Characterizing and Processing Robot-Directed Speech

Speech directed at infants and pets has properties that distinguish it from speech among adults (6). Some of those properties are potentially useful for language learning. By careful design of form and behavior, robots can hope to evoke a similar speech register and take advantage of these properties. We report some preliminary data to support this claim, based on experiments carried out with the infant-like robot Kismet (4). We then show how we can build a language model around an initial vocabulary, perhaps acquired from "cooperati ve" speech, and bootstrap from it to identify further candidate vocabulary items drawn from arbitrary speech in an unsupervised manner. We show how to cast this process in a form that can be largely implemented using a conventional speech recognition system (8), even though such systems are designed with very different applications in mind. This is advantageous since, after decades of research, such systems are expert at making acoustic judgments in a probabilistically sound way from acoustic, phonological, and language models.