Sequence-to-Sequence 모델 기반으로 한 한국어 형태소 분석의 재순위화 모델
A Reranking Model for Korean Morphological Analysis Based on Sequence-to-Sequence Model
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
Sequence-to-sequence(Seq2seq) 모델은 입력열과 출력열의 길이가 다를 경우에도 적용할 수 있는 모델로 한국어 형태소 분석에서 많이 사용되고 있다. 일반적으로 Seq2seq 모델을 이용한 한국어 형태소 분석에서는 원문을 음절 단위로 처리하고 형태소와 품사를 음절 단위로 출력한다. 음절 단위의 형태소 분석은 사전 미등록어 문제를 쉽게 처리할 수 있다는 장점이 있는 반면 형태소 단위의 사전 정보를 반영하지 못한다는 단점이 있다. 본 연구에서는 Seq2seq 모델의 후처리로 재순위화 모델을 추가하여 형태소 분석의 최종 성능을 향상시킬 수 있는 모델을 제안한다. Seq2seq 모델에 빔 서치를 적용하여 K개 형태소 분석 결과를 생성하고 이들 결과의 순위를 재조정하는 재순위화 모델을 적용한다. 재순위화 모델은 기존의 음절 단위 처리에서 반영하지 못했던 형태소 단위의 임베딩 정보와 n-gram 문맥 정보를 활용한다. 제안한 재순위화 모델은 기존 Seq2seq 모델에 비해 약 1.17%의 F1 점수가 향상되었다.
Abstract
A Korean morphological analyzer adopts sequence-to-sequence (seq2seq) model, which can generate an output sequence of different length from an input. In general, a seq2seq based Korean morphological analyzer takes a syllable-unit based sequence as an input, and output a syllable-unit based sequence. Syllable-based morphological analysis has the advantage that unknown words can be easily handled, but has the disadvantages that morpheme-based information is ignored. In this paper, we propose a reranking model as a post-processor of seq2seq model that can improve the accuracy of morphological analysis. The seq2seq based morphological analyzer can generate K results by using a beam-search method. The reranking model exploits morpheme-unit embedding information as well as n-gram of morphemes in order to reorder K results. The experimental results show that the reranking model can improve 1.17% F1 score comparing with the original seq2seq model.
참고문헌 (18)
- S. H. Na, "Deep Learning for Korean POS Tagging", Korea Software Congress, pp.426~428 (2014)
- 이건일, "Sequence-to-sequence 기반 한국어 형태소 분석 및 품사 태깅", 정보과학회논문지, 44(1), pp.57~62 (2017), 한국정보과학회
- H. S. Hwang, "Korean Morphological Analysis using Sequence-to-sequence learning with Copying mechanism", Korea Software Congress, pp.443~445 (2016)
- Y. S. Choi, "Re-ranking Model of Korean Morphological Analysis Results using Neural Networks", pp.2017
- I. Sutskever, "Sequence to sequence learning with neural networks", Advances in neural information processing systems, pp.3104~3112 (2014)
- J. Gu, "Incorporating copying mechanism in sequence-to-sequence learning", ~ (2016)
- F. J. Och, "The alignment template approach to statistical machine translation", Computational Linguistics, 30(4), pp.417~449 (2004)[DOI]
- J. Li, "Mutual information and diverse decoding improve neural machine translation", ~ (2016)
- O. Dusek, "Sequence-to-sequence generation for spoken dialogue via deep syntax trees and strings", Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pp.45~51 (2016)
- J. Chorowski, "Towards better decoding and language model integration in sequence to sequence models", ~ (2016)
- Y. Shao, "Generating high-quality and informative conversation responses with sequence-to-sequence models", Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pp.2200~2209 (2017)
- T. H. Wen, "Stochastic language generation in dialogue using recurrent neural networks with convolutional sentence reranking", ~ (2015)
- T. Mikolov, "Linguistic Regularities in Continuous Space Word Representations", Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp.746~751 (2013)
- C. D. Santos, "Learning character-level representations for part-of-speech tagging", Proceedings of the 31st International Conference on Machine Learning (ICML-14), pp.1818~1826 (2014)
- S. Misawa, "Character-based Bidirectional LSTM-CRF with words and characters for Japanese Named Entity Recognition", Proceedings of the First Workshop on Subword and Character Level Models in NLP, pp.97~102 (2017)
- T. Nakagawa, "A hybrid approach to word segmentation and pos tagging", Proceedings of the 45th annual meeting of the ACL on interactive poster and demonstration sessions, pp.217~220 (2007), Association for Computational Linguistics
- Y. Kim, "Convolutional neural networks for sentence classification", ~ (2014)
- L. Rosasco, "Are loss functions all the same?", Neural Computation, 16(5), pp.1063~1076 (2004)[DOI]