🗣️ 06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications

05. The writing system described the letters and their sounds. This chapter is about the rules that govern how forms are spelled: the spelling of the plural, of the feminine and of verb stems; the spelling of compound words and the use of the hyphen; and the official rectifications of 1990, which change a short list of rules and a few hundred words. French has a spelling system with many exceptions to exceptions: the rules below are the regular part, and each is followed by the cases that do not obey it. The grammar behind the forms (when a plural is needed, when a participle agrees) is in the chapters on nouns, adjectives and verbs (11. Nouns, 13. Verbs, 15. Adjectives and adverbs); here we describe only the written shape.

The paradigms and the basic agreement rules that every reference grammar shares are not attributed to one work (the public-domain grammars we hold, Bescherelle 1848 bescherelle-1848, Clédat 1894 cledat-1894, agree on them). The part on the 1990 rectifications is based on the transcription of the official report (cslf-1990), which we have read in full.

1. What is written and what is not heard

Most of the inflection of French is written and not spoken (see 04. Connected speech: enchaînement, liaison, elision, the mute e and assimilation and 05. The writing system §5): the plural -s of the noun and the adjective is silent, the endings -e, -es, -ent of the verb are silent, and the feminine -e adds a consonant sound only when the consonant before it was silent in the masculine. This makes the spelling of inflection a grammatical skill rather than a phonetic one: the speaker has to know that les petits chats noirs has four plural marks, which are heard in none of the four words (the plural is in the article). The consequences are three:

• the errors of spelling are mostly errors of grammar (a missing -s, a wrong -é/-er/-ez, an unmarked agreement); a study of dictations by children with cerebral palsy or dyslexia finds that a large part of the errors give an existing word that a spell checker cannot detect (antoine-2019);
• the forms are redundant: the plural is marked several times in a noun phrase, and the agreement of the participle with a preceding object is marked only in the spelling;
• a model that works from text can use the written marks as direct evidence of number and gender, but a model that works from speech cannot.

The table gives the inflectional spelling devices at a glance.

2. The plural of nouns and adjectives

2.1 The rule and its frequency

The plural is written by adding -s to the singular. On UD French GSD (UD 2.18, all three files, counted 2026-11-06) the plural nouns (NOUN with Number=Plur) number 18 669; of these, 17 494 (93.7%) end in -s, 417 (2.2%) in -aux, 259 (1.4%) in another -x (jeux 67, prix 41, lieux 28, yeux 26, voix 20), and 499 (2.7%) in something else, almost all of them units and abbreviations (m, km, h, kg, mm, cm, ha) which stay unchanged. Among the -aux plurals the commonest are travaux (57), eaux (40), animaux (28), niveaux (20), réseaux (19), morceaux (16), journaux (12) and oiseaux (11).

2.2 Irregular plurals

Beyond these, a few nouns have irregular plurals by change of stem: œil → yeux, ciel → cieux (and ciels in painting), aïeul → aïeux ('ancestors') and aïeuls ('grandfathers'), madame → mesdames, mademoiselle → mesdemoiselles, monsieur → messieurs, gentilhomme → gentilshommes, bonhomme → bonshommes. The plural of os is /o/ (silent s), of œuf /ø/ and of bœuf /bø/: the written marks do not show the change in the sound, which 05 §5 describes. Surnames stay unmarked (les Dupont), and foreign words take the regular plural in the 1990 spelling (§6.7). Aïeul and ciel are among the few nouns where two plurals have two meanings.

2.3 Plurals of nouns that are used only in the plural or only in the singular

Les gens, les ciseaux, les lunettes, les vacances, les mathématiques, les fiançailles, les funérailles, les environs, les frais, les mœurs, les ténèbres are plural nouns (pluralia tantum); les ciseaux names one tool or several. The nouns gens and gent are an old pair: gens is the plural noun (des gens is masculine; the adjective before it is feminine, de vieilles gens but des gens âgés) (11. Nouns).

2.4 Other numerals and quantifiers

Vingt and cent take -s when they are multiplied and not followed by another number: quatre-vingts but quatre-vingt-trois, deux cents but deux cent trois; mille is invariable (deux mille) and million and milliard are nouns and take the plural (deux millions de personnes). The 1990 rectifications do not change these rules, only the hyphens (§5.1 below, and 12. Determiners, pronouns and numerals).

3. The feminine

The feminine of nouns and adjectives is regularly formed by adding -e to the masculine: grand, grande; petit, petite; ami, amie; vert, verte. This changes the pronunciation when the final consonant was silent: petit /pə.ti/, petite /pə.tit/. The following changes are made at the same time.

The doubling of the final consonant before -e (-el → -elle, -en → -enne, -on → -onne, -et → -ette, -ot → -otte, -s → -sse) does two jobs: after e it marks the open /ɛ/, since e before a single consonant and a vowel is read as a mute e (cruelle, muette, as against the mute e of appeler /a.ple/), and after a nasal vowel it shows that the vowel has become oral and the consonant is heard (bonne /bɔn/, ancienne /ɑ̃.sjɛn/); after s it keeps /s/ from being read /z/ between vowels (grosse, not grose). Where the vowel change is made with an accent instead (complet, complète; secret, secrète; premier, première) the second device is used. The two devices are in competition in verbs too (§4.3).

The feminine of agent nouns has been the subject of recommendations (la professeure, la docteure, l'autrice): the guide of the Fédération Wallonie-Bruxelles (moreau-2014) and the drafting guide of the Swiss Confederation (suisse-2000), both known to us by their cards only, recommend feminine forms of job titles. The debate is outside the scope of the written rule (11. Nouns, where gender-inclusive writing is described).

4. The spelling of verb forms

4.1 Endings

The regular endings are written with letters that are mostly silent.

Three points of spelling are not predictable from the sound:

• The -s, -x, -t of the singular endings: vouloir, pouvoir and valoir are the verbs with -x (je veux, tu peux, je vaux); je vends, tu vends, il vend keep the d; il rompt keeps the p.
• The -e of the first person singular in the -er verbs of the first group (je chante), in ouvrir, offrir, cueillir, souffrir (j'ouvre, j'offre, je cueille), and the 2sg imperative of the same verbs without -s (chante, ouvre, cueille, va), with the -s returning before y and en (chantes-en un, vas-y, cueilles-en).
• The -ent of the third person plural, which is silent and therefore homophonous with the singular (il parle, ils parlent). The -ent is heard only as a liaison /t/ before a vowel (ils parlent‿à Paul, elles prennent‿un taxi), and that liaison is rare in conversation; and -ent in content, couvent is not a verb ending: this is a source of homograph pairs (05 §6).

The endings of the participle (aimé, fini, vendu, mis, dit, fait, ouvert, pris, eu, été) and the agreement of the past participle are described in 13. Verbs. The /e/ endings -é, -er, -ez, -ai, -ais(ait) are the commonest cause of error in French dictation; the counts are in 05 §4.1.

4.2 Verbs with a soft consonant: c/ç, g/ge

The infinitive of the verbs in -cer and -ger has a soft c or g before e, i (the infinitive -er). Before a, o, u the letter would change its value, so the spelling is adjusted: c becomes ç (nous commençons, il commença, commençant) and g becomes ge (nous mangeons, il mangeait, mangeant, mangea). In GSD (counted 2026-11-06) the commonest forms of the first type are reçoit (41), plaçant, aperçoit, commença (7 each), and the verbs of the second type give -geant and -geait forms in great number. The reverse is true of the -guer verbs, which keep the u before a, o (nous naviguons) and of the verbs in -quer, which keep the qu (nous expliquons). In arguer, where the u is pronounced (/aʁ.ɡɥe/), the 1990 report adds a tréma, il argüe, so that the form is not read like il nargue (§6.8).

4.3 Verbs in -yer

The verbs in -oyer and -uyer change y to i before a mute e: je nettoie, tu nettoies, il nettoie, ils nettoient but nous nettoyons, vous nettoyez; j'appuie; future and conditional je nettoierai. The verbs in -ayer have two forms accepted: je paie or je paye, j'essaie or j'essaye, il balaie or il balaye (the -ye forms keep the /ej/ sound); the verbs in -eyer are rare. Envoyer has an irregular future j'enverrai. The y → i alternation is the same as for relai / relayer in the 1990 list (note (i) of the report: relai by comparison with balai / balayer, essai / essayer).

4.4 Verbs in -eler and -eter (and the accent on e)

The infinitive of these verbs has a mute e (appeler, jeter, acheter); before an ending with a mute e, it is read as an open /ɛ/. Two methods mark this: doubling the consonant (j'appelle, tu jettes, il rappelle, elle jette) or an accent on the e (il achète, elle gèle, il pèle, il harcèle, il étiquette or étiquète). Traditionally, appeler and jeter and their families double, acheter, geler, peler, harceler, marteler, modeler, haleter, celer, receler take the accent. The 1990 report remarks that "among the verbs concerned there are few on which all the dictionaries agree". It settles the question: the grave accent for all such verbs except appeler and jeter and their families (rappeler, rejeter, projeter) (§6.4). In GSD (counted 2026-11-06) the forms of the doubling type are the commonest: appelle 24, appelait 8, jette 9, against achète 10 and lève 3 (the 5 tokens étiquette are all the noun). The lemmas with the highest token counts in the treebank, among the VERB forms of -eler/-eter, are appeler 178, acheter 45, rappeler 36, jeter 21, rejeter 18, renouveler 15, racheter 11, projeter 8, démanteler 8.

4.5 Verbs with the accent alternation é/è (céder, semer)

Verbs with an e or an é in the penultimate syllable of the infinitive (semer, mener, lever, peser; céder, régler, répéter, espérer, préférer, considérer) change it to è before a syllable with a mute e: je sème, je mène, je lève, je cède, je règle, il espère, in the present and the imperative singular; but the traditional spelling of the future and the conditional of the é verbs keeps the acute: je céderai, je répéterais, il considérera, although the vowel is now pronounced /ɛ/ by most speakers, as in je cède. The 1990 rectifications change that (§6.3).

4.6 The accent on the participle and the infinitive stem

The vowel of the stem is spelled with an accent according to the same two rules as in nouns (see 05 §2): the accent is acute at the end of a written syllable, grave before a mute e, so the second singular of aimer, tu aimes, has no accent, whereas tu cèdes has one.

5. The spelling of compounds and the hyphen

5.1 The hyphen: uses

The 1990 report begins by listing the uses of the hyphen (analysis 1):

• syntactic uses: inversion of the subject pronoun (dit-il, a-t-il), free coordination (la ligne nord-sud, le rapport qualité-prix), and in numerals, "which is hardly justifiable, only for numerals below one hundred" (vingt-trois but cent trois);
• lexical uses: freely formed compounds (train-train) and fixed sequences (porte-drapeau, va-nu-pied).

The compound competes with fusion (portemanteau, betterave) and with the separated sequence that has become a fixed expression (pomme de terre, compte rendu). The more a sequence is felt as a single word, the more it goes from the separated expression to the hyphenated compound to the fused word (analysis 1).

The report sets out when a hyphen appears in a sequence that has become a compound: (a) when the sequence changes its grammatical class (à propos vs l'à-propos; un ouvre-boîte, un va-et-vient, le non-dit, un après-midi, un chez-soi; sentence nouns like un laissez-passer, le qu'en-dira-t-on); (b) when the meaning or the gender or number differs from that of the sequence (un rouge-gorge, un coq-à-l'âne, un gratte-ciel, un franc-maçon); (c) when an element has grown old and is no longer understood (un rez-de-chaussée, un croc-en-jambe, à vau-l'eau); and (d) when the compound does not respect ordinary morphology and syntax, in archaisms (la grand-rue, un nouveau-né, nu-tête) or in calques (surprise-partie, sud-américain). It adds that the compounds with a learned prefix are usually fused (narcothérapie), and that if all the elements are learned the fusion is compulsory (narcolepsie). The report finds "very numerous hesitations" and "divergences between dictionaries" in the use of the hyphen.

5.2 Hyphen uses that are fixed

On GSD (counted 2026-11-06), of the 905 NOUN tokens that contain a hyphen between letters, the commonest are sud-ouest 24, sud-est 24, nord-ouest 20, nord-est 18, porte-parole 16, week-end 14, rendez-vous 13, chef-lieu 13, vice-président 11, après-midi 11, grand-père 10, demi-finales 9, centre-ville 9. In the same data the inverted forms are tokens that start with a hyphen: -t-il 59, -il 52, -ce 27, -ils 22, -vous 19, -t-elle 18, -elle 18, -on 14, -là 13, -je 8, -nous 7, -ci 7, -t-on 6, -tu 4 (all counted 2026-11-06): the UD analysis separates the hyphenated pronoun from the verb (07). The numerals written in letters with a hyphen are rare in the treebank (36 tokens of the type vingt-X, trente-X … in the three files: dix-sept 4, vingt-quatre 3, trente-quatre 3); the numerals are mostly written in figures.

5.3 Fusion and the lists A–C of the 1990 report

The report proposes to write fused (soudés), without hyphen, the words of three closed lists (III.1–III.3), "mostly old words whose components no longer correspond to the present lexicon or syntax, expressive or onomatopoeic stems, words that have derivatives":

• List A, compounds with a verbal element, 34 words, among them: arrachepied (d'), boutentrain, brisetout, chaussetrappe, clochepied (à), couvrepied, crochepied, croquemitaine, croquemort, faitout, fourretout, mangetout, mêletout, passepartout, piquenique, porteclé, portecrayon, portemine, portemonnaie, portevoix, risquetout, tapecul, tirebouchon, tirebouchonner, tirefond, tournedos, vanupied;
• List B, compounds of nominal and adjectival elements, about 35 words with their derivatives, among them: arcboutant, autostop, bassecour, branlebas, chauvesouris, hautparleur, lieudit, millefeuille, millepatte, platebande, potpourri, quotepart, sagefemme, saufconduit, téléfilm, terreplein, vélopousse;
• List C, onomatopoeias and expressive words: blabla, bouiboui, coincoin, froufrou, grigri, kifkif, mélimélo, pêlemêle, pingpong, prêchiprêcha, tamtam, tohubohu, traintrain, troutrou, tsétsé.

The report also gives list F, of Latin and foreign compounds well established in use and which are not citations: apriori, exlibris, exvoto, statuquo, vadémécum; baseball, basketball, blackout, bluejean, cowboy, fairplay, globetrotteur, handball, harakiri, hotdog, lockout, motocross, ossobuco, pipeline, sidecar, striptease, volleyball, weekend. And the recommendation to lexicographers (IV.2) to extend the fusion to the compounds with contre- and entre- used as prefixes (contrecourant, entraide, entrenuire), with the Latin prefixes extra-, intra-, ultra-, infra- (extraconjugal, ultrafiltration), to compounds on learned stems (autovaccin, cardiovasculaire, électroménager, audiovisuel, macroéconomie, psychomoteur, radioactif, vidéocassette), and in general to the words "well anchored in use and felt as one lexical unit". The hyphen is to be kept when fusion would produce a defective reading, in particular when two vowels would form a diphthong (extra-utérin), and when the composition is free and expresses coordination between two terms (italo-français, anglo-danois, gréco-romain, finno-ougrien) (IV.2 remark). Hyphens may also be used for metaphorical compound nouns (barbe-de-capucin, langue-de-bœuf in botany) (IV.1).

GSD (counted 2026-11-06) shows the situation of the traditional spelling in usage: week-end 15 against weekend 4; après-midi 11 and no aprèsmidi; portefeuille 3 (fused since the eighteenth century), grand-père 11.

5.4 The plural of compounds

The traditional rule was to apply the marks of number "as if the elements were independent words", which yielded subtle and contradictory distinctions: the 1990 report cites des gardes-meubles ('men') against des garde-meubles ('places'), "a mistaken analysis already denounced by Littré", un porte-montre if the object holds one watch and un porte-montres for several; and the contradictions un cure-dent but un cure-ongles, des après-midi but des après-dîners (analysis 2). The report recalls that un mille-feuille is a cake and not a thousand leaves and un ramasse-miettes is one object, so the first element is no longer a verb.

In the GSD treebank the hyphenated plural nouns (not counting the verbal compounds) show the traditional forms in use: demi-finales 9, beaux-arts 5, sous-marins 3, sages-femmes 3, belles-lettres 3, timbres-poste 2, sapeurs-pompiers 2 (counted 2026-11-06).

6. The 1990 rectifications in detail

6.1 Status and background

The report was prepared by the Conseil supérieur de la langue française, after the Prime Minister asked it to formulate "clear and precise proposals" on the spelling of French, to remove uncertainties and contradictions and to allow correct formation of the new words that the sciences require. It was presented to the Prime Minister on 19 June 1990 by Maurice Druon, permanent secretary of the Académie française and chairman of the working group; the experts were led by Bernard Cerquiglini. The Académie française examined the proposals in its session of 3 May 1990 and approved them unanimously, with minor reservations, noting that the adjustments were "in the line of those it had practised in the past, notably in 1740, when the spelling of one word in four was changed, in 1835 …, in 1878 in the 7th edition and in 1935 in the 8th". The Conseil de la langue française of Québec and that of the Communauté française de Belgique gave favourable opinions. The Prime Minister accepted the proposals officially, adding that the Government never meant to legislate on the point ("the language belongs to its users"); the report was published in the Journal officiel, Documents administratifs no. 100, of 6 December 1990 (cslf-1990).

Druon states four principles. First: the proposals must be "firm and flexible"; firm, so that the rectifications form a new norm and that teachers know what to teach; flexible, since "the older generations cannot be asked to unlearn" and the current spelling must remain admitted (the comparison made is to the replacement of oi by ai in j'avais, j'aimais, français in 1835, to which Chateaubriand agreed while continuing to write as he was used to). Second: utility, aiming first at points that cause difficulty even to adults and writers. Third: relying on the "genius of the language", on established uses and tendencies to coherence already visible. Fourth: moderate changes, checked by computer: "the words affected by the modifications, on a page of a novel, even by Proust, could be counted on the fingers of one hand". Druon estimated the number of words affected as three to four thousand of about fifty thousand. The rectifications "do not claim to make spelling simple and rational". The report distinguishes analyses (I), rules (II), particular spellings fixed or modified (III), which are closed lists, and recommendations to lexicographers and creators of neologisms (IV), "not intended, in a first stage, for the individual user, nor for teaching".

6.2 Rule 1: the hyphen in numerals

Compound numerals are joined by hyphens, both below and above one hundred: vingt-quatre, quatre-vingt-neuf, cent-deux, deux-cents, cent-trente-deux, deux-cent-soixante-et-onze, sept-cent-mille-trois-cent-vingt-et-un. The traditional cent trois becomes cent-trois. The agreement of vingt and cent is not changed (deux-cents with -s when final).

6.3 Rules 2–3: plural of compounds; the grave accent

Rule 2 (above §5.4): compounds of a verb and a noun, or of a preposition and a noun, follow the rule of simple words: the plural mark is carried by the second element and only in the plural.

Rule 3 (grave accent). (a) The future and conditional of verbs of the type céder are accentuated like semer: je cèderai, je cèderais, j'allègerai, j'altèrerai, je considèrerai. This brings the spelling in line with the pronunciation (/ɛ/ in modern speech) and with the rule of the accent before a mute e. (b) In the interrogative inversion, the first person singular in -e before je takes the grave: aimè-je, puissè-je.

The report adds in the lists the words of rule 3 that were left out of the Académie's earlier regularisation: list D (accents added where omitted or where pronunciation changed): asséner, bélitre, bésicles, démiurge, gélinotte, québécois, recéler, recépage, recéper, réclusionnaire, réfréner, sèneçon, sénescence, sénestre; and list E (accents modified to follow the general rule): abrègement, affèterie, allègement, allègrement, assèchement, cèleri, complètement (noun), crèmerie, crèteler, crènelage, crèneler, crènelure, évènement, empiètement, fèverole, hébètement, règlementaire, règlementation, and others. The best-known case is événement → évènement, which brings événement into the pattern avènement, élèvement. In GSD (counted 2026-11-06) both are in use: événement 23 and évènement 9.

6.4 Rule 4: the circumflex

The circumflex is kept on a, e, o (where it may indicate "useful distinctions of vowel quality": mâtin / matin, côte / cote, vôtre / votre) and is no longer obligatory on i and u, "where its utility is distinctly more limited" (voûte and doute differ only by the first consonant). Exceptions, where it is kept:

1. verb endings: the first and second person plural of the passé simple (nous suivîmes, nous voulûmes, nous aimâmes; vous suivîtes, vous voulûtes, vous aimâtes), the third person singular of the imperfect subjunctive (qu'il suivît, qu'il voulût, qu'il aimât) and of the pluperfect subjunctive (qu'il eût suivi, il eût voulu, qu'il eût aimé);
2. words where it brings a useful distinction of sense: dû, jeûne, the adjectives mûr and sûr, and the verb croître (because part of its conjugation is homographic with croire). This does not extend to the derivatives and compounds (sûr but sureté; croître but accroitre); dû, mûr and sûr have the accent only in the masculine singular.

The report lists the new regularities: mu (like su, tu, vu, lu), plait (like tait, fait), piqure, surpiqure (like morsure), traine, traitre and derivatives (like gaine, haine, faine), and the adverbs ambigument, assidument, congrument, continument, crument, dument, goulument, incongrument, indument, nument (like absolument, éperdument, ingénument, résolument). The circumflex is also no longer used for loanwords and neologisms (IV.4), except in compounds from words that keep it (repose-flute but allume-dôme). Proper names are not touched (Nîmes, and the adjective nîmois). The report's own commentary on the accent (analysis 3.3) is blunt: its use is "incoherent and arbitrary" and "prevents any systematic or historical teaching"; no adult "will be obliged to give it up". The report spells out plait, traine, traitre, piqure, mu and the adverbs above; the general rule also lets words such as boite, maitre, ile, cout, aout lose the accent, although the report does not list them, and the application to every word is left to the dictionaries. In GSD (counted 2026-11-06) the traditional forms dominate: connaît 25 against connait 6; coût 26 against cout 1; île 91 against ile 1; maître 43 against maitre 1; août 121 against aout 3; paraît 9 and parait 0.

6.5 Rule 5: verbs in -eler and -eter

The grave accent marks the open /ɛ/ for all verbs of this type, except appeler (and rappeler) and jeter (and its family), "the forms of which are the best stabilised in usage": elle ruissèle, elle ruissèlera, j'époussète, j'étiquète, il époussètera, il étiquètera (traditionally ruisselle, étiquette). The nouns in -ement derived from them follow: amoncèlement, bossèlement, chancèlement, cisèlement, cliquètement, craquèlement, craquètement, cuvèlement, dénivèlement, ensorcèlement, étincèlement, grommèlement, martèlement, morcèlement, musèlement, nivèlement, ruissèlement, volètement. The analysis explains the choice: the accent spelling "brings all these verbs back to the model of conjugation of mener" (il mène, elle mènera) (analysis 4).

6.6 Rule 6: the participle of laisser

The past participle of laisser followed by an infinitive is invariable, like that of faire (which is already invariable in that position): elle s'est laissé mourir, elle s'est laissé séduire, je les ai laissé partir, la maison qu'elle a laissé saccager. The traditional elle s'est laissée aller becomes elle s'est laissé aller. This is the only change to participle agreement: the experts judged that the agreement of pronominal verbs could not be changed without changing the general rule, and "it is hardly any more a matter of spelling but of syntax", with the agreements commanded by the sense (the famous series Que d'hommes se sont craints, déplu, détestés, nui, haïs, succédé). Agreement of the past participle is described in 13. Verbs.

6.7 Rule 7: loanwords

Nouns and adjectives of foreign origin have a regular singular and plural: un zakouski, des zakouskis; un ravioli, des raviolis; un graffiti, des graffitis; un lazzi, des lazzis; un confetti, des confettis; un scénario, des scénarios; un jazzman, des jazzmans. The singular is the commonest form, even if it is a plural in the other language. The plural is formed with a silent s: des matchs, des lands, des lieds, des solos, des apparatchiks; also for Latin nouns (des maximums, des médias). Words that keep the value of a quotation are not concerned (des mea culpa). Words ending in s, x, z stay invariable (un boss, des boss; un kibboutz, des kibboutz; un box, des box). Foreign compounds are fused, which simplifies the plural: des covergirls, des bluejeans, des ossobucos, des weekends, des hotdogs. The recommendations (IV.7–IV.9) add that foreign words should be adapted to the French alphabet and spelling (volapük → volapuk, muesli → musli, nirvâna → nirvana, maelström → maelstrom), that the form closest to French be chosen (litchis, ouzbek, bogie, canyon, cascher), and that the nominal suffix -er of anglicisms pronounced /œʁ/ be spelled -eur (débatteur, squatteur, kidnappeur).

6.8 The particular spellings (lists D–H) and anomalies

Besides rules, the report fixes "restricted and limitative" lists of particular spellings.

• The tréma (III.4–III.5): the tréma moves to the vowel that is pronounced in aigüe, ambigüe, exigüe, contigüe, ambigüité, exigüité, contigüité, cigüe (traditional aiguë, ambiguë, exiguë, contiguë, ciguë), and is added where -gue- or -geu- leads to defective pronunciations: argüer (il argüe), gageüre, mangeüre, rongeüre, vergeüre.
• Loanwords with accents (III.9, list G): Latin words artéfact, critérium, déléatur, délirium trémens, desidérata, duodénum, exéat, exéquatur, facsimilé, jéjunum, linoléum, média, mémento, mémorandum, placébo, proscénium, référendum, satisfécit, sénior, spéculum, tépidarium, vadémécum, vélarium, vélum, véto; other languages allégretto, allégro, braséro, candéla, chébec, chéchia, cicérone, condottière, décrescendo, diésel, édelweiss, imprésario, kakémono, méhalla, pédigrée, pérestroïka, péséta, péso, piéta, révolver, séquoia, sombréro, téocalli, trémolo, zarzuéla.
• List H, anomalies (III.10, among them rectifications proposed by the Académie in 1975): absout, absoute (for absous), appâts (for appas), assoir, rassoir, sursoir (for asseoir; the report notes that "the e is no longer pronounced" and that the Académie already writes j'assois), bizut (for bizuth), bonhommie, boursoufflement, boursouffler, boursoufflure, cahutte, charriot, chaussetrappe, combattif, combattivité, cuisseau, déciller (for dessiller), dissout, dissoute, douçâtre (for douceâtre), embattre, exéma (for eczéma), guilde (for ghilde), homéo-, imbécilité, innommé, levreau, nénufar, ognon (for oignon), pagaille (for pagaïe, pagaye), persifflage, persiffler, persiffleur, ponch (for punch), prudhommal, relai (for relais), saccarine, sconse (for skunks), sorgo, sottie, tocade, ventail. The report gives its reasons for each case, for instance exéma "as examen, since the sequence cz is exceptional in French", nénufar "of Arabo-Persian origin; the Académie always wrote it so except in the 8th edition", ognon, "a sequence ign that is a graphic oddity", douçâtre "since ea was an old spelling made useless by the cedilla", charriot by comparison with charrette, relai by comparison with balai / balayer.
• Words in -illier (III.11): joailler, marguiller, ouillère, quincailler, serpillère are written -iller, "as poulailler, where the i after the consonant is not heard".
• Words in -ole (III.12): one l, like bestiole, camisole: barcarole, corole, fumerole, girole, grole, guibole, mariole, with the exception of folle, molle, colle and its compounds.
• Double consonant after a mute e (III.13): interpeler (for interpeller), dentelière, lunetier, prunelier (for dentellière, lunettier, prunellier).
• Recommendations (IV.6, IV.10): verbs in -otter if formed on a base in -otte (botter), in -oter if formed on a base in -ot (garroter) or with the suffix -oter (clignoter, toussoter); and for new words derived from nouns in -an or -on, simple n (-onite, -onologie, -onaire).

A general remark (IV, final) asks the lexicographers to prefer, between competing forms, "the simplest: without circumflex, fused, with simple n, gallicised, with regular plural".

6.9 What happened to them

The rectifications are a recommendation and not an obligation. The Académie said that both spellings would remain admitted "until the new one has entered the use"; the Prime Minister declared that the Government had no intention of legislating; and Druon's report suggests a spelling lexicon to be prepared by the Délégation générale for the use of teachers, printers and lexicographers, and a revision "in thirty years or even before". We have not measured the share of rectified spellings in published French. In the French data we hold, GSD (blogs, news, reviews, Wikipedia), the traditional spellings dominate (counted 2026-11-06, three files). For the circumflex words the traditional forms total 306 tokens (connaît 25, coût 26, île 91, maître 43, août 121) against 12 for the unaccented forms (connait 6, cout 1, ile 1, maitre 1, aout 3), 3.8%; some of these are probably lost accents rather than choices. Évènement (9) is read against événement (23); week-end (15) against weekend (4). Ognon does not occur (oignon 2). No author in the corpus follows the new orthography consistently.

7. Other spelling conventions

• The tréma and the diaeresis in the old spelling: Noël, naïf, aiguë, maïs, haïr, héroïque, Saül; the 1990 rule moves it in aiguë → aigüe (above).
• The apostrophe and the elision are in 04. Connected speech: enchaînement, liaison, elision, the mute e and assimilation §5; the contractions au, aux, du, des are written as one word.
• Abbreviations and capitals are in 07. Punctuation, typography, capitals, abbreviations and numbers.
• Numbers in figures: 1er, 2e (not 2ème), XXe siècle; see 07.
• The spelling of the adverb in -ment: the feminine of the adjective plus -ment (lente → lentement; douce → doucement); -emment, -amment for adjectives in -ent, -ant (évident → évidemment, /evidamɑ̃/; constant → constamment); exceptions lentement, présentement, véhémentement and the -ûment adverbs where the circumflex marks the lost e (assidûment, crûment, dûment; the 1990 report makes them assidument, crument, dument).
• Learned doublets: -tion vs -sion vs -ssion vs -cion; ph vs f; th; ch /k/; y in Greek words. These are in the derivation chapters (17. Derivation II: suffixes).

8. How Mova handles French spelling rules

What exists. Mova has no French morphology, so the rules of §§2–5 are nowhere in the code: it does not produce the plural or the feminine of a French word, it does not conjugate, and it does not know that chevaux is the plural of cheval. French goes through the English parser (en) with the tags of the treebank (en xpos-fill), and the inflected forms come from the training trees: a form is known if it occurred in the training treebank. The GSD figures (UD 2.18, en tag-eval train → dev, 2026-10-22, rerun 2026-11-06 with the same result): 7.2% of the development tokens are unknown forms; the tagger tags them right 84.8% of the time against 96.5% overall. For nouns the figure is 10.07% unseen forms in the development nouns (684 of 6 792 dev noun tokens are forms that never occur as a noun in train; 611 never occur at all; recounted 2026-11-06). The tagger guesses an unseen form from its ending, which is a very good guide in French: -s for plural nouns, -aux, -é for participles, -ement for adverbs, -tion. No spelling variant is normalised: évènement is a different string from événement, ognon from oignon, il connait from il connaît, weekend from week-end.

What the English tokenizer does with hyphens. en/src/tok.rs keeps or splits a hyphenated word by a dictionary learned from the training treebank, with a default (keep) when the word is unknown. From GSD train it learns 1 994 hyphenated words kept whole and 46 split. Compounds like week-end, sud-ouest, grand-mère, porte-parole, chef-lieu stay one token, as in UD French; the inverted pronouns (viens-tu, a-t-il, donne-le-moi), which UD splits (-tu, -t-il), also stay in one piece. en tok-eval on GSD (train → dev and test, 2026-11-06, two runs identical): exact sentence tokenisation 26.1% (dev) and 21.9% (test), token F1 88.84% and 88.16% (04 §9). The cases of §5 are among the misses; the other main class is the elision and contraction.

What would be needed.

1. A normaliser for the spelling variants. The 1990 rectifications are regular enough to be written as a small set of rewrite rules plus the closed lists D, E, H, G and the fusions A–C, F (the official report is in the public domain as an official text and the lists are in §6.3–6.8): rule 1 (numerals), 2 (compound plurals; needs the compound's structure), 3 (futures in -érai), 4 (circumflex on i, u, with the exceptions), 5 (-eler/-eter), 7 (loanword plurals). What cannot be derived by rule is which compounds are fused; that needs the lists. A normaliser should map both spellings to one lemma and keep the original form for generation. The first test is simple: take the GSD text, apply the rectifications automatically, and check that the tagger's accuracy does not fall.
2. A full-form lexicon with morphology (Morphalou, Lefff, GLÀFF: 19. The lexicon: layers, borrowings, multiword expressions and register and Sources) from which the plural and feminine rules can be learned or tested, and from which the exceptions (bals, travaux, bijoux) come as data. The lexicon is a prerequisite also for generation: Mova writes English from a lexicon of forms (en/data/forms.tsv), and French has no such lexicon.
3. A generator of the stem alternations of §4: a short list of verb classes (-cer, -ger, -yer, -eler, -eter, é/è) would be enough, with the 1990 variants as a switch.

Not handled. All of this chapter in generation; the agreement of the past participle (spelling follows syntax: 13. Verbs); the check of spelling in input text (errors like a/à, et/est, which the English tagger would treat as a different word and not as an error).

Sources for this chapter

• Conseil supérieur de la langue française. 1990. Les rectifications de l'orthographe. Journal officiel de la République française, Documents administratifs no 100, 6 December 1990 (official text, transcription). cslf-1990; fr-gram-rectifications-orthographe-1990
• Antoine, Jean-Yves et al. 2019. Ma copie adore le vélo : analyse des besoins réels en correction orthographique sur un corpus de dictées d'enfants. CC BY 4.0 (ACL Anthology). antoine-2019
• Bescherelle, L.-N. 1848. L'art de conjuguer. Public domain. bescherelle-1848
• Clédat, Léon. 1894. Grammaire raisonnée de la langue française. Public domain. cledat-1894
• Moreau, Marie-Louise (ed.). 2014. Mettre au féminin : guide de féminisation des noms de métier, fonction, grade ou titre, 3rd ed. Bruxelles: Fédération Wallonie-Bruxelles. (Card only.) moreau-2014
• Confédération suisse. 2000. Guide de formulation non sexiste des textes administratifs et législatifs de la Confédération. (Card only.) suisse-2000
• UD French GSD, UD 2.18 (CC BY-SA 4.0); counts of 2026-11-06 from UD_French-GSD.
• Mova: en/src/tok.rs, en/src/main.rs (tok-eval, xpos-fill); en tag-eval on GSD train → dev after en xpos-fill, 2026-11-06.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/