Evidence for Classifying Metathesis Patterns as Subsequential

This paper presents a computational analysis of metathesis patterns that distinguishes three categories of metathesis that differ in their computational complexity. These categories are local metathesis, bounded long distance metathesis, and unbounded long distance metathesis. Using the formalism of finite state automata, it is established that the first two categories are subsequential, while the third category is not (in fact, it is not even regular). These terms will be discussed in more detail below, but the overall distinction is one of complexity: the subsequential class is more restrictive than the regular class. Assigning a pattern to the subsequential class then identifies it as less complex than a pattern that is non-regular. Furthermore, the patterns identified as subsequential are robustly attested in the world’s languages, whereas the non-regular patterns are much less common and in fact only attested diachronically. Thus this result suggests an upper bound for how complex a synchronic phonological pattern can be. The outline of this paper is as follows. Section 2 presents the Chomsky Hierarchy of language patterns and discusses recent findings and hypotheses for where to classifying phonological patterns on the hierarchy. Section 3 defines subsequential FSTs and demonstrates how they can be used to describe phonological patterns. Section 4 presents the computational analysis of metathesis patterns. Section 5 discusses the typological implications of the analysis, as well as the implications for learning. Section 6 indicates directions for future work and concludes.