Hey there, I noticed that many functions have recently been moved from discourse-translator to discourse-core. Could you share some context on the long-term plan behind this change?
2 likes
The Discourse AI provider on the translator plugin has been moved to Discourse AI itself. This simplifies configuration by a lot, and allows us to tap on Discourse AIās LLM usage / persona functionalities.
As for the functions moved to Discourse core, they were all experimental features that had been in heavy development from the beginning of this year. Our team has decided to conclude this experiment. It was a successful one and we are seeing many requests for first-class multilingual support.
Sometime this month, we should be sharing an announcement on this feature which works best with Discourse AI.
Main features in Discourse Translator will still be in the plugin (see OP).
10 likes
Please, take in consideration adding DeepL to Discourse Translator and/or Discourse AI (DeepL is pioneer on AI).
Their translations are so good and makes a real difference:
Love your cat, nat! 
[EDIT] Just sharing more data here
Which machine translation provider do the worldās leading language services companies (LSCs) trust to deliver accuracy and efficiency?
The answer, in a new survey of 127 language companies across 28 countries, is clear and resounding: DeepL.
Blind tests also found that DeepLās translations require fewer edits than competitors, with Google Translate needing 2x more edits and ChatGPT-4 needing 3x more edits to achieve the same quality.
In fact, in the 2024 study conducted by Forrester Consulting, āThe Total Economic Impactā¢ of DeepLā, it was discovered that DeepL reduced translation time by 90% and cut translation workloads by 50%āresulting in a 345% ROI.
Its quality is superior because DeepL is created by the people who made Linguee, a dictionary that uses human translations.
8 likes
There are updates to how translations work on Discourse. For more information, please check out the announcement at Content Localization and Automatic Translations for Your Community.
This translator plugin is unrelated to the announcement and will still be maintained, but it would be unlikely that we add new features to this plugin.
5 likes
When it tries to translate from other language to Chinese (zh_CN) with libretranslate, there is a mistake: The translator is unable to translate this postās content (en) to your language (zh_CN), which i guess is because the Chinese language code in libretranslate is zh_Hant which is not matched by zh_CN. How can I fix this problem?
The description of this plugin is a little thin.
Iām wondering about a situation where a forum was primarily English. Could this plugin be set up so that the only translation happening was non-English (of supported languages) into English?
Iād suggest changing horses and looking at the new way translation is done in Discourse, vs this older plugin which will not be getting any new features. See Natās reply two posts up. You can see how well it works here on meta, and by spinning up a hosting trial.
5 likes
Thank you for your response. Does this imply that the new iteration can supersede the existing one, offering greater robustness? Furthermore, are users presented with the option to choose between the two, with the new plugin being the more strongly recommended choice?
1 like
Yes thatās right.
However it is best to choose just one or the other, not both. We should avoid having to let our community users choose and stick to just one user experience - preferably the new automatic translations since the other is deprecating.
3 likes
thanks for your quick reply, the newer one is more friendly and smoothly, and your reply gives clearer info for selection.
1 like
Hello! I expect the costs of running an LLM (for use with Discourse AI) are greater than those of using a specialised translation API (or hosting LibreTranslate).
Do we have any assurance that, even without new features, this plugin will still be maintained for bugs/security? Or should we move to the LLM way? We really donāt want to use Big Tech models, and hosting LLMs is not super cheap 
4 likes
You should switch to Discourse AI for automatic translation when youāre able to, weāll still fix bugs and security issues with this plugin but itās no longer the preferred method.
Automatic translation using AI is the more robust feature now, as it automatically provides translations (rather than on-demand) and includes categories, tags, and can provide translated content to search engines.
There are some models available that will do a fine job of translating that will likely cost less than hosting LibreTranslate indefinitely on a VPS. If you get 10k posts a year it can cost less than $10. Others have had success with Deepseek and Mistral (and other models via OpenRouter) if you want to avoid the larger corporate providers.
3 likes
You could also self-host your own model (Gemini or via Ollama) if youād want to bring down costs on a server with good specs and no need for external APIs.
2 likes