Elamur: **Mova** · [Parse](https://mova.elamur.ai/free/en/parse/) · [Dictionary](https://md.dictionary.elamur.ai/free/en/) · [Translate](https://md.translate.elamur.ai/free/en/) · [Research](https://md.research.elamur.ai/free/en/) · [Groupants](https://md.groupants.elamur.ai/free/en/)

[Mova](https://md.mova.elamur.ai/free/en/) · [this page in HTML](https://mova.elamur.ai/free/en/docs/languages/fr/05-writing-system/)

[home](https://md.mova.elamur.ai/free/en/) / [docs](https://md.mova.elamur.ai/free/en/docs/) / [languages](https://mova.elamur.ai/free/en/docs/languages/) / [fr](https://md.mova.elamur.ai/free/en/docs/languages/fr/) / 05. The writing system

# 05. The writing system

**On this page**

1. [1. The alphabet](#1-the-alphabet)
2. [2. Diacritics](#2-diacritics)
3. [3. Graphemes](#3-graphemes)
4. [4. From sound to spelling](#4-from-sound-to-spelling)
5. [5. Silent letters](#5-silent-letters)
6. [6. Homographs and homophones](#6-homographs-and-homophones)
7. [7. Why French spelling is the way it is](#7-why-french-spelling-is-the-way-it-is)
8. [8. How Mova handles the writing system](#8-how-mova-handles-the-writing-system)
9. [Sources for this chapter](#sources-for-this-chapter)

French is written in the Latin alphabet with a spelling that was settled, in the main, in the sixteenth to eighteenth centuries, when the language had already lost most of the consonants and several of the vowels that it still writes. The result is a deep orthography with two faces. **From sound to spelling** it is very irregular: one phoneme has several spellings and a speaker cannot predict how to write a word they hear. **From spelling to sound** it is much more regular than English: a reader who knows the rules can pronounce nearly any French word correctly, with one large exception, the final consonants. This chapter gives the letters and diacritics, the correspondences in both directions, the silent letters and homophones that make writing a separate system from speech, and the history that explains it. Inflectional spelling and the 1990 rectifications are in [06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/); punctuation and typography in [07. Punctuation, typography, capitals, abbreviations and numbers](https://md.mova.elamur.ai/free/en/docs/languages/fr/07-punctuation-typography/); the sounds in [01. Consonants](https://md.mova.elamur.ai/free/en/docs/languages/fr/01-consonants/) and [02. Vowels](https://md.mova.elamur.ai/free/en/docs/languages/fr/02-vowels/).

## 1. The alphabet

### 1.1 Letters

The alphabet has the **26 letters** of the Latin alphabet, in the usual order. *K*, *w* and *y* are the rare ones: *k* occurs in loans and in a few words of Greek or Germanic origin (*kilo, kiosque, képi, ski, kayak*), *w* in loans (*wagon, week-end, western, watt*) and *y* in words of Greek origin and in a few native ones (*stylo, système, yaourt, y, pays, crayon, voyage*).

| letter | name | letter | name | letter | name |
| --- | --- | --- | --- | --- | --- |
| A a | /a/ | J j | /ʒi/ | S s | /ɛs/ |
| B b | /be/ | K k | /ka/ | T t | /te/ |
| C c | /se/ | L l | /ɛl/ | U u | /y/ |
| D d | /de/ | M m | /ɛm/ | V v | /ve/ |
| E e | /ə/ | N n | /ɛn/ | W w | /dublə ve/ |
| F f | /ɛf/ | O o | /o/ | X x | /iks/ |
| G g | /ʒe/ | P p | /pe/ | Y y | /igʁɛk/ ("Greek i") |
| H h | /aʃ/ | Q q | /ky/ | Z z | /zɛd/ |
| I i | /i/ | R r | /ɛʁ/ |  |  |

The letters *u* and *v*, and *i* and *j*, were not distinguished in print until the seventeenth century: the Académie introduced the distinction in its dictionary of the eighteenth century, following a practice of Dutch printers and a request that Ramus had made in the sixteenth (Brunot 1905, `brunot-1905`). *W* was added from loans and is not used in native words.

### 1.2 Ligatures

Two ligatures are letters of the standard spelling. ***œ*** (*e dans l'o*) stands for /ø/ or /œ/ in *cœur, sœur, œuvre, œuf, vœu, bœuf, œil, chœur*, and for /e/ in a handful of learned words (*œdème, fœtus, œsophage*, with the Greek *oi*). ***æ*** is rare and occurs in Latin words (*et cætera, ex æquo, cæcum*). Both are alphabetised as *oe* and *ae*, and both are often written as two letters in electronic text, which makes the word unfamiliar to a spelling checker. In UD French GSD (UD 2.18, all three files, 400 385 syntactic words, counted 2026-11-06) the ligature *œ* occurs 281 times (*œuvre* 79, *cœur* 43, *œuvres* 36, *sœur* 20, *chœur* 15, *œufs* 12) and the forms of the same words with *oe* (*coeur, oeuvre, soeur, oeil, voeu, boeuf* and their plurals) 23 times, so about 8% of these words are written without the ligature, the usual state of web text; the keyboards of France have no key for *œ*.

### 1.3 Letter frequencies

On GSD (same data, counted 2026-11-06; 1 637 697 letters in the word forms) the commonest letters, counted as letters without a diacritic and both cases together, are *e* 248 408 (15.2%), *s* 125 689 (7.7%), *a* 123 982 (7.6%), *n* 119 971 (7.3%), *i* 116 731 (7.1%), *t* 113 106 (6.9%), *r* 110 657 (6.8%), *l* 105 897 (6.5%), *o* 87 222 (5.3%), *u* 81 313 (5.0%), *d* 67 658 (4.1%). The least common are *w* 1 480, *z* 1 998, *k* 2 866, *j* 5 438 and *y* 5 751. The accented letters are counted separately: *é* 39 392 (2.4%), *à* 10 664, *è* 6 466, *ê* 1 893, *ç* 1 063, *ô* 822, *â* 705, *î* 638, *ù* 392, *û* 336, *œ* 281, *ï* 227, *ë* 65, *ü* 58. The ending of the word matters more than the letter counts: of the 345 183 syntactic words that are not punctuation, numerals or symbols (lower-cased), **31.6% end in *e*** (109 220), 17.3% in *s* (59 598), 11.5% in *t* (39 551) and 7.6% in *n* (26 367) ([01. Consonants](https://md.mova.elamur.ai/free/en/docs/languages/fr/01-consonants/) §4). Three words in five end in one of three letters, *e, s, t*, that are silent in most words.

## 2. Diacritics

French has five diacritics, all on vowel letters except the cedilla.

| mark | name | letters | function |
| --- | --- | --- | --- |
| ´ | accent aigu | é | closed /e/: *été, blé, aimé, côté* |
| ` | accent grave | è, à, ù | on *è*: open /ɛ/ before a syllable with a mute *e* (*père, mère, il achète*); on *à* and *ù*: no sound value, only a distinction between homographs (*à* / *a*, *là* / *la*, *où* / *ou*) |
| ^ | accent circonflexe | â, ê, î, ô, û | a trace of a lost sound (usually a vanished *s*, sometimes a lost vowel); a distinction of homographs; a vowel quality (*â* /ɑ/, *ê* /ɛ/, *ô* /o/) |
| ¨ | tréma (diaeresis) | ë, ï, ü, ÿ | the vowel letter is pronounced on its own, not joined to the letter before: *Noël* /nɔɛl/, *naïf* /naif/, *Saül*, *maïs* /mais/ and *aiguë* /ɛɡy/ |
| ¸ | cédille | ç | /s/ before *a, o, u*: *français, garçon, reçu* |

**The accents on *e*.** *E* is the only letter on which the acute and the grave contrast. The two rules of the 1990 report summarise the practice (`cslf-1990`, analysis 3.2): an *e* takes an acute or a grave accent only at the end of a written syllable (*é/tude*, but *es/poir*, *mé/prise* but *mer/cure*), with exceptions at the end of the word (*accès, progrès*) and in recent compounds (*télé/spectateur*); and it takes a grave, not an acute, when a syllable with a mute *e* follows (*collège, collégien*; *fidèle, fidélité*; *il achète*). The pair *événement* / *évènement* is the standard case where the written syllable rule and the "mute *e* follows" rule disagree ([06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/) §7).

**The circonflexe.** Its history is that of lost letters. In most cases the accent stands in for an *s* that was written in Old French and dropped from the spelling in the eighteenth century: *forêt* (Old French *forest*), *hôpital* (*hospital*), *fenêtre* (*fenestre*), *tête* (*teste*), *île* (*isle*), *maître* (*maistre*), *mât* (*mast*), *août*. Dauzat (`dauzat-1908`) notes that the circumflex is also used in a few words for a vanished vowel (*âge* from *aage*, *sûr* from *seür*, *gaîté* / *gaieté*), and that its use is irregular enough that a family of words may differ: *jeûner* and *déjeuner*, *côte* and *coteau*, *grâce* and *gracieux*, *mêler* and *mélange*. The 1990 report adds that the accent does not even reliably mark a vowel quality (*noirâtre, pédiatre*; *zone, clone, aumône*; *atome, monôme*) and that on *e* it rarely marks a different sound from the grave (*il mêle, il harcèle*; *même, thème*) (`cslf-1990`, analysis 3.3). It also separates homographs (*sur* / *sûr*, *du* / *dû*, *mur* / *mûr*, *cote* / *côte*, *tache* / *tâche*, *jeune* / *jeûne*, *pécheur* / *pêcheur*), and marks the endings of the *passé simple* and the imperfect subjunctive (*nous aimâmes, vous aimâtes, qu'il aimât, qu'il fût*).

**Capitals.** The accents are kept on capital letters in correct typography (*État, École, À bientôt, ÉCOLE*); the omission of an accent on a capital is treated as a fault by the typographic manuals (we hold the Imprimerie nationale lexicon by its card only). In practice they are often dropped, especially on web pages and on keyboards that make capitals with accents hard to type. In UD French GSD (counted 2026-11-06) the preposition *à* at the start of a sentence is written *À* 209 times and *A* with no accent 86 times; so **29.2% of capitalised prepositional *à* are spelled without the accent**, which makes the word identical to the verb form *a* (of the 116 capital *A* tokens, 86 are tagged `ADP`, 10 `PROPN`, 6 `SYM`, 13 `X` and 1 `AUX`). The same occurs with *Etat* (31 times, against 125 of *État*) and *Etats-Unis* (17).

**Letters with diacritics are single letters.** In collation the accent is ignored at first: words are ordered by their bare letters, and accents break ties (the French rule of the dictionaries reads the accents from right to left, giving *cote < côte < coté < côté*). The *ç* is sorted as *c*.

## 3. Graphemes

A **grapheme** is a letter or a group of letters that spells a phoneme. French has many digraphs and some trigraphs; the choice among the possible graphemes for a sound is largely determined by the position in the word and by the history.

### 3.1 Vowel graphemes (letter or group to sound)

| grapheme | sound | context and examples |
| --- | --- | --- |
| *a, à, â* | /a/ (/ɑ/) | *chat, là, pâte* |
| *e* | /ə/ | open syllable, monosyllables: *le, de, petit, regarder* |
| *e* | /ɛ/ | closed syllable: *belle, sec, mer, elle, avec, vert, personne* |
| *é, ez, er, es* | /e/ | *été, parlez, parler, mes, les, des* (monosyllables in *-es*) |
| *è, ê, ai, ei, e + double consonant* | /ɛ/ | *père, tête, lait, peine, belle* |
| *ai* | /e/ or /ɛ/ | *j'ai* /e/, *je parlerai* /e/, *vrai* /ɛ/ (the contrast is lost for many speakers: [02. Vowels](https://md.mova.elamur.ai/free/en/docs/languages/fr/02-vowels/)) |
| *eu, œu* | /ø/ or /œ/ | *peu, feu* /ø/; *peur, sœur, œuf* /œ/ |
| *i, î, y* | /i/ | *vie, île, style* |
| *o, ô, au, eau* | /o/ | *mot, hôpital, chaud, beau* |
| *o* | /ɔ/ | *botte, or, porte* |
| *ou, où, oû* | /u/ | *vous, où, goût* |
| *u, û* | /y/ | *vu, sûr* |
| *oi, oy* | /wa/ | *moi, roi, voyage* |
| *ui* | /ɥi/ | *nuit, lui* |
| *oin* | /wɛ̃/ | *loin, point* |
| *ie* (final) | /i/ | *vie, amie*; in the verb *-ier* /je/ |
| *an, am, en, em, aon* | /ɑ̃/ | *chant, lampe, vent, temps, paon* |
| *in, im, ain, aim, ein, yn, ym* | /ɛ̃/ | *vin, impôt, pain, faim, plein, syndicat, sympathie* |
| *en* | /ɛ̃/ | after *i, é*, in *-ien* and *-éen*: *bien, chien, européen*; in some loans *examen* |
| *on, om* | /ɔ̃/ | *bon, pont, nom* |
| *un, um* | /œ̃/ or /ɛ̃/ | *un, brun, parfum* |

**Nasal vowels.** A vowel letter followed by *n* or *m* is nasal when the *n* or *m* is at the end of the word or before a consonant other than another *n* or *m*: *banc, bon, faim, temps, lampe*. It is oral and the nasal consonant is pronounced when the *n* or *m* is doubled or followed by a vowel: *bonne* /bɔn/, *bonté*, *animal* /a.ni.mal/, *inouï*, *fine* /fin/. The prefixes behave as the rule predicts: *ennemi* /ɛnmi/ keeps the oral vowel and a pronounced /n/ (the digraph *en* is not nasal before another *n*), whereas *emmener* /ɑ̃mne/ has the nasal vowel and a pronounced /m/. Exceptions are mostly Latin or foreign: *examen* /ɛɡzamɛ̃/, *Rhin*, *amen* /amɛn/, *hymen*, *spécimen*, and the *-um* words (*album, maximum, forum*) in /ɔm/.

**The hiatus and the tréma.** A written sequence of two vowels that forms a diphthong-like unit is pronounced as one sound; where it is not, the tréma is placed on the second letter: *naïf, Noël, maïs* ('maize'), *aiguë*. The 1990 report shifts the tréma in *aiguë, ambiguë, exiguë, contiguë* to the *u* that is pronounced (*aigüe* etc.) and extends the use to *gageüre* and *argüer*, where *-geu-* and *-gue-* produce a wrong reading (see [06](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/) §7).

### 3.2 Consonant graphemes (letter or group to sound)

| grapheme | sound | context and examples |
| --- | --- | --- |
| *b* | /b/ | *beau, bébé*; final silent in *plomb*, pronounced in *club, snob, Jacob* |
| *c* | /k/ | before *a, o, u* and a consonant, and at the end of the word: *cadre, col, cuir, clé, sac* |
| *c, ç* | /s/ | before *e, i, y*: *cela, ici, cycle*; *ç* before *a, o, u*: *ça, garçon, reçu* |
| *ch* | /ʃ/ | *chat, riche*; /k/ in learned words (*chaos, chœur, orchestre, technique*) |
| *d* | /d/ | *dame, rond* (final *d* is silent in *grand, pied, nid*) |
| *f, ph* | /f/ | *fer, photo* |
| *g* | /ɡ/ | before *a, o, u* and a consonant: *gare, gorge, gras* |
| *g, j* | /ʒ/ | *g* before *e, i, y*: *gentil, gîte, gym*; *j*: *jour* |
| *ge* | /ʒ/ | before *a, o, u*: *mangeons, geai, gageure* |
| *gu* | /ɡ/ | before *e, i, y*: *guerre, guitare*; /ɡy/ at the end of a word (*aigu* /eɡy/, *ambigu*); /ɡɥ/ in *linguiste* /lɛ̃ɡɥist/, /ɡw/ in *iguane* /iɡwan/ |
| *gn* | /ɲ/ | *montagne, signe*; /ɡn/ in learned words (*stagner, igné, diagnostic*) |
| *h* | — | silent (the two classes of initial *h*: [04](https://md.mova.elamur.ai/free/en/docs/languages/fr/04-connected-speech/) §4); in *ch, ph, th, rh* part of a digraph |
| *j* | /ʒ/ | *je, jeune* |
| *k, c, qu, q, cc* | /k/ | *kilo, cadre, qui, coq, accord* |
| *l, ll* | /l/ | *lit, belle, ville* (but see *ill*) |
| *ill* | /ij/ | *fille, travailler, bouteille* (the contexts after a consonant: /il/ in *ville, mille, tranquille*) |
| *il* (final) | /j/ | *travail, soleil, œil* |
| *m, mm* | /m/ | *mère, pomme* |
| *n, nn* | /n/ | *nez, bonne* |
| *p, pp* | /p/ | *pas, nappe*; *ph* /f/; silent in *sept, baptême, compter, temps* |
| *qu* | /k/ | *qui, que, quatre*; /kw/ in *aquarium, équateur, quadrupède* |
| *r, rr* | /ʁ/ | *rue, terre* |
| *s* | /s/ | initial, double or next to a consonant: *sac, passer, reste, tasse* |
| *s* | /z/ | between vowels: *rose, maison* |
| *ss, ç, sc, t* | /s/ | *tasse, ça, scène, nation, patient* |
| *t* | /t/ | *tête, patte*; /s/ in *-tion, -tie, -tial*: *nation, démocratie, partiel* |
| *th* | /t/ | *thé, mathématiques* |
| *v, w* | /v/ | *vie, wagon* (but /w/ in *whisky, watt, week-end*) |
| *x* | /ks/ | *taxi, expert, excellent, axe* |
| *x* | /ɡz/ | between vowels in *ex-* + vowel: *examen, exemple, exister* |
| *x* | /z/ | liaison and a few numerals: *deux enfants*, *sixième* /sizjɛm/, *dixième* |
| *x* | /s/ | *six, dix, soixante, Bruxelles, Auxerre* |
| *z* | /z/ | *zéro, gaz* (final *z* is silent in *nez, chez, assez*, /z/ in *gaz*) |

**The values of *c*, *g* and *s* in detail.** A *c* is /k/ before *a, o, u* and before a consonant and /s/ before *e, i, y*; the cedilla restores /s/ before *a, o, u* (*ça, garçon, reçu*). A *g* is /ɡ/ in the first context and /ʒ/ in the second; an *e* after *g* before *a, o, u* keeps the soft value (*mangeons, mangeait, nageur*) and a *u* after *g* before *e, i* keeps the hard value (*guerre, guide, fatigue*). This is why verbs in *-cer* and *-ger* change their spelling in inflection (*commençons, mangeons*; [06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/) §4). The letter *s* is /z/ between vowels and /s/ elsewhere, and a double *ss* is needed to write /s/ between vowels: this single rule is why *rose* and *rosse*, *poison* and *poisson*, *désert* ('desert') and *dessert*, *baser* and *basser* are distinct pairs.

## 4. From sound to spelling

French has about 36 phonemes (17 consonants, 3 glides, 12 oral and 4 nasal vowels, counting the marginal ones) and well over a hundred ways of writing them. The distribution is uneven: some phonemes have one dominant spelling, and some have a dozen. A list of the main ones follows.

### 4.1 The vowels

| phoneme | spellings (commonest first) | examples |
| --- | --- | --- |
| /a/ | a, à, â, e (in *femme, solennel*), emm | *chat, là, pâte, femme* |
| /ɑ/ | â, as, a | *pâte, bas, âme* |
| /e/ | é, er, ez, ai, es, et, ée, œ | *été, parler, vous parlez, j'ai, les, et, pensée, œdème* |
| /ɛ/ | e, è, ê, ai, ei, ë, et, es | *elle, père, tête, lait, neige, Noël, jouet, ces* |
| /ə/ | e | *le, petit, chemin* |
| /i/ | i, î, y, ï | *vie, île, style, naïf* |
| /o/ | o, ô, au, eau, aud, ot, op, os | *mot, hôtel, chaud, beau, artichaud, sirop, dos* |
| /ɔ/ | o, au (rare) | *botte, or, Paul, restaurant* |
| /u/ | ou, où, oû, oo, aoû | *vous, où, goût, boom, août* |
| /y/ | u, û, eu, ü | *tu, sûr, eu* (past participle of *avoir*), *Saül* |
| /ø/ | eu, œu, eû | *peu, vœu, jeûne, nœud* |
| /œ/ | eu, œu, ue, œ | *peur, bœuf, cueillir, œil* |
| /ɛ̃/ | in, im, ain, aim, ein, en (after *i, é*), yn, ym | *vin, timbre, pain, faim, plein, bien, lynx, sympa* |
| /ɑ̃/ | an, am, en, em, aon | *chant, lampe, vent, temps, paon* |
| /ɔ̃/ | on, om | *bon, ombre* |
| /œ̃/ | un, um | *un, parfum* |

A very large part of the irregularity of French sound-to-spelling lies in /e/ and /ɛ/, which are written by the many forms in the table, and in the nasals, written with up to six spellings. The vowel /e/ deserves a closer look: of the grammatical endings of the spoken /e/ in GSD (counted 2026-11-06, all three files), the infinitive in *-er* has 3 498 tokens, the past participle in *-é* 5 314 tokens and the finite second person plural in *-ez* 238 tokens. The choice among them is a grammatical decision (*il va manger*, *il a mangé*, *vous mangez*) that the speaker must make from syntax, and the three are among the commonest dictation errors ([06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/) §8, `antoine-2019`).

### 4.2 The consonants

| phoneme | spellings | examples |
| --- | --- | --- |
| /p/ | p, pp | *pas, nappe* |
| /b/ | b, bb | *bon, abbé* |
| /t/ | t, tt, th, d (liaison) | *thé, patte, grand homme* |
| /d/ | d, dd | *dame, addition* |
| /k/ | c, k, qu, q, cc, ch, cqu, ck | *cadre, kilo, qui, coq, accord, chœur, acquérir, ticket* |
| /ɡ/ | g, gu, gg, gh | *gare, guerre, aggraver, ghetto* |
| /f/ | f, ff, ph | *fer, effet, photo* |
| /v/ | v, w | *vie, wagon* |
| /s/ | s, ss, c, ç, sc, t, x | *sac, tasse, cela, ça, scène, nation, six* |
| /z/ | s, z, x | *rose, zéro, deuxième* |
| /ʃ/ | ch, sh, sch | *chat, short, schéma* |
| /ʒ/ | j, g (before *e, i, y*), ge | *je, gens, mangeons* |
| /m/ | m, mm | *mère, pomme* |
| /n/ | n, nn | *nez, bonne* |
| /ɲ/ | gn | *montagne* |
| /ŋ/ | ng | *parking, camping* |
| /l/ | l, ll | *lit, belle* |
| /ʁ/ | r, rr, rh | *rue, terre, rhume* |
| /j/ | y, i, ill, il, ï, ll | *yeux, hier, fille, travail, aïeul* |
| /w/ | ou, w, oi (/wa/) | *oui, whisky, moi* |
| /ɥ/ | u | *nuit, lui* |

The spellings in brackets are not equally likely: for /s/ the main spellings are *s* (initial and next to a consonant) and *ss* (between vowels) and *c* (before *e, i*), and for /ʃ/ the digraph *ch* is the rule, *sh* and *sch* are loans. The spellings of /j/ are the most scattered. Doubled consonant letters (*pp, tt, mm, nn, ll, rr*) rarely double the sound in modern French; they signal an etymological or morphological fact (*addition, appeler*), or the openness of the vowel before (*belle, sotte*), or they are inherited from Latin (*mettre, terre*). The doubling that represents a long consonant is in the conditional and future *rr* of *courir* and *mourir* (*il courrait* against *il courait*) and in learned words (*illégal, irréel, innombrable*) in careful speech.

## 5. Silent letters

A large part of the French spelling is made of letters that are not pronounced. They belong to four groups.

**1. The final consonants.** In most words a final consonant letter is silent: *petit, grand, temps, nez, trop, fils* (the plural), *lourd, gros, blanc, chaud, doigt, sang*. The silent final consonants are *s, t, d, x, z, p, g, c* in many words, *r* in *-er* (infinitive and nouns *boulanger*), and *l* in *fusil, outil, gentil*, *cul*. Pronounced final consonants include *c, r, f, l* in many short words (the old mnemonic CaReFuL: *sac, mer, bref, sel*), but that is a trend and not a rule: *banc, estomac, tabac, porc* (silent in the plural *porcs*), *clef, nerf* (silent in *nerfs*), *cerf*, *bœuf* (silent in the plural *bœufs* /bø/, *œufs* /ø/), *os* (singular /ɔs/, plural /o/). Whether a given word has a pronounced final consonant is a lexical fact, learned word by word. The numerals show the pattern in one word: *six, huit, dix* sound their consonant at a pause (*j'en ai six* /sis/), link before a vowel (*six ans* /si.zɑ̃/) and drop it before a consonant (*six francs* /si fʁɑ̃/); *sept* and *neuf* keep it everywhere for most speakers today (*sept francs* /sɛt fʁɑ̃/).

**2. The final *e*.** The *e* at the end of a word is silent in speech (*petite, table*) but it matters: it shows that the preceding consonant is pronounced (*grand* /ɡʁɑ̃/ vs *grande* /ɡʁɑ̃d/, *petit* vs *petite*), that the preceding *g* is soft (*rage*) and the *c* soft (*race*), that a noun is feminine (a tendency), and that a vowel before is long or open. As §1.3 shows, **31.6% of the words of GSD end in *e***, and the *e* is mute in the large majority. The mute *e* is the mute-*e* of [04](https://md.mova.elamur.ai/free/en/docs/languages/fr/04-connected-speech/) §7: a letter that is pronounced in verse, in the south of France and in some careful speech.

**3. The silent letters inside words.** These are mostly etymological and were added by the sixteenth- and seventeenth-century scholars to bring the French word closer to its Latin or Greek source: *doigt* (Latin *digitus*), *temps* (*tempus*), *corps* (*corpus*), *sept* (*septem*), *baptême* (*baptisma*), *compter* (*computare*), *poids* (the *d* was added because of the false derivation from *pondus*, while the word comes from *pensum*), *vingt*, *legs*, *sculpter*, *dompter*, *pouls*. A group of such words persists in the dictionary as a famous set, and Brunot (`brunot-1905`) called the Académie's attachment to them a fidelity to the "Graeco-Latin spelling". A well-known case is *oignon*, in which *ign* is a graphic archaism for /ɲ/ (the 1990 report proposes *ognon*; [06](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/) §7).

**4. The silent *h*.** *Homme, heure, huile, thé, théâtre, rhume, rhinocéros, exhaler, hôtel*. The *h* also has a role in separating vowels (*trahir, cahier, envahir, ahuri, cohue*) and in the contrast between aspirate and mute initial *h* ([04. Connected speech: enchaînement, liaison, elision, the mute e and assimilation](https://md.mova.elamur.ai/free/en/docs/languages/fr/04-connected-speech/) §4); Martinon notes that "the mute *h* serves absolutely nothing and ought long since to have disappeared from the spelling, or rather never to have been introduced under pretext of etymology" but that "the aspirate *h* is not the same", since it keeps hiatus where elision and liaison would otherwise fuse the words (`martinon-1913`).

## 6. Homographs and homophones

**Homophones.** French has a very large number of words with the same pronunciation and different spelling. The homophones are overwhelmingly a consequence of the silent final consonants and of the equivalence of the vowel spellings.

| sound | words |
| --- | --- |
| /vɛʁ/ | *ver, verre, vers, vert* |
| /sɛ̃/ | *saint, sein, seing, ceint, sain* |
| /sɑ̃/ | *cent, sang, sans, sent, s'en, c'en* |
| /mɛ/ | *mais, mes, met, mets, mai* |
| /ɔ̃/ | *on, ont* (/ɔ̃t/ before a vowel) |
| /sɔ̃/ | *son, sont* |
| /sɛ/ | *c'est, s'est, sais, sait, ces, ses* |
| /kɑ̃/ | *quand, quant, qu'en, camp, Caen* |
| /pwa/ | *poids, pois, poix* |
| /tɑ̃/ | *temps, tan, tant, taon* |
| /vwa/ | *voie, voix, vois, voit* |
| /fwa/ | *foi, foie, fois* |
| /o/ | *o, ô, eau, au, aulx, haut, os (plural)* |
| /a/ | *a, à, as, ah* |
| /u/ | *ou, où, houx, août* (for those who say /u/) |

The grammatical homophones are the practical problem. In UD French GSD (all three files, counted 2026-11-06) the tokens of the commonest pairs are: *a* 2 339 against *à* 10 323; *et* 8 130 against *est* 5 337; *ou* 898 against *où* 392; *on* 547 against *ont* 669; *son* 1 563 against *sont* 1 186; *ce* 1 148 and *se* 1 346; *ces* 407 and *ses* 793; *la* 11 095 against *là* 84; *mais* 833 against *mes* 36. A speech recogniser must choose among them from context, and a spelling corrector must catch the confusion of *a* for *à*, *et* for *est*, *ou* for *où*, *on* for *ont*, *son* for *sont*, which is a **syntactic** error, not an error of form. The ending homophones are also grammatical: *il parle* = *ils parlent* = *tu parles* /paʁl/; *parlé* = *parler* = *parlez* = *parlai* /paʁle/. The countable errors of French dictation follow from this: the corpus of dictations by children with cerebral palsy or dyslexia studied by Antoine et al. (`antoine-2019`) shows that many errors are not lexical, that is, they produce a word that exists and that a dictionary lookup cannot flag: 49% of the errors in the cerebral-palsy sub-corpus and 29% in the dyslexic one (the dyslexic children make more errors that end in a non-word). The authors note that this share is much higher than the 9% they cite for Spanish and the 17% for English. In the same study the spelling correctors tested fail to correct more than half of the errors of the cerebral-palsy sub-corpus, in more than a third of the cases do not even detect them (above all the real-word errors), and leave up to three quarters of the dyslexic errors uncorrected (521 errors in all). Bodard (`bodard-2020`) analyses the errors of dyslexic adults in written corpora in French and compares them to English and Spanish results, to guide specific correction modules.

**Homographs.** The opposite case, one spelling with two pronunciations, is much rarer than in English and arises from the same causes. Most pairs set a noun or adjective against a verb in *-ent*, whose ending is silent: *content* /kɔ̃.tɑ̃/ (glad) against *ils content* /kɔ̃t/ (they tell); *le couvent* /ku.vɑ̃/ against *les poules couvent* /kuv/; *le président* /pʁe.zi.dɑ̃/ against *ils président* /pʁe.zid/; *violent* against *ils violent*. Others are lexical: *les fils* /fis/ (sons) against *les fils* /fil/ (threads), *est* /ɛ/ (is) against *l'est* /ɛst/ (east), *les portions* /pɔʁ.sjɔ̃/ (servings) against *nous portions* /pɔʁ.tjɔ̃/ (we carried), *le as* /ɑs/ against *tu as* /a/. All of these are grammatical or lexical, and in text the tagger sorts them out from the context.

## 7. Why French spelling is the way it is

1. **Latin base.** The first written French, the Strasbourg Oaths (842) and the *Séquence de sainte Eulalie* (c. 880), used Latin letters to write a language that had already changed; the scribes worked out ways to spell the new sounds (*ch* for /tʃ/, later /ʃ/; *ç*, *gn*, *ill*) and by the twelfth century the writing of Old French was close to its pronunciation. Dauzat (`dauzat-1908`) puts the point in general terms: the first scribes tried a phonetic spelling, one sign for one sound; as languages evolved, the spelling followed until the first grammarians arrived who, ignorant of language change, wished to stop it with fixed rules.
2. **The loss of sounds and the keeping of letters.** From Old French to the sixteenth century a large number of sounds disappeared (the final consonants, the preconsonantal *s*, the hiatus vowels, the /ə/ in many positions), but the spelling kept the letters. This is the origin of *forêt* < *forest*, *fête* < *feste*, *doux* < *dous*, and of the circumflex of *mûr* < *meür*.
3. **The humanists' etymological spelling.** In the sixteenth century the scholars and the printers (above all the Parisian ones) added letters to bring words close to Latin and Greek, whether the etymology was right or not: *doigt* with *g* (Latin *digitus*), *poids* with *d* (a false link to *pondus*), *sçavoir* with *c* (a false link to *scire*), *advocat*, *advis*. In Brunot's words about the Académie of 1673: "they wanted this spelling to distinguish *gens de lettres* from the ignorant and simple women", and so it was made "so etymological and so pedantic that it would have sufficed to discredit the Dictionary" (*fresle, charactère, phase, advocat, advis, toy, sçavoir*), with "innumerable contradictions" (*abbattre* and *aborder*, *eschancrer* and *énerver*) (`brunot-1905`). Opposite them stood the phoneticians of the time: Louis Meigret (1550), who proposed a phonetic spelling, was answered by Henri Estienne and others; in Brunot's quotation Meigret asked what would be said of a painter who tried to copy a faulty picture instead of the model (`brunot-1905`).
4. **The Académie française and the fixing of the dictionary.** The first edition of the *Dictionnaire de l'Académie* appeared in 1694 (the decision on spelling was taken on 8 May 1673). The Académie revised its spelling in **1740** (Druon's 1990 report says that in 1740 "the spelling of one word in four was changed"; Brunot says that the revision the Académie entrusted early in 1756 to a "plenipotentiary", the Abbé d'Olivet, touched more than 5 000 of about 18 000 words, with further changes in 1762), in **1835** (the replacement of *oi* by *ai* in *j'avois, françois* → *j'avais, français*, and similar), in **1878** (7th edition) and in **1932–1935** (8th edition) (`cslf-1990`, `brunot-1905`). The same period separated *i* from *j* and *u* from *v*, as Ramus had asked in the sixteenth century and as Dutch printers had done for a hundred years (`brunot-1905`). The grave accent of *mère, frère* is as late: Bourciez (`bourciez-1921`) notes that d'Olivet in 1736 still kept the older writing, that the Académie hesitated in 1740, and that only the dictionary of 1762 writes *è* uniformly in words of this kind. The spelling of the eighteenth century is the one we write.
5. **The school.** Brunot's argument (1905) is that the Revolution made French a compulsory subject of public instruction at all levels and that the schools, which needed books, took the Académie's dictionary as their reference; the Napoleonic Université carried on, with the grammar of Noël and Chapsal; after a generation of schooling "the harm was almost done". Dauzat adds that the importance attached to spelling as a sign of good education is recent: "it is only since the Restoration that spelling became one of the conventional signs of a good education; Victor Hugo is one of the first writers to take care to write the words as the dictionary requires" (`dauzat-1908`). Brunot's open letter to the Minister (1905) states the cost to the school: the primary-school teachers consulted by him voted unanimously for a radical reform; spelling is "the scourge of the school", and examinations, from the primary certificate to the entrance into the postal service, depend on the dictation. His own proposal was a phonetic script of one sign for one sound, with the old spelling taught for reading only for a transition (`brunot-1905`). It was not adopted.
6. **The attempts at reform.** A commission of the Académie in 1893 led by Gréard, Brunot's 1905 campaign, the proposals of 1975 (Académie), and finally the **1990 rectifications** of the Conseil supérieur de la langue française, approved by the Académie française in its session of 3 May 1990 and published in the *Journal officiel* of 6 December 1990 (`cslf-1990`). The report frames them as the continuation of the Académie's own adjustments of 1740, 1835, 1878 and 1935; they leave the old spelling admitted. They are the subject of [06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/).

## 8. How Mova handles the writing system

**What is done.** Mova takes French as written, in Unicode, without a pronunciation layer. The English tokenizer (`en/src/tok.rs`) treats every Unicode alphabetic character as a letter, so *é, è, ç, ô, œ* are letters, and the combining diacritics U+0300–U+036F are accepted as letters too (a decomposed *é* is not cut into two tokens). Lowercasing for features goes through Rust's `to_lowercase`, which maps *É* to *é* and *À* to *à*. `en xpos-fill` maps the UD tags of a French treebank to the Penn Treebank tags the English model knows.

**What is not done.**

- **No Unicode normalisation.** There is no NFC step: a word in decomposed form (*e* + combining acute) is not the same string as the precomposed *é* in the training forms, and is an unseen word for the tagger. GSD is precomposed throughout, so the measurements are not affected.
- **No treatment of lost accents.** The preposition *À* capitalised with no accent (86 of 295 in GSD, 29.2%, counted 2026-11-06) is the word *A*, a different form for the tagger; the same for *Etat* against *État*. These are unseen forms and are also ambiguous with the verb *a*.
- **No ligature normalisation.** *Œuvre* and *oeuvre* are two strings; GSD has 281 of the first kind and 19 of the second kind (counted 2026-11-06), so the second is an unseen form unless the training data also holds it.
- **No grapheme-to-phoneme rules, no phonetic lexicon.** The tables of §3 and §4 are in this chapter and nowhere in code. A set of context rules for the vowel and consonant graphemes (§3), with an exception list for the final consonants, would reach a good pronunciation of most words, but the final consonant and the foreign words need a lexicon; this is the first piece to build for any use of Mova in speech.
- **No homophone model.** The text is disambiguated by its spelling; the homophone problem in §6 reaches Mova only when the input is a transcription of speech or a text with the errors of §6 (*a/à*, *et/est*, *ou/où*, *on/ont*, *son/sont*); the tagger, trained on edited text, has not seen them.
- **Tagger accuracy.** On GSD (`en tag-eval`, train → dev; 2026-10-22, rerun 2026-11-06 twice with the same result) 7.2% of the development tokens (2 570 of 35 721) are unknown to the tagger and it tags them right 84.8% of the time, against 96.5% overall: every spelling variant, every accent error and every informal spelling counts as an unknown word.

**Sources of data.** Letter, accent and homophone counts in this chapter are our own counts on UD French GSD of 2026-11-06 (UD_French-GSD, the three files, 400 385 syntactic words, 268 870 of which consist of letters only).

## Sources for this chapter

- Brunot, Ferdinand. 1905. *La réforme de l'orthographe: lettre ouverte à M. le Ministre de l'Instruction publique.* Paris: A. Colin. Public domain. `brunot-1905`; library `books/b/br/brunot-1905-la-reforme-de-lorthographe`
- Dauzat, Albert. 1908. *La langue française d'aujourd'hui.* Paris: A. Colin. Public domain. `dauzat-1908`; library `books/d/da/dauzat-1908-la-langue-fran-aise`
- Bourciez, Édouard. 1921. *Précis historique de phonétique française*, 5th ed. Paris: Klincksieck. Public domain. `bourciez-1921`; library `books/b/bo/bourciez-1921-precis-historique-de-phonetique`
- Martinon, Philippe. 1913. *Comment on prononce le français.* Paris: Larousse. Public domain. `martinon-1913`
- Conseil supérieur de la langue française. 1990. *Les rectifications de l'orthographe.* Journal officiel, Documents administratifs no 100, 6 December 1990 (transcription). `cslf-1990`; fr-gram-rectifications-orthographe-1990
- Antoine, Jean-Yves, Marion Crochetet, Céline Arbizu, Emmanuelle Lopez, Samuel Pouplin, Amélie Besnier & Mathieu Thebaud. 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`
- Bodard, Johana. 2020. Spécificités des erreurs d'orthographe des personnes dyslexiques : analyse d'un corpus de productions écrites. JEP-TALN-RECITAL 2020. CC BY 4.0 (ACL Anthology). `bodard-2020`
- UD French GSD, UD 2.18 (CC BY-SA 4.0); counts of 2026-11-06.
- Mova: `en/src/tok.rs`, `en/src/main.rs` (`xpos-fill`); `en tag-eval` on GSD train → dev after `en xpos-fill`, 2026-11-06.
- [← previous 04. Connected speech: enchaînement, liaison, elision, the mute e and assimilation](https://md.mova.elamur.ai/free/en/docs/languages/fr/04-connected-speech/)
- [next → 06. Spelling rules: inflection, compounds, hyphens and the 1990 rectifications](https://md.mova.elamur.ai/free/en/docs/languages/fr/06-spelling-rules/)
- [↑ back to Modern Standard French (fr) — a reference manual for Mova](https://md.mova.elamur.ai/free/en/docs/languages/fr/)