from p41 import Chunk, get_chunks text = list(filter([].__ne__, get_chunks())) index = 10 V = list(filter(lambda x : '動詞' in x.get_pos_list(), text[index])) P = [sum(([x.base for x in text[index][y].morphs if x.pos == '助詞'] for y in v.srcs), []) for v in V] S = [[text[index][x].get_sentence() for x in v.srcs if '助詞' in text[index][x].get_pos_list()] for v in V] V = list(map(lambda x : next(y.base for y in x.morphs if y.pos == "動詞"), V)) for v, p, s in zip(V, P, S): if len(p)*len(s) != 0: print('\t'.join([v] + [q for q in p] + [t for t in s]))
from p41 import Chunk, get_chunks text = get_chunks() for chunks in text: [print('{}\t{}'.format(chunks[i].get_sentence(remove=True), chunks[x].get_sentence(remove=True)))\ for i, x in enumerate([x.dst for x in chunks]) if x != -1]