🗣️ 21. Clause structure

A French clause is built round a finite verb with an expressed subject before it. The subject is a noun phrase or a pronoun (the subject clitics je, tu, il, elle, on, nous, vous, ils, elles, the demonstratives ce and ça, or a relative or interrogative word); when there is no participant to name, the subject is the empty il. The objects come after the verb, unless they are pronouns, and then they come before it. The verb may take a predicate (an attribute), which agrees with the subject or with the object. Round this core stand the adjuncts, which French orders more freely. This chapter describes the clause with its subjects, impersonal constructions, objects, attributes and adjuncts, the place of the clitics, and the contexts where the subject follows the verb; it gives counts for written and spoken French from the UD treebanks. Dislocation and clefts, which change the information structure of the clause, are in chapter 29; questions and negation in chapters 26 and 27.

Contents:

1. The clause and its parts
2. Subjects
3. Impersonal constructions
4. The subject in speech: pronouns and doubled subjects
5. Where the subject follows the verb
6. Objects
7. The clitics in the clause
8. Attributes and secondary predicates
9. Adjuncts and the order of complements
10. The clause in the treebanks
11. How Mova handles it
12. Sources

1. The clause and its parts

In the neutral order the clause reads: (adjunct) – subject – [ne – clitics] – verb – (pas) – object – indirect object – adjuncts.

Hier, mon frère ne le lui a pas donné dans la rue. yesterday, my brother NEG it to-him has not given in the street — "Yesterday my brother didn't give it to him in the street."

The same fixed positions show how little freedom a French clause has: only the adjuncts can move. A direct object cannot cross the verb (*Le livre lit Marie does not mean "Marie reads the book"), and the old freedom of Latin and of Old French to put a verb first or last is gone (the medieval verb-second order and its decline are described in larrivee-2019 and larrivee-2025). What French has instead is a set of constructions that rearrange the clause by a different route (chapter 29).

2. Subjects

2.1 What can be a subject

A noun phrase, with its determiner: Le chat dort. Mes amis sont partis. Marie chante. A subject clitic cannot be stressed and cannot be separated from the verb except by other clitics and ne. It is repeated before each coordinated verb that has its own auxiliary or its own tense (Il vient et il part, Il est venu et il repart) and may be left out when the verbs share them (Il lit et écrit). A strong pronoun (moi, toi, lui, elle, nous, vous, eux, elles) can be a subject only when it is coordinated (Lui et moi sommes d'accord), stressed (Moi, je reste; the clitic is kept), or alone (Qui est venu ? — Lui.). The demonstratives ce (before être), ça and cela serve as subjects of any verb in speech (Ça marche); celui-ci, celle-là and the like as subjects with anaphoric reference. A clause (Que tu partes me chagrine) or an infinitive (Partir, c'est mourir un peu) is a subject in writing and in proverbs; in current speech they are replaced by the extraposed construction in §3. A relative or interrogative word (qui) can be the subject of the clause it introduces.

2.2 Obligatoriness

The subject cannot be dropped in a finite clause (except the imperative and some fixed written headings such as Prière de …, Voir plus bas). The speaker who has nothing to refer to uses the empty il (§3). In the spoken corpora the pronominal subjects are the clause; the noun subject is the exception. The counts for the plain nsubj relation (UD 2.18, 2026-11-06):

The pronoun subjects of the written GSD are, in order: il 2 514, qui 1 900 (relative), elle 732, on 526, ils 426, c' 350, je 309, nous 241, j' 217, vous 143, elles 85, cela 79, ce 77, l'on 67. In ParisStories: on 741, je 666, c' 607, j' 418, il 367, elle 221, tu 220, qui 207, ça 158, ils 101. In Rhapsodie: c' 490, je 489, qui 378, vous 313, on 281, j' 240, il 184, ça 143. Two points stand out. The first-person singular, almost absent from written encyclopaedic text (526 je/j' in 400 000 words), is the commonest subject of the spoken narratives (1 084 in 43 000 words); and tu, which is not among the commonest subjects of the written corpus, has 220 in ParisStories. The second is that on is among the commonest subjects of both registers (593 with l'on in GSD, 741 in ParisStories, 291 in Rhapsodie), with the value of an impersonal "one" in the written corpus and often of "we" in the spoken one (see the pronouns chapter and chapter 28).

Agreement. The finite verb agrees in person and number with the subject; after qui it agrees with the antecedent (moi qui suis, vous qui êtes). With several subjects the verb is plural; for the collective subjects and the quantified ones (la plupart des gens sont, un grand nombre de personnes ont) the plural is the rule when a plural noun follows. These facts are in the verb chapter. The treebank has the agreement features, but the parser (§11) does not use them.

2.3 Subject and position

In the plain nsubj relation 97% of the subjects stand before the verb in the written GSD and 99% in the spoken ParisStories (526 and 26 stand after it); a pronoun subject follows only in questions with inversion, in incises and after a few adverbs. Noun subjects follow the verb in 413 cases in GSD (329 common nouns, 84 proper names; §5). The question of whether a French subject can stand after its verb to carry the focus, as the Spanish subject does, has been studied on an interview corpus: in the French data no case of a postverbal subject was found, and the language uses clefts and other constructions instead (cassara-2026; see chapter 29).

3. Impersonal constructions

An impersonal verb has the empty il as subject. The il has no referent: it is not replaced by a noun, it cannot be questioned (qui ?) and it does not trigger agreement with a following noun.

Weather and time. Il pleut, il neige, il fait beau, il fait nuit, il est trois heures, il est tard. In the treebanks il is expl:subj (Il faisait super chaud et on entendait le RER, fr-ud-train_01236).

Existence and occurrence. Il y a, il existe, il reste, il manque, il arrive, il se passe, il suffit, il convient, il paraît, il semble: the verb is followed by the real subject, which is no longer in the nominative position and does not control agreement. The verb stays in the third person singular, even with a plural noun: Il reste deux places, Il est arrivé des accidents. Il y a is the commonest: the verb avoir with the locative clitic y, which is part of the fixed phrase (expl:comp), and it is invariable in the standard: Il y avait des enfants, Il y a eu des problèmes.

fr-ud-dev_00717 · UD French GSD Il y a donc le monde de la réalité comprenant les nobles et les artisans et le monde de la magie. "So there is the world of reality, made up of the nobles and the craftsmen, and the world of magic."

fr-ud-train_00233 · UD French GSD Des édifices de l'époque de Silla il ne reste que les structures en pierre. "Of the buildings of the Silla period only the stone structures remain."

fr-ud-dev_00777 · UD French GSD Il a fallu 33 jours de forage pour atteindre les 33 mineurs. "It took 33 days of drilling to reach the 33 miners."

Necessity and desirability. Il faut, il vaut mieux, il s'agit de, il y va de: Il faut partir, Il faut que tu partes, Il s'agit de la sécurité (fr-ud-dev_00776). With a noun phrase, il faut means "need, take": Il faut deux heures, Il m'a fallu deux heures; the noun phrase is a postverbal complement of the impersonal verb and the verb does not agree with it.

Extraposition. A clause or an infinitive subject can be moved after the predicate and replaced by il (formal) or ce, ça (speech):

fr-ud-dev_00700 · UD French GSD Il est nécessaire de remettre tout cela en état, et de virer le personnel sale et incompétent. "It is necessary to put all this right and to get rid of the dirty and incompetent staff."

fr-ud-dev_00710 · UD French GSD Il se peut même que l'Athenæum et la bibliothèque du Capitole ne fasse qu'un. "It may even be that the Athenaeum and the library of the Capitol are one."

Clausal subjects are rare. The treebank has 224 clausal subjects (csubj) against 22 798 subjects of all kinds in the written GSD; in the French Treebank a study counted 24 sentential and 99 infinitival subjects against about 26 000 nominal ones, with the impersonal variants much more frequent than the sentential subjects (abeille-2019-extracting). A study of extraction from French subjects relates this to information structure: the subject is the preferred topic of most verbs, so the position before the verb suits given material (winckel-2024).

The impersonal passive. A passive verb can have the empty il as subject, with the patient as a postverbal noun phrase or a clause: Il a été décidé que …, Il fut alors décidé qu'à l'avenir … (fr-ud-dev_00635). In GSD there are 28 passive verbs with an expl:subj. Chapter 28 treats it.

Il, ce, ça. In the written norm, il (or ce before être) is the empty subject; in speech ça takes over for the extraposed subject of an adjective and for the weather and time verbs that allow it (Ça caille is familiar). The treebanks have no uniform answer on which of them is an expletive: the expl:subj of GSD covers il 611, c' 187, -il 34, ce 33, -t-il 16, -ce 13, -elle 9; in the spoken ParisStories it is il 249, c' 33, -ce 28, ça 3, and c'est is otherwise a plain nsubj + cop (607 times: §10).

Il impersonal in speech. The empty il is dropped in the spoken forms of the commonest verbs: % Faut partir. Y a personne. Reste plus rien. (il y a is pronounced /ja/). The spoken treebanks write the words in the forms the transcriber heard, and il y a is the commonest impersonal: the y of avoir with expl:comp occurs 212 times in the 42 789 words of ParisStories (50 per 10 000), 202 times in Rhapsodie (46 per 10 000) and 149 times in the 400 385 words of GSD (3.7 per 10 000).

4. The subject in speech: pronouns and doubled subjects

In the spoken language the pronoun subject is obligatory and, with a noun subject, doubled. Subject doubling (also called clitic doubling or subject repetition): the noun phrase stands first, and a subject clitic agreeing with it stands before the verb, without a pause:

ParisStories_2019_peripitiesVoiture_8 euh ma mère elle m'a beaucoup aidée parce que c'est une bonne conductrice. "uh my mother she helped me a lot because she's a good driver."

ParisStories_2019_voyageItalie_14 et, euh, le gars il dit, bah, oui oui pas de soucis ! "and uh, the guy he says, well yes yes no problem!"

ParisStories_2019_concoursEquitation_52 donc du coup, bah, cette reprise elle était magique. "so, well, that ride it was magic."

The UD treebanks annotate the noun phrase as dislocated:subj and the pronoun as nsubj: the same relations as for a true left dislocation with a pause (Cette épreuve, tu la fais costumé à cheval, chapter 29). The two are one family: Bally (bally-1909, §§285–288) describes the spoken French sentence as one in which the subject or the object comes first, "anticipated" or "recalled" by a pronoun, and says that this use of the personal pronouns is the sign of the spoken grammar of modern French: he compares Les enfants cassent tout with Les enfants, ça casse tout and Ça casse tout, les enfants. In ParisStories there are 302 dislocated:subj in 42 789 words (71 per 10 000) and 203 in Rhapsodie (46 per 10 000), against 19 in the 400 385 words of GSD (0.5 per 10 000). Of the ParisStories dislocated subjects, 140 are pronouns (moi, ça, lui …), 130 common nouns and 21 proper names; 272 precede the verb and 24 follow it.

A first person may be stressed and doubled: moi je sais que moi mon lycée, en plus c'était un lycée privé (ParisStories_2019_experienceFac_64). A strong pronoun that is the subject of a finite verb is taken up by the clitic, except in coordination (Lui et moi sommes d'accord) and in an answer. This is one of the facts that make French different from Italian or Spanish, where the clitic is not needed to make a subject.

What status the doubling clitic has — pronoun or agreement marker — is a matter of analysis, not of data. For the tools of this manual the working view is the treebank's: two relations, dislocated:subj and nsubj, with the pronoun as the true subject and the noun phrase as a topic outside the clause.

Subject pronouns in speech. On is the first person plural of speech: it has 741 subjects in ParisStories, where nous has only 4; Rhapsodie, which includes more formal speech, has 291 on/l'on and 74 nous, the written GSD 593 and 241. The choices among tu, vous and on are treated in the chapter on registers.

5. Where the subject follows the verb

A noun subject stands after the verb in a limited set of contexts. In the GSD there are 413 postverbal noun or name subjects (329 common nouns, 84 proper names), classified by the relation of their verb:

The verbs in this position are few and typical: exister 28, venir 15, rester 14, expliquer 14, déclarer 14, indiquer 12, préciser 8, dire 8, annoncer 8, besides the copular wh-clauses (Quel est …, 35). They fall into three groups: the verbs of existence and appearance (exister, venir, rester, arriver, se produire), in which the subject is new to the discourse; the verbs of saying in inserted speech; and the verbs in relative clauses, where the subject is short and the verb carries the heavy complement. The rules are:

Inserted speech. The subject follows the verb in an incise after a quotation: « Je viens », dit-il. « Non », répondit Marie. « Peut-être », fit le professeur. The verb is a verb of speaking, in the present or the past; pronoun subjects are joined to the verb with a hyphen, a /t/ is inserted after a vowel (dit-il, répond-elle, a-t-il dit).

After certain adverbs, in writing. Peut-être, sans doute, aussi (in the meaning "therefore"), ainsi, à peine, du moins, encore, tout au plus, en vain, à plus forte raison, at the head of the clause, require inversion of a subject pronoun or a complex inversion of a noun subject: Peut-être viendra-t-il. Aussi le projet fut-il abandonné. À peine était-il parti qu'il revint. Speech replaces the inversion by que (Peut-être qu'il viendra) or by statement order (Peut-être il viendra).

Stylistic inversion in narrative and descriptive prose. A locative or temporal phrase at the head of the clause can be followed by the verb and then the subject: Dans le jardin poussaient des roses. Autour de la table s'étaient assis les enfants. The verb is one of existence, appearance or position, the subject is a new, indefinite and fairly heavy noun phrase, and the inversion is a stylistic choice with a precise function: it introduces a new referent after the verb. The "presentative" family (il y a, voilà, voici, il reste, reste) belongs to the same group: Reste la question de savoir ….

In relative and comparative clauses. A short verb of a relative clause can precede a noun subject: la maison qu'habitait mon grand-père, les mesures que prend le gouvernement; plus grand que ne le croyait son père. After dont the inversion is nearly the rule when the verb is short: les droits dont jouissent les citoyens. An inverted pronoun subject is not possible in a relative clause (*le livre que lis-je). Acceptability experiments on French (abeille-2019-extracting, winckel-2024) find that relativizing out of a subject noun phrase (un philosophe dont le portrait se trouve au Louvre) is rated as well as or better than relativizing out of an object, contrary to the subject-island constraint, while questioning out of a subject is rated worse; the authors attribute the difference to the discourse function of the two constructions (chapter 25).

In questions with inversion (chapter 26) and in subordinate clauses after que of necessity: Il faut que vienne le jour (literary).

Counts in speech. Postverbal noun subjects are rare in ParisStories (16 pronouns and 10 nouns in the 3 944 nsubj) and in Rhapsodie (31 nouns, 14 pronouns, 6 names): the incises of dialogue (il dit) and the questions (c'est quoi le …?) account for most. Inversion as a stylistic device is a feature of writing.

6. Objects

6.1 The direct object

The direct object has no preposition. It follows the verb and the participle (J'ai vu Marie), and precedes it when it is a clitic (Je l'ai vue). Its relation is obj. The distribution in the treebanks (counts of the obj relation, 2026-11-06):

The preverbal pronouns are the personal clitics (le, la, l', les, me, m', te, se, nous, vous, en), the relative que/qu' and the interrogative que; the postverbal pronouns are the strong forms and the demonstratives (cela, ceci, ça, tout). In GSD the object clitics most often seen are se 370 (the reflexive of se as direct object), le 266, l' 204, les 124, la 106, and the relative que 210 and qu' 233. The spoken corpus has twice the share of pronominal objects, because the clitic is the normal object of a verb whose referent is already in the discourse.

Object omission. French allows an object to be omitted in a few constructions: for verbs used absolutely (Il mange, Il lit), in generic sentences, and in the construction described by Aguila-Multner and Crysmann as the à-infinitival missing-object construction: un livre difficile à lire, une chose facile à faire, where the missing object of the infinitive corresponds to the head noun and the construction behaves like a passive (multner-2022). The object is not omitted in the finite clause: *Je mets sur la table must have le or le livre.

The article after a negation. A direct object with an indefinite or partitive article takes de under negation (Je n'ai pas de voiture). See chapters 27 and 22.

6.2 The indirect object

An indirect object has the preposition à (or de, avec, sur …). The relation is different in the treebank depending on whether it is a noun or a pronoun:

• A clitic lui, leur, me, te, nous, vous, se, y, en is iobj (870 in GSD; all are pronouns): Je lui ai donné le livre (Mova gives lui as iobj: §11). In ParisStories iobj is 358 in 42 789 words, 84 per 10 000, against 22 in GSD. The commonest are lui 270, y 113, en 112, vous 62, dont 57, nous 55, leur 50, se 43 in GSD; me 94, lui 71, m' 49 in ParisStories.
• A noun phrase with à is obl:arg: donner un livre à Marie has Marie as obl:arg with à as case. 3 870 obl:arg in GSD have à, 2 924 have de; the commonest verbs with à are naître (262), situer (135), participer (123), donner (83), mettre (75), appartenir (64).

The consequence for a reader of the trees is that the recipient of donner is iobj when pronominal and obl:arg when nominal. Chapter 23 gives the valency patterns.

Order of the objects. A direct object noun phrase precedes an indirect one: Il donne le livre à sa sœur. A heavy direct object may follow it: Il a donné à sa sœur le livre qu'il venait d'acheter. French has no dative shift (*Il donne sa sœur le livre). In a sentence recall experiment, French speakers recalling a sentence with a direct object and an indirect object tended to put the inanimate direct object before the animate indirect one, contrary to the animate-first tendency found in other languages; the authors attribute it to the canonical mapping of function to role (theme as direct object, recipient as indirect) rather than to animacy (thuilier-2021). In the same study, animacy did affect the choice between active and passive.

6.3 Two pronoun objects

Two object clitics stand together in the fixed order of §7. A pronoun object and a noun object: the pronoun is before the verb and the noun after (Il le donne à sa sœur, Il lui donne le livre).

7. The clitics in the clause

7.1 The cluster and its order

Every clitic of a clause stands before the verb group, in an order fixed by person and function:

Il me le donne. — Il le lui donne. — Il m'y envoie. — Il l'y a mis. — Je lui en ai parlé.

A pronoun of the first or second person and a dative of the third person cannot be combined before the verb: *Il me lui présente, but Il me présente à lui; this is a constraint on the combination (the "person–case constraint"). The cluster follows the verb in the affirmative imperative and the order changes: Donne-le-moi, Dis-le-lui, Vas-y, Donnes-en (with -s before en and y), and with a negative imperative the normal order returns: Ne me le donne pas. Details are in the pronouns chapter.

7.2 The clitics and the verb group

With a compound tense, the clitics precede the auxiliary: Je l'ai vu, Il le lui a donné. With a modal or an aspectual verb followed by an infinitive, the clitics of the infinitive stand before the infinitive: Il veut le voir, Elle peut le lui dire, Je vais t'en parler. A modern French clause does not let the clitic of the infinitive climb in front of the modal; it did in the past — il le veut prendre is cited by Clédat as an archaism from the time when the order of words was freer (cledat-1894) — and the loss of climbing and of interpolation dates from before the beginning of the nineteenth century (olivier-2023). The treebanks confirm the rule: of the object and indirect-object clitics of an infinitive that depends on another verb as xcomp, 355 of 357 stand directly before the infinitive in GSD, 110 of 110 in ParisStories, 80 of 81 in Rhapsodie (counts, 2026-11-06). The two exceptions in GSD are Le pape ne veut pas en entendre parler (fr-ud-train_05390), where en belongs to parler but stands before entendre, a verb of perception, and one with pouvoir.

The one productive exception is the causative with faire and, more weakly, the verbs of perception and laisser: the clitic of the infinitive climbs in front of the finite verb.

D'un coup de pied, il le fait tomber de la main du prêtre, par terre. (fr-ud-train_00044; the treebank annotates tomber as head and fait as aux:caus.) Son nouvel ami le fait engager comme figurant sur le tournage de Runaway Train. (fr-ud-dev_01146)

There are 101 object and indirect-object clitics (including the agents) attached to an infinitive with aux:caus in GSD, all standing before faire. See chapters 23 and 28.

7.3 The clitics and the negation

Ne stands immediately before the cluster, pas after the first verb: Je ne le lui donne pas, Je ne le lui ai pas donné, Il ne veut pas le lui donner. With an infinitive, ne pas precedes the cluster of the infinitive: Il préfère ne pas le lui dire. The negation chapter gives the details.

7.4 Clitics and trees

A clitic is attached in the tree to the verb of which it is an argument, not to the verb it stands beside. In a compound tense the clitic stands before the auxiliary but depends on the participle (Il le lui a donné: le obj, lui iobj, both of donné, with a aux:tense). The arcs of the clitics therefore cross the arcs of the auxiliary and of the subject wherever the scheme makes the auxiliary the head, and do not in UD, which makes the participle the head; this is one reason why SUD, which makes the auxiliary the head, has two and a half times as many non-projective nodes in the French GSD as UD (perrier-2021). The clitic en produces many of the remaining crossings in UD, because it can depend on an object noun or an attribute and not on the verb itself: Il en a pris trois, where en depends on trois (nmod).

8. Attributes and secondary predicates

8.1 The copular clause

Être links a subject to a predicate that can be an adjective, a noun, a pronoun, a prepositional phrase or an adverb. In UD the predicate is the head and être is cop:

The only copula in UD French is être. The predicate adjective agrees in gender and number with the subject. A noun of profession, nationality, rank or religion has no article after être when it is the whole predicate: Louis Bénech était médecin (fr-ud-train_04808), Il est président du conseil de la Banque de Montréal (fr-ud-train_02954), Youcef Boukhantech est chanteur algérien (fr-ud-train_02684); it takes the article when it is modified, or when the subject is ce: C'est un médecin remarquable. The demonstrative c' as subject takes the plural of the predicate noun in careful French (Ce sont mes amis), and the singular in speech (C'est mes amis).

C'est and il est. C'est is used with a noun phrase with a determiner, a strong pronoun and a proper name (C'est un livre, C'est moi, C'est Paul); il est with a bare noun (a profession) and with an adjective whose subject is a specific person or object (Il est grand, Elle est belle, Il est médecin). Before an adjective with a clause or an infinitive subject, written French has il est difficile de partir, spoken French c'est difficile de partir (fr-ud-test_00218: C'est tellement rare de pouvoir dîner au coin du feu !).

8.2 Other linking verbs

The predicate of other verbs is xcomp of the verb. The commonest verbs, in the GSD:

fr-ud-dev_01135 · UD French GSD Sous le rédacteur en chef conservateur Karl Heinz Hagen, la politique devint plus importante dans Bild. "Under the conservative editor-in-chief Karl Heinz Hagen, politics became more important in Bild."

fr-ud-train_07269 · UD French GSD Il semble vraisemblable que cette œuvre comportait quelques restes du style baroque tardif. "It seems likely that this work contained some remains of the late baroque style."

Devenir, rester, demeurer, sembler, paraître, avoir l'air, passer pour link a subject to its predicate; the same verbs with an object (rendre, trouver, juger, nommer, élire, appeler, croire, déclarer, considérer … comme) give a predicate of the object:

Il est élu président de la nouvelle organisation. (fr-ud-train_07252) — the passive of On l'a élu président. La junk-food rend aussi accro que les drogues dures ! (fr-ud-dev_00056) Gordon est considéré comme volage. (fr-ud-train_09834) — with comme before the predicate.

The attribute of the object agrees with it (Je les trouve belles), and the preposition comme or pour may introduce it (On le tient pour un menteur). In UD French the predicate of the object is the xcomp of the verb, and the object is the obj of the verb, not the subject of the predicate. The Subject=ObjRaising attribute in MISC records that the object is the subject of the xcomp (§10).

8.3 Secondary predicates and apposition

A depictive predicate adds a state to the verb: Elle est rentrée fatiguée. Il mange debout. Les enfants dorment tranquilles. The adjective agrees with the subject and is a xcomp or advcl or amod of the verb depending on the treebank. The same adjective as an adjunct of the noun is an apposition set off by commas (Fatigué, il est rentré).

9. Adjuncts and the order of complements

The order of the complements after the verb is object – indirect object – place – time – manner, with the heaviest last:

Il a vu ses amis au café hier soir. — "He saw his friends at the café last night."

but nothing forces it; a time adjunct is just as often at the head of the clause (Hier soir, il a vu ses amis au café). The adjunct at the head of the clause is set off by a comma if it is long; a short one often is not (Hier il est parti). The distribution of the adjuncts in the treebank (GSD, 2026-11-06): advmod 14 330, of which 8 760 depend on a verb (4 726 before it, 4 034 after it); obl:mod 15 399. Both ParisStories and Rhapsodie have far more adverbs per word (604 and 571 per 10 000 words against 358).

Adverbs. A short adverb of manner, degree or frequency stands after the finite verb and, in a compound tense, between the auxiliary and the participle: Il mange souvent du pain, Il a bien mangé, Elle a déjà fini, Il a trop parlé. It cannot stand between the subject and the verb (*Il souvent mange), unlike English. Long or sentence adverbs (malheureusement, heureusement, probablement, évidemment) stand at the head, after the subject set off by commas, or at the end. The adverb chapter gives the lists.

Prepositional adjuncts. A time or place phrase is obl:mod with the preposition as case: en 1995, à Paris, dans la rue, depuis hier, pendant deux ans, sur la table. The commonest prepositions of obl:mod in GSD: en 3 274, à 2 551, dans 1 581, sur 1 065, pour 962, de 676, avec 661, après 279, comme 272, sous 264, depuis 255.

Subordinate adjuncts. A time, cause, concession or condition clause is advcl and has its place at either edge: Quand il arrive, nous partons. Nous partons quand il arrive. Chapter 24 treats them.

The clause in the clause. A verb of speaking or thinking inserted in the middle of a sentence is a clause of its own (parataxis:insert, 183 in GSD): « Je viens, dit-il, tout de suite. »

10. The clause in the treebanks

The conventions that matter for reading a French clause in a UD tree:

1. The head of the clause is the lexical verb; auxiliaries and the copula depend on it. In a clause with a copula the head is the predicate; with il y a the head is the verb avoir and its expl:subj il and expl:comp y are leaves.
2. The subject is nsubj (or nsubj:pass in a passive, nsubj:caus in a causative, csubj for a clause). In the postverbal impersonal construction the real subject is nsubj or obj depending on the treebank. In GSD Il est arrivé un accident has il expl:subj and accident nsubj; in earlier annotations (and in Mova's output, §11) the noun is an obj.
3. Objects. obj for the direct object (noun or clitic), iobj for an indirect object that is a pronoun, obl:arg for a prepositional complement that is a noun. A direct object of a light verb is obj:lvc.
4. A predicate is the head of a copular clause; a predicate with a linking verb other than être is an xcomp.
5. The subject of non-finite verbs is recorded in MISC: Subject=SubjRaising, ObjRaising, OblRaising, Generic, Instantiated. In GSD there are 3 892 SubjRaising, 343 ObjRaising and 251 OblRaising (2026-11-06).
6. The inverted pronoun after the verb in a question (-tu, -il, -t-il) is nsubj, and expl:subj when a noun subject is also present (Le renouveau désiré se produira-t-il ?).
7. Speech. The doubled noun phrase is dislocated:subj, dislocated:obj or dislocated. A false start is reparandum, a filler discourse.
8. c'est. The demonstrative c' before est is nsubj (GSD 350, ParisStories 607, Rhapsodie 490) or expl:subj (187, 33, 82, in clefts and in the impersonal frame); the choice is not stable across treebanks.

11. How Mova handles it

What is done. After the tag mapping of chapter 20, the English parser learns the clause relations from the French trees. We trained it on the GSD training file and parsed raw sentences (en explain, 12 epochs, beam 8, CPU, 2026-11-06). The English tokenizer is used; that is a limit (chapter 20, §12). The results on our sentences were these:

What these show: the parser has learned the core of the clause (subject, object, the auxiliary and the passive) from 354 647 words of French. It is weak where the labels are fine-grained (impersonal il, dislocation) because it cannot see the French subtypes (chapter 20, §12), and where the apostrophe is: every clitic ending in ' (l', m', n', qu', c', j') is joined to the next word by the tokenizer.

What is not done.

• The frames reader (prag/src/frames.rs) builds participants from nsubj, obj and iobj of English trees. For French it would read Il pleut as an event with an agent il, and Il est arrivé un accident as an event with an object accident. What is needed: expl read as no participant (the English reader already does it for it rains and there is), the real subject of an impersonal read from the postverbal noun, and the dislocated noun phrase merged with its pronoun into one participant.
• Agreement. The treebank has the features; the parser's tags do not carry them, so subject-verb agreement is not used to attach the subject or to choose between candidate subjects (compare the English agree module in prag).
• The clitic cluster is learned as a tree shape, the person–case constraint and the order are not checked, and the clitics of a causative (about a hundred in GSD: 107 preverbal clitic pronouns on a verb with aux:caus, 2026-11-06) are learned from the aux:caus frame.
• Predicates. The xcomp predicate and the Subject= attribute of the MISC column are not used: the parser does not predict MISC, and the frames reader has no notion of the predicate of the object.

12. Sources

• Bally, Charles. Traité de stylistique française (1909), §§284–289 on the syntax of the spoken language, dislocation, anticipation and recall: bally-1909.
• Clédat, Léon. Grammaire raisonnée de la langue française (1894), on the infinitive, the subject and the place of the pronoun object: cledat-1894.
• Olivier, Sevdali & Folli, Clitic climbing and restructuring in the history of French (Glossa, 2023): olivier-2023.
• Perrier, Étude des dépendances syntaxiques non projectives en français (TAL, 2021): perrier-2021.
• Cassarà, Subject focus in French and Spanish (Language Science Press, 2026): cassara-2026.
• Winckel, French subject islands: an empirical investigation (Language Science Press, 2024): winckel-2024; Winckel & Abeillé, Extraction out of subject phrases in French: experimental evidence (2019): abeille-2019-extracting.
• Aguila-Multner & Crysmann, French missing object constructions (Glossa, 2022): multner-2022.
• Thuilier, Grant, Crabbé & Abeillé, Word order in French: the role of animacy (Glossa, 2021): thuilier-2021.
• Larrivée, To be or not to be informational: preverbal complements in Medieval French V2 configurations (Glossa, 2019): larrivee-2019; Larrivée et al., The correlation of micro-cues: the case of medieval French verb second (2025): larrivee-2025.
• UD French GSD, ParisStories, Rhapsodie, Sequoia, 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 (explain), en/src/gram.rs, prag/src/frames.rs.
• The full list is in Sources.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/fr/21-clause-structure/