A student repeats a word twenty times and it still sounds almost right. Nothing in the repetition tells them what to change, so the twenty-first attempt is a guess like the first.
What repetition leaves out
Imitation works when the learner can already hear the difference. Most cannot, because the ear filters unfamiliar sounds into the nearest familiar category. A French dental t lands in the English t box, and the box is where the problem stays.
Description does what repetition cannot. Naming the position of the tongue turns an invisible target into an instruction the student can follow at home.
French /t/ is dental: the tongue tip touches the back of the upper teeth. English /t/ is alveolar and aspirated: further back, with a puff of air. Hindi त, Kannada ತ and Bangla ত are already dental.
Diagnosis before drill
Every learner arrives with a different inventory of sounds, and the inventory decides the difficulty. A Kannada speaker and a Bangla speaker need different corrections for the same French word, so a single generic correction wastes both their time.
A student who understands why a sound is difficult can correct themselves after the class ends. That is the point.
A worked example: thé
Take thé, the French word for tea. First, find the nearest sound in the student's first language. Secondly, explain why the French sound is hard for that speaker. Thirdly, build the sound together, over repeated attempts — repetition with a target rather than repetition instead of one.
Why it shows up in the speaking exam
Pronunciation carries marks. An examiner assesses the speaking section of the DELF, the TEF and the TCF, and that examiner listens to how you sound, not only to what you say. A student whose vowels and consonants sit close to native French reads as more fluent than their vocabulary alone suggests.
What to practise this week
Pick five French words beginning with t. Say each one twice: once with the English t, once with the tongue against the teeth. Record both. The gap between them is the accent you are looking for.