SemEval-2012 Task 3: Spatial Role Labeling

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 365–373, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics SemEval-2012 Task 3: Spatial Role Labeling Parisa Kordjamshidi Katholieke Universiteit Leuven parisa.kordjamshidi@ cs.kuleuven.be Steven Bethard University of Colorado steven.bethard@ colorado.edu Marie-Francine Moens Katholieke Universiteit Leuven sien.moens@ cs.kuleuven.be Abstract This SemEval2012 shared task is based on a recently introduced spatial annotation scheme called Spatial Role Labeling. The Spatial Role Labeling task concerns the extraction of main components of the spatial semantics from nat- ural language: trajectors, landmarks and spa- tial indicators. In addition to these major components, the links between them and the general-type of spatial relationships includ- ing region, direction and distance are targeted. The annotated dataset contains about 1213 sentences which describe 612 images of the CLEF IAPR TC-12 Image Benchmark. We have one participant system with two runs. The participant’s runs are compared to the sys- tem in (Kordjamshidi et al., 2011c) which is provided by task organizers. 1 Introduction One of the essential functions of natural language is to talk about spatial relationships between objects. The sentence “Give me the book on AI on the big table behind the wall.” expresses information about the spatial configuration of the objects (book, table, wall) in some space. Particularly, it explains the re- gion occupied by the book with respect to the table and the direction (orientation) of the table with re- spect to the wall. Understanding such spatial utter- ances is a problem in many areas, including robotics, navigation, traffic management, and query answer- ing systems (Tappan, 2004). Linguistic constructs can express highly complex, relational structures of objects, spatial relations be- tween them, and patterns of motion through space relative to some reference point. Compared to nat- ural language, formal spatial models focus on one particular spatial aspect such as orientation, topol- ogy or distance and specify its underlying spatial logic in detail (Hois and Kutz, 2008). These for- mal models enable spatial reasoning that is difficult to perform on natural language expressions. Learning how to map natural language spatial in- formation onto a formal representation is a challeng- ing problem. The complexity of spatial semantics from the cognitive-linguistic point of view on the one hand, the diversity of formal spatial represen- tation models in different applications on the other hand and the gap between the specification level of the two sides has led to the present situation that no well-defined framework for automatic spatial infor- mation extraction exists that can handle all of these aspects. In a previous paper (Kordjamshidi et al., 2010b), we introduced the task of spatial role labeling (SpRL) and proposed an annotation scheme that is language-independent and practically facilitates the application of machine learning techniques. Our framework consists of a set of spatial roles based on the theory of holistic spatial semantics (Zlat- evl, 2007) with the intent of covering the main as- pects of spatial concepts at a course level, includ- ing both static and dynamic spatial semantics. This shared task is …

Content truncated to 500 words