Surnames in capitals: why French text-to-speech spells them out, and how to stop it

If you have ever read a French press release, a town council report or a death notice, you have seen it: "Jean DUPONT, directeur de la communication". French administrative style puts family names in capitals. It is not shouting. It is a convention, used on birth certificates, in wire copy, in formal announcements and in military records, and it does a real job on the page: it tells the reader which word is the surname, which helps when a surname could pass for a first name.

Read aloud by a machine, the same convention can produce "the colonel N, I, C, O, T". This article explains why it happens, shows exactly what our text preparation does about it, and is upfront about the forms it does not handle yet.

Why a voice ends up spelling a name

To a speech engine, a word written entirely in capitals is first of all a possible acronym. Some acronyms are spelled letter by letter, some are pronounced as words, and some are best expanded; we went through those choices in our article on acronyms in French text-to-speech. Faced with "DECHAUX", the engine has no way to know it is a general's name rather than an organisation. It guesses, and sometimes the guess is to spell.

This is not hypothetical. On 18 September 2026, listening to a military history page published by one of our clients, the sentence "Les généraux GILLES, DECHAUX et BODET sont à bord" came out with DECHAUX recited one letter at a time. Everything else in the reading was right, and that one name was all anyone remembered.

Better voices do not fix this, because it is not a knowledge problem. Nobody needs to know who Dechaux was. The question is typographical: in what context does a string of capitals refer to a person?

The rule we apply

Our text preparation, which runs before any voice reads anything, looks for a word in capitals directly after a title that announces a person, and puts it back into normal case. NICOT becomes Nicot, and the voice reads it as a name.

Here is the actual behaviour. Each sentence below was run through the code, not paraphrased from a spec:

• "Le colonel NICOT prend la décision" becomes "Le colonel Nicot". The same holds after capitaine, président, maire, ministre, préfet or docteur.
• "Les généraux GILLES, DECHAUX et BODET": all three names are handled, because the whole list follows the plural title.
• "Monsieur DUPONT et Madame MARTIN", and "Mme LEFEBVRE-MARTIN": the courtesy title is enough, abbreviated or not, and the hyphenated surname keeps its hyphen. How we expand those titles is covered in our piece on Monsieur, Madame and their abbreviations.
• "Me DURAND plaide" becomes "maître Durand", and "le Dr MARTIN" becomes "le docteur Martin": the abbreviation is expanded first, then the name.
• "L'adjoint de NAVARRE" also works, since a deputy is always somebody's deputy.

The rule comes with guards, and they matter as much as the rule. "Le patron de SNCF" does not turn into "Sncf": a known acronym stays an acronym and is expanded as one. "Le col du Galibier" (a mountain pass) is not read as an abbreviation of colonel. And "Ce dossier", where two capitals happen to open a sentence, is left alone. A rule that damages correct text is worse than no rule at all.

What still gets through

These are the forms the rule does not handle today, tested the same way. In each case the name reaches the voice still in capitals, and depending on the word it may be read correctly or spelled out. We cannot promise which.

• First name, then surname: "Jean DUPONT a déclaré", or a press contact line such as "Contact : Marie DURAND". Unfortunately, this is the most common form in a French press release.
• A first name between the title and the surname: "Mme Claire DUBOIS" keeps DUBOIS in capitals, while "Mme DUBOIS" is handled.
• A surname alone at the start of a sentence: "DUPONT a déclaré hier".
• Two-word surnames and particles: "le général DE LATTRE", "le ministre LE MAIRE".
• "M." followed by a capitalised surname: "M. DURAND" is left as it is, abbreviation included.
• Formal announcement wording: "Les époux DURAND".
• Titles outside the list: "le directeur de NAVARRE" is not handled, whereas "l'adjoint de NAVARRE" is.

Why not simply lowercase every capitalised word? Because without context, capitals really are ambiguous. CAF, OTAN, UNESCO and plenty of brand names are not people. A broad rule would get surnames right and break acronyms everywhere, and we would rather leave a rare, visible, fixable error than introduce a systematic one on words that were read correctly.

Two ways to close the gap

Write the name in normal case in the text that gets read. If you control the source, "Jean Dupont" is read without hesitation. Many French newsrooms keep the capitals out of habit; a web article is not a legal document.

Add the name to your pronunciation lexicon. Lexicon entries ignore case, so an entry for "Durand" also fires on "DURAND", and the voice receives the form you entered. The displayed text does not change: read-along text and subtitles still show DURAND exactly as written. For a name that keeps coming back (the mayor, the club chairman, a regular columnist), one entry covers every future article. See the pronunciation lexicon for how entries work.

A quick test for any tool

Have it read three sentences: "Le colonel NICOT prend la décision", "Jean DUPONT a déclaré hier" and "Le patron de SNCF a parlé". The first shows whether the tool knows the convention at all, the second whether it goes beyond the easy case, the third whether it breaks acronyms while trying to help. Passing the first and third already beats a raw reading. The second is where we fail today too, which is why the two fixes above exist.

The cases we do handle are part of the public test record on our French text-to-speech page, where the score is computed from the same sentences our tests run, and drops the day a rule breaks. The forms listed under "what still gets through" are not in that record yet. They are written down here so you know what to watch for in the meantime.

Read next

• Security questionnaire for an audio tool: what to ask, and our answers, including the noes
• Breton, Basque, Corsican, Alsatian: why French text-to-speech trips on regional names
• A woman's or a man's voice for a quote: what French grammar reveals about who is speaking