*清華語言所新第013次專題演講*
演講者：倫敦大學學院 許毅教授 (Yi Xu, University College London )
日期：2016年 5月12日 (星期四)
時間：中午 12時 30分 開始
地點：人社院 B305 室
題目：Make basic research count: Toward theoretical models that are computational and operational (讓基礎研究變得有用: 把理論模型數字化、操作化)
摘要：
In speech science, the currently assumed division of labor is that basic research only needs to generate knowledge that is symbolic or categorical at the phonological level and descriptive or featural at the phonetic level, while the applied areas are expected to come up with their own ways of applying such knowledge. So far, however, such a division has left a large gap between theory and application in areas such as speech technology, speech and language pathology and language teaching. Critically, the gap leaves the validity of our theoretical knowledge questionable. To close this gap, I will argue, basic research needs to propose explicit ways to link sparse linguistic representations to fully continuous behavioral patterns. One possibility is to develop operational computational models that can associate abstract linguistic categories directly to gradient numerical parameters, with which the models can generate fully continuous surface patterns. These patterns can then be directly compared to natural patterns in as much detail as necessary. Such comparison will not only raise the standard of theory evaluation, but also make theoretical knowledge genuinely relevant to applied areas. I will demonstrate the plausibility of this paradigm with our own recent computational modeling work on both prosodic and segmental aspects of speech.