How we write, mark and check the sample essays
We write every sample essay and dialogue with AI. Then a second AI model, made by another company, checks each mark. You see each level as an estimate, with the reason for every mark.
- Written with AI, and labelled where they appear
- Every essay, mark and level checked by a model from another company
- Levels shown as estimates, never as an official mark
L’année prochaine, je pense que vous devez venir avec votre famille.
- First modelMarked it as a mistake, with the fix je vous invite à venir
- Second modelCorrect French. It only sounds like an order.
- ChangeMark removed. The better wording moved to the comment on the task.
How a sample essay is made
The same 4 steps for each of the 9 sample essays, at NCLC 5, 7 and 9.
AI writes the task and the essay
First a task of the same type as FEI’s, then one essay at each target level. Lower-level essays contain mistakes on purpose.
It marks each mistake and gives the reason
Every mistake gets a correction and a reason in one line. The comments sit under 4 headings built from FEI’s list of assessed skills: task, coherence, vocabulary and grammar.
A second model checks everything
A model from another company reads every essay, mark and level, and compares them with FEI’s published rules. On the 9 sample essays it raised 13 points. One changed the page; the other 12 weren’t right when checked against FEI’s pages and standard French. On the speaking dialogues it raised 21 points, and 2 of them changed the page.
A program counts the words
One rule for every essay: a word is a group of characters between spaces that holds a letter or a digit. The title, the greeting and the signature count. Every essay is inside FEI’s limits.
- 1Grammar
Mais aussi c’était difficileMais c’était aussi difficile“Aussi” goes after the verb.
- 2Task and register
Comme ça
Too casual for a newspaper.
Ainsi,
How well AI finds mistakes in French writing
We tested it on 12 real texts by learners of French. Native speakers who volunteer to correct learner texts had marked 115 mistakes in them, not counting notes on style.
A small test. 12 texts isn’t many. Read these numbers as a direction.
The three AI models found 107 to 114 of the 115 mistakes. The free grammar checker found 73, about two thirds.
The models also proposed some wrong fixes: 1 for the best model, 10 for the weakest. Catching wrong fixes like these is the job of the second check on our essays.
A mark counted as found when it covered the same words as the volunteer’s correction. In the strict count (“Strict” in the table), a mark couldn’t cover 3 or more mistakes or run more than 4 words past the mistake.
What a level estimate is, and what it isn’t
Each sample essay shows a level, such as “around NCLC 7”. That level is our estimate. FEI lists the skills it assesses, but it publishes no points per criterion and no marked essays.
So we give a range on the 0 to 20 writing scale, then the NCLC level that IRCC’s table gives for it. Essay B of each task is estimated at 10 or 11 out of 20, which is NCLC 7.
On the real test, FEI gives one writing score for all 3 tasks. We mark each essay as if the whole paper were written at that level.
No teacher or FEI examiner has checked our samples. So the estimate isn’t an official mark, and it can be wrong. Only FEI marks the TCF Canada: all papers are marked centrally at France Éducation international.
| Essay | Score out of 20 | Level |
|---|---|---|
| Essay A | 6 | NCLC 5 |
| Essay B | 10 or 11 | NCLC 7 |
| Essay C | 14 or 15 | NCLC 9 |
Essay A: 6. Essay B: 10 or 11. Essay C: 14 or 15. The real test gives one score for all 3 tasks.
Estimated level of essay B
NCLC 7
AI estimate, not an official FEI mark
Where the facts come from
No fact goes on a page without an official source.
FEI, for the TCF Canada
Task types, word limits, timings, the cases marked “A1 not reached” and the list of assessed skills.
CCI Paris Île-de-France, for the TEF Canada
The test format and how results are given.
IRCC, for NCLC levels
A script copies IRCC’s table into the site. 2,028 automated tests check the calculator against it.
Sources on every page
Each page lists its sources in a fold at the end, with a link to each official page.
| Level | Speaking | Listening | Reading | Writing |
|---|---|---|---|---|
| NCLC 9 | 371–392 | 298–315 | 248–262 | 371–392 |
| NCLC 8 | 349–370 | 280–297 | 233–247 | 349–370 |
| NCLC 7 | 310–348 | 249–279 | 207–232 | 310–348 |
| NCLC 6 | 271–309 | 217–248 | 181–206 | 271–309 |
Report a mistake
Rules and tables change. If you find something wrong on a page, write to us.
Write to us
Give the page, the words that are wrong and, if you have it, a link to the official source.
We check the official page
We compare your message with the page of FEI, CCI Paris Île-de-France or IRCC.
We correct the page
If our page is wrong, we change it. If a sample essay is wrong, we correct the essay and its marks.
Report a mistake by e-mail
Sources
- France Éducation international: TCF Canada (page in English)
- France Éducation international: how a TCF session runs (page in French)
- CCI Paris Ile-de-France: TEF Canada
- IRCC: Express Entry, language test results
- LangCorrect: texts by learners of French, corrected by native-speaker volunteers
- LanguageTool: free public grammar checker