Section: New Results

Allophony and word segmentation in language acquisition models

Participants : Luc Boruta, Benoît Crabbé.

Allophonic rules are responsible for the great variety in phoneme realizations. Infants can not reliably infer abstract word representations without knowledge of their native allophonic grammar. We have explored the hypothesis that some properties of infants' input, referred to as indicators, are correlated with allophony. First, we provide an extensive evaluation of individual indicators that rely on distributional or lexical information. This evaluation relies on a phonetically transcribed corpus, generated automatically from a phonemically transcribed English, French and Japanese child-directed corpus. As such corpora do not exist as such, we used automatically extracted allophonic grammars of various sizes leading to various granularity levels, using our own allophonic rule extraction algorithm [57] . Then, we present a first evaluation of the combination of indicators of different types, considering both logical and numerical combinations schemes [23] . Though distributional and lexical indicators are not redundant, straightforward combinations do not outperform individual indicators.

Models of the acquisition of word segmentation are typically evaluated using phonemically transcribed corpora. Accordingly, they implicitly assume that children know how to undo phonetic variation when they learn to extract words from speech. Moreover, whereas models of language acquisition should perform similarly across languages, evaluation is often limited to English samples. Using the phonetically annotated corpora described above, that cover three typologically different languages, we evaluated the performance of state-of-the-art statistical models given inputs where phonetic variation has not been reduced. We have measured segmentation robustness across different levels of segmental variation, simulating systematic allophonic variation or errors in phoneme recognition. We have shown that these models do not resist an increase in such variations and do not generalize to typologically different languages. From the perspective of early language acquisition, the results strengthen the hypothesis according to which phonological knowledge is acquired in large part before the construction of a lexicon.