MODELS OF CONTINUOUS SPEECH RECOGNITION AND THE CONTENTS OF THE VOCABULARY

Several models of spoken word recognition postulate that recognition i s achieved via a process of competition between lexical hypotheses. Co mpetition not only provides a mechanism for isolated word recognition, it also assists in continuous speech recognition, since it offers a m eans of segmenting continuous input into individual words. We present statistics on the pattern of occurrence of words embedded in the polys yllabic words of the English vocabulary, showing that an overwhelming majority (84%) of polysyllables have shorter words embedded within the m. Positional analyses show that these embeddings are most common at t he onsets of the longer word. Although both phonological and syntactic constraints could rule out some embedded words, they do not remove th e problem. Lexical competition provides a means of dealing with lexica l embedding. It is also supported by a growing body of experimental ev idence. We present results which indicate that competition operates bo th between word candidates that begin at the same point in the input a nd candidates that begin at different points (McQueen, Norris, and Cut ler, 1994; Norris, McQueen, and Cutler, in press). We conclude that le xical competition is an essential component in models of continuous sp eech recognition.