Abstract: Reference resolution is an important problem, one that is essential to understand and successfully handle context of different kinds. This context includes both previous turns and context that pertains to non-conversational entities, such as entities on the user’s screen or those running in the background. While LLMs have been shown to be extremely powerful for a variety of tasks, their use in reference resolution, particularly for non-conversational entities, remains underutilized. This paper demonstrates how LLMs can be used to create an extremely effective system to resolve references of various types, by showing how reference resolution can be converted into a language modeling problem, despite involving forms of entities like those on screen that are not traditionally conducive to being reduced to a text-only modality. We demonstrate large improvements over an existing system with similar functionality across different types of references, with our smallest model obtaining absolute gains of over 5% for on-screen references. We also benchmark against GPT-3.5 and GPT-4, with our smallest model achieving performance comparable to that of GPT-4, and our larger models substantially outperforming it.

Lay summary (by Claude 3 Sonnet): This research is about improving how computers understand what people are referring to when they use words like “it”, “that”, or mention things on their screen. Large language models (like ChatGPT) are very good at understanding human language, but have not been used much for this “reference resolution” problem, especially for things not in the conversation itself like icons on a computer screen. The researchers showed how to turn reference resolution into a language modeling task that large language models can solve. Their system performed much better than an existing system at resolving different kinds of references, with big improvements for things on-screen. It also matched or outperformed GPT-3.5 and GPT-4, which are very capable language models. This shows large language models can be extremely helpful for the important task of understanding what users are referring to in context.