Identifying Physical Object Use in Sentences

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11362 - 11372 December 7-11, 2022 ©2022 Association for Computational Linguistics Identifying Physical Object Use in Sentences Tianyu Jiang and Ellen Riloff School of Computing University of Utah Salt Lake City, UT 84112 {tianyu, riloff}@cs.utah.edu Abstract Commonsense knowledge about the typical functions of physical objects allows people to make inferences during sentence understand- ing. For example, we infer that “Sam enjoyed the book” means that Sam enjoyed reading the book, even though the action is implicit. Prior research has focused on learning the prototyp- ical functions of physical objects in order to enable inferences about implicit actions. But many sentences refer to objects even when they are not used (e.g., “The book fell”). We argue that NLP systems need to recognize whether an object is being used before inferring how the object is used. We define a new task called Ob- ject Use Classification that determines whether a physical object mentioned in a sentence was used or likely will be used. We introduce a new dataset for this task and present a classification model that exploits data augmentation methods and FrameNet when fine-tuning a pre-trained language model. We also show that object use classification combined with knowledge about the prototypical functions of objects has the potential to yield very good inferences about implicit and anticipated actions. 1 Introduction Physical objects play an important role in daily life. People use them for different purposes, for example we use knives for cutting, cars for trans- portation, and books for reading. Recent work by Jiang and Riloff (2021b) argued that most human- made physical artifacts were created for a specific purpose, and that commonsense knowledge about an object’s prototypical function is essential for nat- ural language understanding. For example, “she finished the puzzle” and “she finished the cigarette” implicitly refer to different actions associated with puzzles (solving) vs. cigarettes (smoking). Simi- larly, humans interpret “he used a gun” as a shoot- ing but “he used a knife” as a stabbing based on our knowledge of guns and knives. Jiang and Riloff (2021b) developed a method to learn the prototypical functions for physical arti- facts from text corpora, with the goal of producing a commonsense knowledge resource for physical objects. However an open question is how to ap- ply this knowledge for sentence understanding. It would be risky to assume that objects are always used in the most typical way because objects can be used in atypical ways too. For example, “Max used the knife to open the bottle” probably means that Max popped the top off the bottle with the knife, not that Max cut the bottle. But as we will discuss in Section 4, our study found that physical artifacts are used in the prototypical way most of the time (96%), so it is a very good assumption. A much bigger problem for applying knowledge of prototypical functions is that physical objects are often mentioned when they are not …

Content truncated to 500 words