4 ms·

Mandarin seems to be more phonemic for reading than English. Was this trained on pinyin input?

• int_19h 3y ago
  "For Chinese, we only selected Mandarin words in simplified Chinese and limited to one or two symbols (a.k.a. Hanzis); we then obtained their pronunciation from the CEDICT dataset. ... It is debatable whether Chinese should be included in this study given the term alphabet is usually reserved for largely phonographic systems that have a small number of elements. We decided to include it because our ANN model allowed for alphabets with thousands of graphemes." So it's not really about "phonemic" in the traditional sense where graphemes directly correspond to phonemes. The more accurate description of the generalized metric is "how unambiguously graphemes map to sounds", which is still meaningful even if the mapping is very far from 1:1 as is the case for Chinese. Also, if I understand correctly given what CEDICT is, they basically used pinyin as phonemic transcription (instead of IPA, like for other languages) - i.e. as output. But then again, actual alphabets do this kind of thing too. E.g. Belarusian orthography is almost completely unambiguous (Taraskevica might actually be completely unambiguous even?), but you have quirks like e.g. "я" meaning either /ja/ or /ʲa/ depending on the preceding character, whereas /a/ in other contexts uses a completely different letter "а"; I believe this would score as high as Serbo-Croatian by their metric, although strictly speaking it's less phonemic, because the mapping is not 1:1. So if Chinese characters generally have consistent pronunciations, it ought to score relatively high for reading.
  • otabdeveloper4 3y ago
    Cool. I always suspected that hanzi is actually more phonemic than English, but nice to see this confirmed.