Please note: This PhD seminar will take place online.
Peng Shi, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Jimmy Lin
We focus on the Cross-lingual Text-to-SQL semantic parsing, where the parsers are expected to generate SQL for non-English utterances based on English database schema. Intuitively, English translation as a side information is an effective way for bridging the language gap, while the noise introduced by translation system may effect the parser effectiveness.
In this work, we propose a Representation Mixup Framework (Rex), for effectively exploiting the translations in the cross-lingual Text-to-SQL task. Particularly, it takes General Encoding layer, Transition layer and Target-centric layer to guide information of the English translation flow properly. Experimental results on CSpider and VSpider show that our framework can benefit the cross-lingual training and improve the effectiveness of the semantic parsers, achieving state-of-the-art performance.