Context-aware Neural Machine Translation for English-Japanese Business Scene Dialogues

Proceedings of Machine Translation Summit XIX, Vol. 1: Research Track, pages 272–285 September 4–8, 2023, Macau SAR, China. ©2023 The authors. This article is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). 272 Context-aware Neural Machine Translation for English-Japanese Business Scene Dialogues Sumire Honda pu.sumirehonda@gmail.com Computational Linguistics, University of Potsdam, Potsdam, Germany Patrick Fernandes pfernand@cs.cmu.edu Language Technologies Institute, Carnegie Mellon University, Pittsburgh, USA Instituto de Telecomunicações, Instituto Superior Técnico, Lisbon, Portugal Chrysoula Zerva chrysoula.zerva@tecnico.ulisboa.pt Instituto de Telecomunicações, Instituto Superior Técnico, Lisbon, Portugal Abstract Despite the remarkable advancements in machine translation, the current sentence-level paradigm faces challenges when dealing with highly-contextual languages like Japanese. In this paper, we explore how context-awareness can improve the performance of the current Neural Machine Translation (NMT) models for English-Japanese business dialogues translation, and what kind of context provides meaningful information to improve translation. As business dialogue involves complex discourse phenomena but offers scarce training resources, we adapted a pretrained mBART model, finetuning on multi-sentence dialogue data, which allows us to experiment with different contexts. We investigate the impact of larger context sizes and propose novel context tokens encoding extra-sentential information, such as speaker turn and scene type. We make use of Conditional Cross-Mutual Information (CXMI) to explore how much of the context the model uses and generalise CXMI to study the impact of the extra-sentential context. Overall, we find that models leverage both preceding sentences and extra-sentential context (with CXMI increasing with context size) and we provide a more focused analysis on honorifics translation. Regarding translation quality, increased source-side context paired with scene and speaker information improves the model performance compared to previous work and our context-agnostic baselines, measured in BLEU and COMET metrics. 1 1 Introduction Traditionally NMT models such as Transformers (Maruf et al., 2021) approach the task of machine translation (MT) focusing on individual sentences without considering the surrounding information, such as previous utterances or underlying topics. As a result, the output often lacks discourse coherence and cohesion, which is problematic for MT applications such as chat translation systems (Farajian et al., 2020; Bawden et al., 2018). Thus, it is still an open research question to what degree these models can take advantage of contextual information to produce more accurate translations. To answer this question, several context-aware NMT (Tiedemann and Scherrer, 2017; Voita et al., 2019; Maruf et al., 2019; Xu et al., 2021) studies have been conducted by adding 1Code available at: https://github.com/su0315/discourse_context_mt 273 surrounding sentences to the models and testing if it helps to capture better specific linguistic phenomena requiring context (e.g. coreference resolution). However, there is limited work on discourse or dialogue datasets, and most of it is focused on high-resource or Indo-European (IE) languages (Liu et al., 2021). Therefore, there is a need to investigate how well do the proposed approaches capture discourse phenomena in non-IE or low-resource languages. This work aims to address the aforementioned gap by focusing on English-Japanese (En-Ja) translation for business dialogue scenarios in order to examine if current context-aware …

Content truncated to 500 words