Combining Command-Respond Model and ToBI in Analyzing Second Language English Sentence Prosody

The paper combines the Fujisaki commandresponse model and the ToBI convention to analyze the English sentence prosody produced by American native speakers and Korean and Chinese English learners. The model parameters were extracted by FujiParaEditor for nine short sentences produced by three groups of speakers. The ToBI-tanscribed results were compared with the model parameters. Results show that modelbased approach is a valid complement for judging the pitch and break events in both native and nonnative speech. With the aid of model-generated parameters it is found that both groups of English learners produce shorter prosodic phrase in English sentences but with different prosodic strategies, which is indicative of native language transfer. The Korean subjects invariably produce H-phrasal accent but the Chinese subjects use L-accent in the sentence-medial position. The Chinese subjects accent all content words and sentence-final function words. But the F0 contour produced by Korean subjects shows a narrow pitch range. Both groups have difficulty in acquiring the canonical yes-no question contour (L* H-H%). The majority use L-H% tone for a yes-no question.