Zusammenfassung
This thesis proposes an output-oriented theory of morphology, in which morphological processes are encoded in constraints integrated with and interacting with the familiar Markedness and Faithfulness constraints of Optimality Theory. These constraints require forms with particular syntactic features to display particular phonological properties; they do not form a separate morphological component or module, but represent direct interaction between the syntax and the phonology.
Encoding morphological processes in constraints, which compete and interact with each other and with Markedness and Faithfulness constraints, provides accounts for a number of complex and widely-attested phenomena. Where FIAT constraints are outranked by Markedness constraints, the presence of an affix may be dependent on phonological characteristics of the form; where they are outranked by Input-Output Faithfulness, individual lexical items may form exceptions to general morphological processes. Multiple processes marking the same syntactic characteristics are encoded in potentially incompatible and competing FIAT constraints. This, combined with the effect of Markedness and Output- Output Faithfulness, accounts for phonologically conditioned allomorph selection; combined with the effect of Input-Output Faithfulness constraints, it accounts for unpredictable, lexically determined allomorph selection, and for conjugation or declension classes with unpredictable membership. Since all these rankings may occur in a single grammar, the theory can account for complex systems in which multiple morphological markers, lexical idiosyncrasy, and phonological conditioning all play a role.
The surface-oriented nature of the approach allows FIAT grammars to be learned by uninsightful, inductive processes, even for complex systems riddled with exceptions and irregularities. FIAT constraints originate as inductive generalizations about the structures observable in forms with particular syntactic properties; descriptions of observed strings, which are in most cases true only of a few forms and contradicted elsewhere. Introducing these generalizations into the constraint set and finding a ranking that generates the correct results allows subtle and complex systems to be learned without the learner ever needing to directly apprehend their subtleties.
