SuPar-Kanbun
Tokenizer, POS-Tagger and Dependency-Parser for Classical Chinese Texts (漢文/文言文) with spaCy, Transformers and SuPar.
Basic usage
>>> import suparkanbun
>>> nlp=suparkanbun.load()
>>> doc=nlp("不入虎穴不得虎子")
>>> print(type(doc))
<class 'spacy.tokens.doc.Doc'>
>>> print(suparkanbun.to_conllu(doc))
# text = 不入虎穴不得虎子
1 不 不 ADV v,副詞,否定,無界 Polarity=Neg 2 advmod _ Gloss=not|SpaceAfter=No
2 入 入 VERB v,動詞,行為,移動 _ 0 root _ Gloss=enter|SpaceAfter=No
3 虎 虎 NOUN n,名詞,主体,動物 _ 4 nmod _ Gloss=tiger|SpaceAfter=No
4 穴 穴 NOUN n,名詞,固定物,地形 Case=Loc 2 obj _ Gloss=cave|SpaceAfter=No
5 不 不 ADV v,副詞,否定,無界 Polarity=Neg 6 advmod _ Gloss=not|SpaceAfter=No
6 得 得 VERB v,動詞,行為,得失 _ 2 parataxis _ Gloss=get|SpaceAfter=No
7 虎 虎 NOUN n,名詞,主体,動物 _ 8 nmod _ Gloss=tiger|SpaceAfter=No
8 子 子 NOUN n,名詞,人,関係 _ 6 obj _ Gloss=child|SpaceAfter=No
>>> import deplacy
>>> deplacy.render(doc)
不 ADV <════╗ advmod
入 VERB ═══╗═╝═╗ ROOT
虎 NOUN <╗ ║ ║ nmod
穴 NOUN ═╝<╝ ║ obj
不 ADV <════╗ ║ advmod
得 VERB ═══╗═╝<╝ parataxis
虎 NOUN <╗ ║ nmod
子 NOUN ═╝<╝ obj
suparkanbun.load()
has two options suparkanbun.load(BERT="roberta-classical-chinese-base-char",Danku=False)
. With the option Danku=True
the pipeline tries to segment sentences automatically. Available BERT
options are:
BERT="roberta-classical-chinese-base-char"
utilizes roberta-classical-chinese-base-char (default)BERT="roberta-classical-chinese-large-char"
utilizes roberta-classical-chinese-large-charBERT="guwenbert-base"
utilizes GuwenBERT-baseBERT="guwenbert-large"
utilizes GuwenBERT-largeBERT="sikubert"
utilizes SikuBERTBERT="sikuroberta"
utilizes SikuRoBERTa
Installation for Linux
pip3 install suparkanbun --user
Installation for Cygwin64
Make sure to get python37-devel
python37-pip
python37-cython
python37-numpy
python37-wheel
gcc-g++
mingw64-x86_64-gcc-g++
git
curl
make
cmake
packages, and then:
curl -L https://raw.githubusercontent.com/KoichiYasuoka/CygTorch/master/installer/supar.sh | sh
pip3.7 install suparkanbun --no-build-isolation
Installation for Jupyter Notebook (Google Colaboratory)
!pip install suparkanbun
Try notebook for Google Colaboratory.
Author
Koichi Yasuoka (安岡孝一)
- Downloads last month
- 30
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.