Delving Deeper into Cross-lingual Visual Question Answering

Abstract

Visual question answering (VQA) is one of the crucial vision-and-language tasks. Yet, the bulk of research until recently has focused only on the English language due to the lack of appropriate evaluation resources. Previous work on cross-lingual VQA has reported poor zeroshot transfer performance of current multilingual multimodal Transformers and large gaps to monolingual performance, attributed mostly to misalignment of text embeddings between the source and target languages, without providing any additional deeper analyses. In this work, we delve deeper and address different aspects of cross-lingual VQA holistically, aiming to understand the impact of input data, fine-tuning and evaluation regimes, and interactions between the two modalities in crosslingual setups. 1) We tackle low transfer performance via novel methods that substantially reduce the gap to monolingual English performance, yielding +10 accuracy points over existing methods. 2) We study and dissect crosslingual VQA across different question types of varying complexity, across different multilingual multi-modal Transformers, and in zeroshot and few-shot scenarios. 3) We further conduct extensive analyses on modality biases in training data and models, aimed at understanding why zero-shot performance gaps remain for some question types and languages. We hope that the novel methods and detailed analyses will guide further progress in multilingual VQA.