Abhidip Bhattacharyya

Even though current vision language (V+L) models have achieved success in generating image captions, they often lack specificity and overlook various aspects of the image. Additionally, the attention learned through weak supervision operates opaquely and is difficult to control. To address these limitations, we propose the use of semantic roles as control signals in caption generation. Our hypothesis is that, by incorporating semantic roles as signals, the generated captions can be guided to follow specific predicate argument structures. To validate the effectiveness of our approach, we conducted experiments using data and compared the results with a baseline model VL-BART(CITATION). The experiments showed a significant improvement, with a gain of 45% in Smatch score (Standard NLP evaluation metric for semantic representations), demonstrating the efficacy of our approach. By focusing on specific objects and their associated semantic roles instead of providing a general description, our framework produces captions that exhibit enhanced quality, diversity, and controllability.

Visual Question Answering (VQA) requires a vision-language model to reason over both visual and textual inputs to answer questions about images. In this work, we investigate whether incorporating explicit semantic information, in the form of Abstract Meaning Representation (AMR) graphs, can enhance model performance—particularly in low-resource settings where training data is limited. We augment two vision-language models, LXMERT and BLIP-2, with sentence- and document-level AMRs and evaluate their performance under both full and reduced training data conditions. Our findings show that in well-resourced settings, models (in particular the smaller LXMERT) are negatively impacted by incorporating AMR without specialized training. However, in low-resource settings, AMR proves beneficial: LXMERT achieves up to a 13.1% relative gain using sentence-level AMRs. These results suggest that while addition of AMR can lower the performance in some settings, in a low-resource setting AMR can serve as a useful semantic prior, especially for lower-capacity models trained on limited data.

Discourse relations provide document structure, critical to language understanding and enabling language model performance and ethicality. In this work, we investigate how instruction-tuned Transformer models (LLaMA and Mistral) encode discourse relations, with a focus on the somewhat opposed relations on causation and antithesis. Framing the task as a next-token prediction task and applying a suite of interpretability techniques to test model internals, our findings show that certain early layers make predictive decisions at mid-sequence tokens, while some mid-level layers finalize their decisions closer to the last token. Most of the remaining layers primarily propagate earlier decisions rather than actively influencing them. Additionally, we observe that some layers exhibit a preference for one answer over alternatives, suggesting asymmetric representation of discourse-based reasoning.