CEMIR vs. FISER: a comparative study of reasoning-based and enrichment-based approaches to pragmatic intent recognition
A boundary condition for text-derived multimodal enrichment is suggested: when the relevant task constraints are already available in the textual instruction, synthetic multimodal augmentation may not provide measurable benefit over structured reasoning on the same text.
Ziya Hosseynzadeh, Farjad Fallah, Farnoosh Fallah et al.
· 0 citations