🗣️ 14. Irregular verbs

French has one productive verb class, the verbs in -er, and one closed but still regular one, the verbs in -ir with -issant (13. Verbs). Everything else is the "third group", and the third group is not a class but a list: a few hundred verbs, grouped into a dozen families by the shape of their infinitive and stems. The list holds the commonest verbs of the language. In six French treebanks (§11) the fourteen most frequent verb lemmas, all of them irregular (être, avoir, faire, pouvoir, aller, devoir, dire, voir, prendre, naître, mettre, connaître, savoir, devenir), make 45.6% of all verb and auxiliary tokens. A learner, and a parser, meets them before anything else.

The chapter gives the principal parts that generate the forms (§1), then the paradigms: être and avoir (§2), aller, faire, dire (§3), the modals (§4), venir, tenir, prendre, mettre, voir, recevoir, asseoir (§5), the third-group verbs in -ir (§6), the verbs in -re and -oir by family (§7), the defective and impersonal verbs (§8), the regularities that survive (§9), speech (§10) and the counts (§11). The regular paradigms, the tenses and their uses, and the agreement of the participle are in 13. Verbs.

The paradigms were checked against Bescherelle's L'art de conjuguer (bescherelle-1848), which gives each model verb in full and lists under it the verbs conjugated alike. Where the 1990 rectifications (cslf-1990) allow another spelling it is given as well; the treebanks use the traditional spelling throughout.

1. Principal parts

An irregular verb is described by a small set of forms from which the rest follow by rule.

The rules for the simple tenses, as far as the irregular verbs need them:

• Imperfect: 1pl stem + -ais, -ais, -ait, -ions, -iez, -aient. Only être (ét-) departs from it.
• Present participle: 1pl stem + -ant; exceptions étant, ayant, sachant.
• Future and conditional: future stem + -ai, -as, -a, -ons, -ez, -ont / -ais, -ais, -ait, -ions, -iez, -aient.
• Present subjunctive: 3pl stem + -e, -es, -e, -ions, -iez, -ent; 1pl and 2pl take the 1pl stem where it differs (que nous prenions, que vous veniez). Six verbs have a subjunctive stem of their own: être (soi-), avoir (ai-), faire (fass-), pouvoir (puiss-), savoir (sach-), falloir (faill-); three more have one in the singular and 3pl only: aller (aill-), vouloir (veuill-), valoir (vaill-).
• Passé simple: four vowel types: -a- (the -er verbs only), -i- (-is, -is, -it, -îmes, -îtes, -irent), -u- (-us, -us, -ut, -ûmes, -ûtes, -urent) and -in- (venir, tenir and their compounds: vins, vins, vint, vînmes, vîntes, vinrent).
• Imperative: 2sg, 1pl, 2pl of the present without a subject; a 2sg in -es drops its -s (ouvre, offre, and va), which comes back before y and en (vas-y, offres-en). Four verbs take the subjunctive stem: sois, aie, sache, veuille.

The tables give the six persons as je, tu, il / nous, vous, ils. Compound tenses take avoir unless être is named: j'ai pris, il aura pris; elle est venue (13. Verbs).

2. The auxiliaries être and avoir

Être and avoir are the two most frequent verbs of every corpus: 16 093 and 8 532 VERB or AUX tokens in the six treebanks, together 34.1% of all verb tokens.

2.1 être "to be"

Infinitive être; present participle étant; past participle été, invariable like every participle of an intransitive verb. The compound tenses take avoir (j'ai été). Colloquial French also uses j'ai été as a past of "go" (j'ai été au marché), which normative grammar condemns.

The stems come from three sources: Latin esse (suis, est, sont, sois), stare (the imperfect ét-, Old French estoie) and the perfect fui (fus, fût); the future ser- continues the Old French serai beside estrai. Sommes is the only modern verb form in -mes outside the passé simple; Brunot (brunot-1899) traces the ending -ons of every other first plural to the same Latin sumus, through Old French somes, by analogy. The -tes of êtes has the same kind of rarity: only faites and dites (§3) keep the Latin unstressed ending -itis (facitis, dicitis), which the other verbs replaced by -ez.

In the six treebanks the commonest forms of être (VERB or AUX) are est 8 611, sont 1 648, été 1 380, était 1 354, être 717, the passé simple fut 580 and suis 292; furent has 100. The form soit (273 tokens, 2026-11-06) is the subjunctive (167 as être) and also the conjunction of alternation soit … soit (103 tokens CCONJ).

The 1990 rectifications drop the circumflex on i and u but keep it where it marks a verb ending: the passé simple fûmes, fûtes and the imperfect subjunctive qu'il fût (cslf-1990, II.4.a).

2.2 avoir "to have"

Past participle eu /y/, a spelling kept from Old French eü; the passé simple eus, eut has the same /y/. Present participle ayant. The existential il y a is avoir with the clitic y: il y avait, il y a eu, qu'il y ait; il y a dix ans is "ten years ago". The stems are a-/av-/ai-/ay-/au-/eu- and ont.

3. aller, faire, dire

3.1 aller "to go"

Participle allé (auxiliary être), present participle allant. Three stems: vais, vas, va, vont from Latin vadere, the future ir- from ire, and all-, whose origin Brunot (brunot-1899) gives as one of the etymologies still unresolved in his time (ambulare is the usual proposal). Outside the present, the future and the subjunctive aller is conjugated like an -er verb. Imperative va, but vas-y, and va-t'en. The imperative allez ! is also a discourse marker ("come on").

Aller + infinitive is the near future (il va pleuvoir), 13. Verbs. The compound s'en aller keeps the irregular present (je m'en vais). In the six treebanks (2026-11-06): va 230, aller 142, vais 77, allez 76, allait 45, vont 40.

3.2 faire "to do, to make"

Participle fait, faite; present participle faisant. The ai of faisons, faisais, faisant is pronounced /ə/ (/fəzɔ̃/). Défaire, refaire, satisfaire, contrefaire, surfaire follow it (vous défaites, vous satisfaites). Faire is the causative (faire réparer), the support verb of many fixed expressions (faire peur, faire attention), the weather verb (il fait beau) and the stand-in verb (qu'est-ce que tu fais ?). It is the third verb of the treebanks (1 758 tokens): fait 761, faire 492, font 92, faisait 83, faisant 54, fais 49, fit 43.

3.3 dire "to say"

Participle dit, dite; present participle disant. Among the compounds only redire keeps vous redites; contredire, dédire, interdire, médire, prédire have vous contredisez, interdisez, médisez, prédisez, and maudire follows finir (nous maudissons) except for its participle maudit. The singular passé simple is identical to the present (il dit "says" and "said"), and dit is also the participle. Dire has 668 tokens: dit 317, dire 164, dis 45, disons 21, disait 21, dites 12.

4. The modals and their relatives

Pouvoir, vouloir, savoir, devoir, falloir, valoir share a pattern: a stem that alternates between a stressed form (singular and 3pl) and an unstressed one (1pl, 2pl), an -u- passé simple, a contracted future stem and a participle in -u.

Pouvoir. The first singular puis survives in inversion (puis-je ?, formal) and in the literary je ne puis. The subjunctive puisse is on a stem found nowhere else, in all six persons (que nous puissions). The participle pu is invariable. The conditional is the polite form (pourriez-vous m'aider ?). In the treebanks: peut 433, peuvent 135, pu 104, pourrait 91, pouvait 53, pouvoir 48, peux 41 (verb lemma 1 186; the noun pouvoir has its own 122 NOUN tokens). The passé simple is rare: Elle put prolonger la vie du 15e Karmapa pendant 9 années (fr-ud-train_07288).

Vouloir. Subjunctive veuille, veuilles, veuille, voulions, vouliez, veuillent. Imperative veuille, veuillons, veuillez; the polite veuillez + infinitive (veuillez agréer) is its living use, the older imperative veux, voulez being rare. The conditional je voudrais is the commonest polite request. Fixed uses (vouloir dire, en vouloir à, vouloir bien) are in 19. The lexicon: layers, borrowings, multiword expressions and register. Example: le Maroc ne veut pas les juger (fr-ud-train_00716).

Savoir. Subjunctive and imperative sache, sachons, sachez; present participle sachant. Savoir is to know a fact or how to do something, connaître (§7.3) to be acquainted with. It is one of the few verbs that still take ne alone in formal writing: On ne sait de quel crime ils l'accusèrent au juste (fr-ud-train_01617); je ne saurais vous dire has the conditional in the sense "I could not" (27. Negation).

Devoir. The participle dû has its circumflex only in the masculine singular, which separates it from the article du; the 1990 rectifications keep that accent (cslf-1990, II.4.b). Devoir expresses obligation (je dois partir), probability (il doit être là), debt (je lui dois dix euros) and a planned future in the past (il devait venir). Its only compound is redevoir. 726 tokens: doit 224, doivent 92, devait 67, dû 62, devrait 60.

Falloir is impersonal (§8): il faut, il fallait, il fallut, il faudra, il faudrait, qu'il faille, il a fallu. It takes an infinitive (il faut partir), a subjunctive clause (il faut que tu partes) or a noun phrase (Il faudra encore du temps, fr-ud-train_00391; Il fallut toutefois attendre la saison 1919-1920, fr-ud-train_03955). 301 tokens. Il ne faut pas + infinitive is a prohibition ("one must not"); absence of need is il n'est pas nécessaire de, ce n'est pas la peine de.

Valoir is mostly third person, with the impersonal il vaut mieux (Il vaut mieux réserver car ce petit coin de Naples est bien connu, fr-ud-test_00237) and the fixed vaille que vaille. Équivaloir is conjugated like it; prévaloir too, except for the subjunctive que je prévale.

5. venir, tenir, prendre, mettre, voir, recevoir, asseoir

Venir and tenir. The family is large and regular within itself: devenir, revenir, parvenir, survenir, intervenir, provenir, se souvenir take être; prévenir, subvenir, contrevenir, circonvenir take avoir; convenir takes avoir in "suit" and, in careful usage, être in "agree". The compounds of tenir (retenir, maintenir, obtenir, contenir, soutenir, appartenir, détenir, entretenir, s'abstenir) take avoir. Devenir (371 tokens) is more frequent in the treebanks than venir (285), and both have passé simple forms in written narrative: Quand vint l'âge de la retraite (fr-ud-train_13347), il connut et devint ami de Mushanokoji Saneatsu (fr-ud-train_00371). Venir de + infinitive is the recent past (13. Verbs).

Prendre loses the d of the infinitive outside the singular present and the future, and doubles the n where the stem vowel is stressed: prenons /pʁənɔ̃/, prennent /pʁɛn/. Compounds: apprendre, comprendre, reprendre, surprendre, entreprendre, se méprendre. It is the ninth verb of the treebanks (529 tokens) and the fourth in the passé simple (38), after être, avoir and faire: le M10 prit le nom d'Achilles (fr-ud-dev_00287).

Mettre has one t in the singular present (/mɛ/) and two in the plural (/mɛt/). Compounds: admettre, commettre, compromettre, permettre, promettre, remettre, soumettre, transmettre, omettre, démettre. Se mettre à + infinitive is inchoative: Ce haricot se mit à germer (fr-ud-dev_01302). Mettre 421 tokens, permettre 326.

Voir has voy- /vwaj/ before a pronounced ending and a future verr- (/vɛʁe/, one /ʁ/). Revoir, entrevoir are alike; prévoir has the regular future prévoirai, and pourvoir the future pourvoirai and the passé simple pourvus. The passé simple il vit is homographic with the present of vivre: Il vit surtout dans les buissons (fr-ud-train_00199) is vivre. Voilà (from vois là) is tagged VERB in the treebanks (176 of its 177 tokens, under its own lemma voilà), and voici likewise (15). Voir has 623 tokens.

Recevoir writes ç before o and u to keep /s/ (reçois, reçu, reçut). The family: apercevoir, concevoir, décevoir, percevoir. 232 tokens; Thomas reçut la partie nord-ouest alors que Démétrios recevait la partie sud-est (fr-ud-train_01616) has the passé simple and the imperfect side by side.

Asseoir has two conjugations. Bescherelle gives the older one: j'assieds, tu assieds, il assied, nous asseyons, vous asseyez, ils asseyent; imperfect asseyais; passé simple assis; future assiérai; subjunctive asseye; imperative assieds, asseyons, asseyez; participle assis. The other is j'assois, nous assoyons, ils assoient, j'assoirai, que j'assoie. The 1990 rectifications spell the infinitive assoir (also rassoir, sursoir), since the e is no longer pronounced; surseoir keeps its future surseoirai (cslf-1990, list and note on assoir). The verb is rare: 13 tokens.

6. Third-group verbs in -ir

6.1 partir, sortir, dormir, sentir, servir, mentir

These verbs lose the last consonant of the stem in the singular present (je pars, il part /paʁ/) and keep it in the plural (partons, partent /paʁt/). Imperfect partais, passé simple partis, future partirai, subjunctive parte, participle parti.

Also repartir, consentir, ressentir, pressentir, se repentir, desservir, endormir. Sortir takes être when intransitive (il est sorti) and avoir when transitive (il a sorti les poubelles). Ressortir in the sense "to come under (a jurisdiction)" is conjugated like finir (il ressortit), and asservir follows finir too. In the treebanks: partir 296, sortir 181, servir 117, sentir 56, dormir 23.

6.2 ouvrir, offrir, couvrir, souffrir, cueillir

The present has the endings of the -er verbs: j'ouvre, tu ouvres, il ouvre, nous ouvrons, vous ouvrez, ils ouvrent; imperative ouvre; subjunctive ouvre. The rest is -ir: imperfect ouvrais, passé simple ouvris, future ouvrirai. The participle ends in -ert: ouvert, offert, couvert, souffert. Cueillir (and accueillir, recueillir) has in addition the future cueillerai, as if from -er, and the participle cueilli. Assaillir and tressaillir have the same present but the future assaillirai. In the treebanks: ouvrir 120, offrir 88, accueillir 61.

6.3 courir, mourir, acquérir, fuir

Courir and mourir (with accourir, parcourir, secourir) are the verbs whose future has a pronounced double /ʁʁ/ (je courrai /kuʁʁe/ against the imperfect je courais /kuʁɛ/). Conquérir, requérir, s'enquérir follow acquérir; s'enfuir follows fuir. Mourir has 200 tokens, mostly mort (118): Charles mourut d'une maladie du cœur le 8 juin 1902 (fr-ud-train_10006).

6.4 vêtir, bouillir, haïr

Vêtir: vêts, vêts, vêt, vêtons, vêtent, participle vêtu. Bouillir: je bous, il bout, nous bouillons, participle bouilli. Haïr follows finir except the singular present je hais, tu hais, il hait /ɛ/ without the diaeresis; elsewhere the ï keeps two syllables (haïssons /a.i.sɔ̃/), and so the passé simple nous haïmes has no circumflex.

7. Verbs in -re and -oir

7.1 -indre: craindre, peindre, joindre

All the verbs in -aindre, -eindre, -oindre are alike (plaindre, contraindre, atteindre, éteindre, feindre, teindre, restreindre, rejoindre). The singular has a nasal vowel and no consonant (/kʁɛ̃/); the plural stem has /ɲ/ (craignons /kʁɛɲɔ̃/). The imperfect and subjunctive keep the i after gn: nous craignions, vous craigniez.

7.2 -oudre: résoudre, absoudre, dissoudre, coudre, moudre

Résoudre: résous, résous, résout, résolvons, résolvez, résolvent; résolvais; résolus; résoudrai; résolve; participle résolu. Absoudre and dissoudre conjugate alike but have no passé simple in Bescherelle's table and their participle is absous, absoute; dissous, dissoute, which the 1990 rectifications regularise to absout, dissout after the feminine (cslf-1990). Coudre: couds, cousons, cousis, coudrai, cousu. Moudre: mouds, moulons, moulus, moudrai, moulu. Résoudre has 22 tokens.

7.3 -aître, -oître: connaître, naître, paraître, croître

The traditional spelling puts a circumflex on the i before t (il connaît, connaître, connaîtrai) and nowhere else (je connais, connaissons). The 1990 rectifications drop it (il connait, paraitre, il nait) but keep every circumflex of croître that separates it from croire (je croîs, il croît, crû), not in its compounds (accroitre) (cslf-1990, II.4.b). The family: reconnaître, apparaître, disparaître, comparaître, accroître, décroître. Naître has 457 tokens because GSD is rich in biographies: né 325, née 83, naît 20, naquit 3 (Lazare Boldi, fameux peintre à l'huile et à fresque, naquit à Pistoie en 1624, et mourut en 1703, fr-ud-train_10322; Bryon Anthony McCane naît en 1976, fr-ud-train_00556). Connaître has 393, paraître 58 (Il paraît que oui..., fr-ud-train_00683).

7.4 -uire: conduire and its family

Conduire: conduis, conduis, conduit; conduisons, conduisez, conduisent; conduisais; conduisis; conduirai; conduise; participle conduit. The plural, imperfect, passé simple and subjunctive have an -s- (/kɔ̃dɥiz-/) that the singular lacks. The family is large in the learned vocabulary: produire, traduire, réduire, introduire, séduire, déduire, construire, détruire, instruire, cuire. Luire and nuire have the participles lui, nui. Il conduit and the participle conduit are homographs (Tout ceci nous conduit au désagréable sentiment que …, fr-ud-train_00038). Produire 157 tokens, conduire 73.

7.5 vivre, suivre, écrire, lire, rire, conclure

Compounds: survivre, poursuivre, décrire, inscrire, prescrire, transcrire, élire, relire, sourire, exclure (participle exclu) and inclure (participle inclus, incluse). Nous rions, vous riez in the present and nous riions, vous riiez in the imperfect differ only by one i. Two homographs matter: je suis is être or suivre (292 against 3 in the treebanks), and il vit is vivre or voir (§5). Treebank examples: Chacun vivait dans son village (fr-ud-train_04480), Spinoza y vécut de 1663 à 1670 (fr-ud-train_08637), Elle écrivit aussi pour Girls Aloud (fr-ud-train_01933). Suivre has 200 tokens, écrire 192, vivre 154, lire 60.

7.6 boire, croire, plaire, taire

Boire has three stems (boi-, buv-, boiv-). Croire has croy- before a pronounced ending; its participle cru has no circumflex, unlike crû of croître. Plaire writes il plaît, which the 1990 rectifications align with il tait and il fait as il plait (cslf-1990, II.4); déplaire, complaire are alike. The participle plu (also of pleuvoir) is invariable; tu agrees in elles se sont tues. Croire 136 tokens (personne n'y croyait, fr-ud-train_11852), boire 9.

7.7 -endre, -ondre, -erdre, -ompre, -attre, -aincre

The largest set of -re verbs: vendre, rendre, attendre, entendre, descendre, tendre, répondre, perdre, mordre, tordre, rompre, battre, vaincre. The singular present is -s, -s and, in the third person, the bare stem: il vend, il répond, il perd; rompre adds -t (il rompt), battre has je bats, il bat with one t, and vaincre has il vainc, with qu before a vowel (vainquons, vainquis). The plural keeps the stem consonant (vendons, vendent /vɑ̃d/). All these verbs have a participle in -u (vendu, répondu, perdu, rompu, battu, vaincu) but a passé simple in -i- (vendis, répondis, perdis, rompis, battis, vainquis): the participle does not predict the passé simple here (§9.2).

7.8 envoyer, mouvoir

Envoyer (and renvoyer) is an -oyer verb (envoie, envoyons) with one irregularity, the future enverrai. Mouvoir: meus, meus, meut, mouvons, mouvez, meuvent; passé simple mus; future mouvrai; subjunctive meuve; participle mû, spelled mu after 1990 like su, vu, lu (cslf-1990, II.4). It is rare outside the infinitive and the participle; émouvoir and promouvoir conjugate alike with participles ému, promu.

8. Defective and impersonal verbs

A defective verb lacks persons, tenses or the compound forms; most of them are rare or literary.

The impersonal constructions take the dummy subject il: the weather verbs, falloir, il s'agit de, il y a, il paraît que, il suffit de, il convient de, il importe de, il vaut mieux, il reste, il arrive que, il se peut que. Agir has 178 tokens in the treebanks, at least 146 of them right after the reflexive s', as in Il s'agit du 17e but de la saison (fr-ud-train_00031). The syntax of the impersonal subject is in 21. Clause structure.

9. Regularities among the irregularities

9.1 The future stem

The future stem is the infinitive (without the -e of -re) except in this closed list:

The written rr of verr-, pourr-, enverr-, acquerr- is pronounced as a single /ʁ/; only in courir and mourir (and their compounds) is it a long /ʁʁ/ that separates the future from the imperfect. The -oir verbs lose the vowel before r (devr-, recevr-, saur-), and vouloir, valoir, falloir, venir, tenir insert a d (voudr-, viendr-). The conditional uses the same stem.

9.2 The passé simple

The passé simple is not derivable from the other principal parts, but there are tendencies. A participle in -i, -is, -it goes with an -i- passé simple (parti → partis, pris → pris, mis → mis, dit → dis, fait → fis, conduit → conduisis, écrit → écrivis). A participle in -u goes with -u- in the -oir verbs and in lire, boire, croire, plaire, vivre, connaître, paraître, courir (voulu → voulus, reçu → reçus, lu → lus, vécu → vécus, couru → courus), but with -i- in the -dre verbs, battre, rompre, vaincre and voir (vendu → vendis, battu → battis, vu → vis). Outside both patterns are venir, tenir (-in-), naître (naquis), mourir (mourus), être (fus) and avoir (eus). The imperfect subjunctive is the 2sg passé simple + -se (que je prisse, que je vinsse), with the circumflex in the 3sg (qu'il prît, qu'il vînt). In the six treebanks the imperfect subjunctive has 11 tokens (§11).

9.3 The subjunctive

The present subjunctive is regular on the 3pl stem except for the verbs of §1. Where the 1pl stem differs, 1pl and 2pl use it: que nous voyions, que vous buviez, que nous allions, que nous voulions, que nous valions. The verbs whose own stem goes through all six persons are être (soyons), avoir (ayons), faire (fassions), pouvoir (puissions) and savoir (sachions).

9.4 Stem alternations

• Stressed and unstressed vowel: peux/pouvons, veux/voulons, meurs/mourons, tiens/tenons, reçois/recevons, bois/buvons, dois/devons, acquiers/acquérons.
• Consonant of the plural stem, silent in the singular: part/partent, vend/vendent, met/mettent, résout/résolvent, craint/craignent, écrit/écrivent, conduit/conduisent.
• Glide y before a pronounced ending: voyons, croyons, fuyons, envoyons; i before mute e (je crois, ils croient).
• -ss- in the plural stem: connaissons, naissons, like the -ir verbs of the second group (finissons).

10. In speech

The irregular verbs keep audible distinctions that the -er verbs have lost. Je parle, tu parles, il parle, ils parlent sound alike, but il peut /pø/ and ils peuvent /pœv/, il prend /pʁɑ̃/ and ils prennent /pʁɛn/, il met /mɛ/ and ils mettent /mɛt/, il fait /fɛ/ and ils font /fɔ̃/ do not. Spoken French has three further tendencies:

1. The frequent forms reduce: je suis /ʃɥi/, je sais /ʃɛ/, tu as /ta/, il y a /ja/ (04. Connected speech: enchaînement, liaison, elision, the mute e and assimilation). The passé simple and the imperfect subjunctive are absent.
2. Aller + infinitive replaces the simple future in many uses, so the irregular future stems are more frequent in writing.
3. The distinct subjunctives (que je sache, puisse, fasse, aille, sois, aie) are where the mood is audible; for an -er verb the subjunctive singular and 3pl sound like the present indicative.

11. Counts

Six modern UD French treebanks, GSD, Sequoia, ParisStories, Rhapsodie, FQB and PUD (UD 2.18, all train, dev and test files in UD_French-*/), 606 575 syntactic words, counted on 2026-11-06. ALTS (sixteenth century) and ParTUT (non-commercial) are left out. A verb is a token with UPOS VERB or AUX; the lemma is the treebank's (column 3).

The passé simple is 4.8% of the finite verbs, nearly all in the third person and a third of it être; fut has 580 tokens and furent 100. A tool that reads written French meets the irregular passé simple forms (fut, fit, prit, devint, naquit) more often than the regular ones in -a. The imperfect subjunctive, 11 tokens, matters for generation and literary text only.

12. How Mova handles it

• No French morphology. Mova has no French conjugation tables, lexicon or lemmatiser. Its morphology module, en/src/morph.rs, is English: lemmatisation and inflection over the built-in English dictionary (the AGID paradigms and UD frequencies) with English suffix rules, and its paradigm has eight cells (Slot: base, past, past participle, present participle, third singular, plural, comparative, superlative). A French verb has some fifty simple forms and does not fit it; en morph-eval on a French treebank would measure nothing useful and has not been run.
• What the tagger sees. For French, the treebank's UPOS and FEATS are mapped to Penn tags by en xpos-fill (xpos_of in en/src/main.rs): VB for an infinitive, VBN for a participle, VBD for Tense=Past or Tense=Imp, VBZ for a third singular, VBP for the rest. The rule that would give VBG to a present participle requires Tense=Pres in FEATS, but GSD keeps the participle's tense in the MISC column (Tense[denom]), so all its 13 769 participles become VBN, étant and faisant included. The passé simple and the imperfect are both VBD; the future, the conditional, the subjunctive and the imperative become VBZ or VBP by person. The mood is therefore invisible to the parser, and a check that needs it (the subjunctive after il faut que) cannot be made.
• Unseen forms. The tagger learns from the training treebank; a form not seen there is tagged from its letters and context. Its accuracy on such words is 84.8% (GSD train → dev, UD 2.18, 2026-10-22). The common forms of this chapter (est, a, fait, peut, va) are seen; the rare ones (puissions, vécûmes, dites, assiérai) are mostly not.
• Homographs. Suis (être/suivre), vit (vivre/voir), dit, fait, mis, pris (present or passé simple, and participle) are left to the tagger's context; no lemma is predicted.
• What a fix would be. A conjugation generator in Rust from the principal parts of §1 (a few hundred third-group lemmas with their parts, plus the rules of the first two groups) would give every form with its features; reversed, it would lemmatise and tag forms the treebank lacks. The paradigms of this chapter are its specification, and the lemma and FEATS columns of the six treebanks are its check (a count of mismatches). Sagot (sagot-2019) shows the same design for Old French: lexemes with an inflection class, from which the Alexina framework expands about a million inflected forms; used as an external lexicon it raised a tagger's UPOS accuracy on the Old French SRCMF dev set from 93.8% to 94.8%, and the author notes that irregular verbs were the part still imperfectly described. Nothing of this exists in Mova.

Sources for this chapter

• Bescherelle, L'art de conjuguer (1848 ed.), bescherelle-1848: the model paradigms, asseoir, absoudre, choir, seoir.
• Brunot, Précis de grammaire historique de la langue française (1899), brunot-1899: sommes and the ending -ons, faites/dites from facitis/dicitis, the unresolved etymology of aller.
• Conseil supérieur de la langue française, Les rectifications de l'orthographe (1990), cslf-1990: the circumflex on i and u (II.4), assoir/sursoir, absout/dissout, mu, plait.
• Sagot 2019, Développement d'un lexique morphologique et syntaxique de l'ancien français (TALN, CC BY 4.0), sagot-2019.
• UD 2.18 French GSD, Sequoia, ParisStories, Rhapsodie, FQB, PUD: counts of 2026-11-06; examples quoted from GSD (CC BY-SA 4.0) with sent_id.
• Mova: en/src/main.rs (xpos_fill, xpos_of), en/src/morph.rs; baselines in What every language gets.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/fr/14-irregular-verbs/