From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!usc!sdd.hp.com!uakari.primate.wisc.edu!zaphod.mps.ohio-state.edu!cis.ohio-state.edu!cannelloni.cis.ohio-state.edu!chandra Mon Mar  9 18:32:51 EST 1992
Article 4042 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!usc!sdd.hp.com!uakari.primate.wisc.edu!zaphod.mps.ohio-state.edu!cis.ohio-state.edu!cannelloni.cis.ohio-state.edu!chandra
>From: chandra@cannelloni.cis.ohio-state.edu (B Chandrasekaran)
Newsgroups: comp.ai.philosophy
Subject: Re: Definition of understanding
Message-ID: <CHANDRA.92Feb26131732@cannelloni.cis.ohio-state.edu>
Date: 26 Feb 92 18:17:32 GMT
References: <1992Feb25.011840.24663@beaver.cs.washington.edu>
	<1992Feb25.184610.5199@psych.toronto.edu> <43956@dime.cs.umass.edu>
	<1992Feb26.170232.8676@psych.toronto.edu>
Sender: news@cis.ohio-state.edu (NETnews        )
Organization: Ohio State Computer Science
Lines: 69
In-Reply-To: christo@psych.toronto.edu's message of Wed, 26 Feb 1992 17: 02:32 GMT
Originator: chandra@cannelloni.cis.ohio-state.edu

In article <1992Feb26.170232.8676@psych.toronto.edu> christo@psych.toronto.edu (Christopher Green) writes:

   But we can never understand languages that we don't think we understand.

This goes to the heart of the disagreement that everyone keeps going
round and round on.  Can an agent be mistaken about whether it
understands a language?  Specifically, can an agent actually
understand a language, but think it doesn't and be surprised by its
own behavior when sentences in that language are given?  One group
thinks that our everyday intuition about the word "understand" is 
not very helpful when we are confronted with unusual situations.
Another group, which Jeff Dalton and Christopher Green (and of course
Searle) exemplify, thinks the word has an a priori and clear meaning
and that an agent cannot be mistaken whether it in fact understands
a language.

Just for the record, take the word "mother."  Even though we know what
it means in everyday life (at least used to), consider the following
situations:

i.  A woman (A) who did not give birth to a child, but adopted the child from
the moment of birth, and raised the child.

ii.  The  woman (B) who was a recipient of some other woman's fertilized egg
and gave birth to the child.  

iii. The woman (C) whose egg was extracted, fertilized and implanted in 
the womb of B.

We can't settle which one of the above is the "real" mother: it is no
longer a matter of fact, but of the task at hand.  For genetic analysis,
C would count as the mother, for fetal injury from alcohol ingestion,
B would count as the mother, and for the bonding and the feelings that
arise from it, A would count as the mother.  That is the new situation
breaks the comfortable cocoon in which all the meanings of the word 
nestled comfortably until technology came along to break it.

Question: Is "understanding" such a term?  Do the specifics of the
thought experiment "break" the word and its meanings in a similar
way, so that arguing about whether the person + memorized rules
"really" understand Chinese is similar to arguing about who is the
"real" mother in the above example?

In order to answer it I suggest that we expand on the behavioral 
analysis of the phrase "understand a language."

When I think I understand a language L (say for the sake of the argument,
in addition to a mother tongue, M), the following complex of behavior
(some internal and some external) are seen to happen:

i.  Given sentences in L, I can produce sentences in M, and vice versa,
which preserve meaning. The CR experiment satisfies this sense of
"understand a language."  This can be externalized by producing
typed sentences in M.

ii. Given sentences in L, I can have in my awareness, if I choose,
additional sentences *in L*, which are in some sense consequences of
understanding the given sentences in L.  That is, I can think in L,
similar to my ability to think in M.  The CR experiment in my view
*does not satisfy* this sense of understanding a language.  The man in
the room only "thinks in English". He can't think in L in the sense of
going from a sentence in L to another sentence in L, unmediated by
sentences in M.  

Thus even those who are only interested in "behavioral" definition of
understanding should be able to conclude that some of the
*behavioral* features of "understanding a language" are not present,
while others are.  The question then is: What is the need to insist on
the sense ii above?  


