The syntax of French à and de: an HPSG analysis Anne Abeillé Olivier Bonami Danièle Godard Jesse Tseng Université Paris 7 Université Rennes 2 CNRS Universités Paris 7 & 10 Laboratoire de Linguistique Formelle (UMR 7110) Case 7031 – 2 place Jussieu – 75251 Paris Cedex 05 {abeille|dgodard|jtseng}@linguist.jussieu.fr [email protected] We present a descriptive overview of the uses of the French prepositional forms à and de and the properties of the constructions they appear in. The complexity of the data excludes a unified syntactic and/or semantic treatment, but significant subregularities can be identified. We show that uses of à and de can be grouped in two sets, one where they pattern with ordinary prepositions, and one where they seem to adopt the category of their complement. We show that this can be accounted for in an HPSG analysis which distinguishes ordinary heads from ‘weak’ heads. Finally, we provide an analysis of the portmanteau forms au, aux, du, and des, which are shown to obey an unusual combination of local and non-local restrictions. Keywords: French, preposition, function word, head, head-driven phrase structure grammar 1. Introduction The forms à and de appear in many different syntactic environments, with a variety of functions. They always combine with a following word or phrase, which cannot be extracted or otherwise omitted. The following combinations are possible: (1) a. de/à + NP1 i. Aller de la gare à l’hôtel. Passer directement des hors d’œuvres aux desserts. ii. Je me souviens de ce poème. Les animaux obéissent à leur instinct. b. de/à + PP Il surgit de derrière l’église. Je pense à après le match. c. de/à + VP[inf] Je me souviens d’avoir lu ce poème. Paul songe à y aller. (2) special cases of de a. de + [le + N'] 2 (‘partitives’) i. (mass nouns) Demander de la bière, du pain, de l’aide. ii. (indef. plurals) Acheter des livres. b. de + N' i. (with negation) Je n’ai pas lu de journal. ii. (with quantifiers3) Beaucoup de langues régionales. iii. (variant of 2a with pre-nominal modifier) Acheter de (très) beaux tableaux. c. de + AP/VP[pass] i. Quelqu’un de très fiable, rien de plus intéressant. ii. J’ai encore trois jours de libres. Il y en avait deux de cassés. d. de + Adv Quelque chose de mieux, une page de plus. 1 This class also includes a small number of de+N' constructions (changer de nom, servir de guide), which do not pattern with the N' cases identified below in (2b). 2 This construction is traditionally taken to involve a complex determiner de-LE instead of de plus a definite NP. The weak head analysis introduced below permits a compositional analysis while retaining the intuition that the complete phrase is an NP. 3 Quantifiers also combine with de+NP structures of type (1a)—e.g., beaucoup de ces maisons—which do not have the same properties as this de+N' pattern. ACL-SIGSEM Workshop on the linguistic dimensions of prepositions Toulouse, France, 2003 (3) special cases of à a. à + VP[inf] with missing complement Un livre à lire absolument, plus facile à dire qu’à faire. b. à + [(le) + N'] Un oiseau à bec rouge, pâtes italiennes aux œufs frais. (4) ‘haplology’ of (1a) and (2a) Avoir besoin d’aide, parler de choses sérieuses. We see that the two forms combine with nominal, verbal, and prepositional phrases (and adjectival and adverbial elements for de). In nominal combinations à and de can combine with either a full NP or an N' (with no specifier). In the remainder of this paper we develop an analysis which partitions the uses of à and de into two sets, which we call oblique and nonoblique uses. In oblique uses, illustrated in (1a,b) and (4), à and de pattern with ordinary prepositions. On the other hand, in nonoblique uses, illustrated in (1c), (2) and (3), they differ from other prepositions in a number of ways, thus calling for a special analysis. The syntactic data justifying the distinction between the two types of uses is discussed in section 2. Section 3 presents the descriptive generalizations, and shows that while a number of different classifications of preposition uses have been proposed in the litterature, none corresponds to the empirical observations. Section 4 offers an HPSG analysis of the main data. We conclude by sketching an account of the puzzling distribution of ‘portmanteau’ forms alternating with à and de. 2. Syntactic properties In this section we examine the syntactic behavior of constructions involving à and de, paying particular attention to properties that distinguish two subclasses (‘oblique’ and ‘nonoblique’) of de+nominal constructions. We show that cases where à/de combines with a PP pattern with oblique uses, whereas cases where à/de combine with an VP or AP pattern with nonoblique ones. 2.1 Portmanteau forms One well-known property of à and de is their interaction with the definite article le (and the plural les) to yield the contracted or portmanteau forms du, des, au, and aux (vs. de la and à la with the feminine singular article). This phenomenon is completely regular for all instances of à/de+NP where the NP can start with a definite article (1a, 2a, 3b above)—that is, it is independent of the semantic and grammatical function of the combination. On the other hand, à and de do not give rise to portmanteau forms when they combine with a VP, although the accusative clitics le and les are superficially identical to the forms that trigger contraction in nominal contexts: (5) a. Je cherche à le joindre / *au joindre b. J’essaie de les vendre / *des vendre This might be taken as evidence for distinguishing pre-verbal and pre-nominal à and de, but one could also argue that the pronominal clitics le and les behave differently from the definite article le/les, disallowing contraction. (This is impossible to test directly, since NPs cannot begin with a clitic and VPs cannot begin with an article.) We will provide a detailed discussion and analysis of portmanteau forms in à/de+NP constructions in section 5. 2.2 Pronominal clitics Here we review some well-known data concerning the realization of clitic pronouns corresponding to NPs and VPs marked by à and de. With certain verbs, à-NP complements alternate either with dative (lui/leur) or locative (y) clitics, depending on animacy (6a,b); with other verbs only one clitic is possible irrespective of animacy (6c,d): (6) a. Paul ressemble à Marie lui ressemble (+human) b. Ce tableau ressemble à une tâche y ressemble (–human) c. Paul pense à Marie/à son travail *lui pense / y pense d. Paul a donné un nom à l’enfant/au bateau Paul lui/*y a donné un nom. For à-VP[inf] complements, the clitic y is often possible, but not with all verbs [Gross, 1975]: (7) a. Je songe à partir J’y songe b. Je cherche à venir *J’y cherche VP complements never alternate with lui (or any other dative form). Nominal phrases marked by de can always alternate with e n, whatever their other properties:4 oblique complements (independent of the semantics of de, 8a-c), constructions with negation (8d), constructions with quantifiers (8e), partitives and plural indefinites (8f,g). (8) a. Je viens de Londres J’en viens b. Je me souviens de ce poème Je m’en souviens c. Je veux changer d’hôtel Je veux en changer d. Je n’ai pas d’argent Je n’en ai pas e. J’ai trop de travail J’en ai trop f. Tu as bu de la bière Tu en as bu g. Tu as acheté des livres Tu en as acheté The clitic en is available for de-VPs, except when they alternate with direct NP objects [Gross, 1975; Huot, 1981]: (9) a. Je rêve de venir demain / de cela / *cela J’en rêve b. Venir demain, Paul en rêve. (10) a. Je promets de venir / *de cela / cela *J’en promets b. Venir demain, Paul le promet. The VP data in (8), (9), and (10) indicate that à/de-marked VP[inf] complements do not give rise directly to clitics. Instead, the main verb selects the form le, y, or de according to its NP or PP complementation frame (if it has one). A VP[inf] can be anaphorically linked to this nominal or prepositional clitic.5 2.3 Syntactic functions VPs introduced by à and de are essentially complements: of V (songer à, rêver de partir), of A (facile à lire), of P (avant d’arriver), or of N (fer à repasser). They can also function as modifiers (un conseil à suivre, Il est tout rouge d’avoir couru, des prix à tomber par terre) and subjects (D’apprendre sa venue ne m’a pas fait plaisir). A- and de-NPs also have a wide range of complement and modifier functions. In addition, nominal phrases introduced by de (corresponding to type 2a in the list above) can appear in subject position: (11) a. [Des bijoux] ont été volés b. [De la bière bien fraîche] me ferait du bien. c. *[Sous le lit] est un endroit idéal pour se cacher. 4 There are well-known restrictions on adnominal en, depending on the function of the NP it is dependent on. See e.g. [Milner, 1978]. 5 The same remark applies to the analysis of cleft constructions. The form of the fronted element is determined by the NP/PP subcategorization properties of the main verb: (i) a. Il a osé partir Ce qu’il a osé, c’est partir b. Il a exigé de partir Ce qu’il a exigé, c’est de partir / *Ce dont il a exigé, … c. Il a rêvé de partir Ce dont il a rêvé, c’est (de) partir / *Ce qu’il a rêvé, … d. Il a songé à partir Ce à quoi il a songé, c’est (à) partir / *Ce qu’il a songé, … As indicated in (11c), PPs normally cannot be used as subjects in French. This suggests strongly that de in (11a,b) is not a preposition.6 2.4 Extraction and mobility A fundamental contrast can be observed between à/de-VPs and à/de-NPs with respect to extraction:7 (12) a. Voici les livres que je cherche [à lire] / que je rêve [de traduire] b. *Voici les livres dont je pense [à l’auteur] / dont je rêve [de l’auteur] Extraction out of NPs is impossible in cases like (12b) — corresponding to types (1a) and (4) above — which we referred to earlier as oblique à/de-NPs. These uses of à and de pattern with other French prepositions: *le livre dont j’ai déjeuné [avec l’auteur] / dont je compte [sur l’auteur]. It is possible, however, to extract out of nonoblique de-N' or de-NP structures (types 2a and 2b): partitive/indefinite nominals (13a), in negative contexts (13b), or with a quantifier (13c). Example (13d) shows that nonoblique de-phrases pattern with unmarked NPs. (13) a. Voici l’auteur dont [de nombreux livres] sont ici en vente b. Voici un auteur dont je n’ai pas lu [de livre] c. Voici l’auteur dont j’ai lu beaucoup [de livres] d. Voici un auteur dont j’aime [les livres] (mais pas les poésies) Turning now to extraction of the whole à/de-XP, we observe the same pattern. We compare the behavior of full PPs, ordinary NPs, and à/de-XPs in wh-relativization (14) and topicalization (15). We find again that oblique uses of à and de pattern with ordinary PPs, while nonoblique uses pattern with ordinary NPs. (14) a. Voici un auteur [à qui] parler / [de qui] parler (oblique) b. Voici un auteur [pour qui] voter (ordinary PP) c. *Voici un livre duquel / dont je n’ai pas (nonoblique: negative context) d. *Voici des livres desquels je lis (nonoblique: indefinite/partitive) e. *Voici des livres dont j’ai beaucoup (nonoblique: with quantifier) f. *Voici les livres lesquels j’ai lu (unmarked NP8) (15) a. [À Marie] / [De Marie], je parlerai (oblique) b. [Pour Marie], je voterai toujours (ordinary PP) c. *De livres, je n’ai pas acheté (nonoblique: negative context) d. *De très beaux tableaux, j’ai vu à cette expo (nonoblique: indefinite/partitive) e. *De romans policiers, j’ai lu trop (nonoblique: with quantifier) f. *Marie, je verrai. *Ce livre, j’ai acheté. (unmarked NP9) Once again, VPs marked by à or de pattern with nonobliques. Wh-relativization is excluded because there is no appropriate wh-word, and topicalization is blocked too (16).10 (16) a. *[A quitter ce quartier], je songe depuis longtemps. b. *[De quitter ce quartier], je rêve depuis longtemps. 6 Note also the number and gender agreement in (11a,b). This contrasts clearly with the situation in English, where subject PPs are possible but do not trigger agreement (Between the trees is/*are a good place to rest) 7 Notice that extraction out of bare VPs is possible in French (i), while extraction of VPs introduced by other prepositional forms is subject to speaker’s variation (ii): (i) Voici les livres que j’aimerais [lire] / que je dois [traduire] (ii) %Voici les livres que j’insiste [pour lire] / que je suis partie [sans lire] 8 Unmarked NPs can of course be relativized (e.g., les livres que j’ai achetés) but relative que (like nominative relative qui) is a complementizer, not a wh-word [Kayne, 1975; Godard, 1988]. This construction therefore does not belong with the cases of wh-relativization in (14). 9 We leave aside the highly restricted cases of NP topicalization (e.g., Le chocolat, j’adore). 10 Extraction of à/de-VP[inf] is not impossible in general, however; see the cleft sentences in (i-ii): (i) C’est [à partir] que je songe. (ii) C’est [de partir] que je rêve. 2.4 Wide scope over coordination With certain semantic constraints, à and de can have wide scope over a coordination of NPs in oblique contexts: (17) a. Je pense à [Paul et Marie] (assis côte à côte près de la cheminée) b. Il donne beaucoup à [ses frères et ses sœurs] c. ?J’ai besoin de [cette farine et cette levure] pour mon gâteau The same is true for à and de before PP: (18) a. Il pense toujours à [avant la guerre ou après la libération] b. Il revient de [chez Paul ou chez Marie] In contrast, à and de cannot have wide scope over a coordination of phrasal VPs[inf] [Abeillé and Godard, 1997]; they must be repeated. Note that a single à or de can mark a VP containing a coordination of lexical Vs, as shown in (20). (19) a. Je cherche à lire ce livre et *(à) le traduire pour demain b. Je rêve de lire ce livre et *(de) l’expliquer à mon fils (20) a. Je cherche à [lire et traduire] ce livre pour demain b. Je rêve de [lire et commenter] ce livre à mon fils In contrast to the oblique à/de-NPs examples in (17)-(18), nonoblique uses of de generally cannot take wide scope: (21) a. Pour ce gâteau, il faut de la farine et *(de) la levure b. Il y avait de très grands tableaux et *(des) magnifiques statues (22) a. Il y avait sur la table beaucoup de pain et *(de) vin b. Sur la table, il y avait beaucoup de livres et ?(de) cahiers In these examples, de must be repeated, except (marginally) in the case of a coordination of lexical Ns in the plural (24b). Finally, we give some examples of de-marked APs (type 2c) and adverbs (type 2d): (23) a. quelqu’un de [grand et fort] b. quelqu’un de bon en maths et (de) fort en gym (24) a. quelque chose de [plutôt bien ou plutôt mal] b. une page de plus ou *(de) moins The results are less clear in these cases; wide scope seems possible in many cases, but there is significant variation among speakers. 2.5. Interim conclusion The data presented in this section motivates a two-way partition of à/de-marked phrases. The relevant data is summarized in table (25). (25) (i) (ii) (iii) (iv) (v) 11 “oblique” has the distribution of extraction out of marked phrase topicalization of à/de-phrase wide scope over coordination of phrases semantic contribution “nonoblique” à/de+VP, de+AP/Adv, à/de+NP/PP de+ N'/NP PP VP/AP/NP no yes yes no yes (mostly) no 11 sometimes We do not propose a full semantic analysis of à and de, but we do make the claim that the presence or absence of semantic content is not correlated with the other properties defining the oblique vs. nonoblique distinction in Table (25). See section 3.1 for discussion. 3. Previous analyses Most existing studies only focus on a subset of uses of à and de, without attempting to give a complete account of the two forms. Obviously, we cannot do justice to all previous proposals here. We briefly mention a few authors whose work is most relevant to the present study. 3.1. Semantic analyses In previous litterature, à and de figure prominently in semantic arguments, the central question being whether these prepositions are contentful or semantically empty (or ‘colorless’; see [Spang-Hanssen, 1963] for an early discussion). Although this is not the place for a thourough discussion of the semantics of à and de, a few important remarks are in order. First, some uses of à and de are indisputably semantically potent. For instance, when they combine with the copula as in (26), one must assume that the locative, directional, or possessive relation originates in the preposition—the only alternative being the postulation of three distinct copulas, each subcategorizing for a particular preposition. (26) a. Paul est à/de Paris. b. Ce livre est à Marie. Second, not all semantically potent uses of à or de express the same relation; for instance it seems highly unlikely that locative and possessive à in (26) can be given a unified semantics. Third, some uses of à and de are clearly semantically empty. For example, in nominal complementation structures [N de-NP], de is compatible with every available semantic role.12 These simple observations teach us two important lessons: first, there is no hope of providing a single lexical entry covering all uses of à—which can have a number of unrelated senses, or be empty—or de. Second and most importantly, the contentful/contentless distinction between preposition uses does not correlate simply with the oblique/nonoblique distinction we motivated in section 2: it is clear that some oblique uses are contentless (e.g. uses of de in complements of N).13 3.2. Syntactic analyses [Milner, 1978] draws a distinction among à- and de-NPs (extending an earlier suggestion by [Kayne, 1975] for à-phrases). He analyzes dative à-NPs (those that alternate with dative clitics) and genitive de-NPs (those that do not alternate with de lui—e.g., possessives *la maison de lui, and relational noun arguments *l’ami de lui) as NPs. In these cases à and de are case markers. Other uses of à and de (those allowing de lui or à lui—e.g., un tableau de lui, agent or theme) are analyzed as prepositions. Milner does not account for the fact that the coordination data and the portmanteau forms are the same for both case markers and prepositions. [Huot, 1981] distinguishes two uses of de before VP[inf]: it is either a preposition, when the VP[inf] alternates with a de-NP and pronominalizes as en, or a complementizer, when the VP[inf] alternates with a direct object and pronominalizes as le. On the other hand, à is always treated as a preposition. Huot does not account for the fact that extraction is always possible out of à/de-VP[inf], while it is impossible out of PPs in general. With regard to the morphosyntactic status of à and de, [Miller, 1992] argues for a treatment in terms of affixes, not words. Since à and de are hosted by a variety of categories, they must be analyzed as phrasal affixes—i.e., as prefixes that combine with the first word of the 12 In other cases the appropriate analysis is more disputed, but it is difficult to motivate a semantic treatment. For instance, despite repeated attempts by numerous researchers (see e.g. Cadiot 1997 and references cited therein), no convincing analysis of à and de introducing VPs (1b above) as semantically contentful items has yet been offered. À- and de-marked complements of A (un livre facile à lire vs. il est facile de lire ce livre) would also appear to resist any semantic explanation. 13 As mentioned above, de within the NP is compatible with practically all semantic roles and relations (including identity in la ville de Paris and possession in le chat de Paul). But this semantic content must be associated with the construction as a whole, and not the form de itself, since de alone does not trigger the relevant interpretation [Godard, 1992]. For example, in copular constructions like (26), de cannot express possession: *Ce nounours est de ton bébé. NP. Moreover, Miller assumes that à and de have a case-marking function; for example, the combination à+NP is an NP with a particular case value induced by à. In order to differentiate datives from other à-NPs, he assumes two different case values: à1 and à2. He does not discuss à and de before VP[inf], and so cannot account for differences between marked nominal and verbal phrases. Within HPSG, [Abeillé and Godard, 1997] analyze à and de before VP[inf] as members of the syntactic category marker, with no semantic contribution. They consider it a general property of markers in French that they cannot have wide scope over a coordination of phrases. [Abeillé et al., 1996] extend the analysis to infinitival à relatives (un livre à lire) and à-VP[inf] after “tough” adjectives (difficile à lire). [Abeillé and Godard, 1999] extend the analysis further to dative à-NPs, which they analyze as marked NPs. They provide no account for wide scope of à before NP coordination, and their analysis cannot accommodate semantically contentful uses of à. 4. Proposed HPSG analysis In this section we provide an HPSG analysis of à and de which makes explicit what makes oblique and nonoblique uses different, and what they have in common. The analysis relies crucially on the novel concept of a weak head. 4.1 Oblique uses: true prepositions We treat à and de in their oblique uses as prepositions — i.e., as syntactic heads of category P, selecting a complement and projecting a PP. We propose the following lexical entries, subtypes of the general lexical type prep-word subsuming all French prepositions:14 (27) a. à b. de CAT SLASH HEAD prep à1 ∨ à2 MARKING NP CASE COMPS ∨ PP internal CAT HEAD prep de MARKING CASE NP COMPS MARKING ∨ PP internal unmarked SLASH The HEAD value (inherited from prep-word) indicates the syntactic category (preposition), which percolates to all projections of these lexical entries. The MARKING attribute will be discussed in section 4.3 below. We assume (at least) three CASE values for nouns and nominal projections in French: nominative, accusative, and internal. NP complements of prepositions bear internal case, which must be distinct from accusative because some lexical items (the bare quantifiers tous, deux, etc.) can be complements of prepositions but not of verbs [Abeillé and Godard, 1999]. Prepositions themselves do not have case (and consequently, neither do PPs). The empty SLASH set in (27) prevents extraction of and subextraction out of the preposition’s complement (because heads amalgamate the SLASH information of all their dependents [Sag, 1997]). This feature is inherited from the general lexical type prep-word, because all PPs in French are extraction islands. This accounts for line (ii) of Table (25) for oblique uses of à and de. The data in (14)-(16) above can be handled with constructional constraints. For example, the following implication requires the filler element in a wh-relative clause to be a PP (as in 14a,b): 14 One case that is not covered by (27b) is (4), which is usually analyzed as involving haplology [Gross, 1967; Miller, 1992]. As [Jones, 1996] notes, this can not be a phonological phenomenon, cf. On parlait de De Gaulle. We treat these cases by assuming a special entry for de with an N' complement. (28) wh-relative-cl & head-filler-ph [ NON-HD-DTRS < PP > ] A similar condition can be formulated for the constructions in (15), allowing PPs but not NPs to be topicalized. As for the two remaining properties shown in Table (25), à and de, in their oblique uses, are free to combine with a coordinated complement (like other ordinary lexical heads), and they are certainly free to make a semantic contribution (although they are not required to). 4.2 Nonoblique uses: Weak heads To account for the properties of nonoblique uses, we appeal to the notion of ‘weak head’, which replaces the syntactic category of marker in classical HPSG [Tseng 2002].15 A weak head is a lexical head that shares its syntactic category and other HEAD information with its complement. Recall that the HEAD value in HPSG includes the part of speech (encoded as the type of the value: noun, verb, adj, prep, etc.) and certain features appropriate for each part of speech (such as CASE, VFORM , PFORM, MOD ). The sharing of HEAD values is indicated by the label 1 in the constraint in (29a).16 The weak head à is subject to the further constraint in (29b). (29) a. 1 HEAD MARKING weak-head ⇒ VALENCE b. marked COORD weak-à ⇒ weak-head ∧ 2 SUBJ SPR HEAD MARKING COMPS SUBJ COMPS − COMPS 1 unmarked 2 ⊕3 3 VP[inf ] If the complement of a weak head à or de is a VP[inf], then the weak head itself will have the HEAD value [verb, VFORM inf], and it will project the same kind of phrase. For instance, in (penser) à téléphoner, we analyze à as a weak head, and so the phrase à téléphoner is a VP[inf]. This means that it should have the same grammatical properties as other VPs, and this prediction is confirmed, for example by the extraction data (cf. note 7). The constraint in (29a) also specifies that weak heads inherit the SUBJ and COMPS requirements of their complements. In this example, téléphoner has a non-empty SUBJ list (since its subject is not syntactically realized). The weak head à inherits this SUBJ value, and consequently the marked VP à téléphoner has the same non-empty SUBJ specification. This is crucial for the analysis of argument control (by the verb penser). The COMPS list of téléphoner, on the other hand, is empty, but this is not always the case for VP complements of the weak head à (cf. the cases in 3a above, analyzed in terms of complement inheritance by [Abeillé et al., 1996]). 15 An alternative analysis for similar data is proposed by [Van Eynde, to appear], which distinguishes major and minor prepositions. Like weak heads, minor prepositions introduce a specific marking value and do not contribute any part of speech information to the phrase. Unlike weak heads, they do not have the HEAD function, but a nonhead FUNCTOR function they share with other nonhead selectors. 16 Note that weak heads differ from functional heads in LFG or GB. Although a weak head’s category is underspecified in the lexicon, in any given syntactic context, it has a completely ordinary syntactic category, e.g. N or V. It is important to emphasize that when a weak head inherits a HEAD value of type verb or noun, it does not actually ‘become’ a verb or a noun (i.e., a lexical object of type noun-word or verb-word). Thus it is not surprising that it behaves very differently from a normal noun or verb with respect to complementation, inflection, etc. Similarly, if the weak head de selects an N' or NP complement (note that à is never a weak head in this environment), it will inherit the value [HEAD noun], and project an NP. This explains why these de-marked phrases allow extraction (cf. 13): de is not a preposition in these cases, and so the French PP-island constraint (enforced by [SLASH { }] in 27) is not relevant. The weak head treatment also accounts for the NP-like distribution of nonoblique de-phrases, in particular their occurrence in subject position (see 11, above). In these examples, de is a weak head, and therefore the phrases des bijoux and de la bière bien fraîche are NPs, and completely acceptable as subjects. Note that the weak head de (in contrast to its prepositional counterpart in 27b) does not constrain the CASE value of its complement (which it inherits as its own CASE value). This allows de-marked NPs to appear in nominative, accusative, and internal cases. Valence sharing between de and its N'/NP complement is generally very simple: the COMPS list is always empty, and the SUBJ list is non-empty only in predicative constructions (e.g., considérés comme [des héros]). There is no sharing of the SPR (specifier) list indicated in (29a); the weak head’s SPR list is always empty. This is crucial for the analysis of de+N' phrases, as in (2b.i), repeated here: (2) b. i. Je n’ai pas lu [de journal]. Here, journal has a non-empty SPR list (like all common nouns in French), but the weak head de does not inherit this feature. The resulting phrase de journal is not an incomplete N', but a fully saturated NP, which is what the verb lire requires. The MARKING specifications in (29a) are discussed in the next section. One last detail is the constraint [COORD –] on the complement in (29a), which prevents weak heads from taking wide scope over a coordinated structure. 4.2 Grammatical marking Table (25) shows one way of classifying uses of à and de, which we analyze in terms of a categorial distinction (preposition vs. weak head). But we need to be able to handle phenomena (in particular cliticization) where prepositions and weak heads pattern together. To do this we rely on the MARKING specification. The MARKING feature is familiar from previous work in HPSG; this is the feature that allows phrases containing an explicit marker (e.g., a complementizer) to be distinguished from unmarked phrases. We adopt the proposals of [Tseng, 2002], simplifying the Marking Theory of standard HPSG. In this analysis, the syntactic category marker and the head-marker-str of [Pollard and Sag, 1994] are dispensed with by analyzing markers as weak heads, which provide a marking value to the phrase they head while taking an unmarked complement.17 We assume that Ns and Vs are [MARKING unmarked] in the lexicon, and that each preposition introduces a specific marked value. Thus MARKING takes over the role played by PFORM in earlier HPSG, making it possible for a head to select a complement headed by a specific preposition. For the preposition à, two distinct MARKING values are posited, accounting for the difference in cliticization possibilities: à1corresponds to the clitic lui, and à2 to the clitic y. This approach to marking allows for an account of preposition selection where one and the same preposition is selected on the basis of its semantics in some instances, and on the basis of its lexical identity in others [Tseng, 2001]. Compare the partial lexical entries for the verbs aller and penser, assuming the semantics of [Bonami, 1999] for locative prepositions: aller selects any locative expression of the right semantic type, whereas penser selects a specific MARKING value. (30) a. aller b. penser COMPS 17 KEY locstat-rel COMPS MARKING à2 Note that MARKING is not technically a head feature, since it is not shared between the weak head and its complement. Weak heads à and de also introduce [MARKING marked] values, although they are not prepositions. Moreover, corresponding weak head and prepositions take the same marked value. This is a significant advantage of our analysis, because it allows a very simple explanation of the fact that all de-phrases alternate with the same form en (cf. 8). In our account, (se souvenir) de Marie is a PP, (beaucoup) de livres is an N'[acc], (boire) de la bière is an NP[acc], and (envie) de dormir is a VP, but they all have the same MARKING value, and so they all become en under cliticization. We know of no previous proposal capable of capturing this generalization, which involves phrases that are otherwise so diverse with respect to all other syntactic and semantic features. Note also that the sharing of the same marking value for the weak head and corresponding preposition allows for underspecification of subcategorization information, at least in some cases: for instance the lexical entry in (31b) allows both for PP[à2] complements headed by the preposition à and for VP[à2] complements headed by the weak head à.18 5. The analysis of portmanteau forms In this section we return to the analysis of the portmanteau forms mentioned in §2.1. This phenomenon is traditionally taken to be the result of a surface phonological process ‘contracting’ two adjacent forms into a simpler (idiosyncratic) form: de + le du, à + les aux, etc. The data introduced in (31) below, first brought to light by [Miller, 1992], suggests that the appearance of portmanteau forms is not uniquely conditioned by linear adjacency of the relevant forms, and thus cannot be taken to be a simple surface phenomenon: (31) a. J’ai parlé à la mère et la fille b. *au père et la mère c. *à le père et la mère d. *à la fille et le garçon e. à la fille et l’autre garçon This set of examples involves the preposition à, but the facts are completely parallel for the preposition and the weak head de. Furthermore, the distribution of the plural article les is identical to that of the singular, non-elided form le (e.g., *à la mère et les filles). We can make the following descriptive generalization about the data in (31): à and de cannot combine with a coordination of NPs if any of the conjuncts begins with a contracting form of the definite article (le or les).19 In other words, the only possible forms of the article in any conjunct are la and l’, and we never see any portmanteau forms. So there is an adjacency condition—failure to contract à and de immediately before le or les leads to ungrammaticality (31c)—but the non-adjacent cooccurrence of uncontracted à or de and the form le or les in a non-initial conjunct is also ungrammatical (31d). In this example, the portmanteau form au would have been required, had the preposition à been repeated in the second conjunct (producing the grammatical coordination à la fille et au garçon). It is curious that this ‘virtual portmanteau’ condition should play a role even when à takes wide scope and is not present in the second conjunct. Part of the coordination data is easily explained away if one assumes a lexicalist treatment of portmanteau forms, in which au, aux, du, and de are taken to be lexical items—that is, they are not the result of any post-lexical operation on the syntactic combinations à+le, de+les, and so on. They are syntactic heads (prepositions, or weak heads for some uses of du and des) which select an N' (i.e., an NP lacking a determiner) complement.20 This accounts for (31b) right away, 18 Note that the lexical entries for à in (27a) and (28b) allow for no other type of à-marked phrase. It is not possible to underspecify subcategorization information in the same way we did here for de marked complements: since there are so many types of de-marked XPs, stating e.g. that parler takes a [MARKING de] complement would overgenerate considerably. 19 Assuming in some intuitive sense that (31c), containing le, is the ‘underlying’ form of (31b). 20 The agreement features and (when appropriate) the semantic contribution of the definite article are also built into the lexical entries of portmanteau forms. because we have either a coordination of PP and NP (*[au père] et [la mère]) or a coordination of N' and NP (*au [père] et [la mère]), neither of which is allowed. A problem that the lexicalist treatment does not solve immediately is the blocking of analytic forms such as *de le garçon. For these cases, we need a way to distinguish the portmanteau triggers le and les from the non-triggering forms la and l’. We simply introduce a boolean attribute LE, and we assume that NPs containing le and les are [+LE], while all other NPs are [–LE] (or underspecified for LE). Then, in the lexical entries of à and de, we require the complement to be [–LE], preventing non-contracted sequences as in (31c). The portmanteau forms, on the other hand, require no particular LE value on their complements (since the complement is an N'). To account for the appearance of the appropriate LE value on the NP, we assume that LE is a left edge feature [Miller, 1992; Tseng, 2003]: the value of LE on an NP is identical to that of the NP’s leftmost daughter.21 This accounts directly for the fact that à and de can occur to the left of an NP of the form tout le(s) N. (32) a. Jean parle à tout le monde. b. Je m’occupe de tous les problèmes. In (33a), the determiner le is not the leftmost word in the full NP; thus even if the intermediate NP [le monde], is [+LE], the full NP is [–LE], inheriting its value from tout. Finally, we need an account of the facts in (31c,d): à and de cannot take as a complement a coordination one of whose conjuncts is [+LE]. This can only be done by adopting some account of feature percolation in coordinate structures, a topic that is subject to much debate in current HPSG. In the following we simply sketch how the approach in [Sag, 2003] can be used to account for the current data. The idea in [Sag, 2003] that is crucial for our purpose is the following: a lexical entry can not only put an upper bound, but also a lower bound on the possible values of a feature. For instance, to account for the famous be a Republican and proud of it, Sag assumes the head value hierarchy in (33a). Republican constrains its head value to be higher than noun in the hierarchy, and proud constrains its head value to be higher than adjective. HEAD values are identified in coordination: thus the head value of a Republican and proud of it must be higher than nominal, which is compatible with the selectional requirements of be. b. LE values (33) a. HEAD values pos nominal noun adj boolean verbal prep plus minus verb In a similar vein, we can account for the data in (31c,d) by assuming that le and les constrain their LE feature to be higher than +, whereas all other words constrain their LE feature to be higher than – (35); furthermore we assume that LE is among the (numerous) features whose values are identified in coordination. Thus the coordinated NPs in (31c,d) can only be [L E boolean], which is incompatible with the selectional requirements of à and de (36). (34) a. le: [LE 1 , 1 plus] b. la: [LE 1 , 1 minus] (35) a. à : [COMPS [LE 1 , minus 1 ]] b. de : [COMPS [LE 1 , minus 1 ]] 6. Concluding remarks We have presented an empirical overview of constructions involving the forms à and de in French, and we have offered a number of proposals for their analysis in the framework of HPSG. Our account cross-classifies phrases containing à and de in various ways: according to the syntactic category of the marked element, the syntactic identity of à/de (a true preposition or a 21 Much of the empirical motivation for the EDGE features in [Miller, 1992] has been challenged in subsequent literature, including the present article which analyses as heads à and de which were phrasal (EDGE) affixes under Miller’s analysis. However the EDGE mechanism remains independently needed e.g. for the treatment of liaison and elision in French [Tseng, 2003]. ‘weak head’), or the M A R K I N G value introduced by à/de. This multi-faceted treatment accommodates most, if not all, uses of à and de in French and their properties with respect to a wide range of phenomena, including extraction, coordination, and finally the well-known but problematic portmanteau forms. References Abeillé, A., and D. Godard. 1997. “The syntax of French negative adverbs.” In D. Forget, P. Hirschbuhler, F. Martineau, M.-L. Rivero (eds), Negation and Polarity, pp. 1–27. Amsterdam: J. Benjamins. Abeillé, A., D. Godard, P. Miller, and I. A. Sag. 1996. “French Bounded Dependencies”. In S. Balari et L. Dini (eds), Romance in HPSG, 1–54. Stanford: CSLI Publications. Abeillé, A., and Danièle G. 1999. “A Lexical Approach to Quantifier Floating in French”. In A. Kathol, J.-P. Koenig, and G. Webelhuth (eds), Lexical and Constructional Aspects of Linguistic Explanations, 81–96. Stanford: CSLI Publications. Bonami, O. 1999. Les constructions du verbe. Le cas des GP argumentaux. PhD dissertation, Université Paris 7. Cadiot, P. 1997. Les prépositions abstraites en français. Paris: Armand Colin. Godard, D. 1988. La Syntaxe des relatives en français. Paris: CNRS Editions. Godard, D. 1992. Extraction out of NP in French. Natural Language and Linguistic Theory 10: 130–179. Gross, M. 1967. “Sur une règle de cacophonie”. Langages 7: 105–119. Gross, M. 1975. Méthodes en syntaxe. Paris: Hermann Huot, H. 1981. Constructions infinitives du français : le subordonnant de. Geneva : Droz. Jones, M. A. 1996. Foundations of French syntax. Cambridge: Cambridge University Press. Kayne, Richard S. 1975. French Syntax: The Transformational Cycle. Cambridge, MA : MIT Press. Kim, J.-B., and S. Wechsler (eds). 2003. Proceedings of the 9th International Conference on HPSG. Stanford: CSLI Publications. Miller, P. H. 1992. Clitics and Constituent in Phrase Structure Grammar. New York : Garland. (PhD Dissertation, University of Utrecht, 1991.) Milner, J.-C. 1978. De la syntaxe à l’interprétation. Paris: Le Seuil. Pollard, C. and I. A. Sag. 1994. Head-driven Phrase Structure Grammar. Chicago: The University of Chicago Press and Stanford: CSLI Publications. Sag, I.A. 1997. “English Relative Clause Constructions”. Journal of Linguistics 33: 431–484. Sag, I. A. 2003. “Coordination and Underspecification.” In J.-B. Kim and S. Wechsler (eds), pp. 267–291. Spang-Hanssen, E. 1963. Les Prépositions Incolores du Français Moderne. Copenhagen: G. E. C. Gads Forlag. Tseng, J. 2001. The Selection and Representation of Prepositions. PhD Thesis, University of Edinburgh. Tseng, J. 2002. “Remarks on Marking.” In F. van Eynde, L. Hellan, and D. Beermann (eds), Proceedings of the 8th International Conference on HPSG, pp. 267–283. Stanford: CSLI Publications. Tseng, J. 2003. “EDGE features and French liaison.” In J.-B. Kim and S. Wechsler (eds), pp. 313–333. Van Eynde, Frank. to appear. “Minor adpositions in Dutch”. Journal of Comparative German Linguistics.
© Copyright 2024 Paperzz