Japan Advanced Institute of Science and Technology

Japan Advanced Institute of Science and Technology

JAIST Repository

https://dspace.jaist.ac.jp/

Title

語の組織化

Author(s)

Citation

Issue Date

Type

Text version

URL

Rights

Description

by Multi-Agents

TetsuoOno

Scho ol ofInformation Science,

Japan Advanced Instituteof Scienceand Technology

May9, 1997

Abstract

In this thesis, I prop ose multi-agent models, where agents come to organize a common lan-

guage in the community through a number of exchanges of sentences between them. I study

mechanisms of languagefusion and languagebifurcationbythe mo dels incomputers. Further-

more,I consider acomputationalmodelof languageacquisition.

First,Iproposea childand adult agentmo del. Inthis model, Ipresupp osethatachild agent

who has a primitive grammar set is thrown intoa communityconsisting of adult agents. The

childagentmodieshis/hergrammarandlearnsarenedone,imitatingadult'ssentences,while

theadultagentsalsolo osentheirgrammarinordertoacceptwhatthechildagentsaystolerantly.

I report the result of experimentationwith this mo del, together with grammar representation

and learning methods. I show that this mo del can realize adaptability, which is one of the

importantfeatures of naturallanguage.

Second,I propose an inferable agentmodel. In myarticial agent community,I hypothesize

thateachagentownsdierentgrammarrulesandhasabilityofabductiveandinductiveinference.

In the early stage of the life-span of agents, they try to generate grammar rules ab ductively,

to parse other's sentences, however in the later stage, they try to nd rules inductively from

a numb er of sentence examples they acquired so far. The communicative ability of agents is

measured by energy. The energy score of each agent becomes high if he/she could recognize

others'sentences,orhis/herutterancescouldb erecognizedbyothers. Accordingtothisenergy

score, each agent changes his/her behavior; when the score is high, he/she can increase the

chancesofutterance,andcanhavemoreinuenceuponthegrammarofthewholecommunity. As

eachagentmo dieshis/hergrammarbytheinferences,thecommongrammarinthecommunity

keeps changing dynamically. In my computer simulation, I show that: (1) I can realize the

adaptability and the robustness of the common grammar if I combine ab ductive/ inductive

inferences,(2)therobustnessisincreasedifIadopttheasynchronouscommunitywherethelife-

spans of agents overlap each other, and (3) the grammar fusion and the grammarbifurcation

emergebythelarger-scale, longer-termexperiment.

Finally,I proposea word acquisition modelusing asimilarity-based method. Inthis method,

theimp ortantnotiontomeasuresimilarityamong wordsand sentencesisamutualdependency

of words and sentences. Agents calculate the similarity between words by the metho d. The

agentsareregardedasacquiringawordwhenthey dierentiatethewordintoclustersusing the

similarity. Iexplain a pro cess ofchangeof useof wordsina multi-agentmo del.

Key Words: fusion and bifurcation of language(s), GPSG, abduction, induc-