argosopentech / argos-translate

Open-source offline translation library written in Python

Home Page:https://www.argosopentech.com

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Several models have a repetition issue with single proper nouns to English

rrgeorge opened this issue · comments

Most noticeable with pl, but the issue does seem to be apparent with he, ga, it, fa, sv, and uk as well.
Repeatable example:

for l in he ga it fa pl sv uk
do
  echo $l
  argos-translate --from-lang $l --to-lang "en" "Apple"
done;

This was discovered when translating html on Mastodon where hashtags get encapsulated in their own link elements.

The Polish model is in need of updating; this happens when the training data did not include many short sentences (or single words, like "Apple").

commented

This is a video of the issue happened to me

libretranslate_-_mastodon_4.2.0.mov