Recent advances in multimodal vision and language modeling have predominantly focused on the English language, mostly due to the lack of multilingual multimodal datasets to steer modeling efforts. In this work, we address this gap and provide xGQA, a new multilingual evaluation benchmark for the visual question answering task. We extend the established English GQA dataset to 7 typologically diverse languages, enabling us to detect and explore crucial challenges in cross-lingual visual question answering. We further propose new adapter-based approaches to adapt multimodal transformer-based models to become multilingual, and—vice versa—multilingual models to become multimodal. Our proposed methods outperform current state-of-the-art multilingual multimodal models (e.g., M3P) in zero-shot cross-lingual settings, but the accuracy remains low across the board; a performance drop of around 38 accuracy points in target languages showcases the difficulty of zero-shot cross-lingual transfer for this task. Our results suggest that simple cross-lingual transfer of multimodal models yields latent multilingual multimodal misalignment, calling for more sophisticated methods for vision and multilingual language modeling.

An essential step in FrameNet Semantic Role Labeling is the Frame Identification (FrameId) task, which aims at disambiguating a situation around a predicate. Whilst current FrameId methods rely on textual representations only, we hypothesize that FrameId can profit from a richer understanding of the situational context. Such contextual information can be obtained from common sense knowledge, which is more present in images than in text. In this paper, we extend a state-of-the-art FrameId system in order to effectively leverage multimodal representations. We conduct a comprehensive evaluation on the English FrameNet and its German counterpart SALSA. Our analysis shows that for the German data, textual representations are still competitive with multimodal ones. However on the English data, our multimodal FrameId approach outperforms its unimodal counterpart, setting a new state of the art. Its benefits are particularly apparent in dealing with ambiguous and rare instances, the main source of errors of current systems. For research purposes, we release (a) the implementation of our system, (b) our evaluation splits for SALSA 2.0, and (c) the embeddings for synsets and IMAGINED words.

Current methods for knowledge graph (KG) representation learning focus solely on the structure of the KG and do not exploit any kind of external information, such as visual and linguistic information corresponding to the KG entities. In this paper, we propose a multimodal translation-based approach that defines the energy of a KG triple as the sum of sub-energy functions that leverage both multimodal (visual and linguistic) and structural KG representations. Next, a ranking-based loss is minimized using a simple neural network architecture. Moreover, we introduce a new large-scale dataset for multimodal KG representation learning. We compared the performance of our approach to other baselines on two standard tasks, namely knowledge graph completion and triple classification, using our as well as the WN9-IMG dataset. The results demonstrate that our approach outperforms all baselines on both tasks and datasets.