Academic journal article Perception and Psychophysics

Statistical Computations over a Speech Stream in a Rodent

Academic journal article Perception and Psychophysics

Statistical Computations over a Speech Stream in a Rodent

Article excerpt

Statistical learning is one of the key mechanisms available to human infants and adults when they face the problems of segmenting a speech stream (Saffran, Aslin, & Newport, 1996) and extracting long-distance regularities (Gómez, 2002; Peña, Bonatti, Nespor, & Mehler, 2002). In the present study, we explore statistical learning abilities in rats in the context of speech segmentation experiments. In a series of five experiments, we address whether rats can compute the necessary statistics to be able to segment synthesized speech streams and detect regularities associated with grammatical structures. Our results demonstrate that rats can segment the streams using the frequency of co-occurrence (not transitional probabilities, as human infants do) among items, showing that some basic statistical learning mechanism generalizes over nonprimate species. Nevertheless, rats did not differentiate among test items when the stream was organized over more complex regularities that involved nonadjacent elements and abstract grammar-like rules.

Recent work on statistical learning has suggested that this computational process may be partly responsible for two important aspects of language development: the detection of word boundaries (Aslin, Saffran, & Newport, 1998; Saffran, 2001; Saffran, Aslin, & Newport, 1996; Saffran, Newport, Aslin, Tunick, & Barrueco, 1997) and the extraction of grammar-like structures (Newport & Aslin, 2004; Peña, Bonatti, Nespor, & Mehler, 2002; Saffran, 2001a, 2002). In a series of studies, Saffran, Aslin, and Newport (1996) and Saffran, Newport, and Aslin (1996) showed that human adults and infants could segment a speech stream composed of trisyllabic nonsense words on the basis of statistical regularities alone. In their adult study, the authors presented the stream to the participants for 21 min. After this exposure, the participants could differentiate, at above-chance levels, words that composed the stream from foils. In their study of these abilities in infants, the authors familiarized 8-month-olds with 2 min of the stream. During the test, the infants showed differential responding to words and to foils, reflecting the extraction of sequential regularities. In a later work, Aslin et al. (1998) explored the complexity of the computations performed by the infants. Given the structure of the streams used in the authors' previous studies, the infants could be detecting either frequency of co-occurrence among syllables or their transitional probabilities. Whereas the former implies learning that the two elements tend to appear together in sequences, the latter provides information about how predictable one element is with respect to another1 and better reflects the types of dependencies infants have to learn in natural languages (Aslin et al., 1998). In fact, results showed that the sort of information the 8-month-olds were detecting during the speech segmentation experiments consisted of transitional probabilities, and not just frequency of co-occurrence.

However, it has been demonstrated that the ability to compute the sort of statistics used in this task is not restricted to the detection of word boundaries. It also applies to nonlinguistic stimuli such as tones (Saffran, Johnson, Aslin, & Newport, 1999) and visual shapes (Fiser & Aslin, 2001, 2002a, 2002b; Kirkham, Slemmer, & Johnson, 2002). In addition, the cotton-top tamarin (a New World monkey) can segment a speech stream using statistical cues, just as human adults and infants do (Hauser, Newport, & Aslin, 2001). In that study, following the general procedure used with human infants, the authors familiarized tamarins to the stream for 20 min and tested them on the following day using their orienting responses toward a concealed speaker as the dependent measure. The tamarins were more likely to orient to foils than to words, a result that shows the learning of sequential regularities from exposure to the stream. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.