UTD-SpRL: A Joint Approach to Spatial Role Labeling

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 419–424, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics UTD-SpRL: A Joint Approach to Spatial Role Labeling Kirk Roberts and Sanda M. Harabagiu Human Language Technology Research Institute University of Texas at Dallas Richardson, TX 75083, USA {kirk, sanda}@hlt.utdallas.edu Abstract We present a joint approach for recognizing spatial roles in SemEval-2012 Task 3. Can- didate spatial relations, in the form of triples, are heuristically extracted from sentences with high recall. The joint classification of spatial roles is then cast as a binary classification over the candidates. This joint approach allows for a rich feature set based on the complete rela- tion instead of individual relation arguments. Our best official submission achieves an F1- measure of 0.573 on relation recognition, best in the task and outperforming the previous best result on the same data set (0.500). 1 Introduction A significant amount of spatial information in natu- ral language is encoded in spatial relationships be- tween objects. In this paper, we present our ap- proach for detecting the special case of spatial re- lations evaluated in SemEval-2012 Task 3, Spatial Role Labeling (SpRL) (Kordjamshidi et al., 2012). This task considers the most common type of spa- tial relationships between objects, namely those de- scribed with a spatial preposition (e.g., in, on, over) or a spatial phrase (e.g., in front of, on the left), re- ferred to as the spatial INDICATOR. A spatial INDI- CATOR connects an object of interest (the TRAJEC- TOR) with a grounding location (the LANDMARK). Examples of this type of spatial relationship include: (1) [cars]T parked [in front of]I the [house]L. (2) [bushes]T1 and small [trees]T2 [on]I the [hill]L. (3) a huge [column]L with a [football]T [on top]I. (4) [trees]T [on the right]I. [∅]L SpRL is a type of semantic role labeling (SRL) (Palmer et al., 2010), where the spatial INDICA- TOR is the predicate (or trigger) and the TRAJEC- TOR and LANDMARK are its two arguments. Previ- ous approaches to SpRL (Kordjamshidi et al., 2011) have largely followed the commonly employed SRL pipeline: (1) find predicates (i.e., the INDICATOR), (2) recognize the predicate’s syntactic constituents, and (3) classify the constituent’s role (i.e., TRA- JECTOR, LANDMARK, or neither). The problem with this approach is that arguments are considered largely in isolation. Consider the following: (5) there is a picture on the wall above the bed. This sentence contains three objects (picture, wall, and bed) and two INDICATORs (on and above). Since the most common spatial relation pattern is simply trajector-indicator-landmark (as in Examples (1) and (2)), the triple wall-above-bed is a likely can- didate relation. However, the semantics of these ob- jects invalidates the relation (i.e., walls are beside beds, ceilings are above them). Instead the correct relation is picture-above-bed because the preposi- tion above syntactically attaches to picture instead of wall. Prepositional attachment, however, is a dif- ficult syntactic problem solved largely through the use of semantics, so an understanding of the con- sistency of spatial …

Content truncated to 500 words