Vision-and-Language or Vision-for-Language? On Cross-Modal Influence in Multimodal Transformers
This addresses the issue of evaluating cross-modal integration in multimodal transformers for researchers, revealing that current models may be more vision-for-language than symmetric, which is incremental but clarifies model behavior.
The paper tackles the problem of assessing whether pretrained vision-and-language BERTs truly integrate cross-modal information by proposing a diagnostic method based on cross-modal input ablation. The result shows that these models are not symmetrically cross-modal, with much greater difficulty predicting text when visual information is ablated compared to predicting visual categories when text is ablated.
Pretrained vision-and-language BERTs aim to learn representations that combine information from both modalities. We propose a diagnostic method based on cross-modal input ablation to assess the extent to which these models actually integrate cross-modal information. This method involves ablating inputs from one modality, either entirely or selectively based on cross-modal grounding alignments, and evaluating the model prediction performance on the other modality. Model performance is measured by modality-specific tasks that mirror the model pretraining objectives (e.g. masked language modelling for text). Models that have learned to construct cross-modal representations using both modalities are expected to perform worse when inputs are missing from a modality. We find that recently proposed models have much greater relative difficulty predicting text when visual information is ablated, compared to predicting visual object categories when text is ablated, indicating that these models are not symmetrically cross-modal.