Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!satisfied.elf.com!news.mathworks.com!solaris.cc.vt.edu!insosf1.infonet.net!internet.spss.com!markrose
From: markrose@spss.com (Mark Rosenfelder)
Subject: Re: What's innate? (Was Re: Artificial Neural Networks and Cognition
Message-ID: <D4DDnI.FA5@spss.com>
Sender: news@spss.com
Organization: SPSS Inc
References: <3gtu3i$rf3@mp.cs.niu.edu> <3hr5ap$b6f@agate.berkeley.edu> <D40p5v.6v3@spss.com> <3hs2ls$rj3@agate.berkeley.edu>
Date: Tue, 21 Feb 1995 21:42:53 GMT
Lines: 153

In article <3hs2ls$rj3@agate.berkeley.edu>,
 <jerrybro@uclink2.berkeley.edu> wrote:
>markrose@spss.com (Mark Rosenfelder) wrote:
>> In the '60s, conjunction was widely taken to be pretty much as you've said;
>> "I saw John and Mary" was analyzed as having a deep structure rather like
>> that of "I saw John and I saw Mary".  
>
>Deep structure, eh?  It seems to me likely that we're wired up
>so that, to a limited extent, we can "re-use" input, and so in
>the case of "I saw John and Mary yesterday", in which some words
>are spoken once, we can in effect understand two statements at the
>same time.  Whereas in the case of "I saw John yesterday and I saw
>Mary yesterday," probably we've understood the first part before we
>begin to understand the second part.  If that's so, it's unlikely
>that the common "deep structure" corresponds to some common thing
>happening in the brain.  But isn't that what "deep structure" was
>supposed to be?

Technically, no; deep structure was a theoretical construct, with no claim
about anything that existed in the brain.  Disclaimers to this effect were
generally inserted near the beginning of a book on syntax... how serious
they were is another question.  (UG obviously *does* make claims
about what's in the brain.  But it's a late development of '60s TG,
and it doesn't even have "deep structures" any more.)

I agree that "I saw John and Mary yesterday" and "I saw John yesterday and
I saw Mary yesterday" must be handled differently by the listener.
What about the speaker, however?  Does the speaker have to generate the
long form, then abbreviate it to get the shorter version?  I'd say no;
since conjoined arguments are necessary for some sentences anyway
(e.g. "John and Mary got married"), they're an option at any time.

>Also, I notice that "I saw John and Mary yesterday" seems to
>suggest that the two events don't require separate explanation.
>However, the exact inference drawn depends on the context.  In
>certain contexts, we should decide that John was with Mary, but not
>in others.

Right.  At least, this is true for the listener.  The speaker presumably
knows which interpretation he intends.

>> One complication, seemingly missed in the initial rush of enthusiasm,
>> is that the analysis doesn't work for some sentences; e.g. "John and Mary
>> got married" does not mean "John got married and Mary got married"
>> (or to be precise, it implies that, but it really describes an activity
>> which inherently has a plural subject).
>
>Actually, it *might* be that John and Mary got married to (say) Lisa
>and Bill, respectively.  In a sufficiently clear context (in
>which it was known that John and Mary were brother and sister),
>that would probably be the interpretation.  Still, sure, that they
>got married to each other is the accepted default interpretation.
>Similar cases:
>
>"William and Henry James were brothers (of each other)."
>
>Or, "John and Mary exchanged glances (with each other)."

Good, you'll be a syntactician yet.  :)

>It seems to me that this is likely an outgrowth of a more primary,
>simpler use of "and".  In these cases, a certain part of the
>statement is dropped because it is accepted as a default likely
>meaning in the absence of evidence to the contrary.  This
>little complication would seem to be in response to a very
>practical desire (to save breath).

I'd still hold out for the possibility of conjoined arguments.  Why 
should the subject or object of an action always be one thing?  

>Well, "Who did you see John and..." could be continued as:
>
>"Who did you see John and Mary with?"
>
>The listener might hear it as demanding just such a continuation.  But
>probably he doesn't, and since he doesn't, he has no choice but to
>try to understand "Who did you see John", and he can't.

I don't think that's quite the problem.  Just to emphasize that the Coordinate
Structure Constraint affects other transformations besides "who" questions,
let's consider passivization.  We want to passivize the objects in

   You visited John and Mary.

"John and Mary were visited by you" - no problem.

"John was visited by you and Mary" - an acceptable sentence, but one 
that doesn't mean what we intended: "and Mary" glues itself to "you",
not to "John".  

"Mary was visited by you John and" - word salad.  If we imagine the listener
bopping along word by word, he had stopped, happy as a clam, at "you", since
"Mary was visited by you" is a perfectly good proposition.  "John", however,
throws him for a loop; there's nothing to do with John.  He shrugs and
continues.  "and" doesn't help much; it says that something is being linked
to something, but it's in the wrong order for "you and John".  Perhaps the
sentence is going to be "... by you, John, and Bill".  We keep going--
there's nothing left.  We have some dangling, unmet expectations (Bill never
showed up), plus some constituents (John, and) that never fit in anywhere.
Hooking up "John and" with "Mary" is a bit of a stretch, and doesn't even
seem to occur to people, perhaps because once we've found a nice object for
"visit", namely Mary, we have no expectation that that assignment needs any
modification.

Or take 

   It is easy to please John and Mary.

"John and Mary are easy to please" - no problem.

"John is easy to please and Mary" - similar problems to the word salad
above.  "and Mary" would be likely to be taken as starting a new sentence
entirely, an expectation that leaves us in limbo when the sentence ends.

"Mary is easy to please John and" - "to please John" is a false match,
likely to be signalled as an error because after "Mary is easy to please"
we've already found a plausible object for "please".  

  You saw X and Mary.

"Who did you see and Mary?" - no problems up to the point of processing "Who
did you see"; "and" tells us to link up with something, but what?  The
default expectation would be, with the previous word; we should expect
something like "Who did you see and talk to?"  But the next word is a noun,
confounding this expectation.  Who should we attach Mary to?  The closest
noun phrase is "you", but then it should have been "Who did you and Mary
see?"  And again, "who" is already set as the object of "see", and we
weren't expecting to have to modify that assignment.

  You saw John and X. 

"Who did you see John and?" - Similar mismatches, ambiguities, and unmet
expectations.  After "and" we might expect we're dealing with something
like "Who did you see John and Bill teasing?"  With no continuation, we
might try to attach John with "you" (Who did you and John see?), but again
the syntax is wrong.  

  You visited John and Mary. -->
  It was John and Mary you visited.  (OK)
  It was John you visited and Mary.  (Here we go again)

It's also worth looking at what we *can* do with conjoined phrases.
We can emphasize one of them: "You saw John and *who*?"  We can topicalize 
one of them, leaving a placeholder: "John, you visited him and Mary."
Such sentences cause no problems, since they don't remove one of the
conjoints, and thus leave the intended structure clear.

All this suggests (very informally) that the Coordinate Structure constraint
serves to avoid sentences that are extremely ambiguous or difficult to
process.  



