Popular version of paper 4pSC2
Presented Thursday, December 4, 1997
134th ASA Meeting, San Diego, CA
Embargoed until December 4, 1997
To acquire a language one has to learn the words of the language. In other words, one has to learn which sound patterns go with which meanings. This task is complicated by the fact that in speech, unlike in written language, pauses or spaces rarely occur between words. Instead the acoustic pattern of one word often blends into that of the next word in an utterance. This characteristic of spoken language presents a problem for the language learner: how and when does an infant figure out where one word begins and another one leaves off?
To investigate this issue, Jusczyk and Aslin (1995) devised a method for determining when infants begin to correctly locate words in speech. They familiarized 7.5-month-old English-learning infants with a pair of isolated words, such as "feet" and "bike" (or "cup" and "dog"). The infants listened to each word for about 30 seconds. Then the infants heard a series of four 6-sentence passages. The word "feet" occurred in each sentence of one of the passages, and the word "bike" occurred in another one. The remaining two passages also contained a repeated word ("cup" and "dog"). The infants listened significantly longer to the passages that contained the words that they were familiarized with, suggesting that they detected these words in the utterances. This was the first indication that infants at this age have some ability to segment words in connected speech.
To understand more clearly what kind of information that infants are learning about the sound patterns of words, we conducted additional studies. In one of these, we examined how much information that infants had perceived about the sound patterns of the words that they were familiarized with. For example, does a 7.5-month-old pick up enough information to distinguish a word like "cup" from a very similar sounding word like "cut"? This appears to be the case because when infants were familiarized with a word like "cut" (instead of "cup"), they did not subsequently listen longer to a passage containing "cup" than they did to passages containing other unfamiliar words. We also examined the extent to which 7.5-month-olds can generalize their representations of the sound patterns of words. Infants were familiarized with a pair of words, such as "cup" and "dog" produced by one talker, but then were tested on passages produced by a different talker. The infants showed some limited ability to generalize from speech produced by one talker to that of another. Specifically, they generalized in cases in which the talkers were of the same gender (i.e., either two females or two males). In contrast, 7.5-month-olds did not generalize from a female voice to a male voice, or vice versa. Older infants, 10.5-month-olds, were able to generalize when the familiarization words were produced by a female and the test passages were produced by a male (or vice versa). Thus, when infants are first beginning to segment words from speech, their ability to deal with the acoustic variation that occurs in different talker's productions of the same words is more limited than that of adults.
Finally, we also explored infants' memory for the sound patterns of specific words to determine whether what infants are doing in the laboratory is really an indication of whether they are beginning to build a vocabulary by storing information about sound patterns of words. This time we familiarized the infants with a pair of words, such as "feet" and "bike" on one day, and then, we waited 24 hours before playing them the four test passages. The results suggested that infants did retain some information about the sound patterns that they had heard because they listened significantly longer to the passages containing the words that they had been familiarized with on the day before. Still, at this early stage of language learning, infants' representations of the sound patterns of words are rather fragile, and they do not appear to generalize between words produced by one talker on the first day to the same words produced in passages by a different talker (even one from the same gender) on the second day.
Taken together, the findings suggest that infants are beginning to segment words from speech at about 7.5-months of age. Moreover, they seem to be storing information about these sound patterns in memory, which suggests that they are beginning to build up a vocabulary in their native language at this time. Remarkably, infants are retaining information about these sound patterns even when no effort was made to teach them what these words actually mean. This suggests that infants can begin to learn words in, at least, two different ways. The first way is the one that most people think about word learning, namely, the infants begin with some concept in mind (e.g., a four-legged creature, that barks and wags its tail), and then learn what the label or word is that refers to it. The second way in which infants learn words is by learning a label first (an interesting sound pattern that they retain in memory) and then learning what meaning to attach to it.
Dr. Jusczyk will be staying in San Diego at the Town and Country Hotel, 500 Hotel Circle North. Tel: (619) 291-3584. He will be arriving Wednesday afternoon and leaving on Saturday morning. Prior to and after that time, he can be reached at (410) 516-5165.