LINGUIST List 3.168

Fri 21 Feb 1992

Sum: Brain Research, Selectional Restrictions

Editor for this issue: <>

Directory

1. "Bruce E. Nevin", brain research, Gopnik
2. Arran, Selectional Restrictions: Thanks for all the replies.
3. "Bruce E. Nevin", selection restrictions

I asked

> Is there a good place to look for a survey and summary of research
> into brain function and language capacities (lesions, aphasia, etc.)?

Here are the very helpful responses I received. Thank you both!
As promised, I am posting them here in a summary.

-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-

T. Daniel Seely
Program in Linguistics
Eastern Michigan Univ
Ypsilanti, MI 48198 eng_seelyemunix.emich.edu

Some books on brain research:
Yosef Grodzinsky 1990 "Theoretical Perspective on Language Deficits"
	MIT Press.
Richard M. Restak 1979 "The Brain: the last frontier"
	Warner Books
Philip Lieberman 1984 "The Biology and Evolution of Language"
	Harvard.
William F. Allman 1989 "Apprentices of Wonder: inside the
	neural network revolution" Bantam.

-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-

Dept. of Anatomical Sciences & Neurobiology
Phone: (502) 588-7288

Caplan, David. 1987. _Neurolinguistics and Linguistic Aphasiology:
An Introduction_ New York: Cambridge Univ. Press. A very thorough,
scholarly and readable introductory text by one of the biggies in
this field.

Churchland, Particia Smith. 1986. _Neurophilosophy: Toward a
Unified Science of the Mind/Brain_. Cambridge, MA: MIT Press.
I'm still reading this, but Churchland has done a good job so
far in addressing more general issues of neuroscience and
psychology.

For current research, you must look at the journals _Brain and
Language_ and _Cortex_. Of course _Brain and Language_ is an
obvious choice, but Cortex has articles relevant to neurolinguistics
fairly frequently as well.

Here are a few representative interesting articles that I have
run across recently:

Demeurisse, G. and A. Capon. 1991. Brain activation during a
linguistic task in conduction aphasia. _Cortex_ 27:285-294.

Damasio, H. and A. R. Damasio. 1990. The neural basis of
memory, language and behavioral guidance: Advances with the
lesion method in humans. _Seminars in the Neurosciences_
2:277-286.

Ojemann, G. A. 1990. Organization of language cortex derived
from investigations during neurosurgery. _Seminars in the
Neurosciences_ 2:297-305.

Mesulam, M-M. 1990. Large-scale neurocongitive networks and
distributed processing for attention, language, and memory.
_Annals of Neurology_ 28:597-613.

These references cover a broad range of interests in neurolinguistics.

-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-

I also asked for information about where Gopnik had published her
findings. I got only suggestions for alternative views. These were
from Jeff Snow (SNOWJSHUGSE1.HARVARD.EDU), who recommended Leonard and
Tallal:

Leonard, L. (1987). Language learnability and specific language
	impairment. Applied Psycholinguistics, 10, 179-202.

Tallal, P., Ross, R., Curtiss, S. (1989). Familial aggregation
	in specific language impairment. Journal of Speech and
	Hearing Disorders, 54, 167-173.
-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-=+=-

	Bruce Nevin
	bnbbn.com

I'd really like to thank everyone who replied, I got some very useful
suggestions and examples. At this stage I'm very open to
suggestion, and any more comments and examples would be very welcome.
Below I'll explain what more about what my special interest is, and
the train of thought which led me to the project:

1) If you write a grammer which generates sentences and then look at
 a random sample you find sentences which are not sensible, for
 example:

 Ronnie hits Ronnie with Ronnie.

2) To add a level of semantic filtering, you can use a semantic
 network to make the arguments of 'hits' an animate subject,
 a concrete direct object and an inanimate concrete indirect object.
 Now the grammer will not generate sentences like the one above
 but will generate more sensible things like:

 Ronnie hits the nail with the hammer

3) This introduces selectional restrictions, which most text books
 describe in one paragraph with a couple of examples. The problem
 is that it is very difficult to describe what the selectional
 restriction actually is. Considering my example again:

 a) The house climbed up the stairs.
 b) The road climbed up the stairs.

 I recieved a reply which pointed out that a finer semantic study would
 put 'house' in a point-like category and 'road' in a line-like category. This
 shows the main problem. To describe even a small subset of English
 there need to be a huge number of categories. So a house is a
 point-like, concrete, immovable, heavy etc. Basically the semantic
 categories are trying to capture our intuition of what a house is.
 Then you have two different senses for climb, one taking a point-like
 subject and the other a line like subject. Presumably you would then have
 two different entries for climb in the semantic lexicon. One of my ideas is
 to create a model from the other perspective. To try and capture in the
 nature of a 'road' the fact that it can be thought of as climbing,
 running, going up, where these things mean in some sense 'lying'. As
 in:

 c) 'The road runs between Edinburgh and Glasgow'
 d) 'The road lies between Edinburgh and Glasgow'

 With (d) being being a kind of underlying meaning of (c) because it
 does not break the selectional restrictions. In a database of facts
 such as RUN(Sebastian_Coe,olympics) we would not want to see
 RUN(The_road,Edinburgh,Glasgow) but rather LIE_BETWEEN(The_road,Edinburgh,
 Glasgow).

4) So, as an idea, I would like to model some of the intuition about certain
 objects, for example that a road can be thought of as running because it
 starts in one place and ends up in another. In the example
 'The Rolls_Royce eats gasolene' I would like to represent the intuition
 that a car consumes gasolene and gives out waste products which could be
 analogous to eating (whereas say a table does not).

5) I picture something like:
 a) check to see if normal selectional restrictions are broken
 b) check the knowlege about the objects to find an alternative
 reading which does not break the restrictions.
 c) if none can be found, reject the sentence as not sensible.
 (or in a limited implementation as not understood).

6) The questions are what type of knowlege needs to be stored, how
 to get this knowledge and how to use it to do the things above. To
 get an idea of the kind of knowlege which needs to be stored I'm
 collecting examples of where normal restrictions are broken but
 the sentence is still a meaningful sentence of English. (if you
 can think of more examples like 'The road climbs the hill' or
 'The Rolls_Royce eats gasolene' then please send them to me.

Of course it would be far to ambitious to hope to come up with a
formal model to capture all this, and there are lots of big problems.
(like the artificiality of semantic category names, which seem to be
little better than type1, type2, type3 etc.). However, ideally I would
like to create an implementable model which can extract the meaning from
more sentences and generate fewer nonsense sentences.

At the moment I'm reading as much as I can about the subject, and I'd
welcome any ideas about references to read.

Thanks again for all the ideas and comments,

Arran.

Arran (no last name given, arranling.edinburgh.ac.uk) asks for help
understanding selection restrictions. I think it is helpful to think of
these in terms of normal selection plus analogic extension outside the
normal selection, for metaphor. In what way is a Rolls Royce analogous
to an animal and gasoline analogous to that animal's food?

Another way to get better control of normal selection is to notice that
selection restrictions are more precise and more stable for restricted
subject-matter domains, particularly technical domains. Each such
domain has its specialized sublanguage, characterized by differences of
vocabulary, differences of selection (where a lexical item is the same
as in another domain), and differences of word classes. Metaphor may
then be seen as a process of borrowing from one domain sublanguage into
another. (On sublanguage, the Kittredge & Lehrberger volume on
sublanguage is a start, also Naomi Sager's book on natural language
processing, cp. her discussion of medical informatics. Try Harris's
_The Form of Information in Science_ for a much deeper bite.)

This may help you take another tack with the examples you are gathering.
That phrase using "tack" illustrates what I am talking about. It is
borrowed from nautical (sailing) usage. Language users unaware of this
are currently reinterpreting it as "take another tact." It thereby
becomes a new lexical item, whose meaning now dismoored from its
metaphorical origins is liable to drift with new and unforseen currents
of usage, perhaps those associated with "tactic" and "tactful". Less
metaphorically, selection restrictions appropriate for the image of a
sailboat zig-zagging across a body of water are apt to be violated by
users of this new lexical item "tact."

By the way, I don't find this meaningless:

 2) 'The idea eats the car'

It's perfectly reasonable in, say, a sequel to the Beatles' film "Yellow
Submarine." Might even find a way to make it a colorless green idea.

I can say "rust ate my car". This I would derive from "rust affected my
car as something eating it." (Similarly for acid.) So I work for GM as
a designer. I have a great idea for a new car being developed.
Everybody likes it. The company is committed to it. But over time it
entails many unforeseen design changes in other aspects of the car.
Finally, the car turns out to be too expensive to produce and market.
My colleagues tell me the idea ate the car. They always thought it had
poor sales value--kind of colorless. And it wasn't tested enough, it
was applied to the development process while it was still green. Well,
you can see it coming, so I'll stop there.

The derivation of "rust ate my car" given above is from Harris, _A
Grammar of English on Mathematical Principles._ This is the general form
for derivation of metaphor. Is more formalization required than that?

	Bruce Nevin
	bnbbn.com