Assessing phonetic compliance

This study is part of a broader project investigating the processes involved in the acquisition of new phonetic control regimes in L2 learning. Our main experimental paradigm consists in attempting to 'shape' the speakers' productions in new ways based on what they are made to hear. Quite unsurprisingly, our first results ended up in a large amount of inter-individual variation, even though the participants were matched in several respects including linguistic experience. This is in agreement with evidence deriving from many studies focused on the 'external factors' potentially determining individual variation in L2 phonetic learning, such as age of learning, age of arrival, or amount of L2 use (among others: Piske et al., 2001; Moyer, 2004). It is also to be linked with the results of studies focused on the effects of characteristics inherent to the learner, either in the cognitive domain (Golestani & Zatorre, 2009; Francis & Nusbaum, 2002), or in the broader psychological domain (motivation, personality factors, etc.) (Dörnyei, 2009 ; Cohen & Macaro, 2007).
Moreover, there is a growing body of evidence in favor of the existence of a specific ability to produce and perceive foreign speech sounds, called 'phonetic talent' (Jilka et al., 2007 ; Dogil & Reiterer, 2009). In its strict sense, phonetic talent denotes an 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 depart between initial 'talent' and the other interacting variables that have presided to each individual's language development and may still influence his productions in, e.g. an experimental imitation task. Also, in differential psychology, a separation is clearly made between 'gift', an untrained and spontaneously expressed superior natural ability, and 'talent', that progressively emerges from the transformation of this high aptitude into a well-trained and systematically developed skill (Gagné, 2003). Moreover, to our knowledge, the only attempts reported in the literature to measure individual intrinsic abilities in dealing with foreign sounds are exclusively based on subjective (perceptual) data from native speakers of a different language to the individual's L1 (Reiterer et al., 2011).
Given these conceptual instabilities and methodological weaknesses, we will adopt here a pragmatic view, with no strong hypothesis about innateness. We then focus on the end result of the process, i.e. on the spontaneous ability of adult speakers to accurately produce speech sounds similar to models they are faced with. We call this ability 'phonetic compliance'. We posit that it varies among individuals and can be assessed in terms of gradient. This paper is a methodological account aiming both at studying the feasibility of phonetic compliance assessment (through 3 different quantification methods) and at preparing further research oriented towards a better understanding of the underlying mechanisms.
For this experiment, 10 Belgian French speakers (5 male, 5 female) have been submitted to an imitation task (instructions: 'repeat as faithfully as possible, as if it was a sound from a foreign language') of 94 synthetic vowels regularly spaced in a Mel-scale F1*F2*F3 space (6 repetitions). They also produced 10 realizations of each of the 10 French oral vowels. Formant frequencies have been detected using Praat under the supervision of 2 trained phoneticians.
Basically, the similarity between models given to the speaker and productions imitating the models can be estimated by the deviations of the productions from the targets. For a given speaker in an experiment with S vocoïd models, P productions of each stimulus, and provided that for each vowel, I (Fi) formants are taken into account for each production, the sum of the Euclidean distances target/realization for all stimuli and all productions can express globally the success of the speaker in the task. In formula 1, I1 tends towards zero when the productions of the speaker tend, globally, towards the target in the vocalic space. In other words, I1 is zero when compliance is maximal.
Formula 2 is based upon the same principle, except that in this case, the inverse of the distance (-1/2 exponent) has been taken into account, in order to obtain a number with variations positively correlated with compliance. Furthermore, in this case, the speaker has been calibrated using his/her realizations of L1 vowels. The language has V vocalic phonemes and the speaker has realized v tokens of each. It is therefore possible to identify zones of the vowel space corresponding to usual productions of the speakers, and zones where he/she is not used to produce vocalic sounds. The idea in formula 2 is to give higher reward to the success in imitating when imitation takes place in a region of the vocalic space the speaker does not use in his/her current practice of his/her L1. This is the reason for the weighting by the multiplicative term. It consists in the logarithm of the product of all the distances between a given production and each vowel's cluster centroïd: the multiplicative term tends toward zero when at least one distance production/centroïd tends toward zero. Thus, for a given realization, the product is large if the production resembles the target and if it is produced in a zone far from the ones corresponding with the speaker's L1. I2 may be viewed as a modulation of phonetic compliance assessment through individual L1 phonology.
In formula 3 (where 'var' stands for 'variance'), the similarity between target and production is no more the main point, and the approach is more statistical: it is based upon the analysis of variability in the imitation task. When a speaker tries to imitate a model, he/she produces realizations that fall around it in the reference space. If the speaker's compliance is high, his/her variability around the model in the reference space is random, and if no other source of variance is active, the variability is constant whatever the stimulus. On the other hand, if the speaker is strongly influenced by his/her L1, one can suppose that his/her variability will vary from one stimulus to another, depending on whether the stimulus is close or not to a region of the vowel space present in L1. I3 should therefore tend toward zero (all variances equal) in a speaker with good compliance. Formula 3 uses individual variability in phonetic processing as a source of information on phonetic compliance.
The 3 formulae have been currently applied to the productions of 4 speakers. Results show both convergence and specificities in how the indices characterize phonetic compliance. S4 is the most compliant since his productions are the closest to the models (I1), the furthest apart from his typical L1 realizations (I2), and the most homogenous in their variation (I3) (see Fig.1 for an illustration). S1 and S2 are the less compliant speakers, S2 being sligthly better at leaving her L1 territory (I2) but, doing so, increasing greatly the variation of the variances associated with her realizations of the different models (I3). Based on the results collected on 10 speakers, we will discuss at the conference how to integrate the complementary information given by the 3 indices to further elaborate on the concept of phonetic compliance.
References
Cohen, A.D. & Macaro, E. 2007. Language learner strategies : thirty years of research and practice, OUP.
Dogil, G. & Reiterer, S. 2009. Language Talent and Brain Activity. Trends in Applied Linguistics 1, New York, Mouton de Gruyter.
Dörnyei, Z. 2009. The Psychology of Second Language Acquisition, OUP.
Francis, A.L,. & Nusbaum, H. C. 2002. Selective attention and the acquisition of new phonetic categories. Journal of Experimental Psychology: Human Perception and Performance, 28(2), 349-366.
Golestani, N. & Zatorre, R.J. 2009. Individual differences in the acquisition of second language phonology. Brain and Language 109, 2-3, 55-67.
Jilka, M., Anufryk, V., Baumotte, H., Lewandowska, N., Rota, G. & Reiterer, S. 2007. Assessing Individual Talent in Second Language Production and Perception. 5th ISALSS. Florianópolis, Brazil, 243-258.
Moyer, A. 2004. Age, Accent, and Experience in Second Language Acquisition: An Integrated Approach to Critical Period Inquiry. Clevedon, Multilingual Matters.
Piske, T., MacKay, I. & Flege, J. 2001. Factors affecting degree of foreign accent in an L2: A review. Journal of Phonetics 29, 191-215.