Spatial AMR: Expanded Spatial Annotation in the Context of a Grounded Minecraft Corpus

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4883–4892 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4883 Spatial AMR: Expanded Spatial Annotation in the Context of a Grounded Minecraft Corpus Julia Bonn, Martha Palmer, Jon Cai, Kristin Wright-Bettner University of Colorado at Boulder Boulder, Colorado, USA {julia.bonn, martha.palmer, jon.z.cai, kristin.wrightbettner}@colorado.edu Abstract This paper presents an expansion to the Abstract Meaning Representation (AMR) annotation schema that captures fine-grained semantically and pragmatically derived spatial information in grounded corpora. We describe a new lexical category conceptualization and set of spatial annotation tools built in the context of a multimodal corpus consisting of 185 3D structure-building dialogues between a human architect and human builder in Minecraft. Minecraft provides a particularly beneficial spatial relation-elicitation environment because it automatically tracks locations and orientations of objects and avatars in the space according to an absolute Cartesian coordinate system. Through a two-step process of sentence-level and document-level annotation designed to capture implicit information, we leverage these coordinates and bearings in the AMRs in combination with spatial framework annotation to ground the spatial language in the dialogues to absolute space. Keywords: AMR, Spatial Relations, PropBank, SRL, Semantic Roles, Minecraft, Frame of Reference, Annotation 1. Introduction This paper presents a spatial addendum to the Abstract Meaning Representation (AMR) semantic annotation schema (Banarescu et al., 2013) with accompanying spatial conceptualization, PropBank rolesets (Palmer et al., 2005), and grounded annotated corpus. AMR in its current form represents sentences as directed, rooted, acyclic graphs that capture core surface-level semantics while abstracting away from syntactic specificity. While AMR has not previously approached spatial semantics with much detail (Bonial et al., 2019), the implicit argument marking capability introduced by O’Gorman et al. (2018b) makes AMR an especially good fit for annotating spatial language. Spatial relations are not only semantically complex, they leverage pragmatically essential knowledge about the spatial characteristics of entities and their frameworks to map from linguistic expression onto space and from one spatial framework to another. This addendum takes the fine-grained spatial semantics and object grounding strategies of previous schemata (Pustejovsky, 2017; Gotou et al., 2016; Pustejovsky and Krishnaswamy, 2016; Tellex et al., 2011) and folds them into multi-sentence AMR (MS-AMR). The result is a comprehensive annotation tool that can handle fine- grained explicit and implicit nested spatial relationships that are grounded in quantified space and merged fluidly with event dynamics. Because the spatial annotations are incorporated into the domain-general AMR graphs, this approach also captures information about how spatial relations are expressed in the context of whole sentences and overall discourse. The new set of tools and practices we present span single sentence and multi-sentence annotation. At the single sentence level, we propose a new set of general semantic frames and roles as well as relation-specific rolesets. The rolesets represent the practical application of the new spatial conceptualization we also present here. The conceptualization and rolesets target lexical units from diverse parts of speech (now including prepositions and adverbs) and account …

Content truncated to 500 words