Unsupervised Lexicon Discovery from Acoustic Input

Unsupervised Lexicon Discovery from Acoustic Input Chia-ying Lee1, Timothy J. O’Donnell2, and James Glass1 1Computer Science and Artificial Intelligence Laboratory 2Department of Brain and Cognitive Sciences Massachusetts Institute of Technology Cambridge, MA 02139, USA Abstract We present a model of unsupervised phono- logical lexicon discovery—the problem of simultaneously learning phoneme-like and word-like units from acoustic input. Our model builds on earlier models of unsuper- vised phone-like unit discovery from acous- tic data (Lee and Glass, 2012), and unsuper- vised symbolic lexicon discovery using the Adaptor Grammar framework (Johnson et al., 2006), integrating these earlier approaches us- ing a probabilistic model of phonological vari- ation. We show that the model is competi- tive with state-of-the-art spoken term discov- ery systems, and present analyses exploring the model’s behavior and the kinds of linguis- tic structures it learns. 1 Introduction One of the most basic problems of language acqui- sition is accounting for how children learn the in- ventory of word forms from speech—phonological lexicon discovery. In learning a language, children face a number of challenging, mutually interdepen- dent inference problems. Words are represented in terms of phonemes, the basic phonological units of a language. However, phoneme inventories vary from language to language, and the underlying phonemes which make up individual words often have variable acoustic realizations due to systematic phonetic and phonological variation, dialect differences, speech style, environmental noise, and other factors. To learn the phonological form of words in their lan- guage children must determine the phoneme inven- tory of their language, identify which parts of the acoustic signal correspond to which phonemes— while discounting surface variation in the realiza- tion of individual units—and infer which sequences of phonemes correspond to which words (amongst other challenges). Understanding the solution to this complex joint- learning problem is not only of fundamental sci- entific interest, but also has important applications in Spoken Language Processing (SLP). Even set- ting aside additional grammatical and semantic in- formation available to child learners, there is still a sharp contrast between the type of phonological learning done by humans and current SLP meth- ods. Tasks that involve recognizing words from acoustic input—such as automatic speech recogni- tion and spoken term discovery—only tackle parts of the overall problem, and typically rely on linguistic resources such as phoneme inventories, pronuncia- tion dictionaries, and annotated speech data. Such resources are unavailable for many languages, and expensive to create. Thus, a model that can jointly learn the sound patterns and the lexicon of a lan- guage would open up the possibility of automati- cally developing SLP capabilities for any language. In this paper, we present a first step towards an unsupervised model of phonological lexicon discov- ery that is able to jointly learn, from unannotated speech, an underlying phoneme-like inventory, the pattern of surface realizations of those units, and a set of lexical units for a language. Our model builds on earlier work addressing the unsupervised discov- ery of phone-like units from acoustic data—in par- ticular the Dirichlet Process Hidden Markov Model …

Content truncated to 500 words