Learning Action Conditions from Instructional Manuals for Instruction Understanding

TL;DR

This paper introduces a new task called action condition inference, creating a dataset from instructional manuals and developing models to better understand action pre- and postconditions for AI applications.

Contribution

It presents a large-scale, human-annotated dataset and a weakly supervised approach for inferring action conditions in instructional texts, advancing NLP understanding of complex instructions.

Findings

>20% F1-score improvement with context-aware models

>6% F1-score gain using proposed heuristics

Demonstrates effectiveness of leveraging entire instruction context

Abstract

The ability to infer pre- and postconditions of an action is vital for comprehending complex instructions, and is essential for applications such as autonomous instruction-guided agents and assistive AI that supports humans to perform physical tasks. In this work, we propose a task dubbed action condition inference, and collecting a high-quality, human annotated dataset of preconditions and postconditions of actions in instructional manuals. We propose a weakly supervised approach to automatically construct large-scale training instances from online instructional manuals, and curate a densely human-annotated and validated dataset to study how well the current NLP models can infer action-condition dependencies in the instruction texts. We design two types of models differ by whether contextualized and global information is leveraged, as well as various combinations of heuristics to construct the weak supervisions. Our experimental results show a >20% F1-score improvement with considering the entire instruction contexts and a >6% F1-score benefit with the proposed heuristics.