Grammar as a perceptual system

[bnhpct]

From Bruce Nevin (970927.1630)

Avery Andrews 970919.1258 --

So here's my story about `grammar as a perceptual system'. [...]

Avery Andrew 970916 --

I'm wondering if anyone had any reaction to my posting of this title [...]

Sorry, I'm going to be very slow responding.

I agree that a clear account of "the Describes relation" is central to a PCT
account of language and a prerequisite for answering Bill's question "what
is a rule (really)?" However, I believe we can't take it to be as simple as
both you and Bill are assuming it to be. Following your account:

the Describes relation [works through] a scheme whereby the utterance
[...] is perceived as a complex of
categories and relationships [...] which corresond, in a regular way,
to categories and relationships
in the world. So an utterance will usually in a sense contain a model
or depiction of part of the world (this in certain respects like
Wittgenstein in the Tractatus). Consider for example:

the butter is in the refrigerator

Grammatically this can be resolved into two `noun phrases' (NPs, a
grammatical category), `the butter' and `the refrigerator', connected
by a relationship `is_in', which one might define informally as:

X is_in Y in utterance U iff U is of the form: `X is in Y'

One could do better with a real theory of grammar worked out, for
example I could hack my parser so that if you typed in a sentence
it would tell you which pairs of NPs were in the `is_in' relationship [...]

In fact, the word "in" does not always neatly correlate with a perception
that one could describe as the "is_in" relationship. In other words, "in"
occurs in context of words where any connection to the spatial relationship
called "is_in" is remote or utterly vanished in the etymological mist.
Consider the examples in the preceding sentences. I can look for a 1970s
paper that lays these problems out in some detail, if you like. These are
for the most part problems due to the arbitrariness of language--divergence
and specialization of meanings over time, preservation of frozen expressions
and other archaisms, reinterpretation in ignorance of original meanings, and
so on.

There is another kind of problem. Our main evidence for category perception
is words, and our evidence for perceptions based upon or "using" category
perceptions is for the most part syntactic constructions in language or in
language-dependant systems like mathematics, symbolic logic, and programming
"languages". Maybe it's "level-blindness aphasia" as Bill says; who can
tell, since "I gotta use words when I talk to you" as T.S. Elliot says.
Here's this tool that we use to represent to ourselves what we know (to
remind ourselves later of what we used to know, or because we want someone
else to know, etc.). To indicate what it is that we know and step back and
look at it, we have no other means than language.

"More `this is like that'" complains Bill of my attempt to say what a rule
is. And with justice. That is how language works, by making analogies --
this is like that. What is needed is to go to a level beyond that at which
language is used. What is that? Who can say? Our means of saying is language.

I think no one seriously claims that language is *used* at the postulated
category level or program level or any other level of perception. Each of us
has a self-image, for example, and we control many lower-level perceptions
as means of sustaining it, but putting it into words is not easy and so far
as I know never accurate or complete. Another example, is a tomato a
vegetable or a fruit? The sensible answer is that it does not matter, it's a
food that you put in salads and soups, or perhaps something you avoid
because of allergy--categories for which there is no simple vocabulary item
in English.

What I have been trying to tell you and demonstrate to you is that language
is not in and of our perceptual hierarchy. It is a social tool. As with any
tool, to use it we control various perceptions, including especially the
result of its use, but it is not itself in and of our perceptual hierarchy,
any more than the heft of a hammer and the rhythm of its swing, or the
arrangement of letters on a keyboard, or the strictures of a for loop in a
programming language.

So what is the "the Describes relation"? Actually, I agree with Avery's
account, but not quite as far as seems to go. The correspondence is not
between words and category perceptions, or between syntactic relations of
words and some sort of relations between category perceptions. The
correspondence is between co-occurrance sets of words, on the one hand, and
ranges of co-occurring perceptions on the other. For each word there is
associated in memory a set or sets of familiar, or often heard, or
expectable word dependencies. Spread butter, butter on bread, melt butter,
fresh butter, rancid butter, butter in the refrigerator, and so on. Food in
refrigerator, food spoil, etc. The rather inchoate perceptions associated
with the sets of of word associations -- melt butter, melt ice, melt ice
cream -- are brought into focus, crystallized as it were into more
particular expectations in the intersections of the various association sets
represented in a particular utterance. (Some years ago for my master's
thesis I made some attempt to work out an algorithm for this. One day
perhaps when I've finished this silly membership ritual of a dissertation
I'll get back to it, with an understanding of PCT behind it.)

This is why language works best for us when we both know already know a
great deal about what we are talking about, in the sense of anchoring our
words to our perceptions of particular CEVs to which we each know the other
is referring. Think of Avery and his wife in their kitchen, in which there
is one refrigerator, and one butter dish, with both of which they are both
very familiar. The words they use then seem precise in their reference
because there are no or few alternative candidates.

Indeed, under those circumstances we may omit words entirely. Avery, risen
from his slice of toast on the kitchen table, casting about with a butter
knife in his hand, his wife mutely pointing to the refrigerator with her
left hand as she continues to pour coffee with her right. "Ah," he says, and
opens the refrigerator door.

And it is why language works less well for us the less we can take for
granted -- that is as our confidence that the other controls the same CEVs
as we becomes less warranted.

The sentence "the scientists dug at the base of the tower" evoked for Bill
stick figures which he took to be category perceptions; for me it evoked a
stereotype of men in white lab coats. Both of course are inappropriate, in
that they are not what one present at the scene would have seen; and neither
Bill's perceptions nor mine are likely to be the same as the perceptions
imagined by the author of that piece about an archaeological dig in Rhode
Island--imagined, since he was not present either. In short, there is no
clear evidence of a universal category-perception "mentalese" of which
language is a translation.

Language provides rather rigid and limiting structures onto which we project
our much less structured perceptions. We can use it to remind ourselves, to
re-evoke perceptions from memory, rather in the way that a few jotted key
phrases can bring back something of last night's dream. When we use it to
evoke perceptions from the memories (and imaginations) of others, language
works best when it provides guideposts by which they can experience
something directly for themselves. The communication of expertise.

        Some think an expert is a man who knows everything there is to know
        about his field. I think an expert is a man who knows all the mistakes
        that can be made in his field, and how to avoid them.
                                -- Werner Heisenberg

The failures of language we see in this forum every day. Probably here too.
I doubt that I have given adequate indications for you to experience what I
am talking about directly for yourself. Lacking that, what you make of these
words is somewhat up for grabs. I at least have no control over it. But I
would have that control, through my control of the language in this message,
if language did correlate closely with category and higher levels of
perception. So perhaps I have succeeded in demonstrating at least that. You
tell me.

        Bruce Nevin

[Bill_Powers1]

[From Bill Powers (970928.0557 MDT)]

Bruce Nevin (970927.1630)--

I agree that a clear account of "the Describes relation" is central to a PCT
account of language and a prerequisite for answering Bill's question "what
is a rule (really)?" However, I believe we can't take it to be as simple as
both you [Avery Andrews] and Bill are assuming it to be. Following your

account:

the Describes relation [works through] a scheme whereby the utterance
[...] is perceived as a complex of
categories and relationships [...] which corresond, in a regular way,
to categories and relationships
in the world. So an utterance will usually in a sense contain a model
or depiction of part of the world (this in certain respects like
Wittgenstein in the Tractatus). Consider for example:

the butter is in the refrigerator

Grammatically this can be resolved into two `noun phrases' (NPs, a
grammatical category), `the butter' and `the refrigerator', connected
by a relationship `is_in', which one might define informally as:

X is_in Y in utterance U iff U is of the form: `X is in Y'

In fact, the word "in" does not always neatly correlate with a perception
that one could describe as the "is_in" relationship. In other words, "in"
occurs in context of words where any connection to the spatial relationship
called "is_in" is remote or utterly vanished in the etymological mist.

This encourages me in my opinion that a study of language really can't be a
study of _just_ language. Somehow it's necessary to get outside the
framework of language and see how language relates to that which is not
language. I think you're trying to do that. I think that many of the
constraints on language arise from constraints on the perceived world in
general; once meanings have been assigned to parts of language, the way
those parts relate depends on how their counterparts in non-linguistic
perception prove to be related.

There is another kind of problem. Our main evidence for category perception
is words, and our evidence for perceptions based upon or "using" category
perceptions is for the most part syntactic constructions in language or in
language-dependant systems like mathematics, symbolic logic, and programming
"languages". Maybe it's "level-blindness aphasia" as Bill says; who can
tell, since "I gotta use words when I talk to you" as T.S. Elliot says.
Here's this tool that we use to represent to ourselves what we know (to
remind ourselves later of what we used to know, or because we want someone
else to know, etc.). To indicate what it is that we know and step back and
look at it, we have no other means than language.

It's hard to get across the idea of a non-linguistic category. As I see it,
words and word structures are _names of category perceptions_ -- they are
not themselves category perceptions. When you open a drawer and take out a
screwdriver, you are using a nonlinguistic category perception, the one we
refer to as "a screwdriver," meaning any individual item that looks enough
like the thing we think of as a screwdriver to be used in the same way.
What category perceptions do is make a whole set of lower-level perceptions
interchangeable in terms of their use in controlling still higher-order (or
at least "other") perceptions.

Of course different lower-level perceptions are not in fact
interchangeable. Each one requires specific control processes that adjust
to the individual characteristics of each situation. But when we think in
categories, we don't have to pay attention to the differences; the
lower-order systems handle the differences automatically. The handle
diameter and length of the screwdriver make no difference, because we can
grasp and direct it by suitable variations in our actions, so in some sense
all screwdrivers (with the same shape of tip) seem alike. There are limits,
of course; we can't tighten a screw holding a bookshelf together using a
jeweler's screwdriver. But within the category we name as "a screwdriver"
in a given situation, all the differences that actually exist are simply
ignored. And we _can_ ignore them because they're taken care of by lower
control systems, without thought.

I think no one seriously claims that language is *used* at the postulated
category level or program level or any other level of perception. Each of us
has a self-image, for example, and we control many lower-level perceptions
as means of sustaining it, but putting it into words is not easy and so far
as I know never accurate or complete. Another example, is a tomato a
vegetable or a fruit? The sensible answer is that it does not matter, it's a
food that you put in salads and soups, or perhaps something you avoid
because of allergy--categories for which there is no simple vocabulary item
in English.

This is what I'm getting at. We can deal with "a tomato" as a perceptual
category without giving it a name other than "a tomato." No two tomatoes
are identical, but because we can deal with their different sizes, shapes,
degrees of ripeness, and locations of stems, we can _act as if_ they are
the same at a certain level of perception.

Before we can attach the label "a tomato" to a perception, we must first
have a perception that is the same for any example of a tomato regardless
of the differences at the lower levels. This is the _perceptual_ category
level, where any configuration within a wide range, involving different
sensations and being quite distinguishable from each other at the
configuration level, will yield _the same_ perception -- in fact, the same
_perceptual signal_. The _words_ "a tomato" don't create this sameness;
they only name it. The article "a", or the use of the plural "tomatoes,"
signals that it is a category perception that is being named, rather than
an individual, ideosyncratic, example.

This is how I try to disentangle the word from the object, the map from the
territory.

You say

The sentence "the scientists dug at the base of the tower" evoked for Bill
stick figures which he took to be category perceptions; for me it evoked a
stereotype of men in white lab coats. Both of course are inappropriate, in
that they are not what one present at the scene would have seen; and neither
Bill's perceptions nor mine are likely to be the same as the perceptions
imagined by the author of that piece about an archaeological dig in Rhode
Island--imagined, since he was not present either. In short, there is no
clear evidence of a universal category-perception "mentalese" of which
language is a translation.

But this is exactly what I think of as the evidence. At the category level,
we perceive many different lower-level perceptions as being "the same
thing." When you say "the men", you're naming a category that can be
exemplified by _any_ set of more than one peceived man. When you name only
the category, without at the same time naming other categories that must
exist at the same time ("the black men", "the male scientists", "the men
from the 8087th Engineers"), you leave the listener free to imagine any
specific examples that would lead to the same category perception. When you
say "digging" you refer to a perceived category of processes that could
include the use of any tools from bare hands to dentist's picks to a
backhoe. When you say "at the base of the tower" you again have a choice of
specific situations that lead to the same relationship perception, or even
different ones -- you might imagine digging into the base of the tower
itself, or into the ground near it. And of course you have a wide range of
structures that would all be perceived as the same, and be named "a tower."

Language doesn't show the existence of category perceptions. What shows
their existence is the way we can act as if things that are perceivably
different (at one level) are the same (at another level).

Best,

Bill P.

[bnhpct]

From Bruce Nevin (970928.1304)

Bill Powers (970928.0557 MDT) --

This encourages me in my opinion that a study of language really can't be a
study of _just_ language. Somehow it's necessary to get outside the
framework of language and see how language relates to that which is not
language. I think you're trying to do that. I think that many of the
constraints on language arise from constraints on the perceived world in
general; once meanings have been assigned to parts of language, the way
those parts relate depends on how their counterparts in non-linguistic
perception prove to be related.

Agreed. An example: there are selection-sets of word co-occurrances that we
could call "punctual" vs. "durative". In some languages, the distinction is
formally marked, perhaps by affixes. In English, we see it indirectly in the
complementarity of the selection of some operators (sleep, grow, etc. with
throughout, lasted, etc., but not with was sudden, occurred at, etc.;
complementary with arrive, depart, etc. with sudden, occurred at, etc., but
not easily with throughout, lasted, etc.)

This seems to me to be a property of perceptions as reflected in language,
and not a property of grammar.

However, the expectability of such selections (that is, the high
expectability of some particular word pairing as a member of one of these
word co-occurrence sets) can be the occasion for reducing a word with low
information to an affix or other formal device in some languages, so that
expectations about the world of perceptions become institutionalized as part
of grammar--imperfective or durative verbs vs. perfective or momentaneous
verbs. Noting that other verbs (speak, think) are members of both
co-occurrance sets.

Usage reflects perceived properties of the world. Institutionalized usage,
resulting in grammatical options or even grammatical requirements, can seem
to restrict what is perceivable about the world, by directing attention to
those perceptions that are readily described and by limiting what can easily
be reported to others using language in normal ways. This is the basis of
the Whorf-Sapir hypothesis, or "linguistic relativity," invented by Benjamin
Lee Whorf to attract students to Sapir's linguistics program.
(Unfortunately, Sapir died soon after. Whorf was standing in for him during
his illness.)

It's hard to get across the idea of a non-linguistic category. As I see it,
words and word structures are _names of category perceptions_ -- they are
not themselves category perceptions. When you open a drawer and take out a
screwdriver, you are using a nonlinguistic category perception, the one we
refer to as "a screwdriver," meaning any individual item that looks enough
like the thing we think of as a screwdriver to be used in the same way.
What category perceptions do is make a whole set of lower-level perceptions
interchangeable in terms of their use in controlling still higher-order (or
at least "other") perceptions.

[...]

Language doesn't show the existence of category perceptions. What shows
their existence is the way we can act as if things that are perceivably
different (at one level) are the same (at another level).

I have not seen convincing evidence that there are category perceptions or
that there is a category level of perception. Everything that I have seen
depends upon language or upon symbol systems like mathematics or symbolic
logic which in turn depend upon language.

The claim that words and word structures correlate one for one with category
perceptions is equivalent to a claim that the category level uses language
or that language is in the category level. You believe that you are not
making the latter claim; by making the former, you are.

This claim that the stuff of language correlates with or names category
perceptions is false, or so I have tried to show. There is no such
correspondence. Rather, we project our perceptions onto the structures of
language. "Ah, this is an instance of what they talk about when they say X."
We project association-sets of perceptions onto co-occurrance sets of words,
and locate our immediate perceptions in those association-sets. When we
read, or listen, or converse without projecting controlled or controllable
perceptions onto the words and language structures that we are using, we are
merely playing with association-structures of language in memory, extending,
adjusting exploring their ramifications. This is why Shakespeare has Hamlet
enjoin the players to visualize vivid imagery associated with the words they
speak: because it is not ordinarily done by readers and reciters, because it
takes extra work and concentration to do it, in the actor's craft.

When I open a drawer and take out a screwdriver, I am taking something that
will be part of a perception of screwing in a screw or bolt, and lacking a
screwdriver I may take a knife instead. Or maybe I am taking something that
will be part of a perception of prying the lid out of a paint can, and I
can't find the thingie that the paint store provides that does a much better
job. There is *no* category perception. The categorization is a process by
which I select some particular thing, perhaps opportunistically, in the
course of controlling a perception that normally involves a screwdriver, or
which can involve a screwdriver at a pinch. Because we have a word onto
which we can project many particular instances, we believe there is a
corresponding category perception. It is an illusion.

There is no category perception "tomato" or "dog". There is the process of
naming instances. There are expectable co-occurrances of perceptions (named
bark, wag tail, etc.) and there are expectable word-co-occurrances. The word
co-occurrances include a category word "dog"; the perception co-occurrances
do not include a category perception, other than the words.

Before we can attach the label "a tomato" to a perception, we must first
have a perception that is the same for any example of a tomato regardless
of the differences at the lower levels. This is the _perceptual_ category
level, where any configuration within a wide range, involving different
sensations and being quite distinguishable from each other at the
configuration level, will yield _the same_ perception -- in fact, the same
_perceptual signal_. The _words_ "a tomato" don't create this sameness;
they only name it. The article "a", or the use of the plural "tomatoes,"
signals that it is a category perception that is being named, rather than
an individual, ideosyncratic, example.

This is how I try to disentangle the word from the object, the map from the
territory.

It appears to me that the perceptions of categorizations of what is in the
territory are learned by applying the map that everybody around us uses.
This is why category perceptions differ between speakers of one language and
speakers of another. For example, color names and color perceptions (where
the boundaries are perceived to fall in the rainbow) are not universal.
(Brent Berlin's claims to the contrary have been shown to be wrong.)

You only know that you have a tomato by verifying expectable perceptions. A
trompe-l'oeil fake fruit is too hard, too light, certainly does not meet
expectations on being sliced. This association of perceptions in memory in
the aggregate gives us the impression of there being a single category
perception. There is no evidence for there being a single category
perception that we name "tomato" or "dog" other than the naming itself.

You say

The sentence "the scientists dug at the base of the tower" evoked for Bill
stick figures which he took to be category perceptions; for me it evoked a
stereotype of men in white lab coats. Both of course are inappropriate, in
that they are not what one present at the scene would have seen; and neither
Bill's perceptions nor mine are likely to be the same as the perceptions
imagined by the author of that piece about an archaeological dig in Rhode
Island--imagined, since he was not present either. In short, there is no
clear evidence of a universal category-perception "mentalese" of which
language is a translation.

But this is exactly what I think of as the evidence. At the category level,
we perceive many different lower-level perceptions as being "the same
thing." When you say "the men", you're naming a category that can be
exemplified by _any_ set of more than one peceived man. When you name only
the category, without at the same time naming other categories that must
exist at the same time ("the black men", "the male scientists", "the men
from the 8087th Engineers"), you leave the listener free to imagine any
specific examples that would lead to the same category perception. When you
say "digging" you refer to a perceived category of processes that could
include the use of any tools from bare hands to dentist's picks to a
backhoe. When you say "at the base of the tower" you again have a choice of
specific situations that lead to the same relationship perception, or even
different ones -- you might imagine digging into the base of the tower
itself, or into the ground near it. And of course you have a wide range of
structures that would all be perceived as the same, and be named "a tower."

Repeating what I said earlier, as we read a text like this, we typically do
not associate the words and constructions of words with visual images and
other perceptions. We are adding to and otherwise adjusting associations of
words and constructions in memory, that aspect of our knowledge that is
remembered and recalled using language--"knowledge about" as distinct from
"knowledge how". When asked, or when we have some other reason for it, we
project particular perceptions onto the words and constructions. This is why
your visualizations and mine were particular, each stereotypical in its own
way, and different. In neither case were these visualizations themselves
category perceptions.

Reading a text like that may increase the expectability of certain word
co-occurrances (scientists dig -- ah, archaeologists dig, archaologists are
scientists, I knew that) and adds to particular word associations, at least
as long as I remember them (stone tower in Rhode Island may be pre-colonial,
associated with supposed runic inscriptions in New Hampshire, etc., a quirky
little corner of stuff I have read or heard about and remember). When I
actually visit such a site, or talk with someone about something they've
seen, or stumble across a stone circle in a meadow (I haven't), I'll
remember some of this, probably changed in the process of integrating the
new with the old, making unwarranted generalizations by substituting
classifier words for particular words within the selection sets of the
classifiers, etc. I will project my immediate perceptions onto my remembered
knowledge-about, which will set some of my expectations and in that way
guide my control of perceptions in the new encounter. I think this is
typical of how we use language, the social tool, to recognize the familiar,
to contextualize the unfamiliar, to direct attention, to determine which
perceptions to control in the encounter with a complex environmental variable.

        Bruce Nevin

[Avery_Andrews2]

[Avery Andrews 970929.1501 Eastern Oz Time]
(Bruce Nevin (970927.1630))

Avery Andrews 970919.1258 --
>So here's my story about `grammar as a perceptual system'. [...]
..
Sorry, I'm going to be very slow responding.

me too ...

I agree that a clear account of "the Describes relation" is central to a PCT
account of language and a prerequisite for answering Bill's question "what
is a rule (really)?" However, I believe we can't take it to be as simple as
both you and Bill are assuming it to be. Following your account:
...
...
In fact, the word "in" does not always neatly correlate with a perception
that one could describe as the "is_in" relationship. In other words, "in"
occurs in context of words where any connection to the spatial relationship
called "is_in" is remote or utterly vanished in the etymological mist.
Consider the examples in the preceding sentences. I can look for a 1970s
paper that lays these problems out in some detail, if you like. These are
for the most part problems due to the arbitrariness of language--divergence
and specialization of meanings over time, preservation of frozen expressions
and other archaisms, reinterpretation in ignorance of original meanings, and
so on.

I confess to the charge of oversimplification, but I'm not sure that it's
a vitiating flaw or something that can be fixed. Words such as `in' for
example might be connected not with one Relationship, but with a set
of Relationships, with the most sensible one for the context being chosen
(by a process perhaps sort of like an implementation operator overloading
in C++, but more flexible (operator overloading is a device whereby you
can for example define a data type of `set' and write code to define `+' to
mean union of sets; the compiler figures out whether `a+b' means ordinary
addition or set union).

There is another kind of problem. Our main evidence for category perception
is words, and our evidence for perceptions based upon or "using" category
perceptions is for the most part syntactic constructions in language or in
language-dependant systems like mathematics, symbolic logic, and programming
"languages". Maybe it's "level-blindness aphasia" as Bill says; who can
tell, since "I gotta use words when I talk to you" as T.S. Elliot says.
Here's this tool that we use to represent to ourselves what we know (to
remind ourselves later of what we used to know, or because we want someone
else to know, etc.). To indicate what it is that we know and step back and
look at it, we have no other means than language.

I think I worry about this less than you do, one area where we'll probably
start breaking out of language pretty soon is interaction with computer-game
characters; since the new Dragon speech-rec system is supposed to handle
connected speech, it can't be long before this kind of thing starts showing
up in game engines, and people will start wanting to talk with game chars
about the stuff in the game universe. E.g. `you take this laser cannon
and guard the approach, maybe hiding in those bushes. I'm going to go look
for water in the creek.'

"More `this is like that'" complains Bill of my attempt to say what a rule
is. And with justice. That is how language works, by making analogies --
this is like that. What is needed is to go to a level beyond that at which
language is used. What is that? Who can say? Our means of saying is language.

I think no one seriously claims that language is *used* at the postulated
category level or program level or any other level of perception. Each of us
has a self-image, for example, and we control many lower-level perceptions
as means of sustaining it, but putting it into words is not easy and so far
as I know never accurate or complete. Another example, is a tomato a
vegetable or a fruit? The sensible answer is that it does not matter, it's a
food that you put in salads and soups, or perhaps something you avoid
because of allergy--categories for which there is no simple vocabulary item
in English.

This is a good point because it shows a significant mismatch between the
category system encoded in the English lexicon and the one we actually use in
navigating thru the world. Of course this doesn't mean that the linguistically
encoded categories are different kinds of categories than the ones we operate
on the world through.

So what is the "the Describes relation"? Actually, I agree with Avery's
account, but not quite as far as seems to go. The correspondence is not
between words and category perceptions, or between syntactic relations of
words and some sort of relations between category perceptions. The
correspondence is between co-occurrance sets of words, on the one hand, and
ranges of co-occurring perceptions on the other. For each word there is
associated in memory a set or sets of familiar, or often heard, or
expectable word dependencies. Spread butter, butter on bread, melt butter,
fresh butter, rancid butter, butter in the refrigerator, and so on. Food in
refrigerator, food spoil, etc. The rather inchoate perceptions associated
with the sets of of word associations -- melt butter, melt ice, melt ice
cream -- are brought into focus, crystallized as it were into more
particular expectations in the intersections of the various association sets
represented in a particular utterance. (Some years ago for my master's
thesis I made some attempt to work out an algorithm for this. One day
perhaps when I've finished this silly membership ritual of a dissertation
I'll get back to it, with an understanding of PCT behind it.)

Hmm. I'm imagining that one might get similar effects by (a) having word
meanings be rather abstract (b) collocations. But I think this is a point
of detail that could go one way or another without altering the basic
idea of `grammar as a perceptual system' (hopefully this is why you say
you agree with my account!)

Indeed, under those circumstances we may omit words entirely. Avery, risen
from his slice of toast on the kitchen table, casting about with a butter
knife in his hand, his wife mutely pointing to the refrigerator with her
left hand as she continues to pour coffee with her right. "Ah," he says, and
opens the refrigerator door.

This sort of thing connects to another theme in some of my posting, the
idea that language is often used as a means for controlling mental states,
such as knowledge. When Cindy perceives that I am ignorant of the whereabouts
of the butter, but wish to know where it is, there is a range of things
that she can do about it, some involving language, some not. I'm currently
conjecturing that the essential difference between human language and what
apes can do is that they do not seem to use symbols to control mental states,
but only behavioral results.

Language provides rather rigid and limiting structures onto which we project
our much less structured perceptions.

I think there's an issue here of potential vs. actual structures. My
view of structures (influenced by Penni Sibun, for old timers on this
list who might remember her) is that they tend to be manufactured on a
`just enough and just on time' basis. E.g. if there's a house in your
visual field it just registers as an instance of some category such as
HOUSE, until there's some reason to pay attention to and know more about
it, at which point further detail is manufactured from the scene in
front of you. And this detail is likely to be highly keyed to what it's
wanted for. If we have `imagistic' memory formats (the phenomenon of
`eidetic memory' seems to suggest that we do), then this trick can be
applied to them as well as to actual scenes.

Avery.Andrews@anu.edu.au

[Bill_Powers1]

[From Bill Powers (970929.0600 MDT)]

Bruce Nevin (970928.1304)--

It's hard to get across the idea of a non-linguistic category.

The claim that words and word structures correlate one for one with category
perceptions is equivalent to a claim that the category level uses language
or that language is in the category level. You believe that you are not
making the latter claim; by making the former, you are.

I don't mean that for each category perception there is just one word
associated with it. As far as I know this isn't the case for _any_
perception. That's why reasoning in words is so slippery; you can shift
meanings in the middle of a sentence, not just from one sentence to the
next. Prudence was behaving prudentially, in taking out insurance with
Prudential.

All I'm trying to say, without getting in over my head, is that we seem to
perceive things that are clearly different as being the same in some sense;
that sense is what I am calling category perception. On our recent jaunt,
Mary and I saw several collections of Anasazi (or Old Pueblo, as they say
now) "pots." In one display case there were "examples" of perhaps a dozen
"pots", no two of which looked at all alike. Yet neither of us, not to
mention the display's creator, had any trouble with seeing that the items
were all "pots." Was this simply because we used the same word for all of
them? Or did we use the same word for all of them because we could already
see each pot as an example of a particular category that we _call_ pots?

I do make the claim that one aspect of language is "in" the category level:
the "naming" aspect. A word is just a configuration (written) or event
(spoken) perception. Since I propose that we can make categories out of any
arbitrary collection of perceptions, when we make a word a member of the
same category as some set of nonverbal perceptions, we can use either the
word or a nonverbal perception as an example and thus produce the same
sense of category perception, as in "I [heart] skiing." Or perhaps "I [foot
in cast] skiing." This is an explanation of one phenomenon among those
usually spoken of as "association."

This claim that the stuff of language correlates with or names category
perceptions is false, or so I have tried to show. There is no such
correspondence. Rather, we project our perceptions onto the structures of
language. "Ah, this is an instance of what they talk about when they say X."

"What they talk about" when they say X is what I call a category
perception. The X that they say in words is not that which the X indicates.
The X is not what they talk about; it is a reference to they talk about. An
"instance" is a lower-level perception that is perceived at the category
level as the same as any other instance, in that it gives rise to a sense
of the same category being present. When you use "what they talk about" to
designate the meaning of X, you're trying to separate the category
perception itself from the verbal indicator we use -- to indicate the
meaning of X without saying X.
You're appealing to my ability to perceive the category as a perceptual
experience without the word, just as you would do by pointing to a colored
piece of cloth and saying "That's the color they mean when they say 'mauve'."

We project association-sets of perceptions onto co-occurrance sets of words,
and locate our immediate perceptions in those association-sets. When we
read, or listen, or converse without projecting controlled or controllable
perceptions onto the words and language structures that we are using, we are
merely playing with association-structures of language in memory, extending,
adjusting exploring their ramifications. This is why Shakespeare has Hamlet
enjoin the players to visualize vivid imagery associated with the words they
speak: because it is not ordinarily done by readers and reciters, because it
takes extra work and concentration to do it, in the actor's craft.

There seems to be some kind of model behind what you say here, but I don't
see what it is. What is this process you call "projection?" What is a
language structure, as it exists in a brain? In particular, what is an
"association-set of perceptions"? It seems to me that an association-set
names the same sort of thing I am calling a category. How is an association
formed, and what is its result? According to the conventional view, an
association is simply the process of one perception "evoking" another
(whatever that means). But we would get a similar phenomenon if we said
that a set of items could each evoke a perceptual signal as if the items
were all the same. That signal would indicate the presence of the _set_,
although it would not distinguish one member from another. And that is what
I call a category perception. When I say that a Chihuahua and a Great Dane
are both "dogs", I do not mean that they look the same to me. I mean that
they belong to the same category, which I call "dogs" or "canines" or
"woof-woofs" or "four-footed friends" depending on what _other_ categories
I wish to indicate at the same time.

Are you the only person in the world who can't make sense of what I'm
saying here? Or am I the only one who can?

Best,

Bill P.

[bnhpct]

From Bruce Nevin (970929.1425)

Bill Powers (970929.0600 MDT) --

Are you the only person in the world who can't make sense of what I'm
saying here? Or am I the only one who can?

No, it makes perfect sense to me. It seems to me (1) that it gets into
trouble on the supposed correspondence of category perceptions to words and
word constructions, and (2) that it is not required, that a sufficient
mechanism is memory -- whatever that is -- which must be there with or
without a category level.

For (1) I am on stronger ground, except when I try to communicate to you
what is going on with language and how we can use it to refer to perceptions
and describe them, thence, what a description is, what a mnemonic is, what a
rule is.

For (2) I am on weak ground indeed because I do not have a coherent model to
propose for how memory works or how associations are made, and cannot make
much of a case for how they are involved (as I believe) in the setting of
reference values of controlled or controllable perceptions. My understanding
of how memory works and how to model it is very poor.

It seems to me that an association-set
names the same sort of thing I am calling a category.

Exactly. The idea is that a more primitive mechanism does the work for which
the category level was postulated. I believe it is in memory that we will
find categories, and in the processes of association in memory by which
perceptual input evokes the remembered experiences that amount to expertise,
by which the new experience is integrated with that store, and by which
memory is one basis for choice which of many controllable perceptions to
control -- what to pay attention to -- and the source of reference values
for them.

Memory is not part of the control hierarchy per se but is the source of
reference values of controlled perceptions. It may influence reorganization.
Memory is available at every level of the control hierarchy and in all parts
of the brain where control systems compare remembered reference values to
input signals. Something going on the the limbic system results in some
remembered perceptions having greater urgency or priority than others,
influencing the choice which of many controllable perceptions actually are
controlled. Some kinds of perceptions are especially evocative (whatever
evocation is, however it is to be modelled) -- odors, music -- and in
addition to these everyone has idiosyncratic "triggers" of memory that
reflect personal experience.

It is in memory that association-sets of perceptions are formed, and
selection sets of words and of constructions, and it is in memory that the
association-sets of these two kinds of remembered perceptions, linguistic
and nonlinguistic, are associated.

That's about the extent of what I can say about (2), and pretty flaccid
stuff it is, I confess.

As to (1), which is my motivation for reaching out on this limb, consider
previously noted problems of arbitrariness, historical contingency, analogy
& metaphor, etc., even in the relatively simple case of "dog":

dog
dog fight
dog tired
dogface (soldier)
dog and pony show
put on the dog
that's a real dog (bad car, attractive date, etc.)
it's a dog-eat-dog world
he's bird-dogging
she dogged their footsteps
dogged determination
hot dog
he's hot-dogging
dog star
dog days
dogged down (fastened with a dog)
you lucky dog!
she cleaned and polished the dogs (andirons) before laying the fire
dog latin
dog ear

That's an awful lot of work for reorganization of extremely complex category
perceptions, perhaps not much work at all for memory, depending on how it is
modelled.

On another occasion, maybe I can get into a sketch of how to model how we
understand language, and how we could model a knowledge base of associations
of word-dependencies that is added to and modified in the process of
understanding sentences. In part, this is like Stephen Johnson's work at NYU
and in medical informatics at Columbia Presbyterian, but what I sketched in
my MA thesis goes considerably farther.

        Bruce

[Bill_Powers1]

[From Bill Powers (970929.1954 MDT)]

Bruce Nevin (970929.1425)--

Are you the only person in the world who can't make sense of what I'm
saying here? Or am I the only one who can?

No, it makes perfect sense to me. It seems to me (1) that it gets into
trouble on the supposed correspondence of category perceptions to words and
word constructions, and (2) that it is not required, that a sufficient
mechanism is memory -- whatever that is -- which must be there with or
without a category level.

OK. I'm certainly not going to solve any problems in linguistics. I'll
stick to what I know.

Best,

Bill P.