Activation Across Three Lexicons
Christine Shea, University of Iowa
A fundamental challenge of communicating in more than one language is that the speech signal often calls for different interpretations, depending on which language is being spoken. When multilingual listeners hear words in one of their languages, multiple candidates are activated across all their languages. Sublexically, however, differences do exist and can serve to inhibit unwanted activation if they are perceived by the listener. A well-studied example of such sublexical differences is VOT across languages such as Portuguese and Spanish, compared to English. English has long-lag VOT while Spanish and Portuguese have short-lag VOT for the same phonological categories. Another sublexical difference is vowel nasalization. In Portuguese, vowel nasalization can be phonological while in Spanish and English, it is allophonic. Sublexical ambiguities of this type pose an interesting question for multilingual speech processing. Specifically, what happens when a trilingual (e.g., Spanish-Portuguese- English) listener hears input that is ambiguous between two of her languages? Does ambiguous input activate language-specific lexical representations? To answer these questions, we recruited L1 Spanish and L1 Brazilian Portuguese trilingual participants (English was either L2 or L3) living in Uruguay and Brazil.
We first determined how listeners identify the multilingual stimuli by means of a categorization task. Subsequently, we determined how listeners classify the same input as belonging to specific languages. Stimuli were bisyllabic nonwords of the form [Ce(N).Ca]. The initial consonant was drawn from a [b-p] voicing continuum (-40ms to 40ms, 10ms intervals) and spliced onto one of three vowels: full nasal vowel (contrastive in Portuguese), nasalized vowel (allophonic in English, Spanish and Portuguese) or oral vowel from each language. For example, the nonword [bẽmpa] with -30 VOT included the nasal vowel and negative VOT characteristic of Portuguese while the nonword [phepa], included VOT of at least 30ms, which phonetically corresponds to English. Participants then completed an auditory form priming task in which they heard the nonword syllables (prime), followed by a real-word (target) from English, Spanish or Portuguese and had to identify the language of the target.
Preliminary analysis shows that for phonetically ambiguous primes, RTs were longer and accuracy rates lower compared to non-ambiguous primes. We discuss the relevance of these results for self-organizing models of language selection in multilingual lexical activation.