detect_multiple_languages_of() does not work at all for mixed English, Chinese and Japanese

[xuancong84]

You can easily test out sentences like the following:

The name of that celebrity is 王菲 everything will be classified as English (you can try any Chinese name or any English prefix sentence, it will always get it wrong).

这首歌的名字叫 ロミオとシンデレラ The name of the song is ロミオとシンデレラ everything will be classified as Japanese. FYI, Japanese character set does include almost all traditional Chinese character set, however, a person or a language model can easily tell the boundary between the two chunks, especially if the first chunk is in English.

[ryanheise]

I've found that adding some combination of superfluous languages into the language detector can randomly get it to work. In my case, adding French and German helped it to better detect Japanese and English (but not when I tested on your examples.)

In your first example, the Chinese name might be too short, but in theory I think this ought to be a solvable problem because the character sets of English and Chinese are completely different. That alone ought to give high confidence even on the shortest substrings.

[juntaosun]

可以尝试一下 LangSegment ：

https://huggingface.co/spaces/sunnyboxs/LangSegment

[ryanheise]

@juntaosun That looks interesting. Have you also considered implementing your algorithm on top of lingua?

By the way, you might want to update your example to use "." at the end of the sentence rather than "。" since that is what's used in practice for horizontal text (although your algorithm works fine with both).