The middle dot out loud: what text-to-speech does with French inclusive writing
« étudiant·e·s » reads well with the eyes and badly with a voice. Here is why the French middle dot has no spoken form, what an engine actually does with it, and the only way to get inclusive French that plays without a stumble.
French inclusive writing was designed for the eye. The middle dot in « étudiant·e·s » (students, both genders folded into one word) is a visual shorthand: the reader takes in both forms at a glance and reassembles them in their head. The trouble starts the moment a voice has to read that string of signs aloud, because there is nothing left to reassemble at a glance: you have to pronounce, in order, something that was never meant to be pronounced. It is a textbook case of what happens between the written text and what gets said, which is exactly the ground French text-to-speech works on.
Why the middle dot has no reading
Take « chercheur·euse·s » (researchers, both genders). A person reading aloud does not say "researcher dot euse dot s." They make a silent choice: either they say both full forms, or they read only one. But that choice is written nowhere in the text. The middle dot says "both genders are here," it does not say "here is how to pronounce it." There is no shared spoken convention for reading it, because it was never conceived for speech.
A machine is therefore left in front of a sign with no established sound value. It cannot guess the intent, because the intent is not in the characters. That is a fundamental difference from the other difficulties of read French: a liaison follows a rule, a number has an expected reading, an acronym is spelled out or pronounced according to known cases. The middle dot, by contrast, has no right answer to find, because there is no answer in the text.
What an engine does with it, in plain terms
Faced with « étudiant·e·s », every possible behaviour is imperfect, and it is better to say so than to promise a magic that does not exist.
An engine can ignore the middle dot and read the letters straight through, producing a word that does not exist, something like a full form with an orphan "s" stuck on, or a mush depending on the segments. It can instead pronounce the sign, and then the listener hears "student dot e dot s," which is worse. Or it can treat the dot as a micro-break and chop the word into pieces. None of those three results is what the author of the text had in mind, and that is normal: what they had in mind was not written down.
Our position is therefore simple, and it follows from our stance on quality: we do not claim an engine guesses what the middle dot means. A sign with no spoken form has no right reading to produce, and pretending otherwise would be selling randomness as comprehension.
The form that plays cleanly already exists
The good news is that inclusive French does have a form that reads perfectly aloud, and it is the full doublet. « Les étudiantes et les étudiants » (the female students and the male students) is not a special case for a synthetic voice: it is ordinary French, with its agreements, its liaisons and its punctuation, which the engine handles like any other sentence. Where « étudiant·e·s » has no sound, « étudiantes et étudiants » has one, obvious, and the same one a human would choose.
In other words, the question is not "can the tool read inclusive writing," it is "which inclusive writing can be heard." The doublet can be heard. The middle dot cannot, because it was invented to keep the page light, a goal that makes no sense out loud, where there is no page. Run the test on your own text: the same sentence, once with middle dots, once with doublets, and listen to the difference. It is a one-minute protocol, more telling than any claim, and it fits the logic of our piece on testing a voice in ten minutes: what matters is checked on your hard sentences.
And if the middle dot is already everywhere in your content
Many nonprofits, local governments and universities have adopted the middle dot across their content, and stripping it out everywhere overnight is not realistic. Two safeguards exist.
The first is the pronunciation lexicon: for a form that recurs often, « adhérent·e·s », « citoyen·ne·s », you declare the expected pronunciation once, spelled out, and the engine applies it everywhere afterwards. It is not a guess, it is a decision you make and the tool honours. The second is the clean-up of what goes to synthesis: the same care that strips bare URLs and markup from a text before reading can normalise known inclusive forms. As with English loanwords, the principle holds: what is predictable is fixed cleanly once, rather than hoping the voice sorts it out on its own.
The rule to keep fits on one line. For a text meant to be seen, the middle dot is an editorial choice that is yours to make. For a text meant to be heard, write what you want to be heard.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo