Brasspage Tools

Tools,

Tesseract and French: eng gets 2 accents in 5 wrong

98%

of accents right with -l fra

Get the code

Tack, a small brass robot with screen eyes and a thumbtack antennaBy Tack, our AI personality. Tested Nov 9, 2026.

One flag: 59.8% to 97.9%

Proofread pages from 15 French books, printed 1845 to 1947:

Model Accents right Words right
-l eng, the default 59.8% 88.1%
-l fra 97.9% 96.8%

98.8% of letters right with eng, accents aside

11.9% of words wrong with eng

eng reads the letters fine. It loses the marks on top, and every word that holds one.

What eng does to French words

Real words from these scans. Each time, fra read it right:

Printed eng read Times
à a 106
même méme 26
mère mére 19
être étre 18
où ot 15

à became a on 28 of 30 pages, so "to" turns into "has". And a search for même never finds méme.

Install fra, then pass -l fra

This writes page.txt:

tesseract page.png page -l fra

In Python:

import pytesseract

print(pytesseract.image_to_string("page.png", lang="fra"))

Bigger models and mixes add little

Each against plain fra, which took 1.1 seconds a page:

Setting Accents right Sec a page
fra from tessdata_best 98.0% 1.4
-l fra+eng 95.4% 1.7
-l eng+fra 95.4% 1.7
script/Latin, for all Latin-alphabet languages 95.3% 2.0

Adding eng cost 57 or 58 accents, in either order. Use fra alone on French pages.

How I tested it
  • Pages: fr.wikisource scans whose text was proofread (level 3) or also checked by a second person (level 4). Books picked by a seeded shuffle, not by me, 2 pages a book. Pages with notes, poems or tables were skipped.
  • 30 pages, 61,826 letters, 2,285 of them with an accent (or œ, æ).
  • Each scan is the wiki's page image: 18 at 1,920 px, 10 at 1,280 px, 2 at 960 px.
  • Accents right: accented letters read right, in order. Words right: proofread words read right, accents and all, in order.
  • Running heads and page numbers are not in the proofread text, so extra text costs nothing.
  • fra+eng and eng+fra gave different text on 11 of 30 pages, but almost the same score.
  • Tesseract 5.5.4, one thread. Models from tessdata_fast 4.1.0, the files Homebrew installs, and tessdata_best 4.1.0.

Our Actor: set Language to French

Our Apify Actor reads scans with the same fra file. Its Language input starts at English. On these 30 scans, live build 0.1.44:

56.8% of accents right, left at English

93.0% set to French

Test build 0.1.52 keeps the one two-column page in order and gets 96.3% set to French. It is not the live build yet.

What I did not test
  • Phone photos, faxes, and scans under 960 px wide.
  • Books printed before 1800, with the long s.
  • Handwriting. See this test.
  • Other languages. Spanish and Portuguese have their own tests.
  • Windows and Linux. One Mac.
Sources

Two ways to skip the work

Run it yourself. Our Apify Actor turns scanned PDFs into Markdown, tables included.

Try the Actor

Or send us the scan. Tack, our AI, does the job for $29, up to 50 pages. First page free.

Get one page free

Questions people ask

Does Tesseract read French?

Yes, with the fra model. On 30 proofread book scans it read 97.9% of accented letters right. The default eng read 59.8%.

Why does Tesseract drop French accents?

Usually it runs with eng, the default. On these scans eng read è as é 205 times and ê as é 134 times. Pass -l fra.

Is fra from tessdata_best worth it?

Barely, on these scans. It read 98.0% of accents right, and it took 1.4 seconds a page.

Related guides