Inferring Temporally-Anchored Spatial Knowledge from Semantic Roles

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 452–461, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Inferring Temporally-Anchored Spatial Knowledge from Semantic Roles Eduardo Blanco and Alakananda Vempala Human Intelligence and Language Technologies Lab University of North Texas Denton, TX, 76203 eduardo.blanco@unt.edu, AlakanandaVempala@my.unt.edu Abstract This paper presents a framework to infer spa- tial knowledge from verbal semantic role rep- resentations. First, we generate potential spa- tial knowledge deterministically. Second, we determine whether it can be inferred and a degree of certainty. Inferences capture that something is located or is not located some- where, and temporally anchor this informa- tion. An annotation effort shows that infer- ences are ubiquitous and intuitive to humans. 1 Introduction Extracting semantic relations from text is at the core of text understanding. Semantic relations encode se- mantic connections between words. For example, from (1) Bill couldn’t handle the pressure and quit yesterday, one could extract that the CAUSE of quit was the pressure. Doing so would help answering question Why did Bill quit? and determining that the pressure started before Bill quit. In the past years, computational semantics has re- ceived a significant boost. But extracting all seman- tic relations in text—even in single sentences—is still an elusive goal. Most existing approaches target either a single relation, e.g., PART-WHOLE (Girju et al., 2006), or relations that hold between arguments following some syntactic construction, e.g., posses- sives (Tratz and Hovy, 2013). Among the latter kind, the task of verbal semantic role labeling focuses on extracting semantic links exclusively between verbs and their arguments. PropBank (Palmer et al., 2005) is a popular corpus for this task, and tools to ex- tract verbal semantic roles have been proposed for years (Carreras and M`arquez, 2005). Some semantic relations hold forever, e.g., the CAUSE of event quit in example (1) above is pres- sure. Discussing when this CAUSE holds is some- what artificial: at some point Bill quit, and he did so S NP VP NNP AUX VP John was VBN PP incarcerated THEME LOCATION at Shawshank prison Figure 1: Semantic roles (solid arrows) and addi- tional spatial knowledge (discontinuous arrow). because of the pressure. But LOCATION and other semantic relations often do not hold forever. For ex- ample, while buildings typically have one location during their existence, people and objects such as cars and books do not: they participate in events and as a result their locations change. This paper presents a framework to infer temporally-anchored spatial knowledge from verbal semantic roles. Specifically, our goal is to infer whether something is located somewhere or not lo- cated somewhere, and temporally anchor this spa- tial information. Consider sentence (2) John was incarcerated at Shawshank prison and its semantic roles (Figure 1, solid arrows). Given these roles, we aim at inferring that John had LOCATION Shaw- shank prison during event incarcerated, and that he (probably) did not have this LOCATION before and after (discontinuous arrow). Our intuition is …

Content truncated to 500 words