Second language learners and the variable speech signal

In the article “Age of acquisition and allophony in Spanish-English bilinguals” Barlow (2014) presents production data of /l/ from two groups of Spanish-English bilinguals, who differ on age of acquisition of English (before 5 years or after 6 years of age). Barlow’s contribution is a welcome addition to the relatively understudied field of allophone acquisition by second language learners. In what follows I expand upon issues touched upon by Barlow in her article and comment more generally on why the issue of variability in the speech stream (of which allophones in complementary distribution is but one type) must be addressed differently for L2 learners than for infants acquiring a first language. I restrict the discussion to perception and will primarily address issues key to adult (i.e., individuals who began to acquire their second language after the sound system of their first language is in place) second language acquisition. Research on how adults perceive nonnative sounds has received considerable attention over the past thirty years (see work by Flege and Best for the most influential models of L2 perception) and the vast majority of this work has looked at the way in which non-native sounds assimilate into native-language sound categories, independent of the context in which they occur (for an exception, see Levy and Strange, 2008). In a certain sense, it can be said that much of this research abstracts away from speech perception as it unfolds in real time (McMurray and Jongman, 2011). Part of the challenge realtime speech perception represents for L2 learners involves dealing with the way cooccurring sounds (or abstract contexts such as stress, see Shea and Curtin, 2011) affect each other or lead to variability, whether predictable or indexical in nature. The study of allophone acquisition represents an effort to break away from this tradition and can be included in the broader research program that examines how learners deal with the variability found in the input. Indeed, variability itself is “highly variable” and can be due to individual speaker differences, dialect differences, speech rate, and formality. These kinds of variability are often distinguished from allophonic variability that is the result of phonetic or phonological factors and tend to occur in a more across-the-board fashion in speech. In terms of L1 acquisition, part of learning a language’s sound system necessarily involves learning which sounds contrast and which do not. Research suggests that distributional knowledge and phonetic similarity play a key role in guiding infants toward identifying nonphonemic sounds in their language (see Seidl and Cristia, 2012 for an excellent overview; see Yeung and Werker, 2009, for work showing that a lack of lexical contrast can be used by infants to acquire allophones in non-contrastive distributions as well). For example, in a recent study, Seidl et al. (2009) examined the role of phonemic vs. allophonic contrasts in infant speech perception. They familiarized French-learning 11-month-old and English-learning 11and 4-month-old infants to syllables in which the final consonants conditioned the nasality of the previous vowel. In French, nasality is phonemic while in English it is allophonic. The results showed that French-learning 11-month-olds and English-learning 4montholds had a reliable pattern of preference while English 11-month-olds were insensitive to the patterning, orienting equally to syllables following and violating the familiarized patterns. The authors conclude that language-specific sensitivity to context-driven allophonic contrasts emerges as early as 11 months of age. In contrast, adult native listeners distinguish allophonic contrasts at a phonetic level less accurately than phonemic contrasts. For example, Pegg and Werker (1997), using an AX discrimination task, showed that native English-speaker adults’ performance on the allophonic contrast between voiced [d] and the voiceless unaspirated [t] was better than chance, but nonetheless worse than that on a phonemic contrast (for similar results see Whalen et al., 1997). In addition to perceiving the difference between two different phones, there is another important component to allophonic acquisition: its context-driven nature. Specifically, allophonic perception cannot be truly categorized as such unless the sounds occur in the context in which they are expected (or not, see Shea and Curtin, 2011 for details; Key, 2014). For example, Peperkamp et al. (2001), using the French [χ] [ ] alternations showed that French listeners could discriminate between allophonic segments in CV syllables but as soon as the CV syllables were put into their allophonic contexts, such discrimination disappeared. Thus, to truly speak of “allophone perception” listeners must be aware of the contrast but also the context in which it occurs. The mechanism by which infants build their sound categories is based upon tracking distributional frequencies across the speech stream (Maye et al., 2002). A number of laboratory studies reveal that such learning is possible in both infants and