Single Word Level Research Articles

Mechanisms underlying writing and speech in aphasia

As a parallel to the dual decoding concept for processing of written language we proposed that phonological encoding does not necessarily occur in writing and that the phonemic and graphemic subsystems can be independent on the one-word level. This hypothesis was tested by comparing oral and written performance in a picture-naming task in Broca's and Wernicke's aphasics. In addition, the residual tacit knowledge of the orthographic properties of the names of the pictures was examined with a multiple-choice recognition task. The principal finding is that Broca's aphasics who were better in written than in oral naming showed more graphemically and semantically motivated errors than aphasics who were better in oral than in written naming, the latter producing more phonemically motivated errors. This result supports the dual encoding concept for writing on the singleword level, implying a direct route from the mental lexicon to the graphemic system in parallel with a route mediated by the phonemic system. Multiple-choice recognition was found to be superior to both oral and written performance in both Broca's and Wernicke's aphasics.

A Model for Word-Level Conversion of Arbitrary Text to Speech

While the acoustic correlates of linguistic structures above the word level are of great importance in speech synthesis, it has not been possible to develop a comprehensive theory for these effects. At the single word level, however, the present model shows how any English word can be converted to 1PA symbols with stress marks sufficient to drive a phonemic synthesis-by-rule algorithm. Words are either partitioned into constituent morphemes, or processed by a large set of letter-to-sound rules. The latter rely on the lexicon to contain high-frequency exceptions, and assume that all polymorphemic words have been split up into their constituent morphs. Both monomorphemic and polymorphemic lexical stress rules are applied, and morphophonemic rules provide low-level adjustments. The complete process provides a partial model for native speakers reading aloud, as well as the means to test current linguistic phonological theories.