Racing to Segment

Explaining how infants learn words is one of the central problems in language acquisition. Learning words requires the ability to recognize them in real time, and spoken word-form recognition is an exceedingly complex skill. Because the speech signal is ephemeral, processing is subject to considerable timepressure. Furthermore, words occur as part of the continuous flow of speech. If words were presented in isolation, word-form recognition, and hence word learning, would be greatly simplified, but in reality, infants most often hear phrases of concatenated words (van de Weijer, 2001; Aslin, Woodward, LaMendola & Bever, 1996 but see Brent & Siskind 2001). Therefore, to identify the phonological forms of words, infants must learn to segment those words from fluent speech. Understanding how infants segment word-forms from speech during the early stages of language acquisition may thus illuminate the means by which infants bootstrap their way to parsing speech into meaningful components. As adults we (subjectively) recognize words effortlessly and instantaneously. With access to a vast lexicon, adults can rely on their knowledge of familiar words to map lexical items onto the speech stream. Words may be recognized even before they are completely pronounced (Marslen-Wilson & Welsh, 1978), and listeners can use such recognition to forecast the end of the current word and, hence, the onset of the following word (Cole & Jakimik, 1980). In most research to date, however, it has been assumed that infants lack the requisite lexical knowledge and must therefore rely on alternative means for processing continuous streams of speech. A number of studies have shown that infants are sensitive to a variety of cues in the speech stream including stress (Jusczyk, Houston, & Newsome, 1999), statistical information (Saffran, Aslin & Newport, 1996), and prosodic boundaries (Christophe, Gout, Peperkamp & Morgan, 2003). In keeping with the tradition in this literature, we use “top-down” to refer to