Proper treatment of PUNCTs for KNP

[KoichiYasuoka]

括弧始-PUNCTs, such as ”（" "「" "『" and so on, are not suitable for head tokens in Universal Dependencies.

[KoichiYasuoka]

Oops, "build (juman)" has failed. How do I do?

[KoichiYasuoka]

I'm vague why JUMAN treats "（※" as single midashi with two tokens...

[KoichiYasuoka]

OK, now I understand that "（※" is treated as two tokens by JUMAN, and as single 顔文字 by KNP.

$ echo "（※" | juman
（ （ （ 特殊 1 括弧始 3 * 0 * 0 NIL
※ ※ ※ 特殊 1 記号 5 * 0 * 0 NIL
EOS
$ echo "（※" | juman | knp -tab
# S-ID:1 KNP:4.19-CF1.1 DATE:2020/04/22 SCORE:-0.30231
* -1D <文頭><文末><体言><用言:判><体言止><レベル:C><区切:5-5><ID:（文末）><裸名詞><提題受:30><主節><状態述語>
+ -1D <文頭><文末><体言><用言:判><体言止><レベル:C><区切:5-5><ID:（文末）><裸名詞><提題受:30><主節><状態述語><判定詞><名詞項候補><先行詞候補><用言代表表記:（※/（※><時制-無時制><格解析結果:（※/（※:判0:ガ/U/-/-/-/->
（※ （※ （※ 特殊 1 記号 5 * 0 * 0 NIL <形態素連結-顔文字><顔文字:自動認識><文頭><文末><記英数カ><英記号><記号><自立><内容語><タグ単位始><文節始><文節主辞>
EOS

[tamuhey]

Thank you for your contribution!
Do you have any examples that this PR enables you to handle correctly?
If any, please add them as test cases.

[KoichiYasuoka]

I think test_knp.py already includes very good example "（※" for this PR. It was very hard task for me to pass the test "（※" with 全角, and I got it.

[tamuhey]

I think _modify_head_punct is not tested enough, so please add test cases into test_dependency_parser.py

[tamuhey]

Thank you very much, @KoichiYasuoka!

[tamuhey]

OK

[tamuhey]

@KoichiYasuoka I've publishe new version of camphr, so please check it out.