said that he didn't know what to do since he didn't speak Japanese.'

(41) Ce projet a ete mis au point alors que sont exposes actuellement pour la premiere fois au musee Pouchkine a Moscou 63 toiles de maitres ramenees d'Allemagne a la fin de la guerre, parmi lesquelles des Manet, Degas, Renoir et un "Portrait de Femme'" de Goya qu'on croyait disparu.

'This project was designed while for the first time the Pouchkine museum in Moscow hosts 63 canvas by masters brought back from Germany at the end of the war, among which some Manet, Degas, Renoir, and a portrait of a woman by Goya that we believed [had] disappeared.'

The semantic consistency and the reduced number of the predicates considered in this section constitute strong indications that the constraints that preside over their distribution in CU are independent from their argument structure. Further confirmation comes from the fact that the 114 instances of embedded predicates in the corpus are distributed almost evenly between several grammatical categories, namely 34 adjectives, 21 predicate nominals, 39 transitive verbs, and 20 intransitive verbs. Just like object raising, croire union is not sensitive to the argument structure of the predicate that participates in it. It therefore seems reasonable to propose that the lexical semantics of a given predicate solely determine its selection in the construction.

The conclusions reached in this section cannot be considered definitive because the corpus considered consisted exclusively of journalistic prose. Additional data from other gentes are needed to confirm the results. (22) However, in this specific context, the examination of the usage of object raising and croire union clearly showed that these constructions should be treated as meaningful expressions rather than mete structural templates invoked for diagnostic purposes. Ina way similar to AI, a local account based on their semantic compatibility with the predicates that participate in them was able to provide a satisfactory explanation of the distribution of those predicates. Here again, this kind of local solution is preferable to a structural analysis because the latter fails to capture the semantic import of the constructions, as well as the striking semantic consistency of the predicates that participate in them.

5. Implications for the relevance of UH for French

Because the two constructions presented in Section 4 are considered unaccusative diagnostics, the preferability of the local account considerably reduces the range of application of the Unaccusative Hypothesis in French. Ultimately, if the results obtained for object raising and croire union could be generalized to all other test constructions, its structural motivation would disappear entirely. A thorough analysis of all unaccusative diagnostics is beyond the scope of this article, but this section serves a dual purpose. First, it evaluates in a preliminary fashion how strong the structural motivation for UH is in French by considering the different test discussed in the literature. (23) Secondly, it investigates how the local accounts presented in the preceding sections might be expanded to provide a viable alternative to UH to describe the distribution of intransitives in French grammar.

In addition to object raising and croire union, Legendre (1989) presents the following seven diagnostics to distinguish between unaccusative and unergative predicates: Participial absolutes, reduced relative clauses, cliticization of embedded 3's in causative unions, auxiliary selection, parallel transitive structures, nominalizations, expression of stativity. (24) These diagnostic tests are illustrated in (42)-(55).

5.1. Participial equi and participial absolutes

Unaccusative, but not unergative predicates behave like transitives in both equi and absolute participial constructions. This is illustrated for equi participials in (42)-(44), respectively (53c), (57a), and (61a) in Legendre (1989).

(42) Arrete par la police, Pierre subit une longue interrogation

'Arrested by the police, Peter underwent a long interrogation'

(43) Parti avant l'aube, Pierre est arrive le jour meme a destination

'(Having) left before dawn, Peter arrived to his destination that same day'

(44) * Reagi, le president a ete felicite par la presse

'(Having) reacted, the president was congratulated by the press'

The unaccusative partir 'leave' in (43) behaves like the transitive arreter 'arrest' in (42), thereby indicating that both predicates have an initial object (a 2). The unergative reagir 'react' in (44) is infelicitous because its only initial argument is a subject (a 1). Absolute participials follow a similar pattern, as illustrated in (45)-(47), respectively (74d), (83a), and (87f) in Legendre (1989).

(45) Leurs benefices elimines (par la direction), les employes se mirent en greve

'Their advantages eliminated by the management, the employees went on strike'

(46) La neige fondue, Pierre mit ses skis de cote

'The snow (having) melted, Peter put his skis aside'

(47) * Le candidat parle, l'audience se tut

'The candidate (having) spoken, the audience turned quiet'

The unaccusative fondre 'melt' in (46) behaves in a way similar to the transitive eliminer 'eliminate' in (45). The unergative parler 'speak' in (47) is impossible in the construction.

5.2. Reduced relative clauses

Reduced relative clauses resemble participial clauses because neither the relative pronoun nor the auxiliary are present on the surface. Unaccusative predicates appear in reduced relatives as illustrated in (48), but unergatives are infelicitous, as illustrated in (49). The examples in (48) and (49) are respectively (95a) and (96a) in Legendre 1989. Note that the constraint on the predicate is restricted to reduced relative clauses:

(48) La personne (qui est) morte hier soir....

'The person (who) died last night ...'

(49) * L'homme reagi ... (OK qui a reagi)

'The man reacted ... (OK who reacted)'

5.3. Cliticization of embedded 3's in causative unions

The embedded 3 of unaccusative predicates can be cliticized in a causative construction, but the 3 of unergatives cannot. This is illustrated in (50) and (51), respectively (109) and (104) in Legendre (1989):

(50) a. La peur fera pousser des ailes a Marie

'Fear will make wings grow on Mary'

b. La peur lui fera pousser des ailes

'Fear will make wings grow on her'

(51) a. On fera telephoner Pierre a ses parents

'We will make Peter call his parents'

b. * On leur fera telephoner Pierre

'We will make Peter call them'

The 3 of both unaccusative pousser 'grow' in (50) and unergative telephoner 'call' in (51) can appear in the causative construction if it is not cliticized as in (50a) and (51a). However, only unaccusative predicates can be felicitously cliticized, as (50b) and (51b) show.

5.4. Parallel transitive structures

The presence of a parallel transitive structure as a causative represents another unaccusative test. For example, the transitive form in (52a) acts as the causative form of the intransitive casser 'break' in (52b). The example in (52c) represents an intransitive usage with middle morphology. The examples in (52) are found in (119) in Legendre (1989).

(52) a. Jules a casse la branche

'Jules broke the branch (= made the branch break)'

b. La branche a casse

'The branch broke'

c. La branche s'est cassee

'The branch broke'

Unergative predicates on the other hand do not allow a transitive structure as a causative form. This is illustrated in (53):

(53) a. L'enfant a dormi

'The child slept

b. * La nourrice a dormi l'enfant

'The nanny slept the child'

c. La nourrice a fait dormir l'enfant

'The nanny made the child sleep'

The unergative dormir 'sleep' does not have a transitive causative usage, and its causative meaning must be explicitly marked by the causative construction in (53c).

5.5. Nominalizations

Ruwet (1989) notes that unaccusative verbs allow nominalizations based on their past participle. The examples in (54) are given in (121) in Legendre (1989):
(54) Predicate          Nominalization
     geler 'freeze'     gelee 'frost'
     arriver 'arrive'   arrivee 'arrival'
     entrer 'enter'     entree 'entrance'
     sortir 'go out'    sortie 'exit'

Unergatives do not allow such nominalizations, as shown in (55) (Legendre 1989: 149):
(55) Predicate           Nominalization
     travailler 'work'   * travaillee (travail) 'work'
     agir 'act'          * agi(e) (action) ' act'
     parler 'speak'      * parlee (parole) 'speech'
                         * vieillie (vieillissement) 'aging'

5.6. A brief evaluation of the diagnostics

Most intransitive predicates do not react consistently across the nine unaccusative diagnostics. Compare for example the behavior of augmenter 'increase' in object raising in (56) and croire union in (57):

(56) Les prix seront faciles a faire augmenter

'The prices will be easy to make increase'

(57) * Je croyais les prix augmentes

'I believed the prices increased'

Augmenter behaves like an unaccusative in (56), but as an unergative in (57), making it difficult to clearly establish class membership. In order to remedy this difficulty, Legendre (1989) claims that the nine unaccusative tests should be considered sufficient but not necessary conditions for unaccusativity. In other words, any given French predicate only needs to satisfy one of the nine diagnostics to be classified as unaccusative. This caveat diminishes the explanatory force of the Unaccusative Hypothesis because the appeal of a structural account mostly resides in its range of application across constructions and semantic domains. Furthermore, since participation in one construction represents a sufficient condition for the classification of a predicate in the unaccusative class, it is impossible to independently verify its assignment if it only displays unaccusative behavior in that construction.

In addition to this general point, several of these tests present difficulties that seriously undermine their status as unaccusative diagnostics. The remainder of this section briefly considers the most problematic ones. The presence of a transitive structure cannot be considered a valid test of unaccusativity because, despite the fact that only unaccusative predicates have access to such a form, too many unaccusatives do not. Legendre herself mentions the case of the middle verbs that do not have a causative form despite their unaccusative behavior in the other tests. This is illustrated with s'evanouir 'faint' in (58) ([120] in Legendre 1989):

(58) a. L'enfant s'est evanoui

'The child fainted'

b. * La peur a evanoui l'enfant

'Fear fainted the child'

c. La peur a fait s'evanouir l'enfant

'Fear made the child faint'

This problem is not restricted to reflexive and middle verbs. A large number of otherwise unaccusative predicates do not have corresponding transitive structures either, as illustrated in (59) and (60):

(59) a. Les enfants sont arrives a l'heure

'The children arrived on time'

b. * Jean a arrive les enfants a l'heure

'John arrived the children on time'

c. Jean a fait arriver les enfants a l'heure

'John made the children arrive on time'

(60) a. Marie est restee a la maison

'Mary stayed at home'

b. * Emilie a reste Marie a la maison

'Emily stayed Mary at home'

c. Emilie a fait rester Marie a la maison

'Emily made Mary stay at home'

The predicates that have similar forms for their transitive and intransitive variants seem to be semantically restricted to a certain class of change of state verbs, but their specific characterization is beyond the scope of this article. (25) In any case, the unaccusative structure of the predicates seems to be only remotely relevant to the distribution of transitive/intransitive forms. The use of transitive structures is therefore too limited to be truly convincing as a test construction.

Nominalization also seems difficult to use as a reliable way of testing for unaccusativity because it involves too many diverse layers of linguistic organization, including historical factors and conventionalization, to be uniquely explained by the syntactic structure of the predicate nouns are derived from. Furthermore, several unaccusatives do not have a nominalized form based on their past participle. Legendre herself (1989: 149) provides the following examples: naitre 'be born', * nee (naissance) 'birth'; partir 'leave', * partie (depart) 'departure'; vieillir 'get old' * vieillie (vieillissement) 'aging'. Conversely, enough transitive and even unergative verbs admit nominalizations based on their past participle to cast serious doubt on the validity of this test. Examples include poussee 'push' from pousser 'push', visee 'aim' from viser 'aim', prise 'hold' from prendre 'take', for transitive verbs, as well as criee 'fish market' from the unergative crier 'shout'. (26)

The results of this brief overview indicate that only three tests remain as possible unaccusative diagnostic candidates, namely participial equi and absolutes, reduced relative clauses, and cliticization of embedded 3's in causative union. Further investigation of the actual usage of these test constructions will determine their reliability, but a local analysis that follows the methodology developed in Sections 3 and 4 provides a promising way of approaching the distribution of intransitive predicates in these contexts.

The problems encountered by the individual tests only represent one aspect of the evaluation of the importance of UH in French. Even if we abstract away from the construction-specific issues, the nine unaccusative contexts discussed in this article do not collectively present a convincing case in favor of the validity of the hypothesis. Taken as a group, they might be argued to successfully delineate an unaccusative and an unergative verb classes if one carefully selects which test applies to which predicate. However, this division does not seem particularly insightful because it sheds no particular light onto the principles that govern the behavior of intransitive predicates in general. It reveals no generalities, no natural classes, no privileged semantic environments, and no relation between families of constructions that might enlighten specific areas of French grammar. In fact, one might even argue that the outcome of the solutions based on the unaccusative/unergative split directly results from their primary motivation, namely to validate UH as a structural universal by demonstrating its existence in as many languages as possible. Legendre (1989: 153) makes this goal explicit when she begins the conclusion of her analysis of French unaccusativity in the following way: "The Unaccusative Hypothesis is generally viewed as being one of the most successful contributions of RG. This article provides further support for the

Unaccusative Hypothesis because it demonstrates that a class of French intransitives splits into two sub classes with respect to nine distinct phenomena." The results obtained in this article suggest that Legendre's claim may be overstated. The intransitive predicates do not seem to participate in specific constructions because of their structure, but because they are semantically compatible with them.

5.7. Crosslinguistic potential

Since the distribution of intransitive predicates is best explained by evaluating the semantic compatibility between verbs and constructions, a comprehensive account of their syntactic behavior needs to pay specific attention to every construction they participate in. At first glance, this focus on individual constructions might appear to put local accounts at a serious disadvantage, because it seems to limit their explanatory scope to one single construction at a rime. However, closer scrutiny reveals two facts that contradict this initial impression. First, as illustrated in Section 4, some constructions are so semantically specific that their internal complexity only appears if they constitute the exclusive focus of investigation. Secondly, local solutions do not forsake higher-level generalizations, when the semantic distinctions invoked for one construction are also observed in other situations. These accounts are therefore in a perfect position to precisely evaluate the respective language internal and crosslinguistic generality of the semantic traits invoked in each construction.

One example will suffice to show this crosslinguistic potential. The presence of the field in the scope of predication of some predicates was shown in Section 3 to account for the distribution of intransitives in French active impersonals. This semantic characteristic also figures prominently in the description of English resultatives (Levin and Rappaport Hovav 1995: Chapter 2). The remainder of this section does not pretend to provide an exhaustive analysis of English resultatives. Its purpose is merely to point out the relevance of the field/participant distinction to the description of the construction, and therefore the crosslinguistic potential of the analysis provided in this article.

Levin and Rappaport Hovav (1995: 34) describe the resultative construction as follows: "A resultative phrase is an XP that denotes the state achieved by the referent of the NP it is predicated of as a result of the action denoted by the verb in the resultative construction." Transitive predicates are felicitous in the construction, as illustrated in (61). The examples in (61) are adapted from the ones in (1) in Levin and Rappaport Hovav (1995).

(61) a .... while she soaps me slippery all over

b. Absently, she dipped a finger in the peanut butter and licked it clean

Unergative predicates are only possible if a reflexive pronoun or a post verbal NP follows the predicate, as illustrated in (62). Note that the post verbal NP is often a pronoun (62c) or an inalienably possessed nominal whose possessor is coreferential with the subject (62d). The examples in (62) are adapted from Levin and Rappaport Hovav (1995: 35-36):

(62) a. * Dora shouted hoarse

b. Dora shouted herself hoarse

c. The dog barked him awake

d. Sylvester cried his eyes out

Finally, unaccusatives are also possible with resultatives. This is illustrated in (63), adapted from (19) in Levin and Rappaport Hovav (1995).

(63) a. The river froze solid

b. The gate swung shut

c. The curtain rolled open

However, stative and inherently directed motion predicates are impossible in the construction, as illustrated in (64). The examples in (64a) and (64b) are respectively (51a) and (51b) in Levin and Rappaport Hovav (1995).

(64) a. * Carla remained in the country bored

(Carla became bored by the process of remaining)

b. * Willa arrived breathless

(Willa became breathless by the process of arriving)

The parallel in the distribution of intransitives in French active impersonals and English resultatives is striking. In a preliminary fashion, we might suggest that the field/participant distinction already observed for the French construction also prominently figures in the analysis of the English construction. Because the resultative state denotes the state achieved by the referent of the NP (i.e., the affected participant), it seems reasonable to claim that only the verbs that place their descriptive focus on their participants felicitously occur in the construction. This straightforwardly excludes the stative and inherently directed motion predicates, because these verbs have been shown to be more specifically concerned by the way in which the field is affected by the presence of the participants. Their incompatibility with a construction where the affected participant is in focus is therefore expected

In order to account for the different behavior of the transitive, activity (unergative), and internal change (unaccusative) predicates, we also need to invoke the agentivity of their subject, namely its role relative to the initiation of the action these verbs denote. Transitives constitute a clear case. The subject prototypically codes the initiator of the verbal process, and the object the affected entity. The latter therefore constitutes the only entity the resultative state can be predicated about. The subject of the activity predicates also initiates the verbal process, but because these verbs are intransitive, the affected entity (if relevant) is not usually expressed. The specific cases where the participant represents at the same time the initiator of the verbal process and the entity affected by the state resulting from that process need to be specifically coded. The presence of a reflexive or a post verbal NP formalizes the distinction between the two different roles the referent of the subject plays with respect to the verbal process. Finally, the subject of the verbs of internal change does not initiate that change. It is presented as the recipient of the force responsible for it, as an entity affected by that force. The resultative state can therefore be predicated of that entity.

This example of the French active impersonals and English resultatives clearly illustrates the promise of local accounts at different levels of linguistic analysis. Language internally, the field/participant distinction provides valuable insight into the distribution of intransitives in the French construction. It also elegantly combines with the agentivity of the subject to explain the distribution of predicates in the English construction. Crosslinguistically, it allows us to group together natural classes of constructions (i.e., whether they focus on the participants of the field), but more importantly natural classes of predicates that do not commonly get considered together, either because they are regarded as structurally different (verbs of activity and internal change), or because they constitute subsets of a larger group (inherently directed motion and stative predicates). Perhaps the most desirable aspect of the methodology developed in this article is its capacity of grounding crosslinguistic generalizations into the fertile soil of language specific ecologies.

6. Conclusion

This article has argued that the constraints that govern the selection of intransitive predicates in French are determined by the necessary compatibility between the constructions and their component parts. This lexical semantic analysis has been shown to be preferable to a structural account to explain the distribution of intransitive predicates in active impersonals, as well as the two test constructions of object raising in the context of the faire construction and croire union.

The local character of the analysis proposed in this article makes it not only different from the structural accounts, but also from the other semantic solutions previously suggested in the literature. Semantic analyses of the distribution of intransitive predicates are not new. Several researchers (Van Valin 1990; Kaufman 1995 among others) have argued that unaccusativity is not syntactically but semantically encoded. In perhaps the most specific semantic account of unaccusativity, Van Valin (1990: 221) claims that: "the phenomena which the Unaccusative Hypothesis strives to explain in syntactic terms are better explained in semantic terms." While the proposal developed here agrees with this position, it diverges from Van Valin's account on two critical points, namely the role of the constructions' semantic import, and the intended scope of the semantic characterization of the predicates.

Van Valin (1990: 221) argues that: "two semantic parameters, inherent lexical aspect (Aktionsart) and agentivity, underlie split intransitivity crosslinguistically." Interestingly, his solution does not challenge the validity of the Unaccusative Hypothesis, namely that the syntactic behavior of intransitive predicates across a series of diagnostic constructions needs to be explained by features common to those predicates, it merely gives it a semantic solution by claiming that the combination of two features is responsible for their consistent behavior. It is true that Van Valin's solution is more flexible than the structural accounts. Different languages may be more sensitive to lexical aspect or agentivity, and within single languages, different constructions have similar options (Levin and Rappaport Hovav 1995). However, this increased flexibility is still not sufficient to provide a satisfactory account of the French data. Each test construction exhibits such limited semantic range, and its specificity makes it so different from the other diagnostics that it is impossible to find a set of features (even semantic ones) that would group together all and only the predicates that appear in all of them. Just like the unaccusative/unergative distinction fails to distinguish the relevant classes of predicates, Aktionsart and agentivity are too general to capture the tine grained semantic nuances that govern the predicates' participation in the constructions. The recognition of the meaning of the constructions provides the key aspect of the solution proposed here. It allows the analysis to i) consider the distribution of intransitive predicates as a matter of semantic compatibility rather than structural motivation, and ii) reach the lower level semantic restrictions French requires. The narrow semantic islands encountered in the data reveal that predicate distribution in French is best investigated at the local level of individual constructions.

Rice University

Received 30 July 2004

Revised version received

31 July 2006


Achard, Michel. 1993. Complementation in French: A cognitive perspective. University of California San Diego dissertation.

Achard, Michel. 1998. Representation of cognitive structures: Syntax and semantics of French complements (Cognitive Linguistics Research 11). Berlin & New York: Mouton de Gruyter.

Achard, Michel. 2000. The distribution of French raising constructions. In Lisa Conathan, Jeff Good, Darya Kavitskaya, Alyssa Wulf & Alan Yu (eds.). Proceedings of the 26th Annual Meeting of the Berkeley Linguistics Society (BLS 26), 1-15. Berkeley, CA: Berkeley Linguistics Society.

Barlow, Michael & Suzanne Kemmer. 2000. Usage-Based Models of Language. Stanford: CSLI.

Burzio, Luigi. 1986. Italian syntax: A government binding approach. Dordrecht: Reidel.

Bybee, Joan. 2001. Phonology and language use. Cambridge & New York: Cambridge University Press.

Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.

Cornillie, Bert. 2005. Subjectification as cognitive semantic operation: On Spanish amenazar 'to threaten' and prometer 'to promise'. Paper presented at conference From Gram to Mind, University of Bordeaux.

Cummins, Sarah. 2000. The unaccusative hypothesis and the impersonal construction in French. Canadian Journal of Linguistics 45(3/4). 227-251.

Fauconnier, Gilles. 1983. Generalized union. Communication and Cognition 16. 3-27.

Grimshaw, Jane. 1980. Argument structure. Cambridge, MA: MIT Press.

Gueron, Jacqueline. 1980. On the syntax and semantics of PP extraposition. Linguistic Inquiry 11. 637-678.

Haspelmath, Martin. 1993. More on the typology of inchoative/causative verb alternations. In Bernard Comrie & Maria Polinsky (eds.), Causatives and transitivity, 87-120. Amsterdam: John Benjamins.

Heriau, Michel. 1980. Le verbe impersonnel en francais moderne. Lille: Atelier de reproductions de theses, Universite de Lille III.

Herschensohn, Julia. 1982. The French impersonal as a base generated structure. Studies in Language 6. 193-219.

Herschensohn, Julia. 1996. Case suspension and binary complement structure in French. Amsterdam: John Benjamins.

Kaufmann, Ingrid. 1995. O- and D-predicates: A semantic approach to the unaccusative-unergative distinction. Journal of Semantics 13. 377-427.

Kayne, Richard. 1975. French syntax: The transformational cycle. Cambridge: MIT Press.

Kayne, Richard. 1979. Rightward NP movement in French and English. Linguistic Inquiry 10. 710-719.

Kishimoto, Hideki. 1996. Split intransitivity in Japanese and the unaccusative hypothesis. Language 72. 248-286.

Kuno, Susumu & Ken-ichi Takami. 2004. Functional constraints in grammar: On the unergative-unaccusative distinction. Amsterdam: John Benjamins.

Labelle, Marie. 1992. Change of state and valency. Journal of Linguistics 28. 375-414.

Lakoff, George. 1987. Women, fire, and dangerous things: What categories reveal about the mind. Chicago & London: University of Chicago Press.

Lambrecht, Knud. 1994. Information structure and sentence form: Topic, focus, and the mental representation of discourse referents. Cambridge: Cambridge University Press.

Lambrecht, Knud. 2000. When subjects behave like objects: Ah analysis of the merging of S and O in sentence focus constructions across languages. Studies in Language 24. 611-682.

Langacker, Ronald W. 1985. Observations and speculations on subjectivity. In John Haiman (ed.), Iconicity in syntax, 109-150. Amsterdam: John Benjamins.

Langacker, Ronald. W. 1987. Foundations of cognitive grammar, vol. 1: Theoretical prerequisites. Stanford: Stanford University Press.

Langacker, Ronald W. 1988. A usage-based model. In Brygida Rudzka-Ostyn (ed.), Topics in cognitive linguistics, 127-161. Amsterdam & Philadelphia: John Benjamins.

Langacker, Ronald W. 1990. Subjectification. Cognitive Linguistics 1.5-38.

Langacker, Ronald W. 1991. Foundations of cognitive grammar, vol. 2: Descriptive Application. Stanford, CA: Stanford University Press.

Langacker, Ronald W. 2000. A dynamic usage-based model. In Michael Barlow & Suzanne Kemmer (eds.), Usage-based models of language, 1-63. Stanford, CA: CSLI.

Langacker, Ronald W. 2002. The control cycle: Why grammar is a matter of life and death. Proceedings of the annual meeting of the Japanese Cognitive Linguistics Association 2. 193-220.

Langacker, Ronald. 2004. Aspects of the grammar of finite clames. In Michel Achard & Suzanne Kemmer (eds.), Language, culture, and mind, 535-577. Stanford, CA: CSLI.

Legendre, Geraldine. 1989. Unaccusativity in French. Lingua 79. 95-164.

Legendre, Geraldine. 1990. French impersonal constructions. Natural Language & Linguistic Theory 8. 81-128.

Legendre, Geraldine & Antonella Sorace. 2003. Auxiliaires et intransitivite en francais et dans les langues romanes. In Daniele Godard (ed.), Les langues romanes: Les problemes de la phrase simple, 185-233. Paris: CNRS.

Levin, Beth & Malka Rappaport Hovav. 1995. Unaccusativity: At the syntax-lexical semantics interface. Cambridge: MIT Press.

Martin, Robert. 1970. La transformation impersonnelle. Revue de Linguistique Romane 34. 377-394.

Olie, Annie. 1984. L'hypothese de l'inaccusatif en francais. Lingvisticae Investigations 8. 363-401.

Perlmutter, David. 1978. Impersonal passives and the unaccusative hypothesis. In Jeri J. Jaeger, Anthony C. Woodbury, Farrell Ackerman, Christine Chiarello, Orin D. Gensler, John Kingston, Eve E.. Sweetser, Henry Thompson & Kenneth W. Whistler (eds.), Proceedings of the fourth annual meeting of the Berkeley Linguistics Society (BLS 4). 157-189. Berkeley, CA: Berkeley Linguistics Society.

Perlmutter, David. 1989. Multiattachment and the unaccusative hypothesis: The perfect auxiliary in Italian. Probus 1(1). 63-119.

Pollock, Jean-Yves. 1978. Trace theory and French syntax. In Jay Keyser (ed.), Recent transformational studies in European languages, 65-112. Cambridge, MA: MIT Press.

Postal, Paul. 1982. Arc pair grammar descriptions. In Pauline Jacobson & Geoffrey Pullum (eds.), The nature of syntactic representation, 341-425. Dordrecht: Kluwer.

Postal, Paul. 1984. French indirect object cliticization and SSC/BT. Linguistic Analysis 14. 111-172.

Rosen, Carol. 1981. The relational structure of reflexive clauses: Evidence from Italian. Ph.D Dissertation, Harvard University [Published 1988, New York: Garland].

Rosen, Carol. 1984. The interface between semantic roles and initial grammatical relations. In David Perlmutter & Carol Rosen (eds.), Studies in relational grammar 2, 38-77. Chicago: University of Chicago Press.

Ruwet, Nicolas. 1972. Theorie syntaxique et syntaxe du Francais. Paris: Editions du Seuil.

Ruwet, Nicolas. 1989. Weather verbs and the unaccusative hypothesis. In Carl Kirschner & Janet De Cesaris (eds.), Studies in romance linguistics, 313-345. Amsterdam & Philadelphia: John Benjamins.

Smith, Michael. 1985. An analysis of German dummy subject construction in Cognitive Grammar. In Scott DeLancey & Russell Tomlin (eds.), Proceedings of the first annual meeting ofthe Pacific Linguistics Conference, 412-425. Eugene, OR: University of Oregon Department of Linguistics.

Sorace, Antonella. 2000. Gradients in auxiliary selection with intransitive verbs. Language 76(4). 859-880.

Van Valin, Robert. 1990. Semantic parameters of split intransitivity. Language 66. 221-260.


* Correspondence address: Dept. of Linguistics MS23, Rice University, 6100 Main Street, Houston, TX 77005-1892, USA. E-mail:

(1.) In the generative version of the UH introduced by Burzio (1986), unergatives take a D-Structure subject and no object, while unaccusatives take a D-Structure object (a clause or simple NP) and no subject. The D-Structures in i) and ii) are respectively assigned to unergative and unaccusative predicates:

(i) [[sub.IP] NP [[sub.VP] V]]

(ii) [[sub.IP] e [[sub.VP] V NP]]

In the Minimalist program (Chomsky 1995), the subjects of unergative verbs originate in the specifier of VP, whereas the subjects of unaccusative verbs originate in the direct object position of VP, before moving to the specifier position of TP (= IP).

(2.) The role of semantics in the determination of the unaccusative and unergative classes constitutes the major difference between that various accounts of intransitive predicates based on the Unaccusative Hypothesis. In syntactic/semantic solutions, the position that unaccusativity is "syntactically represented but semantically determined" (Levin and Rappaport Hovav 1995: 30) emphasizes the connection that exists between the semantics of a predicate and its syntactic configuration. Other researchers deny the relevance of semantic factors and argue that the unaccusative/unergative distinction is strictly syntactic. For example, Rosen (1984) notes that unaccusative verbs do not forma homogeneous class crosslinguistically. Furthermore, she points out several mismatches between the semantics of certain predicates and their syntactic assignment. In particular, certain Italian intransitives can be found with both auxiliaries (avere and essere), which classifies them as both unaccusative and unergative by the same diagnostic. The impossibility of systematically relying on semantic structure to predict syntactic behavior leads her to claim that the only characteristic unaccusative verbs share is their structural configuration.

(3.) This construction is sometimes called "Extraposition of Indefinites" (EXI) (Postal 1984; Legendre 1990). It is considered impersonal because the predicate agrees with the "dummy" il 'it' rather than with the "real" subject trois personnes. 'three people'. The cases where the predicate carries middle or passive morphology will not be considered in this article.

(4.) Cummins (2000) also treats this construction as an unaccusative diagnostic in her effort to disprove UH for French. She writes (2000: 240): "Since unergative verbs of diverse semantic classes map to an unaccusative structure, the UH is disproven". Her results, however, fall short of their desired effect because the active impersonal construction is not consistently recognized as an unaccusative diagnostic. Legendre (1989:154) for ex ample, writes: "There is considerable evidence, however, that the EXI construction is not a reliable test for unaccusativity in French".

(5.) Legendre herself (1990: Note 9) admits that "it is unclear how this 'dialect' should be characterized", given the diversity of her consultants. Several of these examples are judged ungrammatical in Labelle (1992). This issue will be considered in more detail in Section 3.4.

(6.) ARTFL is a cooperative enterprise of Analyse et Traitement Informatique de la Langue Francaise (ATILF) of the Centre National de la Recherche Scientifique (CNRS), the Division of Humanities, the Division of Social Sciences, and Electronic Text Services at the University of Chicago. It can be located on the Internet at the following address:

(7.) Two verbs, namely couter 'cost (10), and prendre 'take' (1) do not fit well into any of the five semantic classes. Both verbs are predominantly transitive, and the examples they contribute to the corpus all come from the set expressions il m'en coute 'it costs me', and prendre envie 'take fancy' respectively.

(8.) This verb was included in the corpus despite the presence of the s' marker, indicative of the middle voice, because it constitutes the only possible form for that verb (* ensuivre).

(9.) The verbs in this class are semantically very close to the ones presented in (9). Their difference mainly pertains to the extent to which the beginning or the end of the participant's trajectory occurs inside or outside of the conceptualized scene. This distinction will be discussed further in Section 3.3.

(10.) Cuire 'cook' resembles the other verbs in the class in that the immediately recognizable smell it produces can easily be construed as permeating the scene where the cooking occurs.

(11.) The alternative position by Levin and Rappaport Hovav (1995) that the meaning of unergatives in AI is in some sense "not new" is more difficult to assess. The semantic import of an impersonal construction is certainly not identical to that of a personal construction with the same verb. However, it seems difficult to solely attribute this meaning difference to a change in the verb's lexical content. Section 3.4 shows how the meaning contrast between personal and impersonal structures reflects the two constructions' respective semantics.

(12.) Even though it is not directly apparent in the corpus presented in this article, unergatives cannot be dismissed from AI on the basis of their scarcity. Cummins (2000) reports that in Heriau's (1980) corpus of 700 literary works, unergatives such as courir 'run', roder 'prawl', sonner 'ring' are more frequent than the unaccusatives aller 'go', entrer 'enter', or emerger 'emerge'.

(13.) More specifically, Achard (1998: 282) writes, "I propose that: il ... profiles the abstract setting identifiable as the immediate scope for the existential predication....If the conceptualizer of the sentence C is viewed as peering out through a viewing frame, the immediate scope is that portion of the world being viewed that is subtended by that frame at any given moment. I suggest that il profiles the part of the viewed world subtended by the viewing frame."

(14.) The lack of relevance of the field in the scope of predication of mental activity verbs explains why they are virtually inexistent in active impersonals.

(15.) The indefinite subject constraint does not only apply to unergative predicates, but to some unaccusatives (internal change) as well. This is illustrated in (i) and (ii):

(i) a. Il brunit beaucoup de touristes sur la cote d'Azur 'There tan many tourists on the Riviera'

b. * Il brunit Paul dans le jardin 'There tans Paul in the garden'

(ii) a. Il fond beaucoup de neige au printemps 'There melts a lot of snow in spring

b. * Il a fondu la neige derriere la maison 'There melted the snow behind the house'

The difference between the internal change (unaccusative) pairs in (i) and (ii) parallels that between the examples in (6) and (7) with activity verbs (unergatives), thus illustrating the fact that the indefinite subject constraint also holds for the verbs of internal change. This common syntactic behavior constitutes an argument in favor of an analysis that groups them together despite their different structure.

(16.) The grammatical significance of direct perception versus more indirect perception/ cognition is also attested in other constructions. For example, bare infinitival complements are only felicitous following verbs of involuntary visual perception if the perception is not immediate: * Je vois neiger 'I saw snow' but J'ai vu neiger 2 fois dans ma vie 'I saw snow twice in my life' (Achard 1993).

(17.) The sentences in (14) and (18) also obviously differ in terms of the way in which they structure information (Lambrecht 1994, 2000). The relation between the predicates that most felicitously occur in AI and the kind of information structure strategy the construction codes represents an interesting avenue of research.

(18.) Subjectification has successfully invoked to explain a large range of linguistic phenomena both synchronically and diachronically. These phenomena include among others, the meaning shift from motion verbs to future auxiliaries in English and French (Langacker 1990), the passage from main verbs to modals in English (Langacker 1990), as well as the distinction between the speech act and raising senses of the promise/ threaten verbs in different languages (Achard 1993; Cornillie 2005).

(19.) Some accounts, most notably Levin and Rappaport Hovav (1995) invoke the presence of linking rules that project the verb's lexical properties onto their argument structure. In this sense, both syntax and semantics are invoked to explain the distribution of intransitives. In the Cognitive Grammar framework, syntax and semantics do not constitute separate levels of linguistic organization. Rather, lexical and grammatical structures can all be described as a continuum of symbolic (and hence meaningful) structures. This position does not mean that syntax does not exist, bur that it represents the abstract commonality that complex symbolic expressions share. Consistent with this crucial CG tenet, the analysis presented in this section shows that no specific syntactic level is required to account for the behavior of intransitives in object raising and croire union.

(20.) The causee in (21) is introduced by the preposition a 'to', but the one in (25) is preceded by par 'by'. This distinction is not directly relevant to the issues discussed in this article.

(21.) Predicates such as exister 'exist', or briller 'shine' are infelicitous in the construction despite their stable character because an important characteristic of croire union's meaning specifies that the configuration the raised nominal settles in needs to be the endpoint of an ongoing process.

(22.) Preliminary studies of object raising and croire union in literary texts confirm the results obtained in this article. The range of the predicates is somewhat larger, but their semantic consistency is comparable to the one the constructions display in journalistic prose. Additionally, an anonymous reviewer reports that a Google search of difficile a faire 'difficult to do' largely confirms the results obtained in this article. However, s/he mentions two instances that do not, namely les figues sont difficile a faire murir 'the figs are difficult to make ripen' and le volet manuel est difficile a faire descendre 'manual shutters are difficult to make go down'. Additional data are needed to see if the examples that do not conform to the generalizations proposed in this article are nonetheless organized in semantic clusters.

(23.) Attacks against the explanatory power of UH have recently been made in different languages. For example, Kuno and Takami (2004) have pointed out a large number of counterexamples that challenge the validity of UH for English, Japanese, and Korean. See also Cummins' line of argumentation presented in Note 4.

(24.) Two of these tests will not be considered here, because they clearly do not live up to their diagnostic status. Auxiliary selection has been argued to be a valid test of unaccusativity for Italian (Rosen 1981, 1984; Burzio 1986; Perlmutter 1989). In French, the selection of etre 'be' as an auxiliary should represent a sign that the predicate is unaccusative (Ruwet 1989; Legendre 1989). However, too many unaccusative predicates select avoir 'have' instead of etre 'be' for the test to reliable. Examples of unaccusatives that take avoir include fondre 'melt', cuire 'cook', rougir 'blush' to name just a few. For an analysis of auxiliary choice, see Sorace (2000), Legendre and Sorace (2003). The position that stative verbs should be considered unaccusative also seems so unmotivated that Labelle (1992: 379) calls it "a typical example of a priori assumption". Legendre notes that some of the most frequently attested verbs crosslinguistically such as exister 'exist' and etre 'be' cannot syntactically be distinguished from unergatives by any of the previously mentioned tests. Her reasoning for the inclusion of stativity as an unaccusative test is as follows: "It is well-known, however, from various aspectual studies, that verbs like exist, be denote states while putative unergative verbs like work denote actions. If we can identify a given verb as stative on the basis of certain specific tests, then we can add stativity to our set of tests" (Legendre 1989: 150). Stativity, however, merely identifies a given verb as stative. In the absence of a compelling argument that connects stativity (a semantic property) to unaccusativity (a structural characteristic), the only motivation to include stative verbs among unaccusative predicates is the necessity to identify verbs like exister and etre as unaccusatives.

(25.) A fruitful line of investigation would explore the distribution of middle and active morphology on the intransitive predicates. An interesting distinction is presented in (i) and (ii):

(i) a. Jean a creve le pneu

'John punctured the tire'

b. Le pneu a creve

c. * Le pneu s'est creve

'The tire punctured'

(ii) a. Marie a detache ranneau

'Mary unfastened the ring'

b. L'anneau s'est detache

c. * L'anneau a detache

'The ring unfastened'

The middle and active forms are in complementary distribution in (i) and (ii). One might suggest that the presence or absence of the middle marker may among other things correlate with the manner in which the change of state occurs (sudden and unexpected versus slow, gradual, or in predictable increments for example), but further research is needed to confirm this hypothesis. For a broader discussion of related constructions, see Haspelmath (1993), as well as Labelle (1992) for French data.

(26.) An interesting semantic cluster of vernacular nominalizations indicating intense scolding can be observed. These nominalizations can come from transitive verbs such as engueuler 'scold' (engueulee), ramoner 'sweep a chimney' (ramonee), secouer 'shake' (secouee), or unergatives predicates such as souffler 'blow' (soufflee) or ronfler 'snore' (ronflee). Their presence provides yet another argument against the relevance of the structure of the predicate to the nominalizations.
COPYRIGHT 2009 Walter de Gruyter GmbH & Co. KG
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2009 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Achard, Michel
Publication:Linguistics: an interdisciplinary journal of the language sciences
Article Type:Report
Geographic Code:4EUFR
Date:May 1, 2009
Previous Article:Functional discourse grammar--multifunctional problems and constructional solutions.
Next Article:Null subjects: a reanalysis of the data.

Related Articles
On the event structure of -ung nominals in German(*).
Split intransitivity, linking, and lexical representation: the case of Yukatek Maya *.
Existential sentences with preposed locative phrases and postverbal determinerless subjects in Spanish. *.
Complementation in British and American English: Corpus-Based Studies on Prepositions and Complement Clauses in British and American English.
Intransitivity and the origins of middle voice in Otomi *.
Aligning restricted objects **.
On the middle voice: an interpersonal analysis of the English middle *.
Standing divided: dispositionals and locative predications in two Mayan languages *.
A constructionist approach to complementation: evidence from Ancient Greek *.
Resultative constructions with "implied-result" and "entailed-result" verbs in Thai and English: a contrastive study.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters