一匹 is translated to "一Hiki"

[ScoreUnder]

As the title says.

I've been looking for a library to break kanji down into their readings (preferably hiragana), and my first test with them is to see how they fare with the 〜匹 counters.

For reference, this is the expected output for the first 10:

漢字	ひらがな	Ro-maji
一匹	いっぴき	Ippiki
二匹	にひき	Nihiki
三匹	さんびき	Sanbiki
四匹	よんひき	Yonhiki
五匹	ごひき	Gohiki
六匹	ろっぴき	Roppiki
七匹	ななひき	Nanahiki
八匹	はっぴき	Happiki
九匹	きゅうひき	Kyuuhiki
十匹	じゅっぴき	Juppiki

However, this is the output the program creates:

% ./jakaroma.sh '一匹　二匹　三匹　四匹　五匹　六匹　七匹　八匹　九匹　十匹　1匹　2匹　3匹　4匹　5匹　6匹　7匹　8匹　9匹　10匹'
一Hiki 　二Hiki 　三Hiki 　四Hiki 　五Hiki 　六Hiki 　七Hiki 　八Hiki 　九Hiki 　十Hiki 　1Hiki 　2Hiki 　3Hiki 　4Hiki 　5Hiki 　6Hiki 　7Hiki 　8Hiki 　9Hiki 　10Hiki 

[nicolas-raoul]

Thanks for the detailed feedback!
Do you know whether the same problem appears in Kuromoji?

• If yes, then the problem should be reported to Kuromoji, so that they can fix it.
• If no, we should figure out what we are doing wrong and fix our code.

On the other side, I can imagine cases where someone would prefer 一丁目 to be translated to 1chome rather than Icchome.

Another problem is that the program outputs kanjis (such as 五Hiki in your example), I am not sure why but that's a big problem indeed.

Cheers!

[nicolas-raoul]

Related: atilika/kuromoji#125

Apparently switching to UniDic (might be as simple as modifying pom.xml)would solve that particular case, but it might have lower performance in other areas.