模型:
flair/upos-multi
这是Flair带有的默认多语种通用词性标注模型。
F1-Score: 98.47(覆盖英语、德语、法语、意大利语、荷兰语、波兰语、西班牙语、瑞典语、丹麦语、挪威语、芬兰语和捷克语的12个UD树库)
预测通用词性标记:
| tag | meaning |
|---|---|
| ADJ | adjective |
| ADP | adposition |
| ADV | adverb |
| AUX | auxiliary |
| CCONJ | coordinating conjunction |
| DET | determiner |
| INTJ | interjection |
| NOUN | noun |
| NUM | numeral |
| PART | particle |
| PRON | pronoun |
| PROPN | proper noun |
| PUNCT | punctuation |
| SCONJ | subordinating conjunction |
| SYM | symbol |
| VERB | verb |
| X | other |
基于 Flair embeddings 和LSTM-CRF。
要求: Flair (pip install flair)
from flair.data import Sentence
from flair.models import SequenceTagger
# load tagger
tagger = SequenceTagger.load("flair/upos-multi")
# make example sentence
sentence = Sentence("Ich liebe Berlin, as they say. ")
# predict POS tags
tagger.predict(sentence)
# print sentence
print(sentence)
# iterate over tokens and print the predicted POS label
print("The following POS tags are found:")
for token in sentence:
print(token.get_label("upos"))
这将产生以下输出:
Token[0]: "Ich" → PRON (0.9999) Token[1]: "liebe" → VERB (0.9999) Token[2]: "Berlin" → PROPN (0.9997) Token[3]: "," → PUNCT (1.0) Token[4]: "as" → SCONJ (0.9991) Token[5]: "they" → PRON (0.9998) Token[6]: "say" → VERB (0.9998) Token[7]: "." → PUNCT (1.0)
因此,在多语种句子“我爱柏林,正如他们所说”中,“Ich”和“they”被标记为代词(PRON),而“liebe”和“say”被标记为动词(VERB)。
使用以下Flair脚本进行了此模型的训练:
from flair.data import MultiCorpus
from flair.datasets import UD_ENGLISH, UD_GERMAN, UD_FRENCH, UD_ITALIAN, UD_POLISH, UD_DUTCH, UD_CZECH, \
UD_DANISH, UD_SPANISH, UD_SWEDISH, UD_NORWEGIAN, UD_FINNISH
from flair.embeddings import StackedEmbeddings, FlairEmbeddings
# 1. make a multi corpus consisting of 12 UD treebanks (in_memory=False here because this corpus becomes large)
corpus = MultiCorpus([
UD_ENGLISH(in_memory=False),
UD_GERMAN(in_memory=False),
UD_DUTCH(in_memory=False),
UD_FRENCH(in_memory=False),
UD_ITALIAN(in_memory=False),
UD_SPANISH(in_memory=False),
UD_POLISH(in_memory=False),
UD_CZECH(in_memory=False),
UD_DANISH(in_memory=False),
UD_SWEDISH(in_memory=False),
UD_NORWEGIAN(in_memory=False),
UD_FINNISH(in_memory=False),
])
# 2. what tag do we want to predict?
tag_type = 'upos'
# 3. make the tag dictionary from the corpus
tag_dictionary = corpus.make_tag_dictionary(tag_type=tag_type)
# 4. initialize each embedding we use
embedding_types = [
# contextual string embeddings, forward
FlairEmbeddings('multi-forward'),
# contextual string embeddings, backward
FlairEmbeddings('multi-backward'),
]
# embedding stack consists of Flair and GloVe embeddings
embeddings = StackedEmbeddings(embeddings=embedding_types)
# 5. initialize sequence tagger
from flair.models import SequenceTagger
tagger = SequenceTagger(hidden_size=256,
embeddings=embeddings,
tag_dictionary=tag_dictionary,
tag_type=tag_type,
use_crf=False)
# 6. initialize trainer
from flair.trainers import ModelTrainer
trainer = ModelTrainer(tagger, corpus)
# 7. run training
trainer.train('resources/taggers/upos-multi',
train_with_dev=True,
max_epochs=150)
在使用此模型时,请引用以下论文。
@inproceedings{akbik2018coling,
title={Contextual String Embeddings for Sequence Labeling},
author={Akbik, Alan and Blythe, Duncan and Vollgraf, Roland},
booktitle = {{COLING} 2018, 27th International Conference on Computational Linguistics},
pages = {1638--1649},
year = {2018}
}
Flair问题跟踪器可在 here 处找到。