ckipnlp 0.8.0

Core Pipeline

from ckipnlp.pipeline import CkipPipeline, CkipDocument

pipeline = CkipPipeline()
doc = CkipDocument(raw='中文字喔，啊哈哈哈')

# Word Segmentation
pipeline.get_ws(doc)
print(doc.ws)
for line in doc.ws:
    print(line.to_text())

# Part-of-Speech Tagging
pipeline.get_pos(doc)
print(doc.pos)
for line in doc.pos:
    print(line.to_text())

# Named-Entity Recognition
pipeline.get_ner(doc)
print(doc.ner)

# Sentence Parsing
pipeline.get_parsed(doc)
print(doc.parsed)

################################################################

from ckipnlp.container.util.wspos import WsPosParagraph

# Word Segmentation & Part-of-Speech Tagging
for line in WsPosParagraph.to_text(doc.ws, doc.pos):
    print(line)

Co-Reference Pipeline

from ckipnlp.pipeline import CkipCorefPipeline, CkipDocument

pipeline = CkipCorefPipeline()
doc = CkipDocument(raw='畢卡索他想，完蛋了')

# Co-Reference
corefdoc = pipeline(doc)
print(corefdoc.coref)
for line in corefdoc.coref:
    print(line.to_text())

WS with POS

There are also conversion routines for word-segmentation and POS containers jointly. For example, WsPosToken provides routines for a word (str) with POS-tag (str):

ws_obj, pos_obj = WsPosToken.from_text('中文字(Na)')
plain_text = WsPosToken.to_text(ws_obj, pos_obj)

ws_obj, pos_obj = WsPosToken.from_dict({ 'word': '中文字', 'pos': 'Na', })
dict_obj = WsPosToken.to_dict(ws_obj, pos_obj)

ws_obj, pos_obj = WsPosToken.from_list([ '中文字', 'Na' ])
list_obj = WsPosToken.to_list(ws_obj, pos_obj)

ws_obj, pos_obj = WsPosToken.from_json(json_str)
json_str = WsPosToken.to_json(ws_obj, pos_obj)

Similarly, WsPosSentence/WsPosParagraph provides routines for word-segmented and POS sentence/paragraph (SegSentence/SegParagraph) respectively.

Parsed Tree

In addition to ParsedParagraph, we have implemented tree utilities base on TreeLib.

ParsedTree is the tree structure of a parsed sentence. One may use from_text() and to_text() for plain-text conversion; from_dict(), to_dict() for dictionary-like object conversion; and also from_json(), to_json() for JSON string conversion.

The ParsedTree is a TreeLib tree with ParsedNode as its nodes. The data of these nodes is stored in a ParsedNodeData (accessed by node.data), which is a tuple of role (semantic role), pos (part-of-speech tagging), word.

ParsedTree provides useful methods: get_heads() finds the head words of the sentence; get_relations() extracts all relations in the sentence; get_subjects() returns the subjects of the sentence.

from ckipnlp.container import ParsedTree

# 我的早餐、午餐和晚餐都在那場比賽中被吃掉了
tree_text = 'S(goal:NP(possessor:N‧的(head:Nhaa:我|Head:DE:的)|Head:Nab(DUMMY1:Nab(DUMMY1:Nab:早餐|Head:Caa:、|DUMMY2:Naa:午餐)|Head:Caa:和|DUMMY2:Nab:晚餐))|quantity:Dab:都|condition:PP(Head:P21:在|DUMMY:GP(DUMMY:NP(Head:Nac:比賽)|Head:Ng:中))|agent:PP(Head:P02:被)|Head:VC31:吃掉|aspect:Di:了)'

tree = ParsedTree.from_text(tree_text, normalize=False)

print('Show Tree')
tree.show()

print('Get Heads of {}'.format(tree[5]))
print('-- Semantic --')
for head in tree.get_heads(5, semantic=True): print(repr(head))
print('-- Syntactic --')
for head in tree.get_heads(5, semantic=False): print(repr(head))
print()

print('Get Relations of {}'.format(tree[0]))
print('-- Semantic --')
for rel in tree.get_relations(0, semantic=True): print(repr(rel))
print('-- Syntactic --')
for rel in tree.get_relations(0, semantic=False): print(repr(rel))
print()

# 我和食物真的都很不開心
tree_text = 'S(theme:NP(DUMMY1:NP(Head:Nhaa:我)|Head:Caa:和|DUMMY2:NP(Head:Naa:食物))|evaluation:Dbb:真的|quantity:Dab:都|degree:Dfa:很|negation:Dc:不|Head:VH21:開心)'

tree = ParsedTree.from_text(tree_text, normalize=False)

print('Show Tree')
tree.show()

print('Get get_subjects of {}'.format(tree[0]))
print('-- Semantic --')
for subject in tree.get_subjects(0, semantic=True): print(repr(subject))
print('-- Syntactic --')
for subject in tree.get_subjects(0, semantic=False): print(repr(subject))
print()