MSBART-F infers dependencies with higher accuracy than the comparisons (prediction accuracy: 0.779, which is not fantastic but is pretty good). 

University of Rochester

1. Can we accurately and automatically generate subtasks given an input complex task?
2. Can we correctly identify the temporal dependencies between subtasks?

Research Questions (Learning to Decompose and Organize Complex Tasks)

Zhang, Y., Jauhar, S. K., Kiseleva, J., White, R., &amp; Roth, D. (2021). Learning to Decompose and Organize Complex Tasks. ACL Anthology. https://www.aclweb.org/anthology/2021.naacl-main.217/. 

Learning to Decompose and Organize Complex Tasks

Across the board, multi-source input is significantly better than single-source input in terms of accuracy. MSBART-F performs best on every measurement except Rogue-1 on WKH-R. 

Conclusion: The MSBART-F model (the model that includes input from multiple sources and encodes for relevance, abstraction, and consensus) generates the highest quality subtasks when compared to state-of-the-art variants.

Results: Research Question 1 (Learning to Decompose and Organize Complex Tasks)

Results: Research Question 2 (Learning to Decompose and Organize Complex Tasks)

The reason the proposed model, MSBART-F, was as accurate as it was is that it encoded relevance, abstraction, and consensus .

Learn Before

Related