UK-German Funding Initiative in the Humanities: Prosodic structure at the interface between language and speech

语言冗余和韵律结构都会影响语音的声学显著性和可懂度。韵律结构是语法和语音之间的接口的提议导致Aylett和Turk（2004，2006）和Turk（2010）提出说话者计划话语的韵律结构，以便通过其对声学显着性的影响使可预测性较低的语音部分更容易理解（平滑信号冗余假设，Aylett 2000）。这个项目的目标是：1）测试韵律结构是语言和语音之间的接口的假设，2）在语法的（计算）模型中实现这个接口，以测试和评估我们的架构假设。韵律界面假设的预测是，不同类型的语言冗余（可预测性）度量应该对语音声学具有类似的影响，例如，基本频率和持续时间也就是说，它们都应该影响与韵律突出和边界强度相关的词的属性和部分。该项目的第一链测试的影响，重复，长度，句法，单词bigram，词频和语用语言冗余措施暂停和语音声学的相关部分的话，并比较这些影响的韵律突出和边界强度的已知效果。该项目的第二部分涉及在词汇功能语法（LFG）中对第一部分的效果进行建模。LFG的模块对应结构提供了一个最佳的理论框架，允许语法的不同方面之间的互动。通过在一个计算LFG语法的实现，我们打算在其完整的范围内从语义分析的语音信号测试的正式模型，并扩展的框架来模拟频率和其他可预测性的影响梯度表示的韵律结构。

Both language redundancy and prosodic structure affect characteristics of speech related to acoustic salience and intelligibility. Proposals that prosodic structure is the interface between grammar and speech led Aylett and Turk (2004, 2006) and Turk (2010) to propose that speakers plan the prosodic structure of an utterance in order to make less predictable parts of speech more intelligible via its effects on acoustic saliency (the Smooth Signal Redundancy Hypothesis, Aylett 2000). The goals of this project are 1) to test the hypothesis that prosodic structure is the interface between language and speech, and 2) to implement this interface in a (computational) model of grammar in order to test and evaluate our architectural assumptions. A prediction of the Prosodic Interface Hypothesis is that different types of language redundancy (predictability) measures should have similar effects on speech acoustics, e.g., on the fundamental frequency and duration. That is, they should all affect properties and parts of words that relate to prosodic prominence and boundary strength. The first strand of the project tests the effects of repetition, length, syntactic, word-bigram, word frequency, and pragmatic language redundancy measures on pauses and speech acoustics in relevant parts of words, and compares these effects to known effects of prosodic prominence and boundary strength. The second strand of the project involves modelling the effects of the first part within Lexical Functional Grammar (LFG). The modular correspondence architecture of LFG provides an optimal theoretical framework that allows for an interaction between different aspects of grammar. By means of an implementation in a computational LFG grammar, we intend to test the formal model in its complete range from the semantic analysis to the speech signal, and to extend the framework to model the influence of frequency and other predictability effects on a gradient representation of prosodic structure.