e., pitch, vowel quality, timbre, sociolinguistic variation) and production-specific variables (i.e., prosody) that are not associated with lexical contrast

(e.g., there are no English words that differ only by pitch). As these do not cue phonemic or lexical contrasts, much work in speech perception has been devoted to explaining how listeners are able to overcome such variability to arrive at the underlying meaning (e.g., Perkell & Klatt, 1986). Alternatively, it is possible that the auditory system would need to retain, rather than normalize, multiple forms of acoustic information to arrive at the correct categories check details (Goldinger, 1998; Klatt, 1979; Pierrehumbert, 2003; Pisoni, 1997). Prior work on this has focused on whether listeners use such detail during online perception (Creel, Aslin, & Tanenhaus, 2008; Goldinger, 1998; signaling pathway Johnson, 1990; Ryalls & Pisoni, 1997). Importantly, it has been shown that infants might map both indexical and phonetic information of words in

early word learning (Houston & Jusczyk, 2000). This suggests that irrelevant cues, such as indexical information, may help in the acquisition of speech contrasts. Indeed there is evidence that variability along nonphonemic dimensions may help identify the underlying invariant structure of speech. Singh (2008) has shown that variation in the affective quality of speech improves word segmentation in infancy. Hollich, Jusczyk, and Brent (2002) report that word segmentation abilities are improved by multiple-talker familiarization

in older infants. However, both studies looked at broad segmentation abilities, not at the perception of a single phonetic feature (e.g., voicing) in a highly ambiguous context. This was explicitly tested in Experiment 3. The exemplar set used in Rost and McMurray (2009) was highly variable in noncontrastive aspects of the signal (such as vowel quality or pitch), but the range of variability within these dimensions did not differ between /buk/ and /puk/. If infants NADPH-cytochrome-c2 reductase use highly variable information to isolate relatively invariant elements of the signal, they should succeed at the switch task when exemplars contain lots of variability, but minimal within-category variability in contrastive cues. Recruitment and exclusion criteria were the same as in Experiment 1. Twenty-three infants participated, and data from seven were excluded from analysis for experimenter error (4), fussiness (2), and failure to habituate (1). Sixteen infants (9 boys; M age = 14 months 8 days, range = 13 months 5 days to 15 months 1 day) were included in the experimental analysis. Stimuli consisted of the original set of 54 exemplars recorded from 18 speakers from Rost and McMurray (2009). These were modified to maintain variation in all of the noncriterial (indexical and prosodic) cues but eliminate within-category variation in VOT.