Learning Phonemes: How Far Can the Input Take Us?

Many studies on developmental speech perception (e.g. Werker & Tees 1984, Kuhl et al. 1992) have documented changes in speech perception that occur during an infant’s first year of life. These changes are generally understood to reflect the phonemic structure of the native language (Best et al. 1988, Liberman et al. 1957). There is little research, however, on the phonological abstractness of these initial phonetic categories acquired in infancy. One study by Jusczyk and colleagues (1999) found that 9-month-old infants are sensitive to whether or not a set of sounds shares phonological features, indicating that by this young age infants have already developed featural representations. The question we ask in the present research is whether phonetic categories are initially represented as bearing abstract, contrastive features, or whether additional information or experience is required before a learner will develop featural representations. It might be the case that infants represent speech sounds as bearing potentially contrastive features immediately, as soon as they learn a speech sound category. For example, an infant who learns that /d/ and /t/ are contrastive in their language, might immediately assume that in their language voicing is contrastive. This assumption might be helpful in language acquisition, because some speech sounds are more frequent than others. If alveolar sounds are produced more frequently than velar sounds in a given language, then an infant learning that language might have good evidence that /d/ and /t/ are contrastive before they have heard enough examples to know whether /g/ and /k/ are contrastive. Immediately representing speech sound categories in terms of features could enable infants to bootstrap less robust contrasts from those contrasts that are better represented. However, the above assumption might work against an infant, if the language in question has an asymmetrical phoneme inventory. The language might have a voicing contrast for alveolar sounds, while only having voiceless sounds at labial and velar places of articulation. So hypothesizing that voicing is contrastive in such a language would turn out to be misleading. To avoid this garden path, infants might start with a more conservative hypothesis, and not develop abstract, featural representations until they have good evidence that there are multiple analogous contrasts in the language exhibiting the same feature.