Generating Simulations of Motion Events from Verbal Descriptions

Proceedings of the Third Joint Conference on Lexical and Computational Semantics (*SEM 2014), pages 99–109, Dublin, Ireland, August 23-24 2014. Generating Simulations of Motion Events from Verbal Descriptions James Pustejovsky Computer Science Dept. Brandeis University Waltham, MA USA jamesp@cs.brandeis.edu Nikhil Krishnaswamy Computer Science Dept. Brandeis University Waltham, MA USA nkrishna@brandeis.edu Abstract In this paper, we describe a computational model for motion events in natural lan- guage that maps from linguistic expres- sions, through a dynamic event interpreta- tion, into three-dimensional temporal sim- ulations in a model. Starting with the model from (Pustejovsky and Moszkow- icz, 2011), we analyze motion events us- ing temporally-traced Labelled Transition Systems. We model the distinction be- tween path- and manner-motion in an op- erational semantics, and further distin- guish different types of manner-of-motion verbs in terms of the mereo-topological re- lations that hold throughout the process of movement. From these representations, we generate minimal models, which are realized as three-dimensional simulations in software developed with the game en- gine, Unity. The generated simulations act as a conceptual “debugger” for the se- mantics of different motion verbs: that is, by testing for consistency and infor- mativeness in the model, simulations ex- pose the presuppositions associated with linguistic expressions and their composi- tions. Because the model generation com- ponent is still incomplete, this paper fo- cuses on an implementation which maps directly from linguistic interpretations into the Unity code snippets that create the sim- ulations. 1 Introduction Semantic interpretation requires access to both knowledge about words and how they compose. As the linguistic phenomena associated with lexi- cal semantics have become better understood, sev- eral assumptions have emerged across most mod- els of word meaning. These include the following: (1) a. Lexical meaning involves some sort of “componential analysis”, either through predicative primitives or a system of types. b. The selectional properties of predicators can be explained in terms of these com- ponents; c. An understanding of event semantics and the different role of event participants seems crucial for modeling linguistic ut- terances. As a starting point in lexical semantic analysis, a standard methodology in both theoretical and computational linguistics is to identify features in a corpus that differentiate the data in meaningful ways; meaningful in terms of prior theoretical as- sumptions or in terms of observably differentiated behaviors. Combining these strategies we might, for instance, take a theoretical constraint that we hope to justify through behavioral distinctions in the data. An example of this is the theoretical claim that motion verbs can be meaningfully di- vided into two classes: manner- and path-oriented predicates (Talmy, 1985; Jackendoff, 1983; Talmy, 2000). These constructions can be viewed as en- coding two aspects of meaning: how the move- ment is happening and where it is happening. The former strategy is illustrated in (2a) and the latter in (2b) (where m indicates a manner verb, and p indicates a path verb). (2) a. The ball rolledm. b. The ball crossedp the room. With both of the verb types, adjunction can make …

Content truncated to 500 words