Phonetic Modification of Function Words: Implications for Human and Automatic Speech Processing

功能词的语音修饰：对人类和自动语音处理的影响

• 批准号: 9820126
• 负责人: Stefanie Shattuck-Hufnagel
• 金额: --
• 依托单位: Massachusetts Institute of Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1999
• 资助国家: 美国
• 起止时间: 1999-09-01 至 2002-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9820126&HistoricalAwards=false
• 关键词: Phonetic; Modification; Function; Words; Implications

Shattuck-Hufnagel & Stevens98-20126AbstractIn this study we will survey the function words (grammatical words like articles, conjunctions, auxiliary verbs, pronouns, prepositions etc.) in a corpus of naturally occurring American English speech which has been labelled both phonetically and prosodically, and inventory the way the pronunciation of these words varies in normal speech. This will allow us to test the hypothesis that the phonetic variation of these words is influenced by prosodic structure such as intonational phrasing and pitch accenting.Function words are particularly prone to phonetic modification compared with content words like nouns, verbs and adjectives; see for example ``doncha'' for ``don't you'', ``givem'' for ``give him'', and ``mishu'' for ``miss you''. Although these changes may not seem systematic, we believe they can be described by a relative small number of rules applied to just a few representational structures. For example, results of a pilot study of the function word ``to'' suggest that its modifications are strongly influenced by the distinctive features of the preceding context, the form class of the function word, and the prosody of the spoken utterance. A better understanding of what can happen to function words in running speech will result in improved linguistic theories, cognitive processing models and techniques for automatic speech recognition and synthesis.

本文将对汉语中的虚词（如冠词、连词、助动词、代词、介词等）进行考察。在一个语料库的自然发生的美国英语语音已被标记的语音和韵律，并库存的方式，这些词的发音在正常的讲话。 这将使我们能够验证这一假设，即这些词的语音变化受到韵律结构的影响，如语调短语和音高重音，与名词，动词和形容词等实词相比，虚词特别容易语音修饰;例如，“doncha”代表“don 't you”，“givem”代表“give him”，“mishu”代表“miss you”。 虽然这些变化看起来可能不是系统性的，但我们相信它们可以通过应用于少数表征结构的相对少量的规则来描述。 例如，对虚词“to”的初步研究结果表明，其修饰受到前面语境的显著特征、虚词的形式类别和口语的韵律的强烈影响。更好地理解运行语音中的功能词会发生什么，将导致改进的语言学理论，认知处理模型和自动语音识别和合成技术。