1

如何使用节找出句子中的不定式动词？

例子：

doc = "I need you to find the verbes in this sentence"
en_nlp = stanza.Pipeline('en', processors='tokenize,lemma,mwt,pos,depparse', verbose=False, use_gpu=False)
processed = en_nlp(doc)

print(*[f"id: {word.id}\t word: {word.text}\t POS: {word.pos}\t head id: {word.head}\t head: {sent.words[word.head-1].text if word.head > 0 else 'root'} \t deprel: {word.deprel}" for sent in processed.sentences for word in sent.words], sep='\n')

输出：

id: 1    word: I     POS: PRON   head id: 2  head: need      deprel: nsubj
id: 2    word: need  POS: VERB   head id: 0  head: root      deprel: root
id: 3    word: you   POS: PRON   head id: 2  head: need      deprel: obj
id: 4    word: to    POS: PART   head id: 5  head: find      deprel: mark
id: 5    word: find  POS: VERB   head id: 2  head: need      deprel: xcomp
id: 6    word: the   POS: DET    head id: 7  head: verbes    deprel: det
id: 7    word: verbes    POS: NOUN   head id: 5  head: find      deprel: obj
id: 8    word: in    POS: ADP    head id: 10     head: sentence      deprel: case
id: 9    word: this  POS: DET    head id: 10     head: sentence      deprel: det
id: 10   word: sentence  POS: NOUN   head id: 5  head: find      deprel: obl

但是，在这一行中：

id: 5 word: find POS: VERB head id: 2 head: 需要 deprel: xcomp

我需要说它是一个不定式动词。

4

1 回答 1

0

我有同样的问题，不热衷于闯入分词器并最终调整了节语句。单词。

word.feats 表示 infintival 动词形式，如 id 7 所示，我还没有测试它的可靠性。

test_resp = "He was a little scared to knock on the door"
res = nlp(test_resp)
res.sentences[0].words[4:8]

给了这个

[{
   "id": 5,
   "text": "scared",
   "lemma": "scared",
   "upos": "ADJ",
   "xpos": "JJ",
   "feats": "Degree=Pos",
   "head": 0,
   "deprel": "root",
   "misc": "start_char=16|end_char=22"
 },
 {
   "id": 6,
   "text": "to",
   "lemma": "to",
   "upos": "PART",
   "xpos": "TO",
   "head": 7,
   "deprel": "mark",
   "misc": "start_char=23|end_char=25"
 },
 {
   "id": 7,
   "text": "knock",
   "lemma": "knock",
   "upos": "VERB",
   "xpos": "VB",
   "feats": "VerbForm=Inf",
   "head": 5,
   "deprel": "advcl",
   "misc": "start_char=26|end_char=31"
 },
 {
   "id": 8,
   "text": "on",
   "lemma": "on",
   "upos": "ADP",
   "xpos": "IN",
   "head": 10,
   "deprel": "case",
   "misc": "start_char=32|end_char=34"
 }]

出于我的目的，将字符串“to verb”视为单个词汇项并将 word.text 更新为“to_verb”和动词的字符范围以匹配更为有用。这使动词的 word.lemma 和 word.upos 作为 VERB 保持不变，但需要减少动词的头部和单词位置索引以及后面的单词以考虑删除“to”。

deepcopy 保护原始示例以进行说明，如果可能，最好避免使用它。

import re
import sys
from copy import deepcopy

def patch_inf_verb(processed):
    """hack the parse to treat 'to VERB' as one word"""
 
    # modified sentence
    results = deepcopy(processed)
    
    # regex to captures the text and numerals in  word.misc, 
    # e.g., 'start_char=11|stop_char=13'
    misc_vals_re = re.compile("(start_char=)(\d+)(\|end_char=)(?P<end>\d+)")

    for result in results.sentences:
        for wdx, word in enumerate(result.words):
            
            # peek back for "to"
            if wdx > 0 and word.pos == "VERB":
                one_back =  result.words[wdx - 1]
                if one_back.text.lower() == "to" and one_back.head == word.id:
                    
                    word.text = "to_" + word.text
                    # word.upos = "VERB_INF"  # update upos tag or leave as is

                    # parse verb's character span string
                    vals = misc_vals_re.match(word.misc).groups()
                    assert vals is not None
   
                    # nudge word.misc start_char back to span one-back "to"
                    word.misc = f"{vals[0]}{int(vals[1])-3}{vals[2]}{int(vals[3])}"
                    assert misc_vals_re.match(word.misc) is not None

                    # decrement the indexes for verb position and beyond,
                    # the character spans don't change
                    for tdx in range(len(result.words)):
                        if result.words[tdx].id > wdx: result.words[tdx].id -= 1
                        if result.words[tdx].head > wdx: result.words[tdx].head -= 1
                    
                    # clobber the "to" after
                    del result.words[wdx - 1]
    return results

def format_results(results):
    """results in table format"""
    results_str = '\n'.join(
        [
            "\t".join(
                    [
                        f"{key:5s}: {val}" 
                        for key, val in word.to_dict().items() 
                        if key not in ["lemma", "feats"]
                    ]
                )
                for sent in results.sentences 
                for word in sent.words
            ]
        )
    return results_str

操作示例：

print("python", sys.version)
print("stanza version:", stanza.__version__)

doc = "I need you to find the verbes in this sentence"
en_nlp = stanza.Pipeline('en', processors='tokenize,lemma,mwt,pos,depparse', verbose=False, use_gpu=False)
processed = en_nlp(doc)

print('OP stanza before\n', format_results(processed))

patched_to_verb = patch_inf_verb(processed)
print("after patch_inf_verb\n", format_results(patched_to_verb))

python 3.7.7 (default, Mar 26 2020, 15:48:22) 
[GCC 7.3.0]
stanza version: 1.1.1
OP stanza before
 id   : 1   text : I    upos : PRON xpos : PRP  head : 2    deprel: nsubj   misc : start_char=0|end_char=1
id   : 2    text : need upos : VERB xpos : VBP  head : 0    deprel: root    misc : start_char=2|end_char=6
id   : 3    text : you  upos : PRON xpos : PRP  head : 2    deprel: obj misc : start_char=7|end_char=10
id   : 4    text : to   upos : PART xpos : TO   head : 5    deprel: mark    misc : start_char=11|end_char=13
id   : 5    text : find upos : VERB xpos : VB   head : 2    deprel: xcomp   misc : start_char=14|end_char=18
id   : 6    text : the  upos : DET  xpos : DT   head : 7    deprel: det misc : start_char=19|end_char=22
id   : 7    text : verbes   upos : NOUN xpos : NNS  head : 5    deprel: obj misc : start_char=23|end_char=29
id   : 8    text : in   upos : ADP  xpos : IN   head : 10   deprel: case    misc : start_char=30|end_char=32
id   : 9    text : this upos : DET  xpos : DT   head : 10   deprel: det misc : start_char=33|end_char=37
id   : 10   text : sentence upos : NOUN xpos : NN   head : 5    deprel: obl misc : start_char=38|end_char=46
after patch_inf_verb
 id   : 1   text : I    upos : PRON xpos : PRP  head : 2    deprel: nsubj   misc : start_char=0|end_char=1
id   : 2    text : need upos : VERB xpos : VBP  head : 0    deprel: root    misc : start_char=2|end_char=6
id   : 3    text : you  upos : PRON xpos : PRP  head : 2    deprel: obj misc : start_char=7|end_char=10
id   : 4    text : to_find  upos : VERB xpos : VB   head : 2    deprel: xcomp   misc : start_char=11|end_char=18
id   : 5    text : the  upos : DET  xpos : DT   head : 6    deprel: det misc : start_char=19|end_char=22
id   : 6    text : verbes   upos : NOUN xpos : NNS  head : 4    deprel: obj misc : start_char=23|end_char=29
id   : 7    text : in   upos : ADP  xpos : IN   head : 9    deprel: case    misc : start_char=30|end_char=32
id   : 8    text : this upos : DET  xpos : DT   head : 9    deprel: det misc : start_char=33|end_char=37
id   : 9    text : sentence upos : NOUN xpos : NN   head : 4    deprel: obl misc : start_char=38|end_char=46
于 2020-10-30T23:20:53.190 回答