Dataset improvements

#4
by konstantindobler - opened

I want to raise two cases of low-resources languages where a significant portion of their samples are ambiguous:

  • Ancient Greek (grc) has only one sample which is a single word, "Μεσσίας", which is also a valid Modern Greek (ell) word. (line_id:
    503681-c699394-u9165-l2_000)
  • Nigerian Pidgin (pcm) has only two samples, one of which is just "Sunday, March 2, 2014". (line_id: 480106-c746162-u9053-l0_001)

I would propose to remove these samples from the benchmark.

Sign up or log in to comment