第二语言韵律焦点产出、合成与评价的研究

Focus plays an important role in conveying information in speech communication. It is phonetically realized by increasing duration, pitch and intensity on the focused constituents in a sentence. In some languages, there is a concomitant decrease of pitch and intensity after the focused item, referred to as post-focus compression (PFC). Beijing Mandarin and English are PFC languages of this type. However, Mandarin is a tone language and has lexical tones while English is a non-tone language and has lexical stress. This project investigates the phonetic realization of focus in second language (L2) Mandarin by American English learners and its realization in English by Beijing Mandarin learners. Second language experience is also manipulated in each of the two experiments. Participants in the first experiment vary in age of learning L2 while participants in the second experiment vary in length of residence in the L2-speaking environment. The acoustic results of the speech production are discussed under the framework of the Speech Learning Model (SLM) (Flege, 1995; 2007) with the goal of developing the theory and application of this model from the segmental level to the suprasegmental level in L2 speech research. Acoustic relations between lexical tone and prosodic focus in L2 Mandarin and between lexical stress and prosodic focus in L2 English are analyzed under the framework of the Parallel Encoding and Target Approximation (PENTA) model (Xu, 2005). The speech production is then synthesized in a many-to-one manner with the combinations of communicative functions of the sentences by using the model-based research tool PENTAtrainer2 (Xu & Prom-on, 2014). The original and synthesized target sentences are assessed by native speakers of the target L2 in each experiment in terms of the naturalness of the prosody, the status of the focus and its communicative contexts. This project aims at a comprehensive investigation of the production and assessment of L2 prosodic focus for insight into the effects of first and second language experience on the speech production of L2 prosody, and more importantly, for the purpose of serving L2 teaching and learning, acoustical modeling of L2 speech production, and human and machine recognitions of L2 speech.

本项目针对目前汉语和英语作为第二语言使用的全球化趋势，以及韵律焦点在信息传递上和言语交际中的重要功能，调查美式英语母语者的汉语普通话以及汉语普通话母语者的英语语音产出中的韵律焦点的声学编码，使用基于交际功能和计算机建模的分析合成工具生成源于原始语音产出的合成目的句，并由母语者判别原始语音和合成语音的韵律自然度、焦点状态及其交际语境。本项目旨在利用新技术、发展旧理论，为汉语和英语分别作为第二语言的韵律焦点的产出和评价做一个全面系统的研究，探讨以汉语和英语作为目的语的第二语言韵律焦点的外在表现、内部机制及其语言经验效应，以期拓展第二语言语音习得的理论框架和实证研究的方法与技术，为对外汉语教学和英语外语教学中的语音和语用教学提供实证依据，也为以汉语和英语为二语的大型语料库的语音建模、二语语音产出的评价模式以及人机对话中二语语音识别技术的发展提供参考基础。

随着语音信息技术的发展和人际交流全球化的进程，带有口音的第二语言语音经常在生活中出现，生活在第二语言环境中的人们还需要使用二语与机器对话，因此对二语语音产出的研究势在必行。本项目选择对话中用来表达重要信息的韵律焦点，研究在美国英语母语者学习汉语普通话、以及汉语普通话母语者学习英语时，其第二语言焦点表达的韵律产出情况，以及基于说话人的母语背景和二语经验的韵律声学建模和基频合成，把产出实验中的自然话语和分组建模后的合成话语用来给汉语普通话母语者和美国英语母语者做感知评价，评价任务包括判断句子的焦点状态和给句子的语音自然度评分。. 两个韵律产出实验结果显示，不论普通话为目的语或者英语为目的语，不论母语背景或二语经验，二语学习者都能基本习得到用于编码韵律焦点的母语式的时长变化，其次是音强，最后是调高；在焦点位置方面，二语学习者最容易实现焦点前的韵律变化，其次是焦点上的韵律变化，最不容易实现的是焦点后的韵律变化；综合所有条件，对于二语学习者来说，最难实现的是焦点后的调域压缩和调高降低。同时，二语经验也影响着二语韵律焦点的习得效果，二语经验多的被试的韵律焦点声学数据更接近母语者的数据。个人自然话语和分组合成话语的感知评价实验结果显示，一样的母语背景，二语经验多的的被试，其自然语音产出的韵律焦点状态比较容易被听辩出来，分组合成基频的句子的自然度评分也较高。实验结果还显示，句首和句中焦点的听辩正确率比句末焦点高，这与基频和音强的焦点后压缩辅助突显焦点有关。. 本研究发现的二语韵律习得的声学表现难度层级为二语教学提供了指导性建议，并且把几个二语语音习得理论模型的应用从音段习得提升到超音段层面。本研究建成的两个二语语音小型语料库以及提取到的声学和评价数据，在语音识别领域有一定的应用价值，有助于人机对话中机器通过二语说话人的口音共性判断出二语语音所表达的语义。