I'm trying to use Japanese language tokenizers like mecab, kakasi, kuromoji (the ipadic variant) to help me read Japanese texts.
When they segment sentences that contain an isolated 風 like in the examples below, aforementioned tools all suggest the かぜ (wind) reading, although I - an advanced Japanese learner - would naturally think フウ (manner).
Examples:
- そんな風に色々考えて見ると、...
- ざっとこう云った風なのだ。
- そこで私はどういう風に切り出したものかと迷いながら始めた。
- 警察ではあれをどんな風に解釈したか知りませんが、...
- 僕はこういう風に考えるのですよ。
The native speaker I consulted agrees that the reading should be フウ in all these cases, although oftentimes, one could contrive a context where the かぜ reading would be the correct one, but the likelihood of such a context would be rather low.
So my question: Why do tokenizers do that? Am I and the native speaker biased in a different way than the tokenizers? Who is "right"? By "right" I mean choosing the most likely reading.
N.B. You can tell mecab to give you more than candidate, and when you do so, the フウ reading will show up second or third candidate, depending on the sentence.
風
asカゼ
is listed in the IPA dictionary (for example), in theNoun*.csv
files, and asフウ
only in theSuffix.csv
file. It seems the solver favours picks the Noun category.