人类语音复杂系统的量化及理论研究

To uncover the mathematical and physical laws of human phonemic complex systems is an important fundamental research subject, especially for some frontier scientific problems, such as developing the speech recognition systems, achieving the voice intellectualization, etc. This project is planned to analyze the human phonemic systems (including 10 languages, such as English, Chinese, Khmer, and so on), by using the theories and methods from complex networks and statistical physics. The core research topics and goals of the project are as follows: 1) To construct the Weighted Human Phonemic Networks (WHPN), analyze the characteristics of the network topology and find the universal features, the formation and evolutionary mechanisms of WHPN; 2) To explain the macroscopic (morphemes and words) features of language systems from perspective of the microscopic (phonemes and syllables) features of phonemic systems, and establish the relations between microscopic phonemic systems and macroscopic language systems; 3) To explore the theoretical models for human phonemic systems, and propose a simple probabilistic model with fewer parameters, which would interpret the empirical findings of the phonemic systems. The research of this project is not only beneficial for the understanding of the origins, evolution and self-adaptation of human language, but also provide some theoretical hints on the real-world scientific problems, such as text classification, authorship attribution, speech recognition, and so on.

揭示人类语音复杂系统的数理规律，对于开发语音识别系统、实现语音智能化等前沿科学问题是一项重要基础性研究课题。本项目拟利用复杂网络与统计物理的理论方法研究人类语音系统（包括英语、汉语、高棉语等10种语言），重点研究内容和目标为：1)构建人类语音权重网络，分析权重网络拓扑特性，寻找人类语音权重网的普适规律及其生成演化机制；2)从语音系统微观（音素和音节）特征解释语言系统（语素和词语）宏观规律，建立微观与宏观间的联系；3)探究语音系统的理论模型，从理论上提出简单、参数少的概率统计模型解释语音系统的实证研究结果。本项目的研究不仅有利于理解人类语言的起源、演化、自适应等问题，而且对于文本分类、作者身份识别、语音识别等实际科学问题具有重要的理论意义。

本项目将复杂网络与统计物理思想应用于人类语音及语言复杂系统的研究，主要研究进展包括：1）相对于词语使用频率的Zipf定律的普适性（不依赖于文本、作者），发现语音使用频率的规律依赖于作者和文本，并提出一种理论模型解释该规律；2）发现词语长度分布可以用来对语言进行分类，并利用分级统计分析方法进行了理论解释；3）通过比较文本前半部分和后半部分的统计规律，可用来区分有意义的文本和无意义的文本；4）提出了一种优化汉语学习的思想，构建了汉语语音和构型的双层网络。另外，复杂网络作为研究复杂系统的有力工具，研究复杂网络中的基本问题，发展和优化复杂网络中的基本方法具有很重要应用价值：1）提出了三种社团检测的算法，可以更高效、精确的检测复杂网络的社团结构；2）分析复杂网络上输运流的性质，模拟了实际交通流的输运过程，提出了超车策略，发现该策略可以较大的增大输运效率。