Roni Katzir, T. O’Donnell, E. Rasin

Journal of Language Modelling

Paper

Abstract

Simplicity has long been central to philosophy of science, at least in the sense that all things being equal, a more parsimonious theory is better than a more complex one. In modern linguistics simplicity has played a particularly prominent role, with explicit discussion in Chomsky 1951, 1965, Halle 1962, Chomsky and Halle 1968, andmuch subsequent work. The prominence of simplicity in linguistic theory reflects the importance of learning in this domain: children acquiring a language must choose between many different grammars compatible with the input data, and an intriguing possibility is that their choice, perhaps like that of the scientist, is affected by considerations of simplicity. The present special issue considers the place of simplicity in grammar learning, focusing on recent computational and theoretical linguistic work but very much building on earlier foundations. In addition to discussing the use of simplicity, the papers in this collection touch on some of the challenges involved in turning simplicity from a guiding intuition into a concrete tool. For example, to what extent would such a tool be limited by the observation that simplicity is always stated with respect to a specific frame of reference? Which, if any, of the various notions of simplicity that have been proposed could support successful grammar learning, and would such a notion adequately model how children generalize from the primary linguistic data? Do observed typological generalizations regarding simplicity in linguistic systems arise from general considerations of stability of simple grammars under repeated iterations of learning across generations, or is there (also) a direct pressure for simplicity? Our hope is