Repository navigation
Output language unpredictable #16
Description
When summarizing a Chinese .TXT file with time codes (previously extracted from an M4a file), the long distilled structured output is usually {in English}, but if I omit the time codes for exactly the same file, I get a much shorter text {in Chinese}. Is it possible to specify the language of the output? Thank you for a very helpful summarizer!
Activity
FaaLaoWang commentedon Aug 5, 2026
The issue is hard to reproduce: pure Chinese text seems to be summarized in Chinese. When the unsummarized Chinese text also includes a few acronyms and occasional English words, the text is summarized in English only. Weird!
Wintersta7e commentedon Aug 26, 2026
Fixed on master.
The prompts are all written in English and nothing ever told the model which language to answer in, so it inferred one from the prompt plus the document. That is why a mostly-Chinese file with a few English acronyms came back as a long English summary, while a pure-Chinese one stayed Chinese — your observation was exactly right.
Every request now states the language explicitly. By default it matches the document; you can also pin one, either with the new Summary language picker on the Config screen or with
-l/--languageon the CLI:Any language name is accepted, not only the ones in the picker.
Checked against the model on your case — a Chinese transcript with time codes and acronyms, detailed mode — which went from an entirely English summary to a Chinese one. Ships in v2.0.1. Thanks for the clear report, the "with acronyms vs without" detail is what pinned it down.