6 AG 4: Learning meets acquisition: the learnability of linguistic frameworks from formal and cognitive perspectives

The workshop brings together researchers working on the learnability of linguistic models from a formal point, with those working on the models ‟ cognitive adequacy. In general, studies on the learnability of language account for how grammar and lexicon of a language can be learnt, and by what means. To give an example, considerate progress has been made recently in connectionist-based frameworks such as Opti-mality Theory (Prince & Smolensky 1993) and Harmonic Grammar (Legendre et al. 1990). Most learnability models within OT deal with the learning of the grammar: learning of constraint hierarchies, and learning of constraints themselves. In most of these approaches, lexical information is already given. Other OT approaches tackle the learning of parts of the lexicon. Differences between approaches include whether lexicon and grammar are learnt in turns (offline) or in parallel (online), or whether the OT grammar to be learnt is traditional or stochastic. Current approaches to learnabil-ity within HG include learning constraint weights, by using learning algorithms such as the perceptron algorithm. Yet, formal results have been only seldom tested against empirical data from language acquisition research