Contextual Neural Machine Translation Improves Translation of Cataphoric Pronouns

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5971–5978 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 5971 Contextual Neural Machine Translation Improves Translation of Cataphoric Pronouns KayYen Wong† Sameen Maruf‡ Faculty of Information Technology, Monash University, VIC, Australia †kywon63@student.monash.edu ‡firstname.lastname@monash.edu Gholamreza Haffari‡ Abstract The advent of context-aware NMT has re- sulted in promising improvements in the over- all translation quality and specifically in the translation of discourse phenomena such as pronouns. Previous works have mainly fo- cused on the use of past sentences as con- text with a focus on anaphora translation. In this work, we investigate the effect of future sentences as context by comparing the perfor- mance of a contextual NMT model trained with the future context to the one trained with the past context. Our experiments and evalu- ation, using generic and pronoun-focused au- tomatic metrics, show that the use of future context not only achieves significant improve- ments over the context-agnostic Transformer, but also demonstrates comparable and in some cases improved performance over its counter- part trained on past context. We also perform an evaluation on a targeted cataphora test suite and report significant gains over the context- agnostic Transformer in terms of BLEU. 1 Introduction Standard machine translation (MT) systems typi- cally translate sentences in isolation, ignoring es- sential contextual information, where a word in a sentence may reference other ideas or expressions within a piece of text. This locality assumption hin- ders the accurate translation of referential pronouns, which rely on surrounding contextual information to resolve cross-sentence references. The issue is further exacerbated by differences in pronoun rules between source and target languages, often result- ing in morphological disagreement in the quantity and gender of the subject being referred to (Van- massenhove et al., 2018). Rapid improvements in NMT have led to it re- placing SMT as the dominant paradigm. With this, context-dependent NMT has gained traction, over- coming the locality assumption in SMT through the use of additional contextual information. This has led to improvements in not only the overall trans- lation quality but also pronoun translation (Jean et al., 2017; Bawden et al., 2018; Voita et al., 2018; Miculicich et al., 2018). However, all these works have neglected the context from future sentences, with Voita et al. (2018) reporting it to have a nega- tive effect on the overall translation quality. In this work, we investigate the effect of future context in improving NMT performance. We partic- ularly focus on pronouns and analyse corpora from different domains to discern if the future context could actually aid in their resolution. We find that for the Subtitles domain roughly 16% of the pro- nouns are cataphoric. This finding motivates us to investigate the performance of a context-dependent NMT model (Miculicich et al., 2018) trained on the future context in comparison to its counterpart trained on the past context. We evaluate our mod- els in terms of overall translation quality (BLEU) and also employ three types of …

Content truncated to 500 words