The Insufficiency of Paper-and-Pencil Linguistics: the Case of Finnish Prosody

It is a basic scientific practice to examine a limited amount of data in the light of some theoretical framework and to develop an analysis that accounts for the primary facts and extends to unseen data. If the predictions are correct, the analysis stands and lends further support for the framework in which it is conceived. This paper focuses on a case where the analysis turns out to be wrong. It highlights the need for the formalization and computational implementation of linguistic theories. Paper-and-pencil methods are insufficient to test a theory with real data. The problem is particulary acute for optimality theory (Prince and Smolensky, 1993, Kager, 1999, McCarthy, 2002). The ot framework assumes two levels of representation, a set of inputs and, for each input, a possibly infinite set of outputs. The mapping between the two levels is subject to a set of ranked constraints. It is typically the case that no output candidate satisfies all the constraints. A winning candidate is the one that incurs the fewest violations of the most highly ranked constraint still in play that cannot be satisfied by any of the other surviving output candidates. We consider two closely related ot analyses of Finnish prosody by Elenbaas (1999) and Kiparsky (2003). In both cases the input consists of a sequence of phonemes and the outputs are sequences of metrical