← Blog · 19 November 2026 · Lire en français

Foreign names in French world news: six French reading rules that misfire, and what to do about it

A French text-to-speech voice reads every name as if it were French. On a world news page that means nasal vowels where there are none, silent letters that should be heard, and Polish consonant clusters with no rule at all. Examples from our pronunciation bank, and an honest account of what is fixed automatically and what is not.

If you run the audio for a French-language outlet, the page that will cost you the most correction time is not culture or sport. It is world news. The vocabulary is ordinary; the names are not. Heads of state, newly appointed ministers, cities in the headlines, disputed provinces: they arrive in waves, change week to week, and almost none of them is spelled according to French rules.

A voice tuned for French does exactly what it was built to do: it reads in French. That tuning is what makes it sound natural on a French paragraph, and it is also what makes it fragile on a Ukrainian surname.

The target is not the native pronunciation

Before looking at errors, it helps to agree on what "correct" means. Nobody expects a French news bulletin to say "Xi Jinping" with Mandarin tones, or "Szczecin" the way a Polish speaker would. The target is the adapted pronunciation a French-speaking radio presenter uses: stable, recognisable, and not showy.

The spelling is also an editorial choice that the voice must follow. French papers write "Zelensky", English-language ones "Zelenskyy"; some titles print "Mladic" without the diacritic of "Mladić". Whatever the newsroom chose, the audio has to cope with it.

Six French rules that misfire

Since late September we have been preparing, by theme, names that a French voice tends to get wrong. Several themes are international: world figures, world places, Central and Eastern Europe, Africa, Asia, each with a little over a hundred proposals. Every entry holds the written name, a reading spelled in ordinary French letters (never phonetic alphabet, which the engine does not read), and the reason. Grouping those reasons shows that most errors come from a handful of French rules applied outside their territory.

1. The nasal vowel. In French, "en", "an", "in" before a consonant usually become nasal. In Zelensky ("Zé-lenn-ski"), Orban ("Or-bane") or Xi Jinping ("Chi Djine-ping"), they should not.

2. The silent final consonant. French drops many final consonants. Nouakchott ("Nouak-chotte"), Lagos ("La-gosse") and Eilat ("É-latt") need theirs.

3. The mute e. A French reader swallows many e's. In Tshisekedi ("Tchi-sé-ké-di") or Makeni ("Ma-ké-ni"), every e is a full "é".

4. The French u. Burundi, Enugu, Tusk: the u is "ou" ("Bou-roun-di", "É-nou-gou", "Tousk"), not the rounded French u.

5. The letters x and z. Pinyin x is "ch" (Xiamen, "Chia-mène"); pinyin zh is "dj" (Shenzhen, "Chène-djène"); German z is "ts" (Scholz, "Cholts"; Merz, "Mèrts").

6. Clusters French has no rule for. Polish szcz in Szczecin ("Chtché-tchine"), sz in Olsztyn ("Ol-chtine"), Hungarian cs in Pécs ("Pétche"). Here the voice is not misapplying a rule; it simply has none.

For a desk editor, this list is more useful than any single correction: it lets you spot the risky names in a story before anyone presses play.

What is automatic, and what is not

This is where promises usually slip, so precisely: an entry in the bank does not apply by itself. The bank is a list of proposals. An entry only joins the shared base that applies to every account after it has been played on our test bench, as is and corrected, and approved by a person. Until then it changes no reading anywhere.

The reason is an asymmetry. A mistake in one customer's lexicon affects that customer, who notices and fixes it. A mistake in a shared base would reach every account, including those that never opened the pronunciation screen. And a name being in the bank does not prove the voice gets it wrong: some it does, some it already handles. Correcting what was already right is at best useless and at worst a regression. Only listening settles it. For the same reason, a name that is also an ordinary French word never goes into a shared base: it would change every sentence where the common word appears.

What a newsroom can do today

The tools that work now live in your account.

Your pronunciation lexicon. An entry set once applies to every later article and takes precedence over any shared entry. The steps are in fixing how a word is pronounced. An account holds up to two hundred entries: plenty for the names that keep coming back, and deliberately not a world directory.

Names flagged before reading. Proper names in an article are detected before the audio is made, and uncertain ones come with a suggested reading you listen to before deciding. For a universal name such as Zelensky, said the same way in every newsroom, the question is not put to each account in turn: it goes up once, for a shared decision. See uncertain names flagged before reading.

A routine before big events. An election abroad, a summit, a conflict that opens: the evening before, the duty editor lists the ten names that will recur all week, plays each in a short sentence, and fixes the ones that stumble. Ten minutes of work saves fifty audios to redo.

Ordinary borrowed words ("leitmotiv", "aficionado") are a related but different problem, covered in foreign words in French text to speech.

The test worth running

Take a real article from your world news section, not a demo sentence. Listen while following the text, and mark every name that makes you look up. If you mark more than two, the voice is not what needs changing; the lexicon needs feeding. And if a vendor tells you their voice gets every name in the world right with no setup, ask them to read "Szczecin" in front of you.

Give your articles a voice with WeDispatch

This blog is itself voiced by WeDispatch. Curious how it sounds on your content?

Book a demo

Read next