한국정보통신학회논문지 (Journal of the Korea Institute of Information and Communication Engineering)

  • 제26권5호
  • /
  • Pages.662-667
  • /
  • 2022
  • /
  • 2234-4772(pISSN)
  • /
  • 2288-4165(eISSN)
한국정보통신학회 (The Korea Institute of Information and Commucation Engineering)
권호 표지
DOI QR코드

DOI QR Code

Biaffine Average Attention 모델을 이용한 의미역 결정

Semantic Role Labeling using Biaffine Average Attention Model

  • Nam, Chung-Hyeon (Department of Computer Engineering, Korea University of Technology and Education) ;
  • Jang, Kyung-Sik (Department of Computer Engineering, Korea University of Technology and Education)
  • 투고 : 2021.12.13
  • 심사 : 2022.04.05
  • 발행 : 2022.05.31
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

의미역 결정 작업은 서술어와 문장 내 행위자, 피행위자, 장소, 시간 등 서술어와 관련 있는 논항들을 추출하는 작업이다. 기존 의미역 결정 방법은 문장의 언어학적 특징 추출을 위한 파이프라인을 구축하는데, 파이프라인 내 각 추출 작업들의 오류가 의미역 결정 작업의 성능에 영향을 미치기 때문에 현재는 End-to-End 방법의 신경망 모델을 이용한 방법들이 제안되고 있다. 본 논문에서는 의미역 결정 작업을 위해 Biaffine Average Attention 구조를 이용한 신경망 모델을 제안한다. 제안하는 모델은 기존 연구에서 제안된 특정 시점에 대한 레이블 예측을 위해 주변 시점 정보를 이용하는 LSTM 모델 대신 문장 내 서술어와 논항의 거리에 상관없이 문장 전체 정보에 집중할 수 있는 Biaffine Average Attention 구조로 이루어져 있다. 제안하는 모델의 성능 평가를 위해 F1 점수를 이용하여 기존 연구에서 제안한 BERT 기반의 모델들과 비교하였으며, 76.21%의 성능으로 비교 모델보다 높은 성능을 보였음을 확인하였다.

Semantic role labeling task(SRL) is to extract predicate and arguments such as agent, patient, place, time. In the previously SRL task studies, a pipeline method extracting linguistic features of sentence has been proposed, but in this method, errors of each extraction work in the pipeline affect semantic role labeling performance. Therefore, methods using End-to-End neural network model have recently been proposed. In this paper, we propose a neural network model using the Biaffine Average Attention model for SRL task. The proposed model consists of a structure that can focus on the entire sentence information regardless of the distance between the predicate in the sentence and the arguments, instead of LSTM model that uses the surrounding information for prediction of a specific token proposed in the previous studies. For evaluation, we used F1 scores to compare two models based BERT model that proposed in existing studies using F1 scores, and found that 76.21% performance was higher than comparison models.

키워드

과제정보

This paper was supported by Education and Research Promotion Program of KoreaTech.

참고문헌

  1. L. He, K. Lee, M. Lewis, and L. Zettlemoyer, "Deep Semantic Role Labeling: What Works and What's Next," in Proceeding of the 55th Annual Meeting of the Association for Computational Linguistics, Vancouver, Canada, pp. 473-483, 2017.
  2. L. He, K. Lee, O. Levy, and L. Zettlemoyer, "Jointly Predicting Predicates and Arguments in Neural Semantic Role Labeling," in Proceeding of the 56th Annual Meeting of the Association for Computational Linguistics, Melbourne, Australia, pp. 364-369, 2018.
  3. J. S. Bae and C. K. Lee, "Korean Semantic Role Labeling using Stacked Bidirectional LSTM-CRFs," Journal of KIISE, vol. 44, no. 1, pp. 36-43, Jan. 2017. https://doi.org/10.5626/JOK.2017.44.1.36
  4. K. H. Park and S. H. Na, "A Neural Attention model for Korean Semantic Role Labeling," in Proceeding of the 2017 Korea Software Congress, Busan, South Korea, pp. 512-514, 2019.
  5. J. Devlin, M. W. Chang, K. Lee, K. Toutanova, "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding," in Proceeding of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Minneapolis, US, pp. 4171-4186, 2019.
  6. J. S. Bae, C. K. Lee, S. J. Lim and H. K. Kim, "Korean Semantic Role Labeling with BERT," in Proceeding of the 2019 Korea Computer Congress, Jeju, South Korea, pp. 512-514, 2019.
  7. T. Dozat, C. D. Manning, "A Neural Attention model for Korean Semantic Role Labeling," in Proceeding of the 5th International Conference on Learning Representations, Busan, South Korea, pp. 512-514, 2019.
  8. S. H. Na, J. R. Li, J. H. Shin and K. I. Kim, "Deep Biaffine Attention for Korean Dependency Parsing," in Proceeding of the 2017 Korea Computer Congress, Jeju, South Korea, pp. 584-586, 2017.
  9. J. Yu, B. Bohnet and M. Poesio, "Named Entity Recognition as Dependency Parsing," in Proceeding of the 58th Annual Meeting of the Association for Computational Linguistics, Kuala Lumpur, Malaysia, pp. 6470-6476, 2020.
  10. AI Hub Common sense AI dataset [Internet]. Available: https://www.aihub.or.kr/.
  11. T. Y. Lin, P. Goyal, K. He and P. Dollar, "Focal Loss for Dense Object Detection," in Proceeding of the 2017 IEEE International Conference on Computer Vision(ICCV), Venezia, Italiana, pp. 2980-2988, 2017.