Vision-and-Language or Vision-for-Language? On Cross-Modal Influence in Multimodal Transformers
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Documents
- Fulltext
Final published version, 738 KB, PDF document
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.
Original language | English |
---|---|
Title of host publication | Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing |
Publisher | Association for Computational Linguistics (ACL) |
Publication date | 2021 |
Pages | 9847-9857 |
DOIs | |
Publication status | Published - 2021 |
Event | 2021 Conference on Empirical Methods in Natural Language Processing - Duration: 7 Nov 2021 → 11 Nov 2021 |
Conference
Conference | 2021 Conference on Empirical Methods in Natural Language Processing |
---|---|
Periode | 07/11/2021 → 11/11/2021 |
ID: 301491346