Automatic Extraction of Phonotactics based on Hidden Markov Models and Language Identification
Seiichi Nakagawa, Yoshio Ueda · Institutional Repositories DataBase (IRDB) · 1990
Natural languages were modeled popularly by Markov models. In this paper, they were modeled by HMM (Hidden Markov Model). As applications, the identification of language and the prediction for phoneme/syllable/word-category were performed using HMM. The results show that the HMM extracts automatically the phonotactics and it has also a performance better than the first order Markov model (bigram) and almost the same as the second order Markov model (trigram) on the entropy. From the results, we believe that HMM is useful not only speech recognition but also natural language processing.