#argostranslate — Public Fediverse posts

For offline text natural language translation, argostranslate is a super easy to use python tool:

https://github.com/argosopentech/argos-translate

#foss #argos #argostranslate #python #offlineai

@jerry This seems like a good thread for my security concern.

I noticed a “translate” link on my post. Clicked it to see what would be used. Was disturbed to find that it’s #DeepL. This is a Tor-hostile #Cloudflare site that misleads people into thinking security is taken seriously. Which is the same thing that happends with libretranslate.com.

In principle it’d be most secure to run Argos Translate on the server. But that’s also a heavy process. If translation must be outsourced, I suggest translate.fedilab.app. It’s the same engine as #libretranslate, but without sharing translations with CF.

(edit) guess it’s a good time to republish this:

❌ #GoogleTranslate (#PRISM)
❌ #YandexTranslate [tax(profits) →Putin’s war]
❌ #DeepL (shares sensitive txt w/tech giant; misleads ppl about it)
❌ #LibreTranslate.com (shares sensitive txt w/tech giant; no privacy policy; solicits trust w/buzz phrases “libre”, “free & open”)
🤷 #lingva.ml (FOSS frontend; still Google)
🤷 translate.fedilab.app (no privacy policy but no evidence of mishandling data either)

Running locally:
✅ #ArgosTranslate
✅ #Apertium.org
✅ #translatelocally.com (EU funded)

@publicvoit If the doc is in my mother tongue I simply OCR & store as #DjVu with a scan of the postmarked envelope as the last page. But what if it comes in another language? That’s not covered in your process. I sometimes feed the OCRd txt into a LaTeX doc, format it similar to the original but add a 2nd column for my 1st language & use #ArgosTranslate to fill it in.

@publicvoit If the doc is in my mother tongue I simply OCR & store as #DjVu with a scan of the postmarked envelope as the last page. But what if it comes in another language? That’s not covered in your process. I sometimes feed the OCRd txt into a LaTeX doc, format it similar to the original but add a 2nd column for my 1st language & use #ArgosTranslate to fill it in.

@publicvoit If the doc is in my mother tongue I simply OCR & store as #DjVu with a scan of the postmarked envelope as the last page. But what if it comes in another language? That’s not covered in your process. I sometimes feed the OCRd txt into a LaTeX doc, format it similar to the original but add a 2nd column for my 1st language & use #ArgosTranslate to fill it in.

@publicvoit If the doc is in my mother tongue I simply OCR & store as #DjVu with a scan of the postmarked envelope as the last page. But what if it comes in another language? That’s not covered in your process. I sometimes feed the OCRd txt into a LaTeX doc, format it similar to the original but add a 2nd column for my 1st language & use #ArgosTranslate to fill it in.

I have been using Firefox Translations (aka #bergamot) for some time. It enables client side (local, no web server) in-page translations. It is quite useful and the results are good. What is the experience among #fediverse users? #Argos translate (local) and #LibreTranslate (web) are subjectively better IMHO (not tested) but not well integrated in the browser.

https://github.com/mozilla/firefox-translations/

https://github.com/argosopentech

https://libretranslate.com/

#mozilla #firefox #argostranslate #nlp #machinetranslation

@mousebot yeah iirc #apertium’s language support is sparse and obscure. #ArgosTranslate uses neural networks and Apertium does not, FWIW.

Update:
❌ #GoogleTranslate (#PRISM)
❌ #YandexTranslate [tax(profits) →Putin’s war]
❌ #DeepL (shares sensitive txt w/tech giant; misleads ppl about it)
❌ #LibreTranslate.com (shares sensitive txt w/tech giant; no privacy policy; solicits trust w/buzz phrases “libre”, “free & open”)
🤷 #lingva.ml (FOSS frontend; still Google)
🤷 translate.fedilab.app (no privacy policy but no evidence of mishandling data either)

Running locally:
✅ #ArgosTranslate
✅ #Apertium.org
✅ #translatelocally.com (EU funded)

Trustworthy translation tools:

❌ #GoogleTranslate (#PRISM corp)
❌ #YandexTranslate (tech giant)
❌ #DeepL (shares sensitive queries w/tech giant & goes to great lengths to mislead users about it)
❌ #LibreTranslate.com (shares sensitive text w/tech giant; no privacy policy; solicits trust using buzz phrases like “libre”, “free & open”)
🤷 translate.fedilab.app (no privacy policy but no evidence of mishandling data either)
✅ #ArgosTranslate running locally on your own PC inside #Firejail