🗣️ Modern Standard French (fr) — a reference manual for Mova

This is an open reference grammar of modern standard French (ISO 639-1 fr, ISO 639-3 fra) as it is written and, where the two differ, as it is spoken. It has chapters on the sounds and the spelling, the nominal and verbal systems, the clause, the punctuation, the regional varieties and the lexicon. It is written for two readers at once: a person who reads English and wants a compact account of how French works, and Mova, whose tools read French text as Universal Dependencies trees. The examples are French with English glosses. Every chapter ends with a section, How Mova handles it, that says what Mova's tools do with the point today and where they stop.

The manual is written in our own words. Facts come from public-domain grammars and phonetic manuals (Passy 1906, Nyrop 1902, Martinon 1913, Clédat 1894, Brunot 1899 and others), from the official 1990 orthographic rectifications, and from open-access research. Works under share-alike or non-commercial terms are cited and paraphrased, never copied. The full list, with licences, and the key works we lack, are in Sources.

What Mova can do with French today is small. Mova has no French morphology and no French lexicon of its own. French is read by Mova's own tree parser (en) trained on a French treebank after its tags have been mapped to the tag set the parser knows (en xpos-fill); the model is kept as data, outside the repository. On UD French GSD (UD 2.18, train → dev, 2026-11-06) it gives LAS 87.09 with the parser's own tags and 89.78 with the treebank's gold tags (UAS 91.12 / 92.59), on the parser's relation set, which reads French subtypes such as obl:mod, aux:tense or expl:pass as their base; counted with the French subtypes, LAS is 78.80. The tagger's accuracy on unknown words is 84.8% (2026-10-22). For lemmas and UPOS, the Elephant's French pass uses the treebank itself: a word seen in GSD's training part takes its annotation there, an unseen one an edit script of its tag and suffix; on GSD dev this gives lemma accuracy 97.68% (87.91% on unseen forms), UPOS 97.03%, and a French tokenizer in GSD's conventions reaches word F1 97.87 from the raw sentences (2026-11-06). The Elephant's French word graph is described in wordgraph. The chapters say, construction by construction, what that does and does not cover.

Chapters

French continues spoken Latin; the Latin manual gives the source of its negation, its subjunctive, its avoir tenses and its prepositions de and à (la chapter 13).

Conventions

Examples are given in three lines:

fr-ud-dev_00007 · UD French GSD Elle lutte pour échapper aux tueurs à ses trousses. she fights for escape-INF to-the killers at her heels — "She is fighting to escape the killers on her heels."

• A bold id such as fr-ud-dev_00007 is the sent_id of a sentence of an open treebank (UD French GSD unless named otherwise: ParisStories, Rhapsodie, Sequoia, ParTUT, PUD, FQB). It can be looked up in the UD 2.18 release. Treebank examples are short quotations; the treebanks are CC BY-SA, LGPL-LR or CC BY-NC-SA (see Sources), and the manual does not reproduce any of them in bulk. Their text is kept as published, including its errors and its spacing.
• An example without an id is constructed by us.
• A starred example (*) is ungrammatical in the standard; a question mark (?) marks a doubtful one; % marks a form that is normal in speech and avoided in careful writing.
• The word-for-word gloss is in italics, with hyphens joining the English words that translate one French word and the usual abbreviations (SG, PL, F, M, 1, 2, 3, INF, PTCP, SUBJ, IMP, REFL, CL for a clitic). The translation is in double quotes.

Spelling. French is given in its standard orthography with accents. Where the 1990 rectified spelling differs, the traditional one is used in the examples (it is what the treebanks have) and the rectified one is given in chapter 1.

Phonetic notation. Pronunciation is in IPA between slashes, for example /ʁə/ for the mute e word re. The notation is broad: it shows phonemes, not detail of realisation. The mute e is written (ə) when it is optional.

Grammatical terms. The terms follow the traditional French grammars where they are standard (the passé simple, the conditionnel), with an English equivalent in brackets at first use. The UD labels (nsubj, aux:pass, expl:comp) are in code type. Where the UD French GSD tree differs from the traditional analysis, the section How Mova handles it says so.

Citations. In the text a work is cited as (author-year) with the slug of its card in the research library where a card exists, for example (passy-1906). A citation names the work, not a page. A point cited to a work was checked in the text we hold; the large body of standard description that every reference grammar shares (the paradigms, the basic agreement rules) is not attributed to a single work. Works we know only by their card are marked card only in Sources, and no claim rests on them except where the sentence says "reported in".

Numbers. Every count is measured on our data and given with its date. Treebank counts are on UD 2.18 files as released (2026-10-22 unless said otherwise).

The data behind the examples

Non-commercial and research-only data is used only for tests and measurements, never in what is built.

The Elamur team

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