Common Contracts
1 similar null contracts
♡The Chinese University of Hong KongMarch 1st, 2024
Despite the fact that multilingual agreement (MA) has shown its importance for multilin- gual neural machine translation (MNMT), cur- rent methodologies in the field have two short- ages: (i) require parallel data between mul- tiple language pairs, which is not always re- alistic and (ii) optimize the agreement in an ambiguous direction, which hampers the trans- lation performance. We present Bidirectional Multilingual Agreement via Switched Back- translation (BMA-SBT), a novel and univer- sal multilingual agreement framework for fine- tuning pre-trained MNMT models, which (i) exempts the need for aforementioned parallel data by using a novel method called switched BT that creates synthetic text written in another source language using the translation target and (ii) optimizes the agreement bidirection- ally with the Kullback-Leibler Divergence loss. Experiments indicate that BMA-SBT clearly improves the strong baselines on the task of MNMT with three benchmarks: TED Talks, News, and E
