Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news3.near.net!noc.near.net!pad-thai.cam.ov.com!news.ov.com!news.cerf.net!pagesat.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: <D3yM46.IHn@spss.com>
Sender: news@spss.com
Organization: SPSS Inc
References: <3gtu3i$rf3@mp.cs.niu.edu> <jqbD3r3Cp.CrC@netcom.com> <D3rGKG.16u@spss.com> <jqbD3vq6K.HBx@netcom.com>
Date: Mon, 13 Feb 1995 22:21:41 GMT
Lines: 327

In article <jqbD3vq6K.HBx@netcom.com>, Jim Balter <jqb@netcom.com> wrote:
>In article <D3rGKG.16u@spss.com>, Mark Rosenfelder <markrose@spss.com> wrote:
>>In article <jqbD3r3Cp.CrC@netcom.com>, Jim Balter <jqb@netcom.com> wrote:
>>[re "I saw John and Mary" --> ??"Who did you see John and?"]
>>>"John and Mary" is commutable; both subjects are given equal, or at least
>>>similar, worth.  "John and ... ?" is inherently not commutable; John and Mary
>>>(if that is whom he was with) play radically different *semantic* roles in 
>>>the question, and therefore it is hardly surprising that we do not use a 
>>>syntactic form that gives them equal roles.  
>>
>>John has the same syntactic role in
>>   John flogged Mary.
>>   John was flogged by Mary.
>>(subject position; topic; case), but has a rather different semantic role,
>>as I'm sure you'd agree if you were John.
>>
>>John has the same semantic role in 
>>   John flogged Mary.
>>   Mary was flogged by John.
>>   John's flogging of Mary (was abominable). 
>>yet has very different syntactic characters: subject vs. direct object
>>vs. determiner; topic vs. comment; nominative vs. objective vs. possessive.
>> 
>>So on what grounds do you assume that syntax pays any attention to "equal
>>semantic roles"?
>
>All of this appears to me to be an extravagant quibble.  It is actually the
>commutative *semantics* of "and" that I was commenting upon.  

But you did so by making the claim that "John and Mary [...] play radically 
different *semantic* roles in the question, and therefore it is hardly 
surprising that we do not use a syntactic form that gives them equal roles."
This depends on there being some parallelism between semantic and syntactic
roles: "equal semantic roles" should somehow be reflected in "equal syntactic
roles".  But you have given no grounds for such an assertion, and my examples
cast doubt on there being such a relation.  The same syntactic position can
be filled by very different semantic roles; the same semantic role can be
expressed in very different syntactic forms.

>How does any of this support Pinker's claims?

Who says I'm trying to support Pinker's claim?  I'm discussing your
argument, not Pinker's.

>>For that matter, although we can't say "Who did you see and John?", we
>>*can* say "You saw John and *who*?"  Yet the semantics have not changed;
>>Mary and John are as commutable as they ever were; John and "who" are
>>as semantically mismatched as ever-- why then is one construction 
>>terrible, and the other one just fine?
>
>From "You saw John and Mary" we might derive
>"*Who* saw John and Mary"?
>"You saw *who* and Mary?"
>"You saw John and *who*?"
>"You saw *who* and *who*?"
>"*Who* saw *who and *who*?"
>"You *what* John and Mary?"
>"*Who* *what* *who* and *who*?"
>
>Certainly the last couple of sentences are not normally thought of as
>grammatical, but with the proper stress, they make sense, just like "You
>*what'ed* the *what* *when*?" makes sense, and I'm sure I've used such
>constructions, especially over noisy phone lines.  That replacing various
>nouns and verbs with words that mean <unknown noun> or <unknown verb>
>is "just fine" barely requires explanation.  As for "*Who* did you see and
>John?", if it were just fine it would have to be for a different reason,
>wouldn't it?  

If it *were* just fine, it would (like "You saw John and *who*?") barely
attract comment.  As it is, you can normally form a question by replacing
the questioned component with "who", which is then moved to the front of
the sentence:

  You saw X. -->  Who did you see?
  X saw you.  -->  Who saw you?
  You think X is bald.  -->  Who do you think is bald?

(This can be stated in a way that doesn't imply any mental "transformation",
but it's harder to say that way... if this causes confusion we can discuss
it later.)

If we could likewise go from 

  You saw John and X.  -->  Who did you see John and?

this would be entirely unexceptionable.  Instead, we find that such 
sentences are strange (some people here have claimed, uninterpretable). 

Once linguists noticed this, they also found other oddities with "and"
(and other conjunctions).  For instance, from "It's easy to please John"
you can get "John is easy to please", but from "It's easy to please John 
or Mary" you can't get "John is easy to please or Mary".  Similarly,
"Biff hit John and kissed Mary" can't be transformed into "John was hit
by Biff and kissed Mary" (the latter is OK as a sentence, but means
something else).

This was widely summarized as the "coordinate structure constraint": 
roughly, you can't do something to one conjunct you don't do to the other.

There are, by the way, other places where X -> who won't work:

  You thanked the man who rescued the car X drove into the ditch. -->
  Who did you thank the man who rescued the car drove into the ditch? 

Too complicated, you may say.  However, languages differ on how deep X can
be nested before it can't be extracted by such a transformation.  There
are extractions that are possible in some languages but not others;
a simple answer in terms of buffering won't suffice to explain why.

>Now, we could replace "and" with, say, "with", or "kissing".
>All three are different grammatically, but two are "acceptable",
>namely the two that state a relationship between John and another.
>If this were a programming language, we might talk about binding strength
>or precedence; "with" and "kissing" bind tightly, whereas "and" binds loosely.
>In the C expression, "a + b . c", `.' binds more tightly than `+'.  This isn't
>just a convention, it is due to the differing semantics of the two operators.
>I would think that a set of "Universal Semantics" would sooner explain things
>than a "Universal Grammar".

Remember, all this applies to "or" and "but" as well as "and".  Once you've
generalized your statement to all conjunctions, how different is your idea
from the linguists' "coordinate structure constraint"?

(By the way, Chomsky *does* seem to believe in a universal, innate
semantics.  The only arguments for this that I've seen are however
pitifully unconvincing.)

>>>In syntax driven systems, we might expect forms like "negation I went
>>>to the store" 
>>
>>-- which we do find in many languages, such as Polish-- 
>
>I'll take your word for it.  Now, why isn't that mechanism universal?
>Why is UG so ad hoc?

Is this really a puzzle?  Why is anything evolution comes up with so ad hoc?

>>>or "store the to went I", but in semantic systems the action,
>>>not the statement, is negated, and the negation signal is given first to 
>>>avoid conceptual backtracking; 
>>
>>Oh?  As in "I didn't go to the store?"  The negative comes third, by my count.
>
>Mark, are you being serious?  By first I meant "before the verb (the action)".
>
>>In sentences like "I think not" or "Fear not" (archaic syntax today, but
>>perfectly colloquial in earlier forms of English), the negation comes last.
>
>"I think not" means something like "I think that isn't the case", not
>"I don't think".  As for whether "Fear not" was colloquial, I don't think we
>can judge from its presence in Shakespearean drama and Masterpiece Theatre.
>
>>This semantic constraint of yours begins to look like a mere ad hoc
>>generalization from one dialect of English.
>
>Ah shucks, you mean it's not Universal (tm)?  But perhaps UG can explain
>this drift in English.  Do you think perhaps our brains have been evolving?

Jim, you're missing the point.  You made a claim: X happens in "semantic 
systems" (of which natural languages are presumably examples).  So far as 
I can see, X doesn't happen at all, and so your claim is falsified.
I expect you either to retract it or modify it; instead, you dash off some
non sequiturs.

To put it another way, what exactly are you claiming?  That something 
about semantics *requires* negation to precede the "action"?  But that's
simply false; other languages (colloquial French, for instance), or earlier 
dialects of English, place the negative *after* the main verb.  

The larger point is: where do these semantic constraints of yours come from?
The only backing you give for this one, so far as I could see, was "to avoid
conceptual backtracking".  Again, the French do just fine with a postverbal
negative, so I don't see that "conceptual backtracking" is a problem.

>>>Aside from this, there are ridiculously obvious
>>>syntactic reasons why such forms don't exist: they require buffering (for 
>>>word inversion), or introduction of verbal parentheses in complex forms.  
>>
>>This is a valid objection to word inversion; but not to other conceivable 
>>syntactic operations that don't seem to be exploited by natural languages,
>>such as "Insert 'foo' every other word", or "switch the first two words
>>of the sentence".
>
>"Insert 'foo' every other word" is rather expensive; why should we expect
>it when there are more economical means?  

Chomsky is fond of this argument too; I find it exasperating.  How do you
know what the brain finds expensive or economical?  

>And inverting the first two words
>still requires a one-word buffer, and would produce unmanageble ambiguities
>(When Robin Hood renamed "John Little" to "Little John", this operation
>would have transported him into a radically different possible universe,
>I suppose).  

Come on, you're telling me that a one-word buffer is a problem?  Why can
we handle rules that involve inverting entire *phrases*, then?

And why should the ambiguities be unmanageable?  Did you actually try this?
I did; I took eighty or so sentences and inverted the first two words.
This certainly reveals that word-inversion is incompatible with phrase-
inversion; you couldn't represent negatives this way and continue to form
questions by phrase-inversion.  But other than that, ambiguities are few,
and confined to a small class of pragmatic adverbs.  This is certainly not
unmanageable.  Language users are very good at managing ambiguity.

>Why does the absence of these operation support UG?  Why does the
>absence of *any* particular operation support UG?  Explaining such absences
>is a ridiculous burden to impose.

Such arguments are intended to show that the rules of human grammars are 
not uniformly distributed over the space of possible lexical/syntactic
transformations.  If this is so, it isn't at all ridiculous to expect a
theory of language or cognition to explain it.  (However, that doesn't mean
that I think UG is the only possible explanation of the tendency.)

>>>How would word inversion work for "I ate but didn't drink" 
>>
>>"I ate but drink didn't".
>
>Has anyone claimed that no such forms occur in natural languages?  I thought
>we were talking about sentences being negated by inverting their words?  From
>the previous examples, I would have thought that we were talking about a
>language just like English but in which "I didn't drink" is expressed as
>"drank I" (and "Fear not" is expressed as "Afraid be"?).  But the sentence I
>gave is not the negation of some other, at least not some other that doesn't
>contain a negative (well, perhaps we can express it in
>negation-by-inversion-English as "drink to failed and ate I").  Perhaps
>individual phrases can be negated by inverting them, but that's not the rule
>that was given.  My whole point is that "negate by inverting the words of the
>sentence" is just an intuition pump, it isn't a real, workable, grammatical
>rule.

I think you've failed to shown that-- but I commend you for trying.  

I agree that Chomskyans don't really take the negate-by-inversion rule
very seriously; they don't develop it enough to show that it's a viable
grammatical alternative.  I rather like the attempt to see what such a rule
would really entail.  However, I think you've stopped short of showing that
such a rule would really be unworkable.  

A good (but not decisive) argument against this rule was provided by Scott
Horne on sci.lang: it would make it impossible to negate one-word (or 
word-palindromic) sentences.

>>>All this talk of sentence inversion is a typical misleading intuition
>>>pump.  Give an actual grammar for complex utterances using it, and we can 
>>>talk more about what is wrong with it.  In particular, for strictly semantic
>>>reasons, when talking about what I did or didn't do, the first word out of
>>>my mouth is "I".  Often the sentence begins with "I, um, ...".  
>>
>>What are the "strictly semantic reasons" that, in the same circumstances,
>>force a Spanish or Welsh speaker to utter a verbal morpheme first, or a 
>>Hixkaryana speaker to utter the object first?
>
>C'mon, Mark, cut out the linguistic snobbery.  If you want to make points
>about languages other than the one we are communicating in, you will have to
>be more concrete.  

In Spanish personal pronouns are normally omitted, so if you're talking about 
yourself you begin with a verb: e.g. _Evadi a los cachacos_, "I escaped from
the cops".  The final -i on _evadi_ implies 'I', but the verb root _evad-_
comes first.  Welsh is a VSO language, meaning that the verb always comes
first.  Hixkaryana (an Amerindian language) is one of the relatively few
languages which normally places the subject after the object.

>Even if the syntax of Spanish or Welsh require some other
>phoneme to precede the first person pronoun, why is this relevant, since it
>isn't syntactically universal?  I never said that it would be for strictly
>semantic reasons.  If I say that, when the first word out of my mouth is "I",
>it is for strictly semantic reasons, it doesn't follow that I mean that, when
>the first word out of Ronald Reagan's mouth is "Well, ...", it is for strictly
>semantic reasons, and it is silly for you to imput such a meaning; it
>certainly doesn't further the discussion.  I might well have said "for
>strictly semantic reasons ... the first word out of my mouth, following of
>course any sort of noise words or obligatory syntactic stuff that must precede
>it, is `I'".  The point is that the first word *with specific meaning relevant
>to my intent in making an utterance* that comes out of my mouth is often "I",
>and sentence inversion make that impossible in negated sentences.

Even with this clarification, your statement isn't true of these other
languages.  Since in other languages people get by without starting sentences
with 'I', I don't see that there's any "strictly semantic reasons" for
this behavior in English.  

>BTW, how does one say "I'm walking" in Hixkaryana (whatever that is)?

I don't know, since I've forgotten whether the subject precedes or follows
the verbs (and even if I knew that, pronouns often work differently than
nouns).  

>>I don't mean to be rude,
>
>Not even just a little? :-)
>
>>but what we have here is a set of vague intuitions,
>>not thoroughly checked out against a wide range of situations and languages,
>>against a theory elaborated over the course of fifty years by professional
>>linguists.
>
>Mark, this is a bizarre approach to take in this forum.  Most of the postings
>here are based upon partial knowledge.  I don't think mine are particularly
>more vague or uninformed than others.  I make my arguments with the
>expectation that there are variously cogent counter-arguments that others will
>present.  I welcome you to make them, and to show me my errors.  Many of your
>postings, on this topic and others, have been quite enlightening and have
>helped shaped my views.  But don't expect me to respond to arguments from
>authority.  

It's not exactly an argument from authority when I went on to describe
Chomsky as often "unconvincing or just plain wrong".  The operative 
phrase was "not thoroughly checked out".  It seems to me that many of
the supposed semantic constraints that have been bruited about here
disappear when subjected to closer analysis.  

>I would be very interested in seeing whether Chomsky has responded to the
>sorts of methodological criticisms offered by Minsky in his recent post,
>and if so how.  Do you have any pointers?

No, I'm not at all up to date on my Chomsky reading, and even less on 
others of his school.

>(Note, I still am agnostic regarding the validity of UG.)

So am I.  I think it's an interesting approach and worth continuing, though
it hardly compels belief at this point.  The same can be said of many of the
alternatives to it.
