As one of the lead researchers at the Web Ecology Project- I can confirm that this was also true in 2009 (our finding of a small sample of 1M tweets found it wa

As one of the lead researchers at the Web Ecology Project- I can confirm that this was also true in 2009 (our finding of a small sample of 1M tweets found it was ~61% at the time). It doesn't shock me at all that it could be down to as low as 50% at this point. Jon Beilin of our group wrote and open sourced (MIT/X11-license) a Python language detection module that we used on our Twitter database. http://www.webecologyproject.org/2009/09/code-release-google-language-tool/ http://www.webecologyproject.org/2009/09/code-release-google... We did not find Japanese to be as high up there as this group did. I can't remember, but I think when we ran ours the Twitter.jp was still running as a separate domain perhaps?