This book investigates the learnability of various classes of classical categorial grammars within the Gold paradigm of identification in the limit from positive data. Learning from structure and learning from flat strings are considered. The class of k-valued grammars, for k = 1,2,3,..., is shown to be learnable both from structures and from strings, while the class of least-valued grammars and the class of least-cardinality grammars are shown to be learnable from structures. In proving these learnable results, crucial use is made of a theorem on the concept known as finite elasticity. The learning algorithms used in this work build on Buszkowski and Penn's algorithm for finding categorial grammars from input consisting of functor-argument structures.

This volume is an outgrowth of the second Workshop on Logic, Language and Computation held at Stanford in the spring of 1993. The workshop brought together researchers interested in natural language to discuss the current state of the art at the borderline of logic, linguistics and computer science. The papers in this collection fall into three central research areas of the nineties, namely quantifiers, deduction, and context. Each contribution reflects an ever-growing interest in a more dynamic approach to meaning, which focuses on inference patterns and the interpretation of sentences in the context of a larger discourse. The papers apply either current logical machinery - such as linear logic, generalised quantifier theory, dynamic logic - or formal analyses of the notion of context in discourse to classical linguistic issues, with original and thought-provoking results deserving of a wide audience.