Unambiguous Triggers

Abstract

Triggers for parameter setting may be ambiguous. Strategies for dealing with ambiguity include guessing, parallel processing, and waiting for unambiguous input. The Trigger Learning Algorithm of Gibson and Wexler (1994) is a guessing system. Gibson and Wexler show that under some reasonable assumptions it may never attain the target grammar. I propose instead a deterministic device that waits for unambiguous triggers to set parameters. This learner must be able to detect parametric ambiguity. This is not possible on the classical conception of triggers as sentences that flip parameter switches. It is possible if a trigger (and the parameter value it triggers) is a piece of tree structure, perhaps a single feature, made available by Universal Grammar, and adopted into the learner's grammar if input sentences cannot be parsed without it.