Natural language control of animation of task performance in a physical domain

Jugal Kumar Kalita, University of Pennsylvania

Abstract

We establish a link from natural language statements describing actions to be performed by an agent to a semantic representation suitable for achieving effective control of a computer-driven graphical animation system. A representation scheme based on decompositional analysis is developed emphasizing the requirement that algorithmic implementability of the underlying semantic primitives is our primary concern. Our primitives pertain to mechanical characteristics of the "kernel" tasks denoted by a class of action verbs (verbs whose underlying tasks deal with an agent manipulating one or more objects); they refer to geometric constraints and goals that need to be achieved, kinematic and dynamic characteristics, and certain aspectual characteristics such as repetitiveness of one or more sub-actions, definedness of termination points, etc. We provide lexical entries for a few verbs in terms of such primitives. We also analyze the manner in which prepositional and adverbial modifiers affect the representation as well as the execution of the basic actions denoted by the verbs. Such modifiers either provide values of arguments for the verbs' internal representations, modify default argument values, or provide values of non-obligatory arguments. We obtain semantic representations for a few prepositions and adverbs in a fashion integrable into the scheme for verbal meaning representation. We have developed a system to demonstrate the validity of the results obtained; such a system establishes channels of communication with existing animation software developed at the Graphics Laboratory at the University of Pennsylvania.

Subject Area

Computer science|Artificial intelligence

Recommended Citation

Kalita, Jugal Kumar, "Natural language control of animation of task performance in a physical domain" (1990). Dissertations available from ProQuest. AAI9112582.
https://repository.upenn.edu/dissertations/AAI9112582

Share

COinS