Resolving References in Visually-Grounded Dialogue via Text Generation
This addresses the problem of improving discourse processing for VLMs in dialogue systems, though it is incremental as it builds on existing models and methods.
The paper tackles the challenge of reference resolution in visually-grounded dialogue by fine-tuning an LLM to generate descriptions from linguistic context and using a VLM for zero-shot retrieval, achieving results that exceed baseline performance on a manually annotated dataset.
Vision-language models (VLMs) have shown to be effective at image retrieval based on simple text queries, but text-image retrieval based on conversational input remains a challenge. Consequently, if we want to use VLMs for reference resolution in visually-grounded dialogue, the discourse processing capabilities of these models need to be augmented. To address this issue, we propose fine-tuning a causal large language model (LLM) to generate definite descriptions that summarize coreferential information found in the linguistic context of references. We then use a pretrained VLM to identify referents based on the generated descriptions, zero-shot. We evaluate our approach on a manually annotated dataset of visually-grounded dialogues and achieve results that, on average, exceed the performance of the baselines we compare against. Furthermore, we find that using referent descriptions based on larger context windows has the potential to yield higher returns.