13.07.2015 Views

Dissertation - Michael Becker

Dissertation - Michael Becker

Dissertation - Michael Becker

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

of phonological primitives that constraints are sensitive to, such as marked combinationsof features, preferred alignments of phonological elements, positional faithfulness, etc.Contrasted with traditional generative analyses, the proposal made here “reverses” theeffect of the phonology. Instead of assigning the hidden aspects of bases to their underlyingrepresentation, and then neutralizing them in the unaffixed form, as is done traditionally,I propose that the surface forms of bases are assumed as their underlying form, andany properties of the base that emerge only in suffixed forms are achieved by constraintinteraction. In the simple case of Turkish, where the only hidden property of nominal rootsis the voicing of their final stop, the analysis in terms of cloned constraints is not onlyclearly feasible, it is also the only analysis that allows speakers to capture the variety oflexical trends that the language has.Assuming the base form as the underlying representation has the added benefit ofobviating the search for non-surface-true underlying representations. This search requiresa significant amount of computation, as shown by Tesar (2006) and Merchant (2008), andin parallel lines of work, also by Boersma (2001) and Apoussidou (2007) and by Jarosz(2006), who specifically look at “final-devoicing languages”, i.e. languages like Turkish,where the behavior of root-final stops is hidden in the bare form of the root. In the proposalsmentioned above, the search for the optimal lexicon not only involves a rather large searchspace, it is also done in parallel with a search for a constraint ranking for the language. Inmy proposal, the learner is only trying to learn a constraint ranking, which is shown in Tesar& Smolensky (1998) to be quite efficient, and probably more efficient that searching for aranking and a lexicon. An explicit proof that my approach requires a lighter computationalload, however, is left for future work.4.4.2 Hidden structure in affixes: KoreanIn the discussion of Turkish above, attributing hidden structure of roots to thegrammar was shown to be the only way to make the full range of lexical trends196

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!