On the Representation of Perceptual Knowledge for Understanding Reference Expressions
Recent research has enabled important progress in developing agents aimed at real-world linguistic interaction with humans. Hence, within the general shift of research focus from “information” to “knowledge”, an important question is how to apply large-sc
- PDF / 893,616 Bytes
- 15 Pages / 430 x 660 pts Page_size
- 25 Downloads / 188 Views
Abstract. Recent research has enabled important progress in developing agents aimed at real-world linguistic interaction with humans. Hence, within the general shift of research focus from “information” to “knowledge”, an important question is how to apply large-scale knowledge resources in order to improve agents’ capabilities of linguistic interaction with humans. This paper presents research toward an efficient representation of the necessary perceptual knowledge in dialogue with a particular focus on reference expressions. We generalize an existing formal model of reference expressions involving perceptual grouping in order to account for a number of types of reference expressions that the previous model could not account for. Our model yields an increase in both coverage and accuracy of referent identification − which has been confirmed in preliminary experiments. We outline an algorithm for the future application of this model to other languages, showing how the model can be extended to deal with large-scale multi-language input data. Keywords: representation of perceptual knowledge, perceptual grouping, reference expressions, language-independent systems.
1
Introduction
Recently, the utilization of large-scale knowledge resources (LKR) has been a central issue in achieving progress in different research areas such as analysis of spoken language characteristics, systematization of archeological information or language-learning support systems. In particular, the application of LKR in research in the field of linguistics is a very promising research direction. At the same time, developments in a multitude of research areas like speech recognition, robotics, etc. have enabled important progress in developing agents aimed at real-world interaction with humans. Thus, within this general shift of research focus from “information” to “knowledge”, an important question is how to use large-scale knowledge resources in order to improve agents’ capabilities of interaction with humans through natural language. An important research aim in improving agents’ capabilities of interaction with humans has been to improve T. Tokunaga and A. Ortega (Eds.): LKR 2008, LNAI 4938, pp. 280–294, 2008. c Springer-Verlag Berlin Heidelberg 2008
On the Representation of Perceptual Knowledge
281
their natural language understanding. A fundamental type of human expression − in particular in task-oriented dialogue − are reference expressions. This type of expression is a linguistic entity used to discriminate a specific object from its environment and the rest of the world. Thus, an agent’s capability to handle this type of linguistic expression correctly is an important part of increasing human-agent interaction capabilities. Reference expressions are to a large degree multi-modal; i.e. they include exophoric expressions such as “this one” or “that” in connection with gesturing (e.g.; pointing). It is clear a fuller model of reference expressions must be a multi-modal model including an account of these different channels and how they com
Data Loading...
