MinT injects a letter not present in destination alphabet (en->fa)
Open, LowPublicBUG REPORT
Actions

Assigned To

None

Authored By

	Alfa80
	Jul 29 2025, 5:26 AM

Description

Steps to replicate the issue (include links if applicable):

Launch translation for Policy:Wikimedia Foundation API Usage Guidelines from English to Persian (en->fa).

What happens?:
The suggested text by Mint for Wikimedia (fa:ویکی‌مدیا) is produced as (ویکیمیڈیا) in the suggestion. The latter has the following issues:

The letter ڈ is not present in Persian/Farsi alphabet.

What should have happened instead?:

The correct translation would be ویکی‌مدیا which has a ZWNJ at fourth index, removed 6th index and converted 7th index from ڈ to د (See screenshot)

Software version (on Special:Version page; skip for WMF-hosted wikis like Wikipedia):

Other information (browser name/version, screenshots, etc.):

Event Timeline

Alfa80 created this task.Jul 29 2025, 5:26 AM

Restricted Application added subscribers: Huji, Aklapper. · View Herald TranscriptJul 29 2025, 5:26 AM

MinT uses nllb200-600M for fa (more specifically pes_Arab (see https://gerrit.wikimedia.org/g/mediawiki/services/machinetranslation/+/906100d449d5c4662c8efe1ee5c386e862be8bea/translator/models/languages.py#55).

If this is a problem with the model, there isn't much we can do it. Based on your report it doesn't seem this could be fixed with simple post-processing.

MinT injects a letter not present in destination alphabet (en->fa)Open, LowPublicBUG REPORTActions

Description

Event Timeline

MinT injects a letter not present in destination alphabet (en->fa)
Open, LowPublicBUG REPORT
Actions