BERT-Based Neural Collaborative Filtering and Fixed-Length Contiguous Tokens Explanation

Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing, pages 1–11 December 4 - 7, 2020. c⃝2020 Association for Computational Linguistics 1 Touch Editing: A Flexible One-Time Interaction Approach for Translation Qian Wang1,2 , Jiajun Zhang1,2, Lemao Liu3, Guoping Huang3, Chengqing Zong1,2 1National Laboratory of Pattern Recognition, CASIA, Beijing, China 2University of Chinese Academy of Sciences, Beijing, China 3Tencent AI Lab, Shenzhen, China {qian.wang, jjzhang, cqzong}@nlpr.ia.ac.cn, {redmondliu, donkeyhuang}@tencent.com Abstract We propose a touch-based editing method for translation, which is more flexible than tradi- tional keyboard-mouse-based translation post- editing. This approach relies on touch actions that users perform to indicate translation errors. We present a dual-encoder model to handle the actions and generate refined translations. To mimic the user feedback, we adopt the TER al- gorithm comparing between draft translations and references to automatically extract the sim- ulated actions for training data construction. Experiments on translation datasets with sim- ulated editing actions show that our method significantly improves original translation of Transformer (up to 25.31 BLEU) and outper- forms existing interactive translation methods (up to 16.64 BLEU). We also conduct ex- periments on post-editing dataset to further prove the robustness and effectiveness of our method. 1 Introduction Neural machine translation (NMT) has made great success during the past few years (Sutskever et al., 2014; Bahdanau et al., 2014; Wu et al., 2016; Vaswani et al., 2017), but automatic machine trans- lation is still far from perfect and cannot meet the strict requirements of users in real applications (Petrushkov et al., 2018). Many notable human- machine interaction approaches have been pro- posed for allowing professional translators to im- prove machine translation results (Wuebker et al., 2016; Knowles and Koehn, 2016; Hokamp and Liu, 2017). As an instance of such approaches, post-editing directly requires translators to modify outputs from machine translation (Simard et al., 2007). However, traditional post-editing requires intensive keyboard interaction, which is inconve- nient on mobile devices. Grangier and Auli (2018) suggest a one-time interaction approach with lightweight editing ef- forts, QuickEdit, in which users are asked to simply mark incorrect words in a translation hypothesis for one time in the hope that the system will change them. QuickEdit delivers appealing improvements on draft hypotheses while maintaining the flexibil- ity of human-machine interaction. Unfortunately, only marking incorrect words is far from adequate: for example, it does not indicate the missing infor- mation beyond the original hypothesis, which is a typical issue called under-translation in machine translation (Tu et al., 2016). In this paper, we propose a novel one-time in- teraction method called Touch Editing, which is flexible for users and more adequate for a system to generate better translations. Inspired by human editing process, the proposed method relies on a series of touch-based actions including SUBSTITU- TION, DELETION, INSERTION and REORDERING. These actions do not include lexical information and thus can be flexibly provided by users through various of gestures on touch screen devices. By …

Content truncated to 500 words