Parsing CVCV words into syllables: a perceptual experimental study to test the language dominance effect

Over the past fifty years numerous experimental researches have been published on the syllable. Obviously the interest for this object is not a recent topic, but as international conferences (NYC, USA, 2008; Pescara, Italy, 2013) entirely dedicated to the syllable are organized in this century, not to mention several books on the syllable published quite recently (Gordon, 2006, Duanmu, 2009, Cairns et al. ed., 2011), one can assert it is still a central contemporary topic. The syllable is known to play a role in perception (Mehler et al., 1981, Pallier, 1994), at least for French or Spanish listeners, as a few experiments failed to be reproduced with English data (i.e. Cutler et al., 1986). Phonology would help us understand the differences between French and English syllabification preferences by stating that they behave differently as a function of contexts, probably mainly because of stress pattern differences and vowel quality differences. Not all theories admit that the syllable can play the role of a favored parsing unit since it has not yet been shown that the syllable would have some sort of non variable acoustic characteristics, but neither has it been found for phonemes nor phonetic features - if one agrees it would be almost impossible to observe similar acoustic patterns for similar syllables, one cannot infer from such observation that the syllable does not exist. Although sounds overlap in speech no one says they do not exist. We assume here the syllable would both have a phonological and a phonetic existence (as Ridouane et al., 2011) but we hypothesize that the phonological syllable reference we have in mind would have an impact on the way we are to perceive and parse the phonetic syllables we are given to hear. It is particularly relevant to analyze the notion of syllable from an experimental point of view. In this talk, we will discuss recent experimental results of a perceptual experiment. The corpus was composed of recorded CVCV (pseudo)words read by American English and Parisian French native speakers. The listeners had to choose between 3 given potential answers the one that would best correspond to what they had heard. For example, if a French listener was to hear the word "tipi" pronounced by a French speaker or the word "tippy" pronounced by an American English speaker, he/she had to choose among the three given answers "tip-i", "tip-pi" or "ti-pi", the one which would be the closest to what he/she had perceived. If an American English listener was to listen to the same words, he/she would have to choose between the following answers: "tip-y", "tip-py" or "ti-py". The listeners were English L1 speakers (L2 French) or French L1 speakers (L2 English) with variable language proficiency levels. They all had to answer a questionnaire about various elements which could help us classify them on a gradient bilingual dominance scale (Dunn et al., 2009), among them: their AOA (age of acquisition), self-evaluation of their fluency and frequency of use of each of the languages they could speak, how long participants had lived in L2 speaking countries and how recently, their L1 and L2 in schooling, their L2 use at work and L2 use in the family or friends environment. The results suggest that syllabification preferences differ from one language to the other. Parisian French dominant speakers tend to prefer open syllabification answers (CV-CV) while American English dominant speakers tend to prefer closed (CVC-V) or ambisyllabic (CVC-CV) syllabification answers. We could argue that the diverging behavior of French dominant and American English dominant listeners reflects a difference in their internal syllabic structure at work in perception. Therefore L1 dominant speakers would perceive the syllabification of their L2 with their L1 phonological filter, which is consistent with the results of other psycholinguistics' studies (i.e. Ali et al., 2011, for anglophones perceiving Arabic). As the perceptual experiment was followed most of the time by a speech production recorded reading task of the same corpus, an acoustic analysis of syllable constituents will be carried out. The production and perceptual data are meant to be matched and compared in the future. As we know errors could easily be made in the interpretation of acoustic data (Martin, 2008), we have been fairly cautious about the data and experimental conditions but it is obvious that such results should be validated by further convergent results in similar and different experimental conditions. Nevertheless, we believe we here have a clue as to how Parisian French and American English listeners could prefer parsing CVCV words into syllables. With this empirical method focusing on perception, the dominant language of the listener tends to have an impact on the listener's syllabification preferences