Unsupervised and Semi-supervised Learning of Tone and Pitch Accent

Proceedings of the Human Language Technology Conference of the North American Chapter of the ACL, pages 224–231, New York, June 2006. c⃝2006 Association for Computational Linguistics Unsupervised and Semi-supervised Learning of Tone and Pitch Accent Gina-Anne Levow University of Chicago 1100 E. 58th St. Chicago, IL 60637 USA levow@cs.uchicago.edu Abstract Recognition of tone and intonation is es- sential for speech recognition and lan- guage understanding. However, most ap- proaches to this recognition task have re- lied upon extensive collections of man- ually tagged data obtained at substantial time and financial cost. In this paper, we explore two approaches to tone learn- ing with substantially reductions in train- ing data. We employ both unsupervised clustering and semi-supervised learning to recognize pitch accent in English and tones in Mandarin Chinese. In unsu- pervised Mandarin tone clustering exper- iments, we achieve 57-87% accuracy on materials ranging from broadcast news to clean lab speech. For English pitch accent in broadcast news materials, results reach 78%. In the semi-supervised framework, we achieve Mandarin tone recognition ac- curacies ranging from 70% for broadcast news speech to 94% for read speech, out- performing both Support Vector Machines (SVMs) trained on only the labeled data and the 25% most common class assign- ment level. These results indicate that the intrinsic structure of tone and pitch accent acoustics can be exploited to reduce the need for costly labeled training data for tone learning and recognition. 1 Introduction Tone and intonation play a crucial role across many languages. However, the use and structure of tone varies widely, ranging from lexical tone which de- termines word identity to pitch accent signalling in- formation status. Here we consider the recognition of lexical tones in Mandarin Chinese syllables and pitch accent in English. Although intonation is an integral part of lan- guage and is requisite for understanding, recogni- tion of tone and pitch accent remains a challeng- ing problem. The majority of current approaches to tone recognition in Mandarin and other East Asian tone languages integrate tone identification with the general task of speech recognition within a Hid- den Markov Model framework. In some cases tone recognition is done only implicitly when a word or syllable is constrained jointly by the segmental acoustics and a higher level language model and the word identity determines tone identity. Other strate- gies build explicit and distinct models for the syl- lable final region, the vowel and optionally a final nasal, for each tone. Recent research has demonstrated the importance of contextual and coarticulatory influences on the surface realization of tones.(Xu, 1997; Shen, 1990) The overall shape of the tone or accent can be sub- stantially modified by the local effects of adjacent tone and intonational elements. Furthermore, broad scale phenomena such as topic and phrase struc- ture can affect pitch height, and pitch shape may be variably affected by the presence of boundary tones. These findings have led to explicit modeling of tonal 224 context within the HMM framework. In addition to earlier approaches that employed …

Content truncated to 500 words