DOI QR코드

DOI QR Code

Two-Path Language Modeling Considering Word Order Structure of Korean

한국어의 어순 구조를 고려한 Two-Path 언어모델링

  • 신중휘 (고려대학교 컴퓨터학과) ;
  • 박재현 (고려대학교 컴퓨터학과) ;
  • 이정태 (고려대학교 컴퓨터.전파통신공학과) ;
  • 임해창 (고려대학교 컴퓨터.전파통신공학과)
  • Published : 2008.11.30

Abstract

The n-gram model is appropriate for languages, such as English, in which the word-order is grammatically rigid. However, it is not suitable for Korean in which the word-order is relatively free. Previous work proposed a twoply HMM that reflected the characteristics of Korean but failed to reflect word-order structures among words. In this paper, we define a new segment unit which combines two words in order to reflect the characteristic of word-order among adjacent words that appear in verbal morphemes. Moreover, we propose a two-path language model that estimates probabilities depending on the context based on the proposed segment unit. Experimental results show that the proposed two-path language model yields 25.68% perplexity improvement compared to the previous Korean language models and reduces 94.03% perplexity for the prediction of verbal morphemes where words are combined.

n-gram 모델은 영어와 같이 어순이 문법적으로 제약을 받는 언어에 적합하다. 그러나 어순이 비교적 자유로운 한국어에는 적합하지 않다. 기존 연구는 어절 간 어순의 고려가 어려운 한국어의 특성을 반영한 twoply HMM을 제안했으나, 인접 어절 간 어순 구조를 반영하지 못하였다. 본 논문에서는 용언형태소 사이에 나타나는 인접 어절 간에 어순 특성을 반영하기 위해 두 어절을 결합하는 세그먼트 단위를 정의하고, 제안한 세그먼트 단위에서 문맥에 따라 확률을 달리 추정하는 two-path 언어모델을 제안한다. 그 결과 기존 한국어 언어모델에 비해 제안하는 two-path 언어모델은 기존 연구보다 25.68% 혼잡도를 줄였으며, 어절 간에 결합이 일어나는 경계인 용언형태소에서는 94.03%의 혼잡도를 줄였다.

Keywords

References

  1. F. Jelinek, "Self-organized language modeling for speech recognition", Readings in Speech Recognition, A. Waibel and K. F. Lee, eds., Morgan Kaufmann, 450-506, 1990
  2. P. F. Brown, S. A. Della Pietra, V. J. Della Pietra, F. Jelinek, J. D. Lafferty, R. L. Mercer, and Roossin, P.S. "A statistical approach to machine translation", Computational Linguistics, 16(2), 79-85, 1990
  3. D. Jurafsky and J. H. Martin, Speech and Language Processing (Prentice Hall, 2007) Chap.4, pp.83-121
  4. 김진동, 임희석, 임해창, "Twoply HMM: 한국어의 특성을 고려한 형태소 단위의 품사 태깅 모델", 정보과학회논문지 (B), 24 (12), 1502-1512, 1997
  5. L. Rabiner and B. Juang, "An Introduction to hidden Markov models", ASSP Magazine IEEE Signal Processing, 3(1), 4-16, 1986 https://doi.org/10.1109/MASSP.1986.1165381
  6. E. Arisoy and M. Saraclar, "Lattice extension and rescoring based approaches for LVCSR of turkish", in INTERSPEECH, 1025-1028, 2006
  7. J. Gao, H. Suzuki, and Y. Wen, "Exploiting headword dependency and predictive clustering for language modeling", in EMNLP-2002, 248-256, 2002
  8. M. Creutz, T. Hirsimaki, M. Kurimo, A. Puurula, J. Pylkkonen, V. Siivola, M. Varjokallio, E. Arisoy, M. Saraclar, and A. Stolcke, "Morph-based speech recognition and modeling of out-of-vocabulary words across languages", ACM TSLP, 5(1), 2007
  9. O. Kwon and J. Park, "Korean large vocabulary continuous speech recognition with morpheme-based recognition units", Speech Communication, 39(3-4), 287-300, 2003 https://doi.org/10.1016/S0167-6393(02)00031-6
  10. N. Hideki, Corpus-based approaches to sentence structures (John Benjamins Pub Co., 2005), Chap.3, pp.51-76
  11. S. M. Katz, "Estimation of probabilities from sparse data for the language model component of a speech recognizer", IEEE Transactions on Acoustics, Speech and Signal Processing, 35, 400-401, 1987 https://doi.org/10.1109/TASSP.1987.1165125
  12. A. Stolcke, "SRILM-an extensible language modeling toolkit", in ICSLP-2002, 901-904, 2002