LEARNING THE IMPOSSIBLE - THE ACQUISITION OF POSSIBLE AND IMPOSSIBLE LANGUAGES BY A POLYGLOT SAVANT

We report on the case of a polyglot savant (Christopher) who has a rem arkable talent for learning and translating languages. Building on pre vious work which had established both the range of languages at Christ opher's command and the extent to which his linguistic knowledge was i ntegrated into his cognitive ability, we taught him two new languages for which we controlled the input. We had two main aims: the first was to test the hypothesis (within one version of the Principles and Para meters framework) that parameter resetting is not an option available to the second language learner; the second was to accrue further evide nce for or against Fodor's modularity hypothesis and cast light on the possible range of interactions between linguistic and 'central' cogni tive processes. The languages chosen were Berber, an Afro-Asiatic lang uage spoken in North Africa, and Epun, an invented language deliberate ly devised to contain constructions which violated universal grammatic al principles. In Christopher's acquisition of Berber we gleaned evide nce from a variety of phenomena, including word order, null subjects, that-trace effects, wh-island violations and cliticisation, that his l earning was conditioned by a combination of transfer effects from Engl ish and principles of UG, rather than by the effect of parameter reset ting. In Christopher's acquisition of Epun we began with a core of 'no rmal' constructions, designed to make him feel at home in the new lang uage, and then proceeded to investigate a range of impossible construc tions, both structure-dependent and structure-independent. In the form er case, we concentrated on negative sentences, constructed with no ov ert negative morpheme, and past-tense sentences which involve unattest ed and putatively impossible word-order differences. In the latter cas e, we concentrated on a rle of emphasis that involved counting words, and a form of agreement which again violated putatively universal gene ralisations. In each case we compared Christopher's performance with t hat of a small group of controls. The results were complex, but we thi nk we can justify an interpretation which lends support to both the ma in hypotheses being tested.