Metathesis Is Real, And It Is A Regular Relation

Abstract

Regular relations are mathematical models that are widely used in computational linguistics to generate, recognize, and learn various features of natural languages. While certain natural language phenomena – such as syntactic scrambling, which requires a re-ordering of input elements – cannot be modeled as regular relations, it has been argued that all of the phonological constraints that have been described in the context of Optimality Theory can be, and, thus, that the phonological grammars of all human languages are regular relations; as Ellison (1994) states, "All constraints are regular."

Re-ordering of input segments, or metathesis, does occur at a phonological level. Historically, this phenomenon has been dismissed as simple speaker error (Montreuil, 1981; Hume, 2001), but more recent research has shown that metathesis occurs as a synchronic, predictable phonological process in numerous human languages (Hume, 1998; Hume, 2001). This calls the generalization that all phonological processes are regular relations into doubt, and raises other questions as well. If phonological grammars may include non-regular constraints, why is there no evidence of other non-regular constraints – such as the reversal of all input segments in the output?

This paper will review the evidence for metathesis as a synchronic, productive process that must be included in OT phonologies, and discuss how this evidence supports the existence of limitations on re-ordering that allow metathesis to be modeled as regular relations by defining the domains in which it operates, and proposing a regular family of LINEARITY constraints which describe all known cases of metathesis. Establishing that metathesis is a regular relation supports the claims by researchers such as Heinz (2001a) that the human phonological capacity is limited to regular relations, as well as providing the practical result that the well-known mathematics of these models can also be fruitfully employed to describe linguistic phenomena.