🗣️ 22. Phrases

A French phrase is built round one head word, and the order of the words round it is fixed more tightly than in English. The noun phrase opens with a determiner that is almost never missing, puts most of its adjectives after the noun, and attaches every complement of the noun with a preposition: there are no noun stacks of the English kind (the town hall car park). The adjective phrase puts its degree word before the adjective and its complement after it. The prepositional phrase is a preposition and a noun phrase, an infinitive or, with en, a gerund; a preposition is never left behind at the end of a clause. This chapter describes the noun phrase (determiners, adjectives and their position, complements, modifier clauses, quantity), the phrases headed by pronouns, the adjective phrase and comparison, the adverb phrase and the prepositional phrase, with the UD analysis of each and counts from the treebanks. The forms of the articles, possessives and numerals are in the determiners chapter; the forms and agreement of adjectives in the adjectives chapter; the relative clause in chapter 25.

Contents:

1. Phrases in a French tree
2. The noun phrase: order and determiner
3. Adjectives in the noun phrase
4. Complements of the noun
5. Clauses and participles on the noun
6. Quantity: numerals, quantifiers, partitives
7. Phrases without a noun head
8. The adjective phrase
9. Comparison and the superlative
10. The adverb phrase
11. The prepositional phrase
12. Where attachment goes wrong
13. How Mova handles it
14. Sources

1. Phrases in a French tree

UD has no phrase nodes. A phrase is a head word and everything that depends on it, and the head is the content word: the noun in a noun phrase, the noun (not the preposition) in a prepositional phrase, the adjective in an adjective phrase. Chapter 20 gives the general decisions; for phrases they mean:

A prepositional phrase therefore has three layers when the preposition is contracted with an article: aux tueurs is the syntactic words à les tueurs, with à case and les det of tueurs (chapter 20, §3). In the written GSD the noun phrase is most of the text: det, case, nmod and amod are 44% of all relations, against 22% in the spoken ParisStories (counted 2026-11-06).

2. The noun phrase: order and determiner

The order inside a noun phrase is:

(pre-determiner tout) – determiner – (numeral) – (short adjective) – noun – (adjectives) – (prepositional complements) – (participle or relative clause)

le petit chat noir de ma voisine qui dort au soleil the little cat black of my neighbour who sleeps in-the sun — "my neighbour's little black cat that is sleeping in the sun"

tous les trois grands romans de Balzac all the three great novels of Balzac — "all three of Balzac's great novels"

The determiner is obligatory with a common noun in an argument position, singular or plural, count or mass. Where English has a bare noun, French has an article: Le vin est cher ("Wine is expensive", generic), Il boit du vin ("He drinks wine", partitive), Il a des amis ("He has friends"). In UD French GSD 60 661 determiners stand before 75 330 common nouns; the four commonest determiner lemmas are le (42 903, the definite article in all its forms), un (9 976), the possessive son (4 346) and the demonstrative ce (2 207) (UD 2.18, 2026-11-06).

Bare nouns are limited to a known list of frames (fraser-1921, §330):

• a predicate noun of profession, nationality or status after être, devenir, rester, élire, nommer: Il est médecin, Elle a été élue maire (chapter 21, §8);
• after de of quantity and the de that replaces an indefinite or partitive article under negation (beaucoup de livres, pas de pain; chapter 27, §8);
• after en and in many prepositional idioms: en bois, en voiture, avec soin, sans argent, par cœur, à pied;
• in noun complements that name a kind: une tasse à café, un professeur de musique, une table en bois;
• in apposition and in titles, lists, headlines, addresses: Tarzan, l'homme singe has an article, but Victor Hugo, poète et romancier does not;
• in verb idioms: avoir faim, avoir peur, prendre part, faire attention, rendre service. GSD marks the commonest as light-verb objects (obj:lvc, 554 times: avoir lieu, faire partie, mettre fin);
• in coordinated pairs and proverbs: corps et âme, Pauvreté n'est pas vice.

Determiners do not stack, except the pre-determiner tout (tout le monde, tous les jours, toutes ces raisons) and the numeral after a definite, possessive or demonstrative (les trois mousquetaires, mes deux sœurs, ces quelques mots). GSD analyses this tout as an amod of the noun (368 times before a noun, never after one), not as a determiner. A possessive and an article never combine (*le mon livre); "a friend of mine" is un de mes amis or un ami à moi.

Agreement. The determiner and the adjectives agree with the noun in gender and number. In speech the noun's plural is often audible only on the determiner: /lə ʃa/ le chat, /le ʃa/ les chats.

3. Adjectives in the noun phrase

The rule taught to English speakers — "French adjectives follow the noun" — is a tendency, not a law. The UD counts show how strong it is and how much it depends on the register (adjectives amod of a common noun, UD 2.18, 2026-11-06):

Written prose, with its many classifying adjectives (la politique étrangère, le droit international, une réaction chimique), puts seven in ten after the noun; the spoken narratives, which use few classifying adjectives and many evaluative ones (un petit truc, une grosse voiture, un super moment), put two in three before it.

The adjectives that precede. A short list of frequent, mostly short adjectives precede by default: beau, bon, grand, gros, haut, jeune, joli, long, mauvais, meilleur, moindre, nouveau, petit, vieux, vilain, the ordinals (premier, deuxième, dernier in a series) and autre, même, seul (fraser-1921, §351). In GSD the commonest prenominal adjectives are, with the number after the noun in brackets: premier 696 (9), grand 418 (1), autre 366 (3), tout 368 (0), nouveau 289 (25), même 255 (0), petit 198 (2), nombreux 187 (4), ancien 177 (17), seul 166 (6), bon 157 (4), dernier 153 (40), meilleur 116 (5), second 110 (0), deuxième 88 (0), jeune 90 (8).

The adjectives that follow. Adjectives of colour and shape, of nationality, religion and political affiliation, relational adjectives derived from nouns (le président américain, l'industrie automobile, une crise cardiaque), past and present participles used as adjectives (une porte ouverte, une étoile filante), and any adjective that has a complement or a heavy modifier (une contrée riche en vins, un discours extrêmement long) follow the noun (fraser-1921, §352). A relational adjective cannot precede at all (*l'automobile industrie).

Position changes the meaning of a set of common adjectives. Before the noun the reading is figurative, evaluative or one that classifies the referent within a set; after it, literal and descriptive:

Beyond these lists the choice is graded. A study of adjective placement with French transformer models describes it as the effect of many interacting factors — the adjective's frequency and length, its meaning, the noun, lexicalized pairs (arts premiers "non-Western art" against premiers arts "first arts"), and the other dependents, since an adverb on the adjective pushes it towards postposition — and reports that fine-tuned models choose the attested position well, mostly from the frequency of each adjective, while hard cases with competing factors remain errors for both the models and the native speakers they were compared with (metheniti-2023).

Several adjectives. Each keeps its own place: une jolie petite maison blanche. Two that follow the noun have the more specific one last, the reverse of English: des écrivains français célèbres "famous French writers". Coordinated adjectives go together after the noun: un objet blanc et étincelant.

The adjective in UD. Every attributive adjective is amod of its noun, wherever it stands. The position itself is the only trace of the meaning difference, so a tool that reads ancien must look at the word order, not at the relation.

4. Complements of the noun

The noun complement is a prepositional phrase. The English possessive and the English noun-noun compound are both expressed with a preposition:

In GSD 35 582 words are nmod. Their prepositions (the case dependent, 2026-11-06) are de 23 543 (66%), à 2 154, en 1 279, dans 538, sur 501, pour 449, avec 388, par 244, entre 206, comme 164; 5 366 have none. The prepositionless nmod are mostly dates and numbers (octobre 2001, 26 octobre, Interstate 80: 3 581 of them), the relative pronoun dont (285, chapter 25) and a residue of noun–noun juxtapositions (678 NOUN → NOUN).

Juxtaposed nouns. French has a productive but limited pattern of two nouns side by side, the second qualifying the first: l'homme singe (fr-ud-dev_00040: Tarzan, l'homme singe), un pays membre, une date limite, une voiture balai, un roman fleuve, une pause café. The second noun is invariable or agrees freely (des pays membres), and the pattern is commonest in commerce and the press. GSD has 678 noun-on-noun nmod without a preposition, many of them this pattern. Compounds written with a hyphen (porte-monnaie, timbre-poste) are single words in the treebanks.

The de complement and the article. After de a noun complement has an article when it refers to a particular thing (le livre du professeur "the teacher's book") and none when it names a kind (un livre de professeur "a teacher's book, a book for teachers"). With the article, de + le/les contracts to du/des (chapter 20, §3).

Order of several complements. The complements of a noun follow it, the shortest and most closely bound first: le livre de Paul sur la guerre; an argument of a deverbal noun before a modifier: la destruction de la ville par l'armée en 1945. The agent of a deverbal noun takes par when the patient is also expressed with de (la critique du roman par Sainte-Beuve).

Apposition. A noun phrase set beside another, usually between commas, is appos: Adam, le berceau de la tribu; Tarzan, l'homme singe. Apposition is a mark of written informative prose: 6 327 in GSD (158 per 10 000 words) against 19 per 10 000 in ParisStories.

5. Clauses and participles on the noun

A noun can take a relative clause (chapter 25), a participle phrase, an infinitive and a complement clause:

(Counts of acl and acl:relcl on nouns, pronouns and names, 2026-11-06.)

The à + infinitive on a noun (un livre à lire, une maison à vendre, un problème à résoudre) has a missing object that is the head noun: the book is to be read. A study of these missing-object constructions argues that they behave like a passive of the infinitive — the head noun is the logical object, the logical subject remains available for control — and that the dependency is bounded: a control or raising verb can stand between à and the verb with the missing object (un livre à essayer de lire), but not a finite clause (multner-2022). The same à + infinitive appears after être (Ce livre est à lire), with facile, difficile (un texte facile à lire: the "tough" construction) and as a secondary predicate (J'ai une lettre à écrire).

The de + infinitive on a noun is the complement of a noun of intention, possibility or manner (l'intention de partir, la possibilité de choisir, une façon de parler); its subject is controlled by the noun's possessor or left generic.

Complement clauses on nouns take que and the mood the noun would give its verb: le fait que takes the indicative or, increasingly, the subjunctive; la crainte que the subjunctive; l'idée que, la nouvelle que the indicative (chapter 24).

6. Quantity: numerals, quantifiers, partitives

Numerals precede the noun and follow a definite or possessive determiner: trois livres, les trois livres, mes trois livres. They are nummod (3 640 in GSD). Un, une is the indefinite article and the numeral at once; GSD tags it DET det (9 976) except in counting. Premier and the ordinals are adjectives (amod).

Quantity adverbs with de. Beaucoup de, peu de, assez de, trop de, tant de, autant de, plus de, moins de, combien de take a noun without an article: beaucoup de personnes (fr-ud-test_00139: Ca énerve beaucoup de personnes), peu de temps. The article reappears when the noun is definite: beaucoup des personnes présentes "many of the people present". In GSD the quantity adverb is the head of the phrase: in ils consomment beaucoup de baies (fr-ud-train_00016) beaucoup is the obj of the verb and baies an obl:arg of beaucoup with case de. A tool that looks for the object noun must look one level down.

Quantifier nouns. La plupart "most" takes des and a plural verb: la plupart des cas (fr-ud-train_01641: Dans la plupart des cas), La plupart des gens pensent …. The verb agrees with the plural noun after it, not with plupart. Un grand nombre de, une foule de, la majorité de let the verb agree with either; le peu de with the sense intended.

Partitives with a superlative. La plus intelligente de mes élèves — a superlative with de picks one member of a set. A study of such partitives finds that the superlative can show feminine gender even when the noun inside the de-phrase is a masculine used for a mixed or unspecified group (la plus intelligente de mes gentils professeurs), something the quantified partitives (trois de mes professeurs) do not allow; the authors relate the difference to the structure of the two partitives (sleeman-2016).

The pronoun en of quantity. With a numeral or quantifier and no noun, French requires en before the verb: J'en ai trois "I have three (of them)", Il en reste beaucoup. English can say I have three; French cannot say *J'ai trois. In UD the numeral or quantifier is then the head of the object and en depends on it (the clitic-climbing case of chapter 20, §10).

7. Phrases without a noun head

Demonstrative and possessive pronouns stand for a noun phrase and take the same complements: celui de Paul "Paul's", celle que j'ai vue "the one I saw", ceux qui partent "those who leave", le mien, la sienne, les nôtres. Celui never stands alone: it needs de, a relative clause or -ci/-là. In UD celui is a PRON that heads its complement (nmod, acl:relcl).

Neuter free relatives. Ce qui, ce que, ce dont, ce à quoi head a clause with no noun antecedent: ce que je veux, ce dont j'ai besoin. Ce is the head and the clause its acl:relcl (chapter 25, §7).

Adjectives as heads. An adjective after a determiner can stand for a noun phrase when the noun is understood: Je prends le rouge "I'll take the red one", les riches, le vrai. English needs one; French does not. In the treebanks such a word is tagged NOUN when it is lexicalized (les riches); elsewhere the adjective keeps its tag and heads the phrase.

Numerals and quantifiers alone: Les deux sont partis. Plusieurs pensent que …. Tous sont venus. Tout alone is a pronoun (Tout est prêt), as are chacun, quelqu'un, personne, rien (chapter 27 for the negative ones).

8. The adjective phrase

Degree words precede the adjective: très grand, si grand, trop grand, assez grand, plus grand, moins grand, aussi grand, tout petit, bien content, fort aimable. In GSD the adverbs that modify adjectives (advmod on an ADJ, 2 303 in all) are led by plus 574 and très 383, then ne 134 and pas 110 (the negation of a predicate adjective under être, which is the head of the copular clause). Assez "enough" precedes the adjective (assez grand pour …), unlike English big enough.

Complements follow the adjective, with a preposition fixed by the adjective: fier de, content de, capable de, plein de, sûr de; prêt à, facile à, utile à, fidèle à; fort en, riche en; bon pour, apte à. In GSD the prepositional dependents of adjectives (1 352) take de 411, à 321, dans 111, en 88, pour 87, avec 41. An adjective can take a clause: content que tu sois là (subjunctive, chapter 24), sûr qu'il viendra (indicative), or an infinitive: prêt à partir, capable de comprendre, facile à dire.

An adjective with a complement follows its noun: un homme fier de son fils, never *un fier de son fils homme; English places the adjective first only without a complement (a proud man, a man proud of his son) and so, broadly, does French.

The adjective as a predicate is the head of a copular clause (cop être) or an xcomp of devenir, sembler, rester, rendre, trouver (chapter 21, §8), and it agrees with its subject or its object.

9. Comparison and the superlative

The comparative is analytic: plus (more), moins (less), aussi (as) before the adjective or adverb, and que before the standard (fraser-1921, §345):

Il est plus grand que son frère. — Elle parle moins vite que lui. — C'est aussi cher qu'avant.

Three adjectives have a synthetic comparative: bon → meilleur, mauvais → pire (beside plus mauvais), petit → moindre (abstract: le moindre doute), and two adverbs: bien → mieux, mal → pis (fixed phrases). Plus bon is wrong; plus bien does not exist.

After a negative main clause aussi becomes si in careful style: Il n'est pas si grand que son frère. Before a numeral, "than" is de, not que: plus de trois heures, moins de dix euros.

The standard of comparison can be a noun phrase, a pronoun (strong form: que moi, que lui), a prepositional phrase (plus à Paris qu'à Lyon) or a clause. A comparative clause in careful style has the expletive ne and often the pronoun le for the compared proposition: Il est plus grand que je ne le pensais, plus tard qu'on ne le croit (chapter 27, §9).

fr-ud-dev_01097 · UD French GSD Nous serions deux fois plus nombreux que la tâche ne manquerait point. we would-be two times more numerous than the task NEG would-lack not-at-all — "We would be twice as many, and the task would still not be lacking." (The comparative clause carries a full negation ne … point, not the expletive ne.)

In UD French GSD the standard depends on the degree word, not on the adjective: a noun phrase is the obj of plus/moins/aussi with que as its case (93 on an adverb, 52 on an adjective among 198 comparative que case), and a clause is a ccomp of the degree word with que as mark (26 under plus). In fr-ud-dev_01097 manquerait is the ccomp of plus. A tool must therefore look below plus for the second term of a comparison.

Correlative comparison: Plus il travaille, plus il gagne "The more he works, the more he earns" — no article, no que, the two clauses juxtaposed (GSD fr-ud-test_00102 has a spoken-style example: plus je vois des vidéos de ce jeu, plus je me dis que …).

The superlative is the definite article (or a possessive) before the comparative: le plus grand, la moins chère, mon meilleur ami. When the adjective follows the noun, the article is repeated: le puits le plus profond de France (fr-ud-dev_00044), la ville la plus peuplée; before the noun, once: la plus grande ville. The domain of the superlative takes de, never dans: le plus haut sommet d'Europe. A relative clause after a superlative usually takes the subjunctive: le meilleur film que j'aie vu (chapter 24, §5). The absolute superlative "very" is très, fort, extrêmement or the old suffix -issime (richissime, rarissime), the last one familiar or emphatic.

10. The adverb phrase

Adverbs take degree words (très souvent, assez bien, trop vite, beaucoup plus tard) and a comparison standard (plus vite que lui). Beaucoup modifies a comparative, never a plain adjective: beaucoup plus grand, but très grand (*beaucoup grand). Adverbs in -ment take complements with the preposition of their adjective: conformément à la loi, indépendamment de son âge, contrairement à ce qu'il croit. Their place in the clause is in the adjectives and adverbs chapter and in chapter 21, §9. In UD all of them are advmod; GSD has 14 330, ParisStories 604 per 10 000 words against GSD's 358, because the spoken language uses adverbs where writing uses noun phrases (là, après, bon, vraiment, trop).

11. The prepositional phrase

The prepositions. French has a small set of simple prepositions and a large set of compound ones. In GSD 57 917 words are case; the commonest lemmas are de 29 090, à 9 223, en 5 037, dans 3 135, par 2 288, sur 1 678, pour 1 555, avec 1 353, comme 530 (2026-11-06). De and à are two thirds of all prepositions and both are grammatical as much as lexical: de marks the noun complement, the agent of some passives, the partitive and the infinitive after many verbs; à the indirect object, the purpose of a noun (tasse à café) and the infinitive after others (chapter 23).

Compound prepositions are written as several words and annotated as one: the first word heads, the others are fixed (à partir de, grâce à, à cause de, au lieu de, près de, afin de, en face de, au cours de, quant à). In GSD 3 243 words are fixed; the commonest second elements are que (408 + 165 qu', in conjunctions such as alors que, bien que), plus (238), partir (117, à partir de), effet (103, en effet).

Repetition. À, de and en are repeated before each coordinated noun phrase: Il parle à Paul et à Marie, une table de bois et de fer. Other prepositions may be shared: pour Paul et Marie. In the written norm the repetition of à and de is a rule; journalistic French does not always keep it.

What the preposition governs. A noun phrase (sur la table), a strong pronoun (avec lui, pour moi, chez eux — never a clitic), an infinitive (pour partir, sans dire, avant de partir, après avoir mangé: après takes the perfect infinitive), an adverb (depuis hier, pour toujours, d'ici). En is the one preposition that takes the present participle, making the gerund: en sortant, en lisant. A preposition never takes a finite clause directly: it needs que or ce que (avant qu'il parte, à ce que je sache), and most prepositions form a conjunction this way (pour que, sans que, depuis que; chapter 24).

No stranding. A French preposition cannot be left at the end of the clause, as English to in the woman I spoke to. In relatives and questions the whole phrase moves: la femme à qui j'ai parlé, À qui parles-tu ? (chapters 25 and 26). In speech, prepositions such as avec, pour, contre, dessus, dedans can be used without a complement when the object is recoverable: % Il est venu avec. % Je suis pour. (avec without a complement is regional and familiar; pour and contre are standard). These are not stranded prepositions but prepositions used intransitively, as adverbs.

Place names. En before feminine countries and before masculine ones beginning with a vowel (en France, en Iran), au before other masculine countries (au Japon), aux before plural ones (aux États-Unis), à before cities (à Paris); "from" is de, du, des in the same distribution.

12. Where attachment goes wrong

Noun or verb? A prepositional phrase after an object can attach to the noun (nmod) or to the verb (obl): Il a vu l'homme avec le télescope. French has the same ambiguity as English. The de-phrase almost always attaches to the nearest noun, so it is the least ambiguous; à, avec, pour, dans, sur are the hard cases.

Chains of de. le directeur de la banque de France de Lyon — each de-phrase can attach to any noun before it. The chains are a mark of administrative French and the main source of nmod errors in a parser.

De + article and the indefinite des. After a verb, des can be the indefinite plural article (Il a des amis) or the contraction of de + les (Il parle des amis); GSD splits only the second. A tokenizer has to decide between them from the verb's valency — parler de takes de, avoir does not — which is the reason Mova's English tokenizer leaves des unsplit (§13).

Adjective or participle? Les documents imprimés hier — imprimés is a participle with its own adjunct (acl), not an adjective (amod). A participle with a complement or an agent is acl.

Postposed adjective or predicate? Il a trouvé la porte ouverte — "he found the door open" (xcomp of trouver) or "he found the open door" (amod). The two readings have the same words and order.

13. How Mova handles it

What is done. The parser learns the noun-phrase relations from the GSD training trees through en xpos-fill (chapter 20, §12). amod, det, nmod, nummod, case, appos and fixed are all in Mova's relation set and are kept as they are. The tag mapping gives the determiners DT (possessives PRP$), adjectives JJ, numerals CD, so the tagger loses gender and number, and with it the agreement that ties an adjective to its noun.

We parsed our own sentences through the English tokenizer with the parser trained on the GSD training file (en explain, 12 epochs, beam 8, CPU, 2026-11-06):

The noun phrase in isolation is parsed well; the errors come from outside it — the tokenizer (C'est, elided articles, des and du unsplit) and the comparison standard, whose GSD analysis is unusual.

What is not done.

• Adjective position and meaning. Nothing in Mova reads the position of ancien, grand, propre, seul as a difference of meaning; the frames reader (prag/src/frames.rs) would treat both orders alike.
• Agreement. Gender and number are in the treebank's FEATS but not in the PTB tags the tagger predicts, so the parser cannot use agreement to choose between two candidate heads of an adjective (la table et le buffet anciens).
• The contractions. The tokenizer does not split du, des, au, aux (chapter 20, §12), so in raw text the article inside a prepositional phrase is never a word of its own, and des as an article and des as de + les cannot be told apart. The treebank's 11 022 multiword tokens are the data for a French splitting rule, which is not written.
• Quantity. En + numeral and de + bare noun after quantity adverbs are learned only as tree shapes; the frames reader does not turn J'en ai trois into a quantity of a known referent.
• Comparison. The English frames reader has no comparison frame for French; the GSD attachment of the standard below plus would have to be read specially.

14. Sources

• Fraser, Squair & Coleman, The New Fraser and Squair Complete French Grammar (1921), §§330 (omission of the article), 345–349 (comparison), 351–353 (position of adjectives): fraser-1921.
• Metheniti, Van de Cruys, Kerkri, Thuilier & Hathout, "Chère maison" or "maison chère"? Transformer-based prediction of adjective placement in French (Findings of EACL 2023): metheniti-2023.
• Aguila-Multner & Crysmann, French missing object constructions (Glossa, 2022): multner-2022.
• Sleeman & Ihsane, Gender mismatches in partitive constructions with superlatives in French (Glossa, 2016): sleeman-2016.
• UD French GSD, Sequoia, ParisStories, Rhapsodie, UD 2.18; sentences cited by sent_id; counts made with awk over the CoNLL-U files, 2026-11-06.
• Mova: en/src/main.rs (xpos_of, explain), en/src/gram.rs (Rel), prag/src/frames.rs.
• The full list is in Sources.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/fr/22-phrases/