← Blog · 10 September 2026 · Lire en français

Non-English foreign words in French text read aloud

Leitmotiv, aficionado, ciao, Autobahn, a cappella: French text is full of words borrowed from elsewhere. What a French voice does with them, why it is not the same problem as English loanwords, and how to take back control.

English is not the only language that invites itself into French text. A food column talks about tapas and antipasti, a music review about crescendo and a cappella, a report about a kibboutz or an Autobahn, a gossip brief about paparazzi and dolce vita. These words are French now, in the sense that they are written without quotation marks and without a second thought, but they keep an Italian, Spanish, German, Hebrew or Japanese origin. Handed to a synthetic voice tuned for French, they raise a particular problem, different from the one English words raise, and worth naming precisely.

Why this is not the loanword problem

An English word in French text mainly raises a question of source pronunciation: should you say "meeting" the English way or the French way, and how far. That debate is familiar, and a modern French voice often gets by with a deliberately French-ified pronunciation. We covered it separately, in English loanwords.

Other languages shift the problem. First, because the engine has seen these words far less than their English cousins: "leitmotiv" is rare, and a voice that has not met it enough applies French reading rules by default, sometimes with comic results. Second, because French-ification is not always the right answer here: nobody says "paella" fully the French way, but nobody says it fully the Spanish way either. Each word has settled, through use, on a point of balance, and that point is written nowhere. A voice cannot guess it; it applies a rule, and the French rule gives, on these words, a reading that ranges from correct to plainly wrong.

What a French voice actually does

Take a few cases and what you hear when nothing is corrected.

  • "a cappella": read as two French words, the double p survives, but the stress falls the French way and the word is lost. Many engines cope because it is common in music.
  • "Autobahn": the German "au" is not the French "au", and the aspirated "h" in the middle does not exist in French. A French voice reads a flat "oto-ban", which stays understandable.
  • "kibboutz": the "ou" is fine, but the word is rare, and depending on the engine the final "tz" is reduced or leaned on.
  • "leitmotiv": the classic trap. The spontaneous French reading gives something like "laïtmotif", far from the German with its voiced final "v".
  • "jalapeño": the Spanish "ñ" is not on a French keyboard, and if it is typed "jalapeno", the voice will say "jalapèno" instead of the palatal "jalapégno".

None of these cases is a disaster on its own. The problem appears when a foreign word is central: an article on Italian cooking where "bruschetta" comes back ten times, a portrait of a composer where "Weltschmerz" is the keyword. There, a wrong reading repeated is no longer an anecdote, it is a nuisance.

Take back control without becoming a phonetician

The answer is not to hope the engine knows every word of every language: that is a bottomless pit, and a tool that claimed to handle it all would be over-promising. The honest answer is a pronunciation lexicon. You spot the few words that matter for your site, you tell it how to read them, and the tool applies your instruction everywhere, without you having to come back to it. That is exactly the mechanism described in fixing pronunciation with a lexicon, and it is the same gesture as for a difficult proper noun.

This point joins our stance, set out on the French text to speech page: we would rather read one language well and give you the means to correct its edge cases than promise perfect coverage of thirty languages that nobody delivers. A rare Italian or German word is not a bug to hide, it is a known case, and real quality is measured by what the tool lets you do about it.

The practical rule

Write normally, with the original accents and characters when you have them to hand ("jalapeño" rather than "jalapeno" already helps the engine). Then, before narrating in bulk, reread the list of foreign words that come up often on your site, listen to them once, and fix in the lexicon the two or three that genuinely snag. That sorting takes a few minutes and it holds over time, because these words do not change. It is the same discipline as for proper nouns and place names: you do not correct everything, you correct what matters, and you correct it once and for all.

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