Footnotes and note markers: what French text to speech does with them

Scholarly writing, legal commentary, long-form journalism all lean on the note: a small superscript number after a word, and the detail waits at the foot of the page or the end of the article. On the page the device is clean. The eye sees the number, knows it points elsewhere, and chooses whether to go down or read on. None of that exists for the ear. A text to speech engine does not see a superscript, it sees a character, and it has to decide what to do with it. This article looks at what actually happens, on examples you can reproduce, and what you should write to be read well.

The note marker, a digit that was never meant to be spoken

Here is the most common case, and the most jarring. A note marker is often a superscript digit stuck to the word it follows:

> Le traité de Rome1 fixe les bases du marché commun.

On screen the "1" is small, raised, and nobody mistakes it for a number. But when the page is turned into text for reading, the superscript loses its height: all that remains is a plain "1" pressed against "Rome." The engine reads what it is handed. So it says "le traité de Rome un fixe les bases," and the listener hears a number that makes no sense in the sentence. The same thing hits stacked markers: "la Commission2,3" comes out as "la Commission deux trois." This is not a pronunciation error by the engine, which reads the digit it is given correctly; it is a digit that should never have reached it. On how an engine reads an isolated number, our piece on reading numbers aloud in French explains why "1" becomes "un" rather than "premier."

What becomes of the note itself

The second problem is quieter. The content of the note does exist in the text, usually gathered at the end of the article. So when the engine reaches the bottom, it strings together, with no transition, a run of references cut off from their context: "1. Journal officiel, 25 mars 1957. 2. Voir chapitre précédent." The listener, who was following a continuous argument, drops without warning into a dry list whose numbers no longer connect to anything they can remember. On the page you glance back in an instant; listening, you do not glance back, you endure. A footnote rests entirely on the reader's ability to make an instant round trip between the body and the margin. Listening is linear: it makes no round trip.

The asterisk and other markers

Not every note marker is a digit. The asterisk is common, especially for a single note. An engine does one of two things with it: either it ignores it, which is the behaviour you want, or it names it, and the listener hears "le tarif de base astérisque est indiqué hors taxes." The second outcome is rare on a lone asterisk, but it grows likely when the symbol repeats (two or three asterisks to separate note levels) or mixes with other signs. The dagger, the bracketed marker such as "[note]," the bare hyphen all share the fate: they are silent typographic conventions on the page, but an engine has no reason to guess that an opening bracket announces a note rather than something else. They are, at bottom, typographic noise, that class of characters that carry meaning to the eye and none to the ear, which we covered more broadly in what should not be read aloud.

Writing notes that listen well

The good news is that the fix is editorial, not technical, and it does not mean giving up notes in the spoken version. The principle: a piece of information the listener needs in order to follow the sentence belongs in the sentence, not in the margin. If the note pins down a date, a figure, a decisive source, fold it into the running text, in parentheses or a short aside. If the note is only a bibliographic reference or an internal cross-reference, it adds nothing to the listening and is better dropped from the audio version, without the written version changing a word. This is what writing for two channels means: the eye and the ear do not ask for the same thing, and a good article is willing to answer them differently.

That leaves the case where the note contains a term the engine mishandles, a foreign proper name, an initialism, a Latin reference. Here again you do not touch the visible text, which must stay exact. You go through the pronunciation lexicon: you declare the intended reading of the term once, and it applies everywhere the term appears, notes included. The text stays what the reader sees, the correction lives beside it.

What to take away

The note marker is a classic listening trap because it is invisible to the person writing: the superscript "1" does not look like a problem until you hear it pronounced "un" in the middle of a sentence. The remedy comes down to one question, worth asking the moment you add a note: does the listener need this to follow along? If yes, it joins the thread; if no, it stays on the page. Checking that kind of detail, sentence by sentence, is exactly what separates a text that reads from a text that listens, and it is the work we do on reading French aloud, set out on the French text to speech page.

Read next

• An audio player in the page or in an iframe: what that choice decides
• Image captions and photo credits: what French text to speech does with them
• WCAG 2.2 and an audio version: what the standard requires, and what it does not