Is Infants' Learning of Sound Patterns Constrained by Phonological Features?

Abstract

Abstract Phonological patterns in languages often involve groups of sounds rather than individual sounds, which may be explained if phonology operates on the abstract features shared by those groups (CitationTroubetzkoy, 1939/1969; CitationChomsky & Halle, 1968). Such abstract features may be present in the developing grammar either because they are part of a Universal Grammar included in the genetic endowment of humans (e.g., CitationHale, Kissock and Reiss, 2006), or plausibly because infants induce features from their linguistic experience (e.g., CitationMielke, 2004). A first experiment tested 7-month-old infants' learning of an artificial grammar pattern involving either a set of sounds defined by a phonological feature, or a set of sounds that cannot be described with a single feature—an "arbitrary" set. Infants were able to induce the constraint and generalize it to a novel sound only for the set that shared the phonological feature. A second study showed that infants' inability to learn the arbitrary grouping was not due to their inability to encode a constraint on some of the sounds involved. ACKNOWLEDGMENTS This work was supported by NICHD R03 HD046463-0 to AS and from funds from Purdue University to AC and AS. Portions of this work were presented at the First International Conference on the Acquisition of Spanish as a First and Second Language in Rosario, Argentina; the 2007 Linguistics Society of America Meeting in Anaheim, California; and the 32nd University of Pennsylvania Linguistics Colloquium in Philadelphia, Pennsylvania; as well as at informal meetings at Purdue University, the University of California at Berkeley, and the University of Illinois at Urbana-Champaign. We would like to thank these audiences, as well as Alexander Francis, Jeff Mielke, Elliott Moreton, and three anonymous reviewers for helpful comments, though all mistakes remain our own. Notes 1An interesting question is whether infants only remain sensitive to features that serve a discriminatory function or whether they also pay attention to those involved in allophonic variation, which may be defined broadly as features involved in the positional variability of sounds. There is some indication that, although discriminability is reduced (CitationGoto, 1971; CitationPallier, Bosch, & Sebastian-Galles, 1997; CitationPegg & Werker, 1997), adults may still be sensitive to features involved in allophonic variation although less so than speakers of languages for which the same values are contrastive (CitationBernard et al., 2006). 2It should be apparent, however, that it is not the case that perceptual explanations must rely on simple acoustic descriptions. For example, infants learning the voiced-initial rule in Saffran and Thiessen's study cannot have encoded something like "VOT was less than 20 ms", given that there is overlap between the VOT values of voiced and voiceless consonants across place of articulation; e.g., voiced velar stops may have a VOT value larger than voiceless labials stops. On the contrary, for this particular case, infants might have relied on the auditory discontinuity that underlies the short lag-long lag boundary, a discontinuity that varies with place of articulation. That is, even chinchillas perceive the categorical boundary between voiced and voiceless stops at different VOT values depending on the place of articulation, suggesting that this discontinuity may respond to a basic property of mammals' auditory systems (CitationKuhl & Miller, 1978). 3In word-initial position, however, this cue would be absent in most dialects of English, where even phonologically voiced stops show no energy whatsoever during closure. This is the case in our stimuli, where only one token in one familiarization exhibited some prevoicing. Therefore, in order to propose that our infants' joint category of stops and nasals is based on the acoustic similarity of absence of energy above the second harmonic, we would have to assume (a) that our infants hear prevoiced stops in their ambient language, upon which they form their 'stop category'; and (b) that during the study they access an abstract representation such that e.g. the b[short lag] and k[long lag] they hear during familiarization both map onto that 'stop category' that is characterized by a voicing lead, and is thus similar to nasal consonants. Finally, while the presence of energy above the second harmonic may be a good cue to the presence of a fricative, the absence of that energy may not justify categorizing stops and nasals together, since a similar argument could be used to categorize stops and vowels together, neither of which has a high concentration of energy above the third harmonic. We thank Alexander Francis for pointing this out to us. 4One reviewer suggested that the degree to which familiarization and test onsets overlapped in place of articulation may explain our results for 7-month-olds. For instance, in Order A of the natural condition, the legal test onset /k/ shared its phonetic place of articulation with the familiarization onset /g/ while the illegal test onset /Σ/ did not, so that sheer novelty effect of place of articulation would predict the result in this condition. As an analysis of the stimuli in the Appendix shows, this explanation, however, would predict either order effects or a null result in both conditions, given that only in Order A of the natural condition and Order B of the arbitrary one is there a difference in number of segments that overlap in place of articulation. In Order B of the same condition, stop-initial trials share two (coronal and velar) but so do fricative-initial trials (coronal and labial). In Order A of the fricative and nasals condition, there is no place overlap between familiarization and test items, while for the Order B stop-initial trials the coronal place is shared, but fricative-initial trials share both the coronal and the labial. In other words, if one collapses across order within each condition and calculates the difference, the Natural condition should show a preference for illegal trials (Illegal 2, Legal 3) and the Arbitrary condition should show a preference for legal trials (Illegal 1, Legal 2).