Annotating Temporally-Anchored Spatial Knowledge on Top of OntoNotes Semantic Roles

Annotating Temporally-Anchored Spatial Knowledge on Top of OntoNotes Semantic Roles Alakananda Vempala and Eduardo Blanco Human Intelligence and Language Technologies Lab University of North Texas AlakanandaVempala@my.unt.edu, eduardo.blanco@unt.edu Abstract This paper presents a two-step methodology to annotate spatial knowledge on top of OntoNotes semantic roles. First, we manipulate semantic roles to automatically generate potential additional spatial knowledge. Second, we crowdsource annotations with Amazon Mechanical Turk to either validate or discard the potential additional spatial knowledge. The resulting annotations indicate whether entities are or are not located somewhere with a degree of certainty, and temporally anchor this spatial information. Crowdsourcing experiments show that the additional spatial knowledge is ubiquitous and intuitive to humans, and experimental results show that it can be inferred automatically using standard supervised machine learning techniques. Keywords: spatial knowledge, temporally-anchored knowledge, semantic inference 1. Introduction Extracting meaning from text has received considerable at- tention in the last decade. In particular, semantic role la- beling and efforts focused on spatial meaning—both cor- pora development and automatic tools—have become pop- ular. Semantic roles capture semantic links between pred- icates and their arguments; they capture who did what to whom, how, when and where (Baker et al., 1998; Palmer et al., 2005). Efforts targeting spatial meaning use special- ized relations such as TRAJECTOR and LANDMARK (Kord- jamshidi et al., 2011; Kolomiyets et al., 2013), or define subtasks such as identifying spatial elements and spatial signals (Pustejovsky et al., 2015). There are several corpora with semantic role annotations, e.g., FrameNet (Baker et al., 1998), PropBank (Palmer et al., 2005), and OntoNotes (Hovy et al., 2006). While se- mantic roles are useful, there is much more meaning in all but the simplest statements. Consider the sentence John drove to San Francisco for a doctor’s appointment and the semantic roles annotated in OntoNotes (Figure 1, solid ar- rows). On top of these valuable semantic roles, one can infer that John had LOCATION San Francisco for a short period of time after drove (more precisely, during the doc- tor’s appointment), but probably not long after, long be- fore or during drove. This additional knowledge is intuitive to humans although it is disregarded by existing tools and highly ambiguous: if John drove home to San Francisco af- ter a vacation in Colorado, it is reasonable to believe that he had LOCATION San Francisco well after drove. This paper presents (1) annotations of temporally-anchored spatial knowledge on top of OntoNotes semantic roles, and (2) experiments to extract this knowledge automatically. We release a new resource1 that annotates where entities are and are not located, and temporally anchor this infor- mation. Additionally, we incorporate certainty levels since there is often evidence that something is (or is not) located somewhere, but one cannot fully commit. 1Available at http://hilt.cse.unt.edu/ Annotating Temporally-Anchored Spatial Knowledge on Top of OntoNotes Semantic Roles Alakananda Vempala and Eduardo Blanco Human Intelligence and Language Technologies Lab University of North Texas AlakanandaVempala@my.unt.edu, eduardo.blanco@unt.edu Abstract This paper presents a framework to infer additional meaning from existing semantic role representations. …

Content truncated to 500 words