CSE 473 Autumn 1998, Copyright, S. Tanimoto, Univ. of Washington 
Introduction to Artificial Intelligence (Oct 26, 1998)

"Representation of Meaning in Multilingual Systems"


 

Rationale for Multilingual Systems

The Internet provides a direct electronic means for connecting people across the world in new ways.
Although many people understand English, language barriers remain, especially for younger users.
Traditional "international" languages such as Esperanto and Bliss Symbols have significant learning curves.
They don't take advantage of computation.
Integrated language systems that work with concrete representations of meaning offer a different approach with a lot of potential.
 
 


 

Ontology Design
 

Ontology: (There are a variety of definitions for this term).  A set of meanings.
 In formal language theory, we define a language to be a set of sentences.
An ontology could be thought of as the corresponding set of semantic interpretations.

Design criteria:
1. Useful in serving the purpose of the system
2. Based upon a small set of component ideas the combine effectively
    (simplicity,  orthogonality)

Important sub-ontologies should cover:

time  (days, dates, minutes, hours, absolute and relative)
space (places and relationships among them)
people (names, relationships)
events (transportation, visitation, communication)
modes (narration, thinking, dreaming)
communication (parties to a message)
attitude (happy, sad)

metalanguage (definitions of new elements)
 
 

Interplay of Visual and Semantic Representations
 

Visual representations offer "language-free" suggestions of semantics.

Computational objects can have both an appearance (visual) and an internal semantics (the behaviors defined by the methods associated with the objects).

The two get mixed up (in beneficial ways) when the behaviors cause visual changes, as in animations.

Some objects can be defined using a combination of visual and textual features.
 e. g., a person can be described using a face image (visual) and a name (textual).
Additional definition can be effected by posing predicates using frames.
 
 
 
 

Translation from Internal Representations to Natural Language
 

The internal semantics is concrete and can, in principle, be translated into any language without any increate in ambiguity.

Primary challenges involve designing the translation mechanisms to be simple and efficient, yet reasonably effective in handling a variety of natural languages.

A number of the trickier parts of language generation -- handling verb tenses, for example, can be taken care of some extent by providing means to establish sub-narratives with distinct time frames, and then sticking to present-tense literal translations.

Language extensibility while striving for orthogonality, full semantic support, and complete translation ability is an interesting and difficult challenge.
 
 
 
 

Demonstrations
 

Representation of transportation events.

Representation of dates and time

Representation of thinking via sub-narratives

Introducing new people and places

Representation of space with the containment relation

Representation of travelling groups, visitation, and side trips

Representation of dreaming with nested sub-narratives
 

Research Questions
 

Can an entire general purpose language with concrete computer semantics and animation be successful?

How can language extension be designed to permit a diverse group of people to cooperate on a single, good version of the language?

How can new "words" be created from existing words and yet have concrete meanings that are more than the sum of the parts?
 
 
 


 
 

Last modified: October 26, 1998

Steve Tanimoto

tanimoto@cs.washington.edu