现代汉语普通话数字化样本库的设计与建设.pdfVIP

现代汉语普通话数字化样本库的设计与建设* 侯敏, 邹煜, 滕永林, 何伟, 张佳 中国传媒大学国家语言资源监测与研究中心有声媒体语言分中心 北京 100024 E-mail: houminxx@263.net 摘 要：为真实地再现以及永久保存现代汉语普通话的历史面貌，我们构建了现代汉语普通话数字化样本库。 该样本库具有历时性、开放性、跨媒体性、共享性的特点。样本采集的原则是选择具有权威性的说话人和具 有常态性的话语。根据语言样态、语料载体、历史分期所做的样本分类体系可保证样本采集的全面性，以达 到历史的覆盖性。样本的数据加工包括元数据标注、文本转写、语音数据标注、文本数据标注等项内容。 关键词：现代汉语普通话；数字化样本库；样本采集；数据加工 Construction and Analysis of Chinese Mandarin Digital Multi-modal Corpus Hou Min, Zou Yu, Teng Yonglin, He Wei, Zhang Jia Broadcast Media Language Branch, National Language Resources Monitoring and Research Center, Communication University of China, Beijing 100024 E-mail: houminxx@263.net Abstract: Its goal is to represent the development of Chinese Mandarin in the past century, and store its representative sample in digital form permanently. This paper present our work which designed and established a Chinese Mandarin Digital Multi-modal Corpus (CMDMC), it’s a dynamic miniature model (or speech museum) with diachronic, opened, cross-media and sharable features. The materials generated by the authoritative speakers (e.g. announcer at TV or radio station, actor/actress in movie or drama) with normality are required samples if we can get them. In our study, we also proposed the classification according to the speaking style, medium and historical periods. The finished corpus has about 1500 hours video or audio files with metadata tagging; there are more annotation has done in a core sub-corpus with about 30 hours programs or movies, the data processing includes aligned transcription, POS tagging, discourse marks, phonetic segmentation and prosodic labeling etc. Furthermore, a managemen