Abstract
Charles Darwin hypothesized (in 1871) that mastery of language is possible on the basis of an " instinctive tendency " to learn languages, a characteristic of our species. This instinctive tendency has been the subject of extensive experimental research in recent decades. A first significant result has shown that newborns have a preference for human speech over other sounds or noises of comparable complexity, enabling selective attention to language facts from the outset (Voloumanos and Werker, 2007). But babies' abilities go far beyond the language/non-language distinction. A key figure in the study of " the initial cognitive state " of the child is Jacques Mehler, founder of a school of psycholinguists in France and Italy that has had a major influence on international research in recent decades. Mehler and his students have shown that, right from the start, babies are able to distinguish languages (even languages they've never heard!) on the basis of certain rhythmic properties that are immediately accessible to them (Mehler et al., 1988; Mehler and Dupoux, 1990). But how does the baby learn the specific properties of the language (or languages, in the case of multilingualism) to which he is exposed ? For example, the distinctive features used to distinguish words in his future language ? Janet Werker has shown experimentally that this learning takes place " by forgetting " (Werker and Tees, 1984) : at birth, the baby is sensitive to all the major distinguishing features used by human languages, but by the age of 8-10 months, it will only be sensitive to the features used by the language to which it is exposed. For example, the " English-speaking " baby loses, or " forgets " the distinction between dental t and retroflex t, to which he was nevertheless sensitive at birth, whereas the baby exposed to Hindi retains this contrast, which is used by his target language. This conception of learning as a selection among mind-generated options is surprising in relation to the models of the empiricist tradition, but in line with the rationalist tradition. It is also compatible and consistent with models of learning adopted by neuroscientists such as Jean-Pierre Changeux (1981, 2007) and Stanislas Dehaene (2017) on the basis of well-established experimental data in cognitive neuroscience.
