Choice-75: A Dataset on Decision Branching in Script Learning
This addresses the problem of enabling machines to reason about narratives with implicit information for AI researchers, but it is incremental as it builds on existing script learning by adding branching scenarios.
The authors tackled the problem of script learning by introducing Choice-75, a dataset with 75 scripts and over 600 scenarios that challenge systems to handle decision branching, and found that current large language models show decent performance but have notable headroom in hard scenarios.
Script learning studies how stereotypical events unfold, enabling machines to reason about narratives with implicit information. Previous works mostly consider a script as a linear sequence of events while ignoring the potential branches that arise due to people's circumstantial choices. We hence propose Choice-75, the first benchmark that challenges intelligent systems to make decisions given descriptive scenarios, containing 75 scripts and more than 600 scenarios. We also present preliminary results with current large language models (LLM). Although they demonstrate overall decent performance, there is still notable headroom in hard scenarios.