🗣️ 11. Nouns

A French noun has a grammatical gender, masculine or feminine, that it keeps in all its uses, and a number, singular or plural, that it takes according to the sense. There is no case (the Old French two-case declension is gone, §5.1) and no neuter. Gender and number show on the words around the noun (the determiner, the adjective, the pronoun that refers back to it, sometimes the past participle) far more than on the noun itself, and in speech the plural of most nouns is not heard at all. This chapter covers gender (§1), how to guess it (§2), forms that exist in both genders (§3), the feminine of nouns of persons and animals (§4), the plural and its exceptions (§5), compounds (§6), proper names (§7), count, mass and plural-only nouns (§8), the gender of acronyms, loans and clippings (§9), and feminisation and inclusive writing (§10); §11 says how Mova handles nouns. Agreement of the words around the noun is in 12, 13 and 15; the word class in 10; noun-forming suffixes in 17.

1. Gender

1.1 Two genders

Latin had three genders; French has two. Brunot (brunot-1899, §§181–185) separates the logical neuter (a thing named as neither male nor female) from the grammatical neuter (a form of its own). The grammatical neuter disappeared and its nouns passed to the masculine or the feminine; the logical neuter survived, expressed mostly by the masculine and by pronouns: ce, cela, ça, quoi, the le of je le suis, the impersonal il (12). The traces he lists are Old French feminine collectives made from Latin neuter plurals (arme, brace, charre, doie from arma, brachia, carra, digita), the numeral mille, the old neuter adjectives pis and mieux (now adverbs), and pronouns such as the o of oïl > oui and ce. Some neuter plurals in -a were taken for feminine singulars because they ended like rosa: folia → la feuille, gaudia → la joie, cornua → la corne, and by analogy from adjectives mirabilia → la merveille, paria → la paire; they later took a regular plural in -s (les feuilles, les merveilles; §195).

Every noun has one of the two genders, whether it names a thing or a person: le livre, la table, le soleil, la lune. For things the gender is arbitrary from the speaker's point of view (German has die Sonne, der Mond); for persons and animals with a sex it usually follows the sex, with the exceptions of §3.3 and §10.

1.2 Where gender shows

• The determiner: le, un, ce, mon, quel against la, une, cette, ma, quelle. Before a vowel the definite article loses the contrast (l'ami, l'amie), the possessive uses the masculine form (mon amie, not *ma amie), and the masculine demonstrative has its own form (cet arbre).
• The adjective: un grand homme, une grande femme.
• The pronoun that refers to the noun: il / elle, le / la, celui / celle, lequel / laquelle.
• The past participle where it agrees (13).

On the noun itself gender is marked only by some derivational suffixes (§2.2) and by the feminine of nouns of persons and animals (§4).

In UD French GSD (UD 2.18, train + dev + test, counted 2026-11-06), 75 175 of the 75 330 NOUN tokens carry Gender: 41 742 masculine (55.5%), 33 433 feminine (44.5%). Of the 155 without it, 134 are plural with Number only, 17 singular with Number only and 4 have neither. By gender and number: masculine singular 30 689, feminine singular 25 654, masculine plural 10 961, feminine plural 7 574.

2. Guessing the gender

The gender of a noun naming a thing is lexical, but the form says a lot about it. Sahai and Sharma (sahai-2021) trained classifiers of French noun gender on different kinds of input. The written form alone (the last letters above all) gives 92.5% accuracy; the pronunciation alone 81%; the meaning alone (semantic features of the noun's English translation) 77%; spelling plus meaning 94.0%, and adding the pronunciation does not help. Spelling beats sound because many endings that differ in writing sound alike (their example: lit /li/ "bed", masculine, and lie /li/ "dregs", feminine). Meaning helps where the spelling misleads: poix "pitch" and ardeur "ardour" are feminine although they look masculine. The authors also explain the neural model by small decision trees over endings and semantic classes, a form a learner can use.

Neural language models learn gender as an abstract category. Sukumaran, Houghton and Kazanina (sukumaran-2024) gave an LSTM and a small transformer a made-up noun in one or two sentences with le or la and tested agreement in other contexts (adjective, participle, lequel/laquelle). Both generalised, with a bias for the masculine: feminine nouns stayed less accurate even after ten examples. Twenty-five native speakers given the same one-shot task showed the masculine bias too.

2.1 By meaning

Male persons and animals are masculine and female ones feminine where the language has a pair (un garçon / une fille, un cheval / une jument). Days, months, seasons, languages, colours, metals and most trees are masculine (le lundi, le français, le rouge, le fer, le chêne). Sciences in -ique are feminine (la physique, la musique, la logique). Country, region and river names in -e are mostly feminine (la France, la Bretagne, la Seine), with exceptions (le Mexique, le Cambodge, le Zimbabwe, le Rhône); the others are masculine (le Canada, le Portugal, le Japon, le Danube). In GSD pays is masculine in all 289 tokens; ville (321), commune (301), région (272) and équipe (233) are feminine in all theirs, département masculine in all 161 (lemma counts, 2026-11-06).

2.2 By ending

The endings below decide most often. The counts are GSD NOUN tokens by the ending of the lemma (UD 2.18, 2026-11-06); an ending is matched on the lemma, so -ité nouns are also inside -té.

The feminine -eau tokens are almost all eau itself; the feminine -eur ones are the abstract nouns. Other tendencies: -oir, -ier, -in, -on, -al, -o, -at mostly masculine; -ade, -ette, -ine, -ille, -aison, -erie, -ière, -ude mostly feminine.

How strong this is: matching the fifteen endings of the table in the order -tion, -ité, -té, -ment, -age, -ance, -ence, -esse, -ure, -ie, -isme, -eau, -ée, -eur, -sion (first match wins) covers 22 313 of the 75 175 gendered noun tokens of GSD (29.7%) and predicts the right gender for 21 033 of them (94.3%). The crude rule "lemma ends in -e → feminine, otherwise masculine" is right on 63.2% of the -e tokens and on 70.6% of the rest (2026-11-06).

Meaning can override the ending: a metonymic use takes the gender of the hidden noun (un Picasso "a painting", une Peugeot "a car").

2.3 What a program should do

The gender of each noun has to be learned with the word, and a dictionary gives it for every entry. For a system that produces or checks agreement the order of evidence is: a lexicon; for an unknown noun, the agreeing words around it in the sentence (determiner, adjective, participle); then the ending table; then the masculine default; for a noun of a person, the sex of the referent where it is known.

3. Words with two genders

3.1 Homonyms that differ in gender

Some forms are two nouns, one of each gender, with different meanings:

A lexicon must store them as separate entries; the determiner or an agreeing word decides which is meant.

3.2 Gender that changes with number or position

Amour, délice and orgue are masculine in the singular and may be feminine in the plural in literary style (un grand amour, de folles amours; les grandes orgues of a church). Gens "people" has no singular; by the classical rule an adjective placed before it is feminine and one after it masculine (de bonnes gens, des gens heureux), a rule now often relaxed. GSD does not carry the traditional gender of gens: of its 30 tokens 21 have Number=Plur only, 8 are masculine plural and 1 feminine plural, as in Les gens sont accueillants (fr-ud-train_01230, masculine noun and predicate). Après-midi is masculine in the dictionaries, but une après-midi is also used.

3.3 One gender for both sexes

Some nouns of persons have one gender whatever the sex: feminine la personne, la victime, la vedette, la recrue, la sentinelle; masculine le témoin, le mannequin, le bébé, un individu, un ange. Agreement follows the grammatical gender: La victime est un homme, Elle est un excellent témoin. Most animal names work the same way: une souris, une girafe, une grenouille, un crocodile, un papillon, un moustique; the sex is added with mâle or femelle (une girafe mâle).

Epicene nouns are the opposite case: one form, both genders, the determiner telling the sex (un/une élève, un/une artiste, un/une collègue, un/une enfant). GSD often leaves them unmarked or tags them by default: 129 of the 148 tokens of enfant have Number and no Gender, while all 41 tokens of élève are masculine (lemma counts, 2026-11-06). Only 134 of the 9 466 GSD noun lemmas occur with both genders; most are such epicenes or annotation variation (not checked one by one), since a pair like directeur / directrice has two lemmas.

4. The feminine of nouns of persons and animals

The feminine is derived from the masculine by the same rules as the adjective (15). The base rule adds -e; a final consonant silent in the masculine is then pronounced, and several consonants double in writing (Brunot, brunot-1899, on the feminine in -e).

The -teur / -trice nouns go back to Latin -tor / -trix; where a French verb in -ter stands beside the noun, the feminine is usually -teuse (chanteur goes with chanter, acheteur with acheter → acheteuse). The -esse suffix was productive in the Middle Ages and is now closed; newer feminines of titles use -e (maire → une maire) or -eure (§10.2).

In speech the feminine is audible only when the -e uncovers a consonant or changes the vowel: marchand /maʁʃɑ̃/, marchande /maʁʃɑ̃d/; chien /ʃjɛ̃/, chienne /ʃjɛn/. Ami / amie (/ami/), employé / employée (/ɑ̃plwaje/) and Daniel / Danielle sound the same. So for a large part of the vocabulary gender is written, not heard, and a system that writes French from speech must take it from context (01. Consonants, 02. Vowels).

5. Number

5.1 The regular plural

The regular plural adds a silent -s: le livre / les livres. In speech the number of the noun is carried by the determiner (le / les, un / des, ce / ces, mon / mes) and, before a vowel, by liaison (les‿amis, 04. Connected speech: enchaînement, liaison, elision, the mute e and assimilation); a following adjective is not audibly plural either (des livres rouges), unless noun or adjective has a stem change (un cheval / des chevaux).

The -s is the old oblique plural. Old French had two cases, a subject case and an oblique case; in the main masculine class the -s marked the subject singular (murs) and the oblique plural (murs). When the case system broke down the oblique forms survived, and -s became the plural mark (Brunot, brunot-1899, §§190–206).

GSD's 18 669 plural noun tokens (NOUN with Number=Plur, 2026-11-06) end as follows:

In 1 096 of these tokens the plural form equals the lemma: the units (km, h) and the nouns in -s, -x that do not change, led by pays (118), fois (57), mois (50), prix (41), cours (31), gens (30) and cas (27).

5.2 The exceptions

• Nouns in -s, -x, -z do not change: un fils / des fils, une voix / des voix, un nez / des nez.
• Nouns in -au, -eau, -eu take -x: un tuyau / des tuyaux, un bateau / des bateaux, un jeu / des jeux. Exceptions in -s: landau, sarrau, bleu, pneu, émeu, and lieu "pollack" (lieus; lieu "place" has lieux).
• Nouns in -ou take -s (des trous, des clous, des fous) except seven in -x: bijou, caillou, chou, genou, hibou, joujou, pou.
• Nouns in -al make -aux (un cheval / des chevaux, un journal / des journaux; aux journaux, fr-ud-dev_00755); a few take -s: bal, carnaval, chacal, festival, récital, régal. GSD has animaux 28, journaux 12, chevaux 11 tokens.
• Nouns in -ail take -s (des détails, des éventails) except bail, corail, émail, soupirail, travail, vantail, vitrail, which make -aux. Travaux is the most frequent -aux plural of GSD (57 tokens; les nouveaux travaux, fr-ud-test_00286); travails survives only in technical senses.
• Irregular: un œil / des yeux (26 GSD tokens of yeux), le ciel / les cieux (beside des ciels in painting and weather), un aïeul / des aïeux "ancestors" (aïeuls "grandparents"), and the compounds with a determiner or adjective inside, where both parts change: monsieur / messieurs, madame / mesdames, mademoiselle / mesdemoiselles, bonhomme / bonshommes, gentilhomme / gentilshommes.
• Loss of a pronounced consonant: un œuf /œf/ / des œufs /ø/, un bœuf /bœf/ / des bœufs /bø/; for os the traditional plural /o/ now competes with /ɔs/.

The -x has a scribal origin that Brunot (brunot-1899, §206) traces. Before -s, Old French l became u, so chevals became chevaus, as Latin alba became aube. Scribes abbreviated the group -us with a sign that came to be written x: chevax. The sign was then read as the letter x, taken as a plural mark, and since a u was heard it was written back: chevaux. The Renaissance added the etymological l (chevaulx), dropped again in the seventeenth century. This is why the -x plurals cluster after au, eu, ou.

The audible plurals are therefore few: -al → -aux, the seven -ail → -aux, œil / yeux, ciel / cieux, œuf, bœuf (and os), and the titles. Everywhere else the noun's number is heard only through its determiner.

5.3 Plurals with another sense

Some plurals mean more than "several": des affaires "business, belongings", les échecs "chess", les lettres "literature", les cieux "heaven", des lunettes "glasses", des jumelles "binoculars", les vacances "holidays" (une vacance "a vacancy"). A mass noun in the plural names kinds or portions: des vins "kinds of wine", deux cafés "two cups of coffee".

6. Compound nouns

6.1 The plural of hyphenated compounds

The traditional rule marks the elements that are nouns or adjectives and that can be plural in the sense of the whole:

Because the verb + noun type depends on the sense (un compte-gouttes has an -s even in the singular; des gardes-chasse or garde-chasses), the 1990 rectifications (cslf-1990) replaced it with one rule: compounds of verb + noun and preposition + noun are treated like simple nouns and take the plural mark only when plural, on the last element: un pèse-lettre / des pèse-lettres, un cure-dent / des cure-dents, un après-midi / des après-midis (06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications).

GSD has 946 NOUN tokens with a hyphen (2026-11-06), kept as one token. The frequent lemmas are the compass points (sud-ouest, sud-est 24 each, nord-ouest 20, nord-est 18), porte-parole and demi-finale 16, week-end 15, rendez-vous and chef-lieu 13, vice-président 12, après-midi 11, grand-père 10; in the plural demi-finales 9, beaux-arts 5, sous-marins, sages-femmes, belles-lettres, années-lumière 3 each, timbres-poste and sapeurs-pompiers 2 each. Un gratte-ciel is in fr-ud-train_04898. Word formation by compounding is in 18.

6.2 Compounds written as one word

A welded compound inflects like a simple noun: un portefeuille / des portefeuilles, un passeport / des passeports, un parapluie / des parapluies. Bonhomme / bonshommes keeps the old plural of its first part. The 1990 rectifications welded a list of compounds (among them portemonnaie, weekend, piquenique); the treebanks use the traditional spellings. Compounds that contain a phrase are invariable: des tête-à-tête, des laissez-passer, des je-ne-sais-quoi.

7. Proper names

Family names do not take the plural: les Dupont, les Martin; dynasties may (les Bourbons). Plural place names take a plural article: les États-Unis, les Pays-Bas, les Philippines (aux États-Unis, fr-ud-dev_00569). Nouns of nationality are common nouns written with a capital (un Français, les Italiens, une Québécoise); the adjective and the language name are lower case (un livre français, le français). Inhabitants of towns take -ien, -ais, -ois, -ain and others (Parisien, Lyonnais, Marseillais; 17).

Names of persons have no gender of their own; the title or determiner shows it: Madame Tellier (fr-ud-test_00263), where Madame is a NOUN, Gender=Fem, and Tellier depends on it as flat:name. Place names choose their preposition by gender and number: en France, en Iran (feminine, or masculine beginning with a vowel), au Canada (other masculine), aux États-Unis (plural), à Paris (towns) (22. Phrases). Some names carry an article (Le Havre, La Rochelle, La Haye); names built on a common noun keep its gender (la Banque de France, l'Assemblée nationale).

In GSD only 3 214 of the 27 693 PROPN tokens carry Gender (88.4% do not); France has Gender=Fem in 78 of its 308 tokens. Multi-word names are linked by flat:name (7 005 tokens with that relation; 2026-11-06).

8. Count, mass, collective and plural-only nouns

Count and mass. A count noun takes un and numerals (un livre, trois livres); a mass noun takes the partitive (du pain, de l'eau, 12). Many nouns have both uses with a shift of sense (du poulet "chicken meat" / un poulet; de la bière / une bière "a glass of beer"). English furniture is a mass noun counted by its pieces; its French translation meubles is a count noun. Le, Bale and Barner (le-2026) used this contrast: French speakers judging "who has more meubles" and English speakers judging "who has more furniture" both compared by number of pieces, so the individuation belongs to the word's meaning and does not depend on count or mass syntax. French also has object-mass nouns of its own (le mobilier, la vaisselle, le linge), which take mass syntax and denote sets of objects.

Collective nouns (foule, groupe, équipe, famille) take a singular verb; with a plural complement the verb may be plural (une foule de gens se pressait or se pressaient). La plupart with a plural complement takes the plural (la plupart des gens pensent).

Plural-only nouns: gens, funérailles, fiançailles, obsèques, mœurs, frais "costs", vivres, environs, alentours, archives, arrhes, ténèbres; ciseaux, lunettes and vacances in their usual senses. Nouns without a plural in the basic sense are mass and abstract nouns (le courage, la faim, le sang); a plural shifts the sense (les eaux "waters, spa").

Number words used as nouns are masculine and invariable (les deux, un dix); millier, million, milliard are nouns and take de (des milliers de personnes).

9. Gender of acronyms, loans and clippings

An acronym takes the gender of the head noun of its full form: la SNCF (société), le SMIC (salaire), la CGT (confédération), l'OTAN (feminine, organisation), le PDG (président-directeur général). A loanword gets a gender by analogy with a French noun of similar meaning or ending, with the masculine as default: un week-end, un e-mail (like courriel), un smartphone, une start-up (entreprise), une interview. A clipped word keeps the gender of the full word: la photo (photographie), la télé, la moto, le vélo, la pub; le/la prof is epicene. Letters and numbers used as nouns are masculine: un a, le b, un dix.

10. Nouns of persons: feminisation and inclusive writing

10.1 The generic masculine

Traditional grammar uses the masculine as the unmarked gender: for a mixed group (les étudiants), for a person whose sex is unknown or irrelevant (le candidat doit…), and as the dictionary form.

10.2 Feminine titles

For a woman a title may be feminised (une directrice, une avocate, une ministre, une professeure, une autrice) or the masculine kept with a feminine determiner (une professeur), or wholly masculine (Madame le Ministre … elle). Usage differs by country. Québec adopted feminine forms early; the French Community of Belgium publishes a guide of feminine forms (Mettre au féminin, moreau-2014, known by its card); the Swiss federal administration has drafting guidelines (suisse-2000, card only); the Académie française, long opposed, accepted the feminisation of professions in a 2019 report (reported, not read). In GSD ministre is masculine in 99 tokens and feminine in 7 (2026-11-06); in elle est nommée ministre de la Culture (fr-ud-dev_01095) the noun with a female referent and no determiner is tagged masculine. The treebank thus marks the form's grammar, not the referent's sex.

10.3 Inclusive writing

Écriture inclusive covers several devices; Grouin (grouin-2022), following Alpheratz, lists them: coordination of both forms (toutes et tous, les Françaises et les Français); feminine titles; epicene nouns (personne, membre, journaliste); both endings in one word joined by a median dot or a substitute (les doctorant·e·s, lecteur·ice, chercheur.euse, étudiant-e, étudiant(e)); and neutral coinages: the pronouns iel, iels, the demonstratives cellui, celleux, words such as frœur "sibling" and adelphité for fraternité, endings with æ (blessæ).

Grouin measured TreeTagger and spaCy on a small corpus of political texts (56 inclusive spans, 177 tokens) against the same spans in standard French. Part-of-speech errors rose by three to seven points, almost all of it on the merged forms and the neutral coinages: on the median-dot forms TreeTagger mistagged 8 of 17 tokens and spaCy 12 of 17, while coordination caused no extra errors. TreeTagger could not lemmatise the merged or coined forms at all; spaCy copied the merged ones and guessed lemmas for some coinages by removing a plural mark. Named-entity models lost precision in the inclusive contexts. The author's conclusion is that the tools must be retrained on such text.

Three other open papers treat the topic as an NLP problem. Doyen and Todirascu (doyen-2025) built GeNRe, a system that neutralises the gender of French text with collective nouns, whose gender is fixed (les étudiants → le corps étudiant): a rule-based system with a dictionary, two models fine-tuned on its output, and an instruction-tuned large model that, given the dictionary, came close to the rules. Flesch and Burnett (flesch-2025) collected texts by non-binary authors and found three patterns: non-binary markers, one grammatical gender throughout, or both genders; gender-inference systems that read je suis + agreeing adjective assume a binary. Pironneau (pironneau-2022) describes a commercial writing checker that flags text elements that work against an equal representation of women and men.

GSD has no inclusive forms: the median dot occurs once, as a punctuation token (2026-11-06). A parser trained only on GSD has never seen étudiant·e·s or iel. Their typography is in 07. Punctuation, typography, capitals, abbreviations and numbers.

10.4 The 1990 rectifications and nouns

The rectifications (cslf-1990; recommended, not compulsory) touch nouns in four ways: the plural of verb + noun and preposition + noun compounds (§6.1); the welding of a list of compounds (§6.2); a regular plural in -s for loanwords, the singular being the most frequent form even if it is a plural in the source language (des matchs, des solos, des scénarios, des confettis, des raviolis; but des mea culpa, kept as a quotation, and nouns in -s, -x, -z stay invariable: des boss); and accents that follow pronunciation (évènement, cèleri), with the circumflex optional on i and u (maitre). The treebanks follow the traditional spelling, and the examples of this chapter do too (06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications).

11. How Mova handles nouns

There is no French noun morphology in Mova: no analyser or generator for the feminine and the plural, no table of irregular plurals, no gender lexicon. Gender and number exist only as annotations in the treebank the parser is trained on.

The tags. en xpos-fill (functions xpos_fill and xpos_of in en/src/main.rs) gives every word of a French treebank a Penn-style tag computed from UPOS and FEATS, because the French treebanks have no XPOS. For nouns: NOUN → NNS if Number=Plur, otherwise NN; PROPN → NNPS if Number=Plur, otherwise NNP. Gender is dropped. The tagger (en/src/ptag.rs, with the suffix model of en/src/tag.rs as prior for unknown words) and the parser therefore learn number and gender only indirectly, from the forms and neighbours in the training data: the suffix model, trained on French, learns that an unknown word in -s or -x after les is likely NNS, and the word la before a noun is a feature like any other word. Agreement is never checked.

Measured (UD French GSD, UD 2.18):

UD conventions that the parser inherits from GSD: nouns carry Gender and Number, epicenes often Number only; du, des, au, aux are multiword tokens split into ADP + DET (10); a hyphenated compound is one token; a title before a name is the head, with the name as flat:name.

Not done.

• Feminine and plural formation, the irregular plurals, the compound plurals, in either direction.
• Gender for unseen nouns: neither the ending table nor the evidence of agreeing words is used.
• Agreement checking between noun, determiner, adjective and participle.
• Inclusive forms: étudiant·e·s has never been seen in training and would be tokenised and tagged by guesswork; the result is unmeasured.
• The homonyms of §3.1 are not distinguished (no lexicon, no sense).

Next steps this chapter supports. A form → lemma + features lexicon from the lemma and FEATS columns of the open French treebanks (it would cover only attested forms); the rules of §§4–6 as a generator and analyser with the exception lists of §5.2, checked against the 75 175 gendered noun tokens of GSD; and the ending table of §2.2 as the fallback guess, whose expected accuracy on its covered tokens is the 94.3% measured above.

Sources for this chapter

• Brunot, F. 1899, Précis de grammaire historique de la langue française (brunot-1899, public domain): §§181–185 (the neuter and its traces), §195 (neuter plurals made feminine singulars), §§190–206 (the two-case declension, the origin of the plural s, the x of chevaux); the feminine in -e.
• Conseil supérieur de la langue française 1990, Les rectifications de l'orthographe (cslf-1990, official text): plural of compounds, welding, plural of loanwords, accents.
• Sahai, S. & Sharma, D. 2021, Predicting and explaining French grammatical gender (sahai-2021, CC BY 4.0).
• Sukumaran, P., Houghton, C. & Kazanina, N. 2024, Investigating grammatical abstraction in language models using few-shot learning of novel noun gender (sukumaran-2024, CC BY 4.0).
• Le, K. N., Bale, A. & Barner, D. 2026, Object-mass nouns specify individuation lexically: evidence from English and French (le-2026, CC BY 4.0, Glossa).
• Grouin, C. 2022, Impact du français inclusif sur les outils du TAL (grouin-2022, CC BY 4.0).
• Doyen, E. & Todirascu, A. 2025, GeNRe (doyen-2025, CC BY 4.0); Flesch, M. & Burnett, H. 2025, Characterizing non-binary French (flesch-2025, CC BY 4.0); Pironneau, M. 2022, Aider à une rédaction plus inclusive (pironneau-2022, CC BY 4.0).
• Moreau 2014 (moreau-2014) and the Swiss federal guide (suisse-2000): known by their cards only.
• Counts: UD 2.18 French GSD, fr_gsd-ud-{train,dev,test}.conllu in UD_French-GSD, counted 2026-11-06 over the word lines (lemma counts unless a form is named); the train → dev figures of 2026-10-22 are from the parse-eval baselines. Examples quoted from GSD (CC BY-SA 4.0) with their sent_id.
• Mova: en/src/main.rs (xpos_fill, xpos_of), en/src/ptag.rs, en/src/tag.rs; baselines in What every language gets.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/fr/11-nouns/