Neural Correlates of Processing Syntax in Music and ... - PubMan

Language Perception 57
were shown to use this sequential statistics to discover word-like units in absence of
other acoustic cues to word boundaries (Saffran, Aslin, & Newport, 1996). After repeated
exposure to nonsense words with recurring sound patterns, infants can make
generalizations about syllable structure (Saffran & Thiessen, 2003), stress (Gerken,
2004) and phonotactic patterns (Chambers, Onishi, & Fisher, 2003). Infants exploit such
knowledge to segment words from fluent speech. Thus, they are able to parse speech
and discover potential words, presumably before they understand their meaning
(Saffran, 2002; Saffran et al., 1996).
Saffran (2001b) demonstrated that statistical learning mechanisms can generate wordlike
units. She familiarized eight-month-old infants with nonsense words, and tested the
infants' responses to words versus part-words that were embedded either in simple English
sentences, in matched nonsense frames, or non-linguistic frames generated from
tone sequences. Infants’ listening preferences were affected by the context in which the
learned items were embedded. Saffran and Wilson (2003) demonstrated that the same
learning mechanisms can be exploited for several learning stages: the mental representations
produced by a statistically based word segmentation procedure are represented not
as sets of syllables, but form new units that are available to serve as the input to subsequent
learning processes.
These statistical learning mechanisms may interact with prosodic cues. Jusczyk and
Aslin (1995) found that 7½ month old infants successfully segmented words with the
more common stress pattern, but they missegmented words with uncommon stress pattern.
In contrast, 10½ month old infants segmented words with both kinds of stress
pattern appropriately. That is, infants use prosodic as well as statistical strategies to
identify words, but the reliance on these change with age: up to 7 months they rather use
statistical than prosodic cues, and around 9 months they start to identify words predominantly
on the basis of initial word stress (Thiessen & Saffran, 2003). When conflicting
with statistical cues 8 to 9 month old infants seem to favour prosodic stress cues
(Johnson & Jusczyk, 2001) or phonotactic cues (Mattys, Jusczyk, Luce, & Morgan,
1999). Recently, research on adult speech processing focused on how listeners integrated
probabilistic information from numerous sources (Seidenberg, 1997). The use of
language-specific phonetic and phonotactic cues together with prosodic cues (as wordstress)
was demonstrated to improve performance in computational models of word
segmentation (Swingley, 2005). Moreover, certain prosodic features in continuous
speech, such as stress pattern, pauses or vowel lengthening may cue boundaries between
words and phrases (Gleitman, Gleitman, Landau, & Wanner, 1988; Morgan, 1996;
Morgan, Meier, & Newport, 1987).