School of Computing & Intelligent Systems, Faculty of Engineering
University of Ulster, Magee
Northern Ireland
Abstract
One of the most difficult tasks in Computing and Artificial Intelligence
is to develop systems which can interpret and tell stories.
We are developing a computer system called CONFUCIUS which focuses on story
interpretation and multimodal presentation. The long-term goal
is for CONFUCIUS to receive input natural
language stories or (play/movie) scripts and present them with 3D
animation, speech and non-speech audio (sound effects).
CONFUCIUS' knowledge base contains language and visual knowledge
(e.g. dictionary, syntactic parser) and visual knowledge
(3D models of characters, props and animations of actions).
Currently CONFUCIUS is able to visualise single sentences which
contain action verbs, e.g. "Paul left the gym",
and "Eunice gave Paul a loaf of bread."
CONFUCIUS has three levels of multimodal semantic representation
with an XML-based high-level
semantics, intermediate-level semantics represented in
Lexical Visual Semantic Representation based on Jackdoff's
Lexical Conceptual Structure (LCS) and
enhanced with Schank's scripts and low-level semantics in VRML.
CONFUCIUS is implemented in Java and VRML,
uses the H-anim standard for character modelling and animation,
and has potential applications
in computer games, movie/drama production and training and education.