KUL-Eval: A Combinatory Categorial Grammar Approach for Improving Semantic Parsing of Robot Commands using Spatial Context

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 385–390, Dublin, Ireland, August 23-24, 2014. KUL-Eval: A Combinatory Categorial Grammar Approach for Improving Semantic Parsing of Robot Commands using Spatial Context Willem Mattelaer, Mathias Verbeke and Davide Nitti Department of Computer Science, KU Leuven, Belgium willem.mattelaer@gmail.com mathias.verbeke@cs.kuleuven.be davide.nitti@cs.kuleuven.be Abstract When executing commands, a robot has a certain level of contextual knowledge about the environment in which it oper- ates. Taking this knowledge into account can be beneficial to disambiguate com- mands with multiple interpretations. We present an approach that uses combina- tory categorial grammars for improving the semantic parsing of robot commands that takes into account the spatial context of the robot. The results indicate a clear improvement over non-contextual seman- tic parsing. This work was done in the context of the SemEval-2014 task on su- pervised semantic parsing of spatial robot commands. 1 Introduction One of the long-standing goals of robotics is to build autonomous robots that are able to perform everyday tasks. Two important requirements to achieve this are an efficient way of communicating with the robot, and transforming these commands such that the robot is able to capture their mean- ing. Furthermore, this needs to be consistent with the context in which the robot is operating, i.e., the robot’s belief. Semantic parsing focuses on translating natural language (NL) into a formal representation that captures the meaning of the sentence. Most of the current semantic parsing approaches are non- contextual, i.e., they do not take into account the context in which the command sentence should be executed. This can lead to erroneous parses, most often due to ambiguity in the original sentence. Consider the following example sentence “Move This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ the pyramid on the blue cube on the gray cube”. This sentence has two valid interpretations. Either the robot needs to move the pyramid that is cur- rently standing on the blue cube and put it on the gray cube, or move the pyramid and place it on the blue cube that is standing on the gray cube. Humans will decide on the correct interpreta- tion by taking into account the context. For in- stance, by looking at Figure 1, it is clear that the second interpretation is not possible, because there is no blue cube on top of a gray cube. However, there is a pyramid on top of a blue cube, making the first interpretation possible. The goal of this paper is to improve on non-contextual semantic parsing by tailoring the context to guide the parser. In this way, part of the ambiguity that causes mul- tiple interpretations can be resolved. Figure 1: Possible situation (taken from (Dukes, 2013b)). Our approach consists of two steps. First, non- contextual semantic parsing using combinatory categorial grammars (CCG) (Steedman, 1996; Steedman, 2000) is performed on the sentence. This returns multiple possible …

Content truncated to 500 words