The Role of Semantic Parsing in Understanding Procedural Text

Findings of the Association for Computational Linguistics: EACL 2023, pages 1837–1849 May 2-6, 2023 ©2023 Association for Computational Linguistics The Role of Semantic Parsing in Understanding Procedural Text Hossein Rajaby Faghihi1, Parisa Kordjamshidi1, Choh Man Teng2, and James Allen2 1 Michigan State University, 2 Florida Institute for Human and Machine Cognition {rajabyfa, kordjams}@msu.edu, {jallen, cmteng}@ihmc.org Abstract In this paper, we investigate whether symbolic semantic representations, extracted from deep semantic parsers, can help reasoning over the states of involved entities in a procedural text. We consider a deep semantic parser (TRIPS) and semantic role labeling as two sources of semantic parsing knowledge. First, we propose PROPOLIS, a symbolic parsing-based proce- dural reasoning framework. Second, we inte- grate semantic parsing information into state- of-the-art neural models to conduct procedural reasoning. Our experiments indicate that ex- plicitly incorporating such semantic knowledge improves procedural understanding. This paper presents new metrics for evaluating procedural reasoning tasks that clarify the challenges and identify differences among neural, symbolic, and integrated models. 1 Introduction Procedural reasoning is the ability to track entities and understand their evolution given a sequence of actions (Tandon et al., 2020). This kind of reason- ing is crucial in understanding recipes (Bosselut et al., 2018; Yagcioglu et al., 2018), manuals and tu- torials (Tandon et al., 2020; Wu et al., 2022), cyber- security text (Pal et al., 2021), natural events (Tan- don et al., 2020), and even stories (Storks et al., 2021). An example of a procedural text in the natu- ral event domain, its entities of interest, and their state changes are shown in Figure 1. Inferring actions and their impact on entities in- volved in a procedural text can be challenging in various aspects. First, there are dependencies be- tween steps to be considered in predicting a plausi- ble action set. For instance, an entity destroyed at step t of the process cannot be moved again at step t + 1. Second, some sentences contain ambiguous local signals by including multiple action verbs. For example, "The oxygen is consumed in the pro- cess of forming carbon dioxide.", where the oxygen Process Participants Sentences plant animal bone oil Before the process begins ? ? - - 1. Plants and animals die in a watery environment watery environment watery environment - - 2. Over time, sediments build over sediment sediment - - 3. The body decomposes sediment - sediment - 4. Gradually buried material becomes oil - - - sediment Figure 1: An example of procedural text and its annota- tion (location of objects). ‘-’ means the entity does not exist; ‘?’ means the entity’s location is unknown. is being destroyed, and the carbon dioxide is be- ing created. Third, the sentences are incomplete in some steps. For instance, a step of the process might only indicate "is buried in mud", which can- not be understood without context. Fourth, finding the properties of some entities may require reason- ing over both the global context and local relations. For instance, in the sentences “1. Magma …

Content truncated to 500 words