the text “Hello China" is detected to 'it'

[gaowenxin95]

when l detect ”Hello China"
print(langid.classify(”Hello China"))
the result :
('it', -37.309250354766846)
@Paczesiowa @pquentin @martinth @jnothman @saffsd

[pquentin]

This can happen on short texts, try a longer one

[gaowenxin95]

This can happen on short texts, try a longer one

thanks

l try a sentence contain five words still detect wrong
like this:"hello China you are great"

('it', -31.29085063934326)

when contain six word like this "hello China you are my sunshine"
its right

('en', -49.038776874542236)

another like this "hello China hello China hello China "
its wrong

('it', -27.979979038238525)

l would like to know how many words should l try at least in the sentence?
@pquentin @martinth @jnothman

[KoenVanDuin]

I am dealing with the same issue. In my case, inputting larger pieces of text is no problem, but I want to know what increase of text volume increases the reliability in which extent.
Moreover, does it have to be a real text, or is a bunch of words from the language also fine?
Lastly, I wonder what the returned negative coefficient says about the reliability of the translation. I couldn't find information about what this number actually means.

Many thanks in advance.

[ffreemt]

Try my fastlid: pip install fastlid

Fast and accurate, dependent on fasttext though (Windows systems without a C compiler can use fasttext*,whl available at https://www.lfd.uci.edu/~gohlke/pythonlibs/) .

fastlid also tries to imitate two of langid's functionalities.

[yuviabhi]

Having the same issue.
The text Our fifth module explains some key calculus skills is detected as 'no' though it have 8 words.
In another example, the text (with 4 words) Discover some angle relationships is detectesd as 'sw' but when I changed the text to Discover some angle relationships between them (with 6 words) then it is detected as 'en' as expected..
So what is the minumum word we need to detect?

[everdrone]

+1