Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!bloom-beacon!RESEARCH.ATT.COM!dlm
From: dlm@RESEARCH.ATT.COM
Newsgroups: comp.ai.digest
Subject: Computer Modelling of Child Language Learning
Message-ID: <19880718042701.8.NICK@HOWARD-JOHNSONS.LCS.MIT.EDU>
Date: 18 Jul 88 04:27:00 GMT
Sender: daemon@bloom-beacon.MIT.EDU
Organization: The Internet
Lines: 50
Approved: ailist@ai.ai.mit.edu

From: dlm@research.att.com
Date: Sat, 9 Jul 88 10:54 EDT
>From: allegra!dlm (D.L.McGuinness)
To: arpa!mc.lcs.mit.edu!AIList
Subject: Computer Modelling of Child Language Learning



               How Do Children Learn to Judge Grammaticallity?
                                     or
      Research Issues for Computer Modelling of Child Language Learning

                       Thursday, July 14, 1988, 10:30 am
		  AT&T Bell Laboratories - Murray Hill 3D-436

                              Mallory Selfridge
                        The University of Connecticut

    Development of a successful computer model of child language learning
    would  have  important  implications  for  the development of natural
    language interfaces to computers.  However, no such fully  successful
    model  has yet been developed, and ongoing research is taking several
    different approaches.  The purpose of this talk is  to  identify  the
    most  promising  approach  and  the most important research issues it
    suggests.  This talk first discusses the problem of developing a com-
    puter  model  of  child language learning and argues that the primary
    questions are those of accounting  for  empirical  data  rather  than
    abstract  questions from theoretical linguistics.  It then identifies
    a set of several linguistically-motivated  questions,  including  the
    question of how children learn to judge grammaticallity, and suggests
    that they should be answered as side-effects of computational mechan-
    isms  required  to account for empirical data.  The "grammar acquisi-
    tion" approach to child language learning is then  reviewed,  and  is
    judged to be undesirably abstract and of uncertain promise.  Then, an
    example of a "semantic" approach  to  child  language  learning,  the
    CHILD  program,  is considered, and its performance in accounting for
    empirical data is described.  Further, CHILD's ability  to  learn  to
    judge   grammaticallity   is   described,   and  answers  to  set  of
    linguistically-motivated questions are proposed  as  side-effects  of
    CHILD's mechanisms.  This talk concludes that the "semantic" approach
    to computer models of child language learning is the most  promising,
    and  identifies  as important research issues a) the investigation of
    the relationship  between  language  and  memory  processes;  b)  the
    development of non-linguistic representations of syntactic knowledge;
    c) the investigation of the process  whereby  the  child  infers  the
    meaning of an incompletely understood utterance; and d) the identifi-
    cation and  investigation  of  additional  empirical  data  on  child
    language learning.

    SPONSOR:  Bruce Ballard -  allegra!bwb