Computational Linguistic Motivations for a Finite-State Machine Hierarchy

Abstract

Given the prominence of finite-state techniques in natural language processing, toolkits are required that can provide finite-state capability for applications in this domain. For example, work in the context of Multi-Agent Speech Recognition requires such finite-state capability. In particular, an interface is required both for processing finite-state representations and for automatically acquiring such representations. The finite-state representations are the traditional finite-state machines together with their stochastic equivalents. The representations are specified in XML which is chosen both for clarity and for reusability and portability across different platforms and applications. Motivated by these requirements this paper presents a Finite-State Machine Hierarchy. The hierarchy is an extendible object-oriented inheritance tree where each class in the hierarchy represents a particular finite-state machine. The hierarchy has been used to deliver the required interfaces with support for XML marked up machine structures.