Yash Lal will be presenting his proposal on reasoning in narratives and plans this coming Monday. Come join us. Details below.

Time: Monday, November 4 at 12:30pm

Venue: NCS 120 or Zoom link

Title: Using Causal Knowledge to Improve Reasoning in Narratives and Plans
 

Abstract: Humans reason about everyday situations by making commonsense-based inferences, derived not only from explicitly stated information but also from implicit, unstated knowledge. Consider the sentence, John walked by a man peering into the hood of a car in the driveway with tools around him. If one were to ask “Why was John peering into the hood of a car?", humans can provide a range of inferential answers such as the engine was broken (cause), he was finding the problem in the engine (effect), he was working to fix the car (goal), and he wanted to use the car going forward (motivation). To infer such answers, humans use explicit information, as well as their causal knowledge of everyday situations. In this thesis, I investigate whether NLP models have different aspects of such causal knowledge and how to improve their understanding of narratives and plans. For each domain, I create a resource to benchmark how much causal knowledge NLP models possess to reason about them. Then, for each domain, I develop methods to improve their comprehension of such knowledge.