2018/08/19 by He Bai, Bai, He, Yu Zhou +9 · 1 citation
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Multimodal Machine Learning Applications #Natural Language Processing Techniques #Reinforcement Learning in Robotics #Speech and dialogue systems #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1808.06167
openalex publication_date 2018/08/19 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
To deploy a spoken language understanding (SLU) model to a new language,\nlanguage transferring is desired to avoid the trouble of acquiring and labeling\na new big SLU corpus. Translating the original SLU corpus into the target\nlanguage is an attractive strategy. However, SLU corpora consist of plenty of\nsemantic labels (slots), which general-purpose translators cannot handle well,\nnot to mention additional culture differences. This paper focuses on the\nlanguage transferring task given a tiny in-domain parallel SLU corpus. The\nin-domain parallel corpus can be used as the first adaptation on the general\ntranslator. But more importantly, we show how to use reinforcement learning\n(RL) to further finetune the adapted translator, where translated sentences\nwith more proper slot tags receive higher rewards. We evaluate our approach on\nChinese to English language transferring for SLU systems. The experimental\nresults show that the generated English SLU corpus via adaptation and\nreinforcement learning gives us over 97% in the slot F1 score and over 84%\naccuracy in domain classification. It demonstrates the effectiveness of the\nproposed language transferring method. Compared with naive translation, our\nproposed method improves domain classification accuracy by relatively 22%, and\nthe slot filling F1 score by relatively more than 71%.\n