Building a corpus of phonemic lexicons to study information theoretic universals

人类语言的使用反映了人类交际的本质。例如，频繁的单词往往比不频繁的单词发音更少，这有助于快速发音和理解。然而，人们对更细粒度的区别知之甚少。例如，英语中的/k/发音比/p/发音要多。这是否反映了人类语言的属性及其生理和知觉性质，还是历史上的偶然？回答这些问题需要关于许多语言中单词的频率和语音组成的比较数据。该项目将在现有文本来源和词频列表的基础上，以近200种资源不足的语言提供单词的语音构成。语音词表将为理解人类语言提供宝贵的资源，并为资源匮乏的语言提供急需的语言资源。该项目的成果将公之于众，便于查阅，从而协助记录和教授处理后的语文，并协助建立文本-语音转换引擎等计算语言资源。包括训练有素的本科生和研究生在内的研究团队将制定规则，将字母转换为多种语言的音素表示法。然后，该团队将从在线圣经、报纸和电影字幕等公开来源收集文本资源和词频列表。这些规则将分别应用于每个来源，并将产生的音素表示法公开提供，这样不仅研究人员，而且普通公众都能够使用这些数据并与之互动。研究人员将继续使用这些数据来调查声音的信息论属性是否具有分布普遍性：声音是否倾向于跨语言提供类似数量的信息量，如果是，它们的信息内容是否与它们的语音特性相关？普遍性是一个古老的问题，不同语言属性的异同可以为语言的使用提供新的见解。具体地说，研究人员将使用信息论属性来预测低信息或其他以前研究的语音属性是否可能促进这些语言中辅音的减弱。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Human language use reflects the nature of human communication. For instance, frequent words tend to have fewer sounds than infrequent ones, which facilitates quick production and understanding. However, little is known about more fine-grained distinctions. For instance, English has more /k/ than /p/ sounds. Does that reflect a property of human language and its physiological and perceptual nature or a historical accident? Answering such questions requires comparative data on the frequency and phonological makeup of words in many languages. This project will build on existing textual sources and word frequency lists to provide the phonological makeup of words in close to 200 low-resource languages. The phonological word lists will provide an invaluable resource to the understanding of human language and provide much-needed linguistic resources to low-resource languages. The outputs of the project will be made public and easily accessible, thereby assisting in documenting and teaching the processed languages, and in building computational linguistic resources such as text-to-speech engines. The research team, including trained undergraduate and graduate students, will create rules to translate alphabets to phonemic representation for multiple languages. The team will then collect textual resources and word frequency lists from publicly available sources such as online Bibles, newspapers, and movie subtitles. The rules will be applied separately to each source and the resulting phonological representations will be made publicly available, such that not only researchers but also the general public will be able to use and interact with the data. The researchers will proceed to use the data to investigate whether the information theoretic properties of sounds have distributional universality: do sounds tend to provide similar amounts of information cross-linguistically, and if so, does their information content correlate with their phonetic properties? Universality is an age-old question, and the similarities and differences of properties across language can provide new insights into language use. Specifically, the researchers will use information theoretic properties to predict whether low information or other previously studied phonological properties are likely to promote consonant weakening in those languages.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.