[GH-ISSUE #5235] [BUG]: embedding / retriever does not work with other language than english #4981

Closed
opened 2026-06-05 14:51:16 -04:00 by yindo · 1 comment
Owner

Originally created by @sancelot on GitHub (Mar 19, 2026).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5235

How are you running AnythingLLM?

Local development

What happened?

Hi,

I uploaded an english document in the workspace. When I ask questions in french language regarding the document I don't get any accurate results.

I am using native multilingual-e5-small.

Are there known steps to reproduce?

upload a document in english that describe some command :
command AA

command BB

....

ask in french "decris moi la commande AA" => it does not find it (note:I used a bigger document than this....3500 lines.)

Originally created by @sancelot on GitHub (Mar 19, 2026). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5235 ### How are you running AnythingLLM? Local development ### What happened? Hi, I uploaded an english document in the workspace. When I ask questions in french language regarding the document I don't get any accurate results. I am using native multilingual-e5-small. ### Are there known steps to reproduce? upload a document in english that describe some command : command AA command BB .... ask in french "decris moi la commande AA" => it does not find it (note:I used a bigger document than this....3500 lines.)
yindo added the possible bug label 2026-06-05 14:51:16 -04:00
yindo closed this issue 2026-06-05 14:51:16 -04:00
Author
Owner

@timothycarambat commented on GitHub (Mar 19, 2026):

This would make sense, the model is multi-lingual but that semantic match still matters and cross-language still can have tradeoffs. This is just the nature of embedding and semantic search.

That being said, you can improve results (possibly marginally with multi-lang) with ReRanking
https://docs.anythingllm.com/chatting-with-documents/introduction#vector-database-settings--search-preference-reranking

<!-- gh-comment-id:4092121560 --> @timothycarambat commented on GitHub (Mar 19, 2026): This would make sense, the model is multi-lingual but that semantic match still matters and cross-language still can have tradeoffs. This is just the nature of embedding and semantic search. That being said, you can improve results (possibly marginally with multi-lang) with ReRanking https://docs.anythingllm.com/chatting-with-documents/introduction#vector-database-settings--search-preference-reranking
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#4981