Bon, ben, quoi, voilà: what do these little French words actually do?
These little words manage the dialogue. What three corpus studies measure, and which ones to adopt first.
By Dorian Savreux · Founder of Frenchee
They are not verbal tics. In a corpus of dialogues presented at the 2024 Congrès mondial de linguistique française, the agreement markers produced by the listener rise from 103 to 254 occurrences per thousand words between the first and the fourth round of the same game, until they are a quarter of everything that listener says.
If they are not empty words, what do they do?
Textbooks ignore « bon », « ben », « quoi » and « voilà », and learners treat them as filler. Linguists call them discourse markers: Anne Abeillé and Benoît Crabbé annotated a corpus of spoken French using a list of 115 markers.
Their job is not to add meaning but to manage the dialogue. « Ben », « donc » and « alors » announce what is coming; others signal that you are following or that you agree; others close an utterance, which is the everyday use of « quoi » and « voilà ». A learner who produces none is understood, but the other person never knows whether they may carry on. If speed is still the obstacle, fast spoken French covers the other half of the problem.
If you speak English you already own this machinery: "you know", "I mean" and "right" do much the same work. The trap is translating word for word, since "well" is not « bon » and "right?" is not « quoi ». What transfers is the habit of signalling, not the words.
Why does the listener talk so much?
Ludivine Crible recorded eighteen dialogues between thirty-six native speakers from France and Belgium, almost nine hours over Zoom with cameras off: the players describe and identify abstract shapes across four rounds, and only the listener is analysed. She counts thirty-eight types of alignment marker for 4,082 occurrences: « ouais » 1,016 times, « OK » 931, « voilà » 50.
The figure that matters is density. Per thousand words it is 103 in the first round, then 135, 196 and 254, each value recomputing from the counts in the same table. In the final round these markers make up 25 % of everything the listener produces, even though the round has shrunk from seventeen minutes to two.
As the players come to understand each other, there is less to describe and almost nothing left but confirming, and the functions shift with it: in round one, final agreement accounts for only 36.54 % of uses; by the last round it approaches 80 % and disagreement falls to 3 %.
What exactly does « bon » do?
Joy Peltier and Diana Ranson recorded 227 occurrences of « bon » in nine and a half hours of conversation with thirty-seven speakers. The arithmetic is ours: roughly one occurrence every two and a half minutes.
They separate textual functions, which organise the discourse, from attitudinal ones, which convey what the speaker thinks. The textual ones dominate, 66.5 % against 33.5 %, and « bon » most often marks continuation, 38.3 %, above all the move into a sub-topic, 16 %.
The practical lesson is that « bon » is almost never a judgement but a signal of articulation: it says "I'm picking up again" or "I'm moving to a detail". The authors add that their analysis of its place in the turn «n'a pas discerné de patron net entre ses fonctions et sa position», no clear pattern between function and position. There is no slot to memorise, only a usage to hear.
Why do learners use them less, and which ones should you adopt first?
Delin Deng studied « quoi » among thirty-seven speakers of Chinese origin living in France, across twenty-seven hours of interviews from 2014 and 2015. She found only 272 occurrences of « quoi », of which 165 as a discourse marker, or one every 9 minutes 48 seconds.
The benchmark comes from an older study she cites, not a measurement of her own: among natives « quoi » would recur about once every two minutes, roughly five times as often. Nobody will correct a learner who uses it too little: it is not a grammatical error but a shortfall in naturalness, and it is audible.
So start with the reception signals, which carry no risk: « ouais », « d'accord », « c'est bon » and « voilà » stand alone as a reply, and are the most frequent in Crible's corpus. They modify no sentence, so cannot be badly built. Next come the articulation markers: « bon » and « alors » go at the front of what you are about to say, « quoi » attaches to the end.
Three habits are enough. Pick two markers, not ten, and use them until they come without thinking. Listen to where natives put them rather than learning a rule of position, because there is none. And accept that these words belong to ordinary spoken French, the variety that drops « que », that repeats the subject and where the « ne » disappears. Our level test will tell you where you stand.
What these figures do not tell you
They do not tell you that an ordinary conversation resembles these corpora. Crible's is a task game between strangers, over Zoom and without video, in which reaching agreement is the whole point: its densities measure that setup, not a dinner with friends. The coding was done by one person, the author.
The 227 instances of « bon » in Peltier and Ranson come from nine and a half hours kept out of twenty-five recorded at Montpellier and Rognes in 2005 and 2006, classified separately by the two authors then settled together, with no reported agreement rate. Deng's 165 « quoi » are a count she judges too small: «il est impossible à ce stade de donner des interprétations conclusives», no conclusive interpretation is possible at this stage. Her subgroups are lopsided, 147 occurrences among fifteen men against 18 among twenty-two women, and she offers two explanations without choosing.
Three cautions. The percentages concern the occurrences recorded, not speakers: they say how these words are distributed when they appear. None of the studies measures learners other than Chinese speakers. And « ben », despite the title, appears in none of the counts: it is described as a structuring marker, not measured. We also set aside two figures: Crible's article states "more than 35000 words" where its own table totals 32,416, so we keep only the frequencies, computed on that table; and nothing on the position of « bon », since we could not read that table in full.
For an exam, let us be blunt: in the DELF, the DALF or the TCF these markers add nothing and may harm the speaking paper, which expects a formal register. They serve conversation, not the diploma.
A list takes two minutes to read, but knowing whether your « voilà » landed at the right moment takes someone who is listening and reacting. At Frenchee, classes are live with qualified native teachers, one to one or in a small group of three to five students, with no subscription: that is the setting where these signals are acquired, since they only exist between two people.
