Arabic Morpho-syntax for Text-to-Speech

摘要

Attempts to provide text-to-speech systems for Arabic are considerably hampered by the fact that written Modern Standard Arabic omits a great deal of information about short vowels, and a certain amount of other phonetically relevant information. We argue that it is possible to recover this information by exploiting a mixture of morphological, syntactic and shallow semantic information. To do this effectively, the various levels of processing have to be carefully coordinated. We do that by delaying evaluation of constraints for as long as possible, so that we do not explore unnecessary hypotheses.