Categorical patterns, phonetic integration and within-category variability in the recognition of spoken words in a second language

It has been shown on several occasions that speech perception in a second language (L2) is often tainted by the phonetic and phonological structures of a listener’s native language (L1). As languages vary in their phonological inventories, listeners often have to develop new phonological categories during acquisition in order to recognize words and phonemes in their L2. Recent evidence on the perception of coarticulation and phonetic variability (Beddor et al., 2013; McMurray et al., 2008) has shown that not only coarticulatory cues are perceived and used by listeners during spoken word recognition, but also that speech perception can be gradient. For example, McMurray et al. (2008) found that small variations in VOT (5 ms steps on a 0-40 ms continuum) led to gradient patterns of fixations in a visual world paradigm: the longer the VOT values, the greater the proportions of fixations to the voiceless target (e.g. peach, as opposed to beach), and inversely for shorter VOT values. This yielded to the hypothesis of continuous integration of fine-grained phonetic information during spoken word recognition, as listeners did not display strictly categorical patterns of spoken word recognition. These studies were conducted with L1 listeners and phenomena that regularly occur in their L1. However, as L2 speech perception requires learning at a later age, phonetic details that are stored by L1 speakers for a contrast may be different from those stored by L2 speakers for the same contrast. In this study, we use the contrast in Canadian French (CF) between phonological nasal vowels (Ṽ; dent /dã/ ‘tooth’) and coarticulatory nasalized vowels (VN; dame /dam/ ‘lady’) to investigate L2 perception (L1 English listeners) of variable phonetic realization and gradience in word recognition. The interest of this contrast lies in the fact that CF uses vowel nasality as a contrastive property while English vowel nasalization does not contrast for lexical items. Several phonetic cues serve to distinguish Ṽs and VNs in CF. For example, some Ṽs can be optionally diphthongized, and the tongue position differs between nasal and oral counterparts of the same vowel (Delvaux, 2006). One of the main differences between the two vowel types, however, is the timing of velum lowering. Phonological Ṽs are generally nasalized (i.e. the oral and nasal tracts are coupled together) for a longer period of time than coarticulatory VN sequences (Desmeules-Trudel, 2015). However, there is also within-category variability in CF which means that duration of nasalization can overlap for both Ṽs and VNs, leading to potentially similar realizations of both vowel types. We used nine Ṽ-VN word pairs as stimuli, and cross-spliced (part of) Ṽs onto corresponding VNs to create 45 target words. This led to nine five-step continua of words that were nasalized for 0% of their duration, 20%, 50%, 80% and 100%. For example, in the 50N condition, the target word contained the initial consonant of the VN word, 50% of the vowel from the VN word, 50% of the vowel from the Ṽ word, and a 50 ms nasal coda from the VN word.