The Characterization of Metathesis in Phonological
(Based on Hume 1998, 2000)
Variation in the linear ordering of elements is typical in the domain of syntax, but comparatively striking in phonology, differing in nature from most other phonological processes which are typically defined in terms of a single sound, or target, which undergoes a change in a specified context. Thus, the change from /nb/ to [mb] can be described in simple terms as place assimilation of the target /n/ in the context of a following /b/, thereby yielding [mb]; or, in traditional linear formalism, /n/ -> [m]/ __[b]. In contrast, the reversal of sounds such as /sk/ -> [ks], as attested in Faroese, defies such a simple formalism given that metathesis seems to involve two targets, with each essentially providing the context for the other. Due in part to the distinct nature of the process, metathesis has traditionally posed a challenge to theoretists attempting to develop a unified and predictive account in phonological theory.
In Chomsky & Halle's (1968) seminal work in generative phonology, metathesis is formally described by means of transformational notation, as in (1).
(1) Transformational notation:
1 2 -> 2 1 Output: [ks]
Unrestricted rewrite rules such as this are excessively powerful and unconstrained; virtually any operation could be formally described in these terms, whether attested or not, such as the unattested metathesis rule shown in (2). This shows a case in which the first consonant in a word metathesizes with the final consonant of a word over a string of five segments. All regular cases of synchronic metathesis involve strictly adjacent segments (see Hume 1998, Mielke & Hume 2000, for related discussion).
(2) C1V2C3V 4C5V6C7 -> C7V 2 C3V4C5V6C1
The fact that linear formalism was inadequate to represent metathesis is not a sufficient argument for rejecting metathesis as a basic operation. Deletion also required an unrestricted rewrite rule, yet it is unlikely that one would doubt the existence of deletion as a phonological process.
Metathesis has resisted a unified, explanatory treatment in nonlinear phonology despite advancements in the formalism used to account for many other processes, such as assimilation (Clements 1985) and dissimilation (Odden 1987). Unlike these phenomena, there is no unique formalism for characterizing metathesis as a primitive rule-type. Instead 'metathesis effects' have been derived by a variety of different means including:
Optimality Theory (OT), a constraint-based theory of phonology, provides a promising approach to the theoretical analysis of metathesis since not only are segment reversals possible between an input and output in the theory, they are predicted to exist (Prince & Smolensky 1993; but especially McCarthy & Prince 1995, McCarthy 1995). Metathesis results in part from a mismatch in the linear ordering of sounds between input and output, formally encoded as a violation of the constraint Linearity. Thus, unlike rule-based approaches, there is no longer a principled reason to reject the existence of metathesis; indeed, within an OT framework there is just the contrary. Since earlier theoretical frameworks have been unsuccessful in providing an explanatory account of the process, the study of metathesis constitutes a good testing-ground for a constraint-based approach to phonology.
As just noted, in Optimality Theory (OT), an output form showing metathesis violates the constraint Linearity, stated in (3). Informally stated, Linearity is violated when there is a mismatch in linear ordering relations between a string of segments in the output, and the corresponding string in the input (and vice versa).
(3) Linearity: "No Metathesis" (McCarthy
& Prince, 1995; McCarthy 1995)
S1 is consistent with the precedence structure of S2, and vice versa.
A necessary requirement for metathesis to occur is for Linearity to be ranked below other faithfulness constraints, such as Max (don't delete), Dep (don't epenthesize), etc. Thus, metathesis, rather than, e.g., deletion or epenthesis, is the strategy used to repair some ill-formed structure. The ill-formedness of a given structure is formally described by means of a markedness constraint. This constraint is necessarily more highly ranked than Linearity, thus, forcing a violation. This is illustrated in (4) where Q is assumed to be some markedness constraint. The structure of segments in the first candidate, while being faithful to the input, violates Q. By reversing the order of the segments y and x in the second candidate, constraint Q is satisfied. Although this reordering causes a violation of Linearity, the ranking of this constraint below Q means that the second candidate is selected as the output.
(4) Constraint Q forces a violation of Linearity: metathesis applies.