In multi-task settings, combining multiple modalities allows task representations to draw on signals that a single modality alone would not capture. However, individual tasks may depend disproportionately on particular modalities, a structure that conventional fusion-based architectures do not model explicitly, instead enforcing a unified representation space. Consequently, a task's gradients can affect parameters tied to modalities it barely relies on, leaving features that compromise between competing demands. Although prior work has explored task-weighting and loss-balancing strategies, their effectiveness is limited by the absence of explicit modality separation, preventing models from optimising each modality according to its task-specific importance.
Proposed Method: MOSAIC
To address these limitations, we propose MODality-Separated Architecture with Inter-modal Cross-attention (MOSAIC), which preserves unimodal representations while enabling controlled interaction between them, and uses modality-specific losses to route gradients to their corresponding partitions.
Experimental Results
On the Corpus of Everyday Japanese Conversation (CEJC), MOSAIC improves average macro-F1 by 7.4 points across four tasks with a 72% reduction in backbone parameters relative to a fusion-based baseline. Our results demonstrate that modality partitioning yields consistent gains across tasks while requiring far fewer backbone parameters than a fusion-based design.