Abstract :
[en] This paper investigates speaker-specific abilities to acquire the phonetics and phonology of a foreign language. Performances in second language (L2) phonetic learning exhibit a large amount of interindividual variation (for a review, see Piske et al., 2001; Ehrman et al., 2003), which causes can be either related to contextual factors affecting each individual's linguistic experience (e.g., 'age' factors, Flege & MacKay, 2011) or to 'internal factors', i.e. factors that are inherent to the learner, such as cognitive abilities (Golestani & Zatorre, 2009), psychological aspects (Dörnyei, 2009), and linguistic abilities (Dogil & Reiterer, 2009). Contextual and internal factors are considered as interacting with each other in a complex way, as well as interacting with the specificities of the achievements that are required from the learner in a given situation. The latest, interactionist, models of second language learning aptitude propose a componential framework for mapping the interactions between tasks demands, context properties and learner factors (Robinson, 2009).
Recently, Jilka and collaborators proposed a comprehensive approach for testing 'phonetic talent', which assesses phonetic abilities of adult experienced L2 learners with a special emphasis on pronunciation (Jilka, 2009). Phonetic talent denotes a largely innate, neurobiologically grounded, individual skill which is part of general language aptitude, but may be separated from other specific linguistic skills such as grammatical talent in L2. Although phonetic talent is an appealing concept, its objective assessment is hindered by the difficulty to distinguish initial predisposition from the other interacting variables that have presided to each individual's language development and may still influence his productions in a specific task, independently of his proficiency in any L2.
In this paper, we adopt a pragmatic, task-oriented view, with no strong hypothesis about innateness nor any attempt at enlightening the complex etiology of phonetic proficiency. We thus focus on the end result of the process, i.e. on the spontaneous ability of adult speakers to accurately reproduce non familiar speech sounds they are faced with, notwithstanding the differences between these sounds and the sounds they are used to process. This paper is an exploratory methodological account aimed at developing assessment tools for distinguishing individuals in terms of this specific ability.
The participants were 10 native French speakers from Belgium, 5 female, 5 male, aged 24 to 42, who were administered a comprehensive 'linguistic questionnaire' in which they detailed their knowledge and experience with foreign languages. The experimental paradigm comprised several parts administered in a single session, including the production of L1 sounds to be used as control sounds in data analysis, a task of reproduction ('as faithful as possible, as if it was a sound from a foreign language') of the vowel stimuli set and a task of reproduction of the consonant stimuli set. A complementary data collection was carried out a few weeks later on the vowel stimuli set only (four speakers only).
The vowel stimuli set was made of 94 synthesized vowels (Klatt, 1980) that were evenly distributed over a mel scale F1*F2*F3 acoustic space. Total vowel duration (200ms) and F0 contour (from 110Hz to 90Hz) were kept constant across stimuli. The consonant stimuli set was made of 9 [ka] synthetic syllables, where [k] varied in voice onset time (VOT) from virtually nil VOT [k] to long positive VOT [kh]. The [a] vowel was the same across stimuli, with a duration of 160ms and an appropriate descending F0 contour. The speech productions from the participants were segmented manually. Formant frequencies were first automatically detected using Praat, then manually verified by two trained phoneticians.
Data analysis consisted in (group and individual) comparisons between the acoustic properties of the stimuli and those measured in the reproductions. Comparisons were carried out through linear regression analysis (see Fig.1 and Fig.2 for group results) as well as distance-based measurements, i.e. the absolute difference (in ms) between Stimulus VOT and Response VOT for consonant productions and, for vowel productions, the euclidean distance between the stimulus and the response in the three-dimensional F1*F2*F3 acoustic space defined in mels ('Index1'). The complementary data collection provided the additional data (3 more repetitions of each vowel stimulus per speaker) necessary for the computation of two additional indices of the efficiency of the speakers in the reproduction task: 'Index2', which was designed to take into account the speakers' ability in the reproduction task to produce vowels which formant frequencies dramatically differ from those of their L1 targets, and 'Index3', which assesses the distribution of the responses over the acoustic space.
Data analysis resulted in the development of 12 complementary indicators that allowed to compare individuals in terms of overall, and specific aspects of, their performances in the vowel and consonant reproduction tasks (for an illustration, see Fig.3).
We will discuss at the conference the possible mathematical refinements that may benefit the quantification techniques used in this paper, as well as the usefulness of the methods developed here to investigate the individual ability to reproduce non familiar speech sounds, in relation with the cognitive mechanisms involved and other, potentially related, behavioral factors such as individual strategies in L2 learning, or propensity to phonetic convergence in L1.
