DEP: Beyond Epenthesis

1 Introduction

The correspondence-theoretic constraint DEP (McCarthy and Prince 1995) is often informally described as a constraint against epenthesis. While DEP does militate against epenthesis in input-output mappings, it has a more technical sense when applied to other domains, such as base-reduplicant correspondence and output-output correspondence. The goal of this squib is to show that DEP constraints do more than block epenthesis. In the domain of base-reduplicant correspondence, the high ranking of DEP_BR can result in underapplication of deletion, as in the case of Tonkawa. In the domain of output-output correspondence, DEP_OO is known to require overapplication of deletion in paradigmatically related forms. All of these effects follow from McCarthy and Prince's definition of DEP as a family of faithfulness constraints that require a match between two strings that stand in correspondence.

This discussion sheds light on a long-standing ambiguity in classical Optimality Theory (OT): does the treatment of epenthesis really require a faithfulness constraint, or is markedness sufficient? On the face of it, epenthesis is an unfaithful mapping, and as such should violate faithfulness. As Prince and Smolensky (1993/2004) and McCarthy and Prince (1993) point out, however, epenthesis typically compounds markedness violations, as well; for example, if a candidate contains an epenthetic [h], it incurs one more violation of the constraint against the feature [spread glottis] than candidates without [h]-epenthesis. If we follow Prince and Smolensky (1993/2004), McCarthy and Prince (1993), and Zoll (1996) in assuming that CON includes *STRUC constraints against all structure, then faithfulness constraints against insertion seem to be redundant. This purported redundancy has been pointed out and criticized continuously since the earliest days of OT (Myers 1994, Bernhardt and Stemberger 1998, Causley 1999, Urbanczyk 2006). Most of the arguments against DEP rest on the claim that *STRUC constraints are independently necessary outside the domain of epenthesis, whereas DEP is at best redundant and at worst harmful; blocking epenthesis in this view is a matter of structural markedness, not faithfulness. Here, I present arguments that DEP is far from redundant; in fact, a *STRUC reanalysis of the patterns discussed here would be impossible.

2 Why DEP Is Not Equivalent to *STRUC

To appreciate the difference between DEP and *STRUC, we should start by looking at the definition of DEP. McCarthy and Prince's (1995) correspondence theory of faithfulness establishes a correspondence relation between pairs of representations (strings), including input-output (IO), base-reduplicant (BR), and output-output (OO). The DEP family of constraints is then defined as follows:

(1) DEP
Every element of S₂ has a correspondent in S₁. Range ()
= S₂.
(McCarthy and Prince 1995)

By virtue of its definition, DEP is capable of distinguishing two identical segments in different candidates on the basis of their correspondence relations alone. This is something that markedness constraints (including *STRUC) cannot do, since markedness constraints have access to just one level of representation: the output. For input-output correspondence, this means simply that no segments of the output can be inserted—there can be no mappings of the form /p₁k₂a₃/ → [p₁ik₂a₃], where [i] is an epenthetic vowel without lexical affiliation. For other types of correspondence, DEP really need not have anything to do with this type of epenthesis at all, as will be demonstrated next.

3 DEP_BR Blocks Syncope in Tonkawa

In the domain of base-reduplicant correspondence, DEP is violated by segments that appear in the reduplicant morpheme (S₂) but not in the string that constitutes the reduplicative base (S₁). From the point of view of the base, such segments are "inserted," but they are not epenthetic in the same sense that a vowel that has no morphological affiliation is epenthetic. Thus, DEP_BR can be crucially active in two distinct circumstances: (a) if it blocks "true" epenthesis of nonlexical material in the reduplicant,¹ and (b) if it blocks deletion of lexical material...