Mapping function to form : Adjective positions in French

Lingvisticæ Investigationes 29:1 (2006), 43–60.
issn 0378–4169 / e-issn 1569–9927 © John Benjamins Publishing Company

Abstract

The position and meaning of Modern French adjectives is discussed in the generative syntax framework. The classical analysis of Cinque (1995), by N-raising to some functional position, is rejected for a number of reasons, notably its inability to account for changes in position since the Old French period. After a brief discussion of some typological and diachronic facts, two ‘semantic’ analyses are considered, those of Bouchard (1998) and Larson (1998). Based on Larson’s mapping model, the present study proposes an analysis of the position and meaning of adjectives in French in terms of movement of ‘restrictive’ and ‘evaluative’ adjectives to the pre-N position and the presence of logical operators of existence and genericity within DP. This analysis is supported by both semantic and syntactic facts: constraints on the determiner, as well as on modification, stress and coordination, defective semantic features.

Résumé
La position et le sens des adjectifs en français moderne sont examinés dans le cadre de la grammaire générative. L’analyse classique de Cinque (1995) par montée de N à une position
fonctionnelle est rejetée pour plusieurs raisons, notamment son incapacité à rendre compte des variations de position constatées depuis l’époque ancienne. Après une brève discussion
typologique et diachronique, deux analyses « sémantiques » sont considérées, celles de Bouchard (1998) et de Larson (1998). A partir du modèle proposé par ce dernier, une hypothèse
est proposée qui tente d’expliquer la variation de sens des adjectifs français, d’une part par le déplacement de certains adjectifs ‘restrictifs’ ou d’évaluation subjective vers la position
pré-nominale, d’autre part par la présence d’opérateurs logiques d’existence et de généricité à l’intérieur du SN. Cette analyse est étayée par des faits sémantiques et syntaxiques : contraintes sur le déterminant, impossibilité de modifier, accentuer ou coordonner les adjectifs pré-N, traits sémantiques ‘défectueux’.

Introduction
For Cinque (1995), usually considered the ‘classic’ study of adjective positions in Generative Syntax,[1] the difference in the position of N’s relative to A’s in Romance as opposed to Germanic is due to an N-raising mechanism, which applies in the former but not in the latter. Assuming a universal basic order, derived from Kayne’s (1994) Linear Correspondence Axiom (LCA), all A’s are claimed to be pre-N at D-structure. Cinque claims that adjectives are ordered across languages in a series of functional projections such as Size, Color, Nationality, etc. and that this universal hierarchy is determined by the selectional properties of the various functional heads. Romance N’s are then assumed to raise to an intermediate F0 (since Romance NP’s generally show A-N-A order) to check some semantic feature of N against the features of F0. Germanic N’s are argued to remain in situ, since they are claimed not to possess this feature, thus deriving the surface order A (A) N.

(1) [DP un [FP petit [F′ balloni [FP rouge [NP ti ]]]]]
(2) [DP a [FP little [FP red [NP ball ]]]]

Bernstein (1991, 1993) proposes a similar analysis for Romance adjectives, but argues that N-raising is triggered by the need for French N’s to raise to Number0, as opposed to English Ns which remain in situ, since Number is assumed to be strong in French and weak in English. Within the same theoretical framework, Bouchard (1998, 2002) has taken a radically different position. He proposes abandoning the LCA altogether in favor of a single operation, Associate, which assumes a fundamental asymmetry between Functor and Dependent elements and states that “the functor
precedes or follows the dependent”. Arbitrariness, he argues, is the ultimately economical system. Bouchard claims that word order in DP is then dependent on the semantic relationship between A and N and the distribution of Number morphology.

The present paper will attempt to show that neither of these analyses is adequate to account for adjective distribution in Modern French. In the first section, I will briefly look at some typological data indicating that adjective/noun order across languages is unpredictable approximately 50% of the time. In the second section I will consider diachronic data for French showing that word order has changed from A-N to N-A over the centuries, while at the same time number agreement on nouns and adjectives has almost completely disappeared. Finally, I will return to the analyses mentioned above and attempt to propose a more satisfactory account, not only for adjective-noun order in Modern French, but also for the relative order of adjectives in the pre-N as well as the post-N positions. To do this I will propose a ‘mapping theory’ derived in part from Larson (1998).

1. The Typological Approach
There have been a number of attempts to find a universal rule for adjective distribution across languages. According to Greenberg’s (1966) Universal Principle 18:

« When the descriptive adjective precedes the noun, the demonstrative and the numeral, with overwhelmingly more than chance frequency, do likewise. »[2]

Turning this statement around, one would expect that a language which systematically places demonstratives and numerals before the noun should do the same with adjectives. This is clearly not the case for French, where the vast majority of attributive adjectives follow the noun (see the statistics from Goyens (1994) in the following section), although all determiners must precede it. Hawkins (1983)[3] takes a more flexible approach, proposing a Mobility Principle, according to which certain categories, like adjectives, demonstratives and numerals, move around more easily within DP than others, such as relative clauses or possessives. Finally, Rijhkoff (2002) found the following in a study of 52 representative languages:[4]

• Out of 14 VO languages examined, the modifier5 follows the noun in 8 cases;
• Out of 30 OV languages examined, the modifier precedes the noun in 16 cases;
• Out of 5 ‘mixed’ languages (i.e. where both VO et OV orders are found), two use both the AN order and the NA order, two prefer the AN order and
  only one uses the NA order.

As this study shows, adjective-noun order, according to the various criteria proposed in the literature, is only predictable 50% of the time. The first conclusion then is that one should be careful before making any sweeping claims about adjective distribution. Secondly, as we’ll see in the next section, word order can change over time.

2. The Diachronic Approach
It is generally agreed that Latin was an OV language.[6] According to C. Marchello-Nizia (1995) this continued to be the dominant order throughout the Proto-Romance period. She quotes Richeter (1903) and de Dardel (1989)[7] in support of this idea and gives the following example from an Italo-Romance text written around the 9th century.

(3) Se pareba boues, alba pratalia araba / albo uersorio teneta, negro semen seminaba[8] (he prepared the oxen, plowed the white fields / he held the white plow,
sowed the black seed)

As we can see, verbs follow their objects and nouns follow their adjectives in this sample. What about Gallo-Romance? In the first written text available, the Serments de Strasbourg (ca. 842), likewise verbs follow their objects (4–5 and nouns follow their adjectives[9] (6).

(4) …et podir me dunat (and give me the power)
(5) …in o quid il mi altresi fazet (let him do as much for me)
(6) Pro Deo amur et pro Christian poblo et nostro commun salvament (for the love of God and for the salvation of the Christian people and for
our own)

For Yvon (1946–47:290), this text corresponds without a doubt to “l’état ancien de la langue”, which may be due in part to the fact that it is a near word-for-word translation of the oaths of allegiance sworn in Latin at the time of Charlemagne.[10]

In another text, the Cantilène de Sainte Eulalie, written only a half-century later, the situation is quite different. This time the verb is found in the second position (V2) of the sentence 60% of the time (7–9) and adjectives, numerals and possessives are found in both pre- and post-N position (10–11).

(7) Buona pulcella fut Eulalia (Eulalie was a good girl)
(8) Enz enl fou lo getterent com arde tost (they threw her into the fire like a burning brand)
(9) La polle sempre non amast lo Deo menestier ((brought) the girl not ever to love the service of God)
(10) A-N : mals conselliers; buona pulcella; Deo inimi; Deo menestier; arde tost (evil counselors; good girl; God’s enemies; God’s service; burning brand)
(11) N-A : nom chrestiien; rex pagiens; manatce regiel; figure de colomb (Christian name, pagan king, royal decree, dove’s face)

Goyens (1994) studied approximately 2500 NPs in a 13th century French translation of a Latin legal text. Out of 323 of the NPs from her corpus containing an
adjective or a participle, I found:

• 219 NPs with a modifier in pre-N position;
• 45 NPs with a modifier in post-N position;
• 33 NPs with modifiers in both positions;
• 25 NPs with 2 coordinated modifiers in post-N position;
• 1 NP with 2 coordinated modifiers in pre-N position.

This corresponds roughly to the overall statistics given in Boucher (2003) for classic 12–13th century Old French. Goyens then compares the first translation to a second, 19th century translation of the same text. This time, out of 339 NPs containing an adjective or a participle, I found:

• 39 NPs with a modifier in pre-N position;
• 254 NPs with a modifier in post-N position;
• 3 NPs with modifiers in both positions;
• 43 NPs with coordinated modifiers in post-N position
• 0 NPs with coordinated modifiers in pre-N position.

The facts speak for themselves: while Old French placed adjectival and participial modifiers before the noun in an overwhelming majority of cases, Modern French places them after the noun with equal frequency. This raises a number of questions, such as why the change in position took place, or what explains the overall flexibility of adjective positions over time. Boucher (2003) offers a partial reply to these questions. In the present paper I will concentrate on the 15% of adjectives found in the pre-N position in Modern French. In the following section I will look briefly at the analyses mentioned in the introduction and then offer my own explanation for the facts.

3. Movement versus Mapping
Cinque’s analysis of adjective distribution in Romance assumes the presence of some abstract semantic feature associated with the various adjective classes (Size, Color, Age, etc.) which somehow triggers N-raising in Romance but not in Germanic languages. All adjectives are generated in the specifier positions of the functional projections associated with each of the classes and agree with the abstract F0 associated with the head position. Romance nouns possess a semantic feature associated with one of these heads, say Size, and must raise to F0 to check it. Germanic nouns, for some mysterious reason, do not possess such a feature in their make-up and therefore need not move.

I find this analysis unsatisfactory for a number of reasons. First, the feature he posits seems artificial and stipulative. Why should Romance nouns possess such a feature but not Germanic nouns? And why should they possess a Size feature and not a Color, or Nationality feature? Secondly, what in this analysis can explain the change in adjective positions in the transition from Old to Modern French observed above? As for the role of Number in triggering N-raising, as proposed by Bernstein, the diachronic facts clearly argue against such an approach. If indeed it is the strength of number morphology on French nouns which forces them to raise to the left of most adjectives, then why is it that Old French nouns, which had fully active number and case morphology, ‘remained in situ’, in the terms of Bernstein and Cinque, 85% of the time, while Modern French nouns, which have almost completely lost active number morphology, are forced to raise 85%
of the time?

The analysis proposed in Bouchard (1998, 2002) seems more satisfactory but still leaves a number of questions unanswered. For Bouchard, the internal distribution of noun phrases depends on two factors: (a) The nature of the semantic relation between A and N; (b) The scope of number morphology. On his analysis, pre-N adjectives in French have a different semantic relationship
with the noun than post-N adjectives. The former he calls ‘non-intersective’, by which he means that, rather than predicate some quality of the entity referred to by the noun, they interact with the semantic content of this noun and modify one of its internal components. Post-N adjectives he terms ‘intersective’, meaning that the quality predicated of the object by N intersects with the
quality predicated by A. (12–13) are examples of ‘non-intersective’ adjectives; (14–15) are examples of ‘intersective’ adjectives in French.

(12) le futur président
[D-A-N [D le] [A-N [A futur] [N président]]]
futur modifies the ‘time interval’ component of président
(13) de parfaits scélérats
[D-A-N [D de] [A-N [A parfaits] [N scélérats]]]
parfait (perfect) modifies the ‘characteristic feature’ component of
scélérats (scoundrels)
(14) la voiture rouge
[D-N [D-N [D la] [N voiture]] [A rouge]]
the set of ‘red things’ and the set of ‘car things’ intersect
(15) les diamants volés
[D-N [D-N [D les] [N diamants]] [A volés]]
the set of ‘stolen things’ and the set of ‘diamond things’ intersect

The reason why ‘non-intersective’ adjectives must occupy the pre-N position, on Bouchard’s analysis, is that they form a complex functor with N, which then merges to D to form a referential entity. Post-N adjectives then merge with this complex referential entity. This is what the labeled brackets above are meant to express. Now, if it were sufficient to freely Associate As and Ns, as Bouchard suggests, and let semantics control surface word order, then why should there be any difference in adjective distribution across languages? Why should English, for instance, place all adjectives in the pre-N position? For Bouchard the answer to this question lies in the different ways in which number is encoded.

In English, number features are found on the plural affix. Since this is a bound morpheme, Williams’ (1981) Right-Hand Head Rule applies and number scopes leftward. In French however, number morphology on N and A has become phonologically inactive, outside of irregular nouns like cheval > chevaux and liaison contexts. Number is encoded on the determiner, creating
what Miller (1991) calls a left-edge inflection. Since number is expressed in syntax rather than in morphology in French, Williams’ Right-Hand Head Rule does not apply and number scopes rightward. This is called the Number Scope Condition by Bouchard. In English DPs, the number affix closes off the N projection, forming a complex referential head [N+Num]. Post-N adjectives
cannot interact semantically with the noun, so all As must be pre-N.

Bouchard’s analysis has several interesting aspects for the present study: (a) it opens up the possibility of a mapping analysis which relates the semantics of adjectives to their syntactic positions; (b) it suggests a way to explain the historical change in adjective distribution observed for French in Section 2. If number morphology in French has weakened over the centuries, then Bouchard’s
Number Scope Condition could explain the move towards the post-N position: Old French had strong number morphology (or ‘semi-strong’; see Boucher 2003) and most adjectives were pre-N, as in Modern English, while Modern French has weak number morphology and most adjectives are post-N. On the other hand, a number of difficult problems remain to be solved.

First of all, post-N adjectives are possible in English, as we’ll see below. Secondly, the set of pre-N adjectives in French is only partially accounted for by Bouchard’s ‘non-intersective’ class. His analysis does not account for the different pre- and post-N positions available to adjectives in French as well as in English and in other languages.

4. Qualification of N: A topological approach
Larson (1998) proposes an interesting variation on the sort of analysis found in Bouchard (1998). Beginning with a set of adjectives which allow an ‘intersective’ reading (16a) and an ‘adverbial’ reading (16b), Larson develops a theory based in part on Deising’s (1992) Mapping Hypothesis.

(16) Olga is a beautiful dancer.
a. Olga is a dancer and Olga is beautiful.
b. Olga dances beautifully.

In Deising’s approach, the syntactic domain is divided into three sub-domains, which correspond to three sorts of semantic relations.

(17) [ Quantification [ Restriction [ Nuclear Scope ]]]
[ Subject NPDÉF [ VP [ AdvP ]]]

The ‘quantification domain’ is the domain of the ‘subject’ in the syntactic tree, that is, definite DPs for the most part. The ‘restriction domain’ is the domain of the ‘predicate’, that is, the verb phrase. Finally, the nuclear scope domain is occupied by adverbials. For Larson the surface word order of a particular set of verbal constructions, so-called middle constructions like ‘this book reads easily’ or ‘those cakes sold quickly’, corresponds exactly to this schema and he claims that this is in fact the deep structure of VPs in general (18).11 By extending this analysis to the nominal domain, he obtains (19). This gives him the basis for analyzing the various positions and meanings of adjectives in English.

(18) [VP ballerinas [V′ dance [ADVP beautifully]]]
(19) [NP Olga [N′ dancer [AP beautiful]]

4. Individual- versus Stage-level predicates
Larson points out that post-N adjectives are possible in English, but only with an ‘extrinsic’ interpretation, corresponding to temporary, situation-specific properties, which he calls, following Carlson (1972), stage-level predicates.

(20) The stars visible tonight include Cappella, Betelgeuse, etc.
(21) The rivers navigable at the moment include the Nile and the Amazon.
(22) The jewels stolen from the shop the other day were on the table.

This meaning contrasts with that of the same adjective in pre-N position, where it expresses a permanent, ‘intrinsic’ property of N, which he terms individuallevel predicates.

(23) The visible stars in the western hemisphere include Cappella, Betelgeuse, etc.
(24) The navigable rivers in the world include the Nile, the Mississippi, etc.
(25) The stolen jewels were on the table.

Given the deep-structure analysis for adjectives in (19) and the different readings and positions illustrated by (20–22) as opposed to (23–25), Larson suggests that the former might be obtained by leaving the adjective in situ, with the latter being derived through an ‘A-raising’ operation. If this analysis is on the right track, it would constitute a radical departure from traditional thinking
in generative syntax circles, which generally assumes the correctness of the Kayne-Cinque analysis, where As are universally generated in pre-N position and differences in surface order are obtained through N-raising, as seen above. The following sections will offer some empirical proof that Larson’s analysis is the right one.

4.2 Mapping meaning to form
Larson observes that both sorts of adjectives may be found in the pre-N position in English.

(26) The invisible visible stars include Capella, Betelgeuse, etc.

He points out that, while the order S-level / I-level is possible, as in (26), which means that some of the intrinsically visible stars are momentarily invisible, the order I-level / S-level, as in (27), results in agrammaticality. I would add that the same results can be obtained by alternating the pre- and post-N positions for I-level as opposed to S-level adjectives (28–29).

(27) * The visible invisible stars include Cappella, Betelgeuse, etc.
(28) The visible stars invisible tonight.
(29) * The invisible stars visible tonight.

These observations lead Larson to propose the following organization of the semantico-syntactic space of the noun phrase.

(30) [NP [AP S-level ] [N′ [N′ [AP I-level ] N] [AP S-level ]]]

In order to account for the differences in meaning, he introduces two logical operators into this schema, the existential operator, ∃, and a generic operator, Γ, borrowed from Chierchia (1995).

(31) [DP ∃ [ AP [ Γ [ AP N ] ] AP ] ]

The adjective which is immediately adjacent to N falls under the scope of Γ, giving it an ‘intrinsic’, ‘generic’ reading, while the adjectives outside the scope of Γ fall under the scope of the existential operator, giving them an ‘extrinsic’, ‘momentary’, reading, linked to a specific spatio-temporal situation. How does this schema apply to the French noun phrase?

5. Adjective positions in French
French predicative adjectives are generally placed to the right of the noun, with I-level adjectives next to the noun and S-level adjectives further to the right (32). A change in this order results in agrammaticality (33).

(32) Les étoiles visibles invisibles en ce moment.
(33) * Les étoiles invisibles visibles en ce moment.

This makes it look, at first glance, as if the French NP is simply the mirror image of an English NP like (26). However the facts are somewhat more complicated. First of all, while in some cases S-level adjectives may be found in the pre-N position, this only seems to be compatible with a speaker-oriented reading.

(34) Pierre faisait de visibles efforts pour paraître calme.
(Pierre was making visible attempts to look calm)
(35) * Pierre voyait de visibles étoiles dans le ciel.
(Pierre saw visible stars in the sky)

That the French NP is not simply the English NP turned around is shown by the fact that speaker-oriented adjectives in English are impossible in the post-N position (36).

(36) * Peter made attempts visible to remain calm.

This suggests that the schema in (31) may be too simple to cover all the cases of adjective modification. However, for the moment I will simply adapt this schema to French by making two small but important modifications: a) I-level adjectives must be placed immediately to the right of the noun; b) S-level adjectives can be found either to the right of the I-level adjective, after the noun,
or to the left of the noun but then only with a speaker-oriented reading. This gives us (37).

(37) [DP ∃ [NP Speaker-Oriented AP [N′ Γ [N′ N [I-level AP]] [S-level AP]]]]

However, this schema does not account for all the possibilities of nominal modification offered by the French system. In addition to speaker-oriented adjectives, the pre-N domain may include (at least) the following:

I. Bouchard’s ‘non-intersective’ adjectives: futur, soi-disant, parfait, etc.
II. mono and bi-syllabic adjectives of size and quality: petit,, grand, beau,
mauvais, etc.

III. comparative, ordinal, cardinal and restrictive adjectives: meilleur, premier,
seul, etc.
IV. any predicative adjective which changes meaning when moved to the pre-N position: pauvre, ancien, vieux, etc.

The change of position for the fourth group of adjectives seems to correspond the type of change of meaning observed above for visible, that is, the intersective meaning becomes speaker-oriented. The heterogeneity of this list is only apparent. In fact, all of these adjectives fall fairly neatly into two groups:

• Restrictive Adjectives, which operate a qualitative or quantitative restriction on the denotation of the noun; this group would include types I and III above, that is, ‘non-intersective’ and ‘quantifier’ adjectives;
• Evaluative Adjectives, which express some sort of evaluation by the speaker; this group would include types II and IV above, that is, speaker-oriented adjectives as well as subjective evaluations of color, size and so on.

The relative order of the two sub-classes in the pre-N position seems to be: Restrictive Adjectives > Evaluative Adjectives. The opposite order results in agrammaticality.

(38) Mon premier petit bateau de pêche.
(my first little fishing-boat)
(39) * Mon petit premier bateau de pêche.
(my little first fishing-boat)
(40) Un soi-disant grand homme d’Etat.
(a so-called great man of state)
(41) * Un grand soi-disant homme d’Etat.
(a great so-called man of state)

In order to take these facts into account, we need to make the following changes in the schema in (37).

(42) [DP∃ [NP [N′ Restrictive AP [N′ Evaluative AP[N′ Γ N [I-level AP]] [Slevel AP]]]]

Restrictive and evaluative adjectives in pre-N position, as well as stage-level adjectives on the outer edge of the post-N position, fall under the scope of the existential operator, while individual-level adjectives placed immediately to the right of N fall under the scope of the generic operator. This representation of adjective positions in French suggests the possibility of a ‘cartographic’ study of the ‘left-periphery’ of the French DP along the lines of what Rizzi (1997) has proposed for CP and IP. However, as it stands, it is certainly only a rough sketch of what actually goes on in the pre-N zone of the French DP. One thing that suggests the need for more research is the fact that moving adjectives to the pre-N position is subject to certain syntactic constraints.

5. Syntactic Contraints on the left periphery
If the noun is in the plural, the presence of a determiner is usually blocked.

(43) Pierre faisait de (*des) visibles efforts pour paraître calme.
(Pierre made visible efforts/*some visible efforts to appear calm)
(44) Après d’ (*des) interminables heures d’attente.
(after waiting for hours/*for some hours)

Conversely, when the adjective is in the post-N position, the determiner is
obligatory (45–46), unless the DP falls under the scope of a negative operator

(47) or of certain degree adverbs like bien peu or autant which work like determiners (48–49).
(45) Pierre faisait (*de) des efforts visibles pour paraître calme.
(46) Après (*de) des heures interminables d’attente.
(47) Il n’y avait pas de (*du) vin dans l’armoire. (there was no wine in the cupboard)
(48) Il y avait bien peu de (*des) fruits frais dans la corbeille. (there was very little fruit in the basket)
(49) Nous avons autant d’ (*des) élèves que de professeurs. (we have as many pupils as teachers here)

This suggests that the adjective in the pre-N position may have actually raised to the determiner (D) position, as suggested by Longobardi (1994) for nouns like casa in Romance, or alternatively that restriction and determination are incompatible for semantic reasons which remain to be determined.

5.2 Determiners and pre-N adjectives
Larson (1998) points out that a similar phenomenon is observed in Norwegian. Svenonius (1994)12 has shown that nouns ending with the definite affix -et usually require the collocation of the definite article det (50). However, DPs containing certain adjectives like samme (same) or første (first) are exceptions to this rule (51–52).

(50) * (Det) viktige motet.
(*(the) important meeting-DEF)
(51) Samme trøtte maten.
(same boring food-DEF)
(52) Første viktige motet.
(first important meeting-DEF)

Interestingly, these same adjectives must precede any other adjective in the DP, as in (51–52). If they do not, then the determiner is required (53).

(53) * (Det) viktige første motet.
(*(the) important first meeting)

Svenonius suggests calling these adjectives ‘determiner-adjectives’. Larson claims that such adjectives raise to the D position in Norwegian, where they act like quantifiers. The parallel with French restrictive adjectives is obvious. It seems that in French the presence of an evaluative or a restrictive adjective (or a negative or a degree operator) before a plural noun validates indefinite or partitive reference. In the terms of the schema in (17), moving an adjective from the post-N to the pre-N position in French corresponds to moving it from the domain of nuclear scope to the domain of quantification.

5.3 Structural deficiency
The A-raising hypothesis receives additional support from another set of facts. Cardinaletti and Starck (1999) argue that the strong/weak paradigm usually associated with Romance pronouns can be extended to other grammatical categories. For instance, the French adverb bien has two readings, a manner reading (54) and a degree reading (55).

(54) Il l’a fait bien.
(He did it well, i.e. in a satisfactory manner)
(55) Il l’a bien fait.
(He indeed did it, i.e. it is undoubtedly true that he did it)

Cardinaletti & Starck argue that these two readings correspond to a structurally non-deficient as opposed to a structurally deficient form. In their terms, ‘deficient’ forms are semantically underspecified, as opposed to ‘non-deficient’ forms. Likewise, clitic pronouns are semantically underspecified, since they may be interpreted indifferently as [+human] or [-human] (56a), contrary to strong pronouns, which carry the [+human] feature (56b).

(56) a. Je l’ai invité / Je l’ai acheté.
(I invited him/her / I bought it)
b. J’ai invité lui (mais pas elle) / *J’ai acheté lui (mais pas elle).
(I invited him (but not her) / *I bought it (but not her))

Crucially, as shown in Boucher (2003b), the deficient version of bien shows the diagnostics generally associated with clitic pronouns (Kayne (1975)). It must leave its base position (55) and can be neither modified (57) nor coordinated (58).[13]

(57) * Il l’a très bien fait (vs Il l’a fait très bien)
(*He very indeed did it) (vs He did it very well)
(58) * Il l’a bien et vite fait (vs Il l’a fait bien et vite)
(*He indeed and quickly did it) (vs He did it well and fast)

Although judgements differ among native speakers of French, it would seem that pre-N adjectives show the same diagnostic properties, which, if this analysis is on the right track, suggest their semantically defective status. This is to be expected if they have raised to a determiner (like) position, or, alternatively, occupy some functional position higher than N. (59–60a) are examples with a structurally deficient pre-N adjective; (59–60b) are examples with its base-position, non-deficient equivalent.

(59) a. * Cet absolument parfait imbecile !
(*This very perfect fool)
b. Ta solution est absolument parfaite.
(Your solution is absolutely perfect)
(60) a. * Ce parfait et aimable imbécile !
(*This perfect and likable fool)
b. Ta solution est parfaite et ingénieuse.
(Your solution is perfect and ingenious)

6. Conclusion
We rejected the analysis in Cinque (1995), which claims that the A-N-A in French is obtained by raising the noun out of its base position to some functional Mapping function to form a head between D0 and N0. This analysis, as we have seen, is unsatisfactory for a number of reasons. After having examined some typological and diachronic facts concerning adjective positions, we observed that it is difficult to find a universal rule for adjective distribution. We then turned to semantics for a solution. Bouchard (1998) argues for an analysis in which adjective distribution
is obtained through a combination of semantic and syntactic constraints.

While providing a promising alternative, notably in terms of accounting for the diachronic facts, this approach only accounts for a small number of pre-N adjectives in French. Larson (1998) was seen to be more promising, since his ‘mapping analysis’, based in part on Deising (1992), allowed us to cover a much broader range of problems. Using this analysis, we came to the following
conclusions.

The nominal space in French is organised, as in English, in terms of the proximity of the adjective to the noun. The area immediately to the right of N is reserved for I-level adjectives, while S-level adjectives are found in the ‘right periphery’.

The pre-N domain is occupied by adjectives which seem to operate some sort of quantitative or qualitative restriction on the denotation of the noun, or to express some sort of evaluation by the speaker; this pre-N area is ordered, with restrictive adjectives preceding evaluative adjectives.

Prenominal adjectives are subject to a certain number of syntactic constrains: non collocation with determiners, absence of modification, stress or coordination.. These were analysed as deriving, directly and indirectly, from the syntactic position they occupy in the DP: certain ‘determiner-adjectives’ may occupy the D position itself, or a similar position in the left periphery of the noun phrase; certain evaluative adjectives may be in a head-adjoined position to the left of N. The latter conclusion is conjectured to be due to their semantically ‘defective’ status.

Deriving the above properties by raising the adjective from a post-N base position to a pre-N position may in fact turn out to be the wrong solution. Much future research on the left periphery of DP is still needed before any definite conclusion can be reached. However, within current understanding of the relationship between syntax and semantics, it seems to be the best solution
available.

Notes
1. See references to Cinque’s work in most major textbooks on generative syntax, such as Radford (1997), Haegeman & Guéron (1999).
2. J. Greenberg 1966. Universals of Language, MIT Press, p. 86.
3. John Hawkins. 1983. Word order universals: quantitative analyses of linguistic structure. New York: Academic Press, as quoted by J. Rijkhoff (2002).
4. In fact, only 49 languages are examined here, the other three being extinct.
5. The term ‘modifier’ is used instead of ‘adjective’; since it is not always clear that there is a category of free adjectives in every language discussed.
6. Serbat (1980:141): “Le trait le plus frappant dans l’ordre des mots (en latin classique) c’est sans doute que le verbe est le plus souvent à la fin de la proposition.»
7. Elise Richeter (1903). Zur Entwicklung der romanischen Wortstellung aus der Lateinishen, Halle; Roland de Dardel (1989). „L‘hypothèse d‘une base OVS en protoroman“ in Probus 1,
121–143.
8. Taken from the Enigma de Verone, 8–9th century, in Monteverdi, Testi volgari italiani dei primi tempo.
9. See Boucher (2003) for a more detailed study of this as well as the following text.

10. See de Murcin (1815).
11. This analysis assumes the VP-internal subject hypothesis proposed in Larson (1988), in which subjects are first projected in the Spec (specifier) position of VP before raising to their
final syntactic position in Spec, TP. The VP-shell analysis developed by Larson posits other, intermediate positions to account for verbal constructions such as causatives, di-transitives,
and so on.
12. Peter Svenonius. 1994. On the Structural Location of the Attributive Adjective. In E. Duncan, D. Farkas, and P Spaelti (eds.) Proceeding of the Twelfth West Coast Conference
on Formal Linguistics. Stanford, CA: Stanford Linguistics Association, as quoted in Larson 1998.
13. Clitic pronouns also differ from strong pronouns in not being able to receive emphatic stress. This diagnostic is difficult to test for the opposition between functional and lexical
adverbs and adjectives in French, since French allows both types of items to be stressed: C’est LA solution (‘That’s THE solution’). I’ll leave this complex problem for another study.

References
Bernstein, Judy. 1993. Topics in the syntax of nominal structure across Romance. PhD Dissertation, City University of New York, New York.
Bouchard, Denis. 2002. Perfect tools: from functional motivation to substantive motivation. Talk at GLOW–2002, Amsterdam and Utrecht, April, 2002
Bouchard, Denis. 1998. The distribution and interpretation of adjectives in French: A consequence of Bare Phrase Structure. Probus 10, 139–183.
Boucher, Paul. 2003a. Perfect adjective positions in French: a diachronic perspective. In Martine Coene & Yves d’Hulst (eds.). Proceedings of the International Colloquium on
Romance Linguistics, Antwerp: http://webhost.ua.ac.be/apil/apil107/file04.PDF.
Boucher, Paul. 2003b. Adjective positions in Old and Modern French. Talk at Western Conference of Linguistics, September 2003, University of Arizona, Tucson.
Cardinaletti, Anna and Michal Starcke. 1999. The typology of structural deficiency. In Henk van Reimsdijk (ed.) Clitics in the Languages of Europe. Mouton de Gruyter, 145–233.
Cinque, Guglielmo. 1995. On the evidence for partial N-movement in the Romance DP. Italian Syntax and Universal Grammar. Cambridge University Press.
de Murçin. 1815. Serments prêtés à Strasbourg en 842 par Charles-le-Chauve, Louis-le-Germanique et leurs armées respectives, Paris: Didot l’Ainé.
Deising, Molly. 1992. Indefinites. Cambridge, MA: MIT Press
Goyens, Michèle. 1994. Emergence et évolution du syntagme nominal en français. Bern.
Kayne, Richard. 1975. French Syntax: the Transformational Cycle. Cambridge: MIT Press
Kayne, Richard. 1994. The Antisymmetry of Syntax. MIT Press.
Larson, Richard. 1998. Events and Modification in Nominals. In D. Strolovitch and A. Lawson (eds.) Proceedings from Semantics and Linguistic Theory (SALT) VIII.Cornell University,
Ithaca, N.Y.
Longobardi, G. 1994. Reference and proper names: a theory of N-movement in syntax and logical form. Linguistic Inquiry, 25, 609–665.
Marchello-Nizia, Chr. 1995. L’évolution du français: ordre des mots, démonstratifs, accent tonique. Paris: Colin.
Rijkhoff, Jan.2002. The Noun Phrase. Oxford University Press.
Rizzi, L. 1997. The fine structure of the left periphery. In L. Haegeman (ed), Elements of Grammar. Dordrecht: Kluwer, 281–337.
Yvon, Henri. 1946–47. L’assiette du nom des origines au XIV° siècle. Romania 49, 289–316.

Author’s adress:
Paul Boucher UFR Lettres, langues et sciences humaines
Université d’Angers
11 Bd Lavoisier
49045 Angers, cedex 01 (France)
boucher.paul@wanadoo.fr