Copy Link
Add to Bookmark
Report
NL-KR Digest Volume 03 No. 05
NL-KR Digest (7/28/87 21:23:47) Volume 3 Number 5
Today's Topics:
re: Sanskrit Conference
Re: An Unsearchable Problem
Natural Kinds (Re: AIList Digest V5 #186)
Journal Issue - Planning (Int. J. for AI in Engineering)
CHAT
----------------------------------------------------------------------
Date: Tue, 14 Jul 87 20:09 EDT
From: Bob Widlansky <bwidlans%zodiac@ads.arpa>
Subject: re: Sanskrit Conference
Recently, I read a short intriguing article about the
First International Conference on Knowledge Representation and
Inference in Sanskrit (held in Bangalore, India between December
20-22, 1986) in AI Magazine.
Does anyone know where I can get a copy of the proceedings?
If you do, please contact me.
Thank you,
Bob -- bwidlans@ads.ARPA
------------------------------
Date: Fri, 17 Jul 87 12:11 EDT
From: Bruce Nevin <bnevin@cch.bbn.com>
Subject: Re: An Unsearchable Problem
Re: An Unsearchable Problem
(NL-KR 3.4, Peter Berke of the UCLA Computer Science Department
<berke@CS.UCLA.EDU>)
The question of logical types is critical and often ignored. I have
some suggestions for strengthening your presentation.
Why coin a word `equivocacy' (with spurious allusiveness to words like
`advocacy') when there already exist the words `equivocalness' and
`equivocality'? The road to obfuscation is paved with new coinages.
There is another reason the distinction between ambiguity and
equivocalness is not well founded. The term `ambiguity' is well
established for cases where alternative clear interpretations are
available. What you want is a term for cases where a word or image has
no clear interpretation.
Moreover, you have to provide some better examples if we are to
understand and use your distinction.
The discussion of `IF a THEN IF b THEN c ELSE d' vs isolated `else'
itself equivocates (uses duality of reference in a misleading way): the
use of ELSE in context is in a computer language, and the claim that
`else' has meaning in isolation is with reference to natural language.
Is it not the case that isolated ELSE has no meaning in a computer
language? I think you have to recast the first case in natural language
terms if you want to contrast it with the second. Another reason is
that any computer language has provision for specifying the scope of
conditionals precisely to preclude ambiguity of this kind. The
ambiguity *only* arises in written forms of natural languages, which use
intonation to specify scope in this sort of expression. Thus, you have
provided no example of an expression for which there is *no* clear
interpretation.
Any expression sufficiently out of context is uninterpretable. `Every.'
Conversely, any such expression may occur in a prior or concurrent
context such that it is totally unambiguous. `Do mean "each" or
"every"?' Even `colorless green ideas sleep furiously' has many times
been embedded in a context that supports reasonable interpretation,
notably (with striking use of irony) in a poem by Dell Hymes in one of
his papers on communicative competence. Given an expression or image
and sufficient need to interpret it, humans *invent* sufficient context
to make a reasonable interpretation. A large measure of the process of
understanding is the process of contextualization. Can this process of
placing things in an context so that they are capable of interpretation
be be accomplished by goal-search in a state space? I don't know, but
it is a different problem from the one you found insoluble.
Look at your problem definition again, substituting terms: `Even if we
can simulate [lack of interpretable context] by a host of [alternative
interpretations in alternative contexts], we don't thereby remove [the
lack of context], we push it "down" a level, and then ignore it.'
You note that `"Representing something" in a given formalism cannot be
represented in that formalism.' This may be true of formal languages.
However, note that this is because formal languages depend on a
metalanguage hierarchy external to themselves. They lack the means to
talk about themselves (at least in any comprehensive way). It is not
true of natural language, and therefore not true of our ordinary means
of expressing our understandings with natural language.
In particular, your argument re inability to use language to represent
characteristics of language and language entities is reduced in scope by
the fact that natural language does contain its own metalanguage. Thus,
one *can* characterize `word', `language', etc. using words in a
descriptive (perhaps operational) way.
The important point remains that the meaning `carried by' words cannot
be expressed adequately by other words. Meaning or information is not a
language entity (like words, sentences, etc.) but is something that
language `carries'. Because it is not part of the object language,
meaning or information is beyond the descriptive scope of the
metalanguage hierarchy that is native to each natural language. (I put
the word `carries' in quotes because it is a metaphor.)
This applies of course to translations of one language into another. A
translation of this posting into Armenian would be a representation of
the meaning or information in it in a sense, but only in the same sense
that the original text is also such a representation. Note in
particular that translation into a formal language has no special claim
to being a semantic representation, pace many approaches to formal
semantics. I have previously mentioned Harris, Ryckman, Gottfried, et
al. as demonstrating a naturalist theory of information and formal
semantics distinct from the usual rationalist/realist theories of formal
semantics.
I do not see how it follows that we cannot apply our ordinary means of
understanding to the process of understanding itself. We can tie
ourselves into dreadful knots by ignorance of the distinctions of
logical type involved, but because we are under no compunction to ignore
those distinctions we can retain clarity. Whatever we can understand,
we can describe (model). Indeed, in general, if we cannot describe
something, we conclude that we have not understood it, if we cannot
describe something fully, we conclude that we have not understood it
fully, and we conclude that we have understood a thing to the extent
that we can describe it. By claiming that we cannot describe the
process of understanding, you are really claiming that we cannot
understand the process of understanding. This would be a strong claim
indeed, stronger than Goedel's demonstration of undecidability.
One possible support for such a claim would be the evident requirement
of an infinite regress of meta-*understandings.
In the case of natural language, the infinite regress of metalanguages
reduces to four levels. The fourth metalanguage suffices for all higher
levels, differing only in reference. Harris tells a joke to illustrate
this. A colleague delivered a paper in his native language at an
international conference. For publication, the abstract had to be in
English, not his native language. In an opening footnote, he said:
I wish to thank U. for helping me with the translation of the
abstract.
I wish to thank V. for helping me with the translation of the
previous sentence.
I wish to thank W. for helping me with the translation of the
previous sentence.
The analogy is thus:
Paper in native language : object language
Abstract in English : metalanguage 1
Thank U. for abstract : metalanguage 2
Thank V. for previous : metalanguage 3
Thank W. for previous : metalanguage 4
The thanking of Z suffices for the implicit infinite regress of
thankyous, differing only in reference as to the object sentence, not in
reference to W and not in its form.
------------------------------
Date: Wed, 22 Jul 87 10:43 EDT
From: MINSKY%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU
Subject: Natural Kinds (Re: AIList Digest V5 #186)
[Excerpted from AIList
Note, I have not been reposting the entire Natual Kinds discussion, for
though the subject is relevant to this list, most of the postings have not
been. I will continue to filter for NL/KR relevant postings. - BWM]
About natural kinds. In "The Society of Mind", pp123-129, I propose a
way to deal with Wittgenstein's problem of defining terms like "game"-
or "chair". The basic idea was to probe further into what
Wittgenstein was trying to do when he talked about "family
resemblances" and tried to describe a game in terms of properties, the
way one might treat members of a human family: build, features, colour
of eyes, gait, temperament, etc.
In my view, Wittgenstein missed the point because he focussed on
"structure" only. What we have to do is also take into account the
"function", "goal", or "intended use" of the definition. My trick is
to catch the idea between two descriptions, structural and functional.
Consider a chair, for example.
STRUCTURE: A chair usually has a seat, back, and legs - but
any of them can be changed in so many ways that it is hard
to make a definition to catch them all.
FUNCTION: A chair is intended to be used to keep one's bottom
about 14 inches off the floor, to support one's back
comfortably, and to provide space to bend the knees.
If you understand BOTH of these, then you can make sense of that list
of structural features - seat, back, and legs - and engage your other
worldly knowledge to decide when a given object might serve well as a
chair. This also helps us understand how to deal with "toy chair" and
such matters. Is a toy chair a chair? The answer depends on what you
want to use it for. It is a chair, for example, for a suitable toy
person, or for reminding people of "real" chairs, or etc.
In other words, we should not worship Wittgenstein's final defeat, in
which he speaks about vague resemblances - and, in effect, gives up
hope of dealing with such subjects logically. I suspect he simply
wasn't ready to deal with intentions - because nothing comparable to
Newell and Simon's GPS theory of goals, or McCarthy's meta-predicate
(Want P) was yet available.
I would appreciate comments, because I think this may be an important
theory, and no one seems to have noticed it. I just noticed, myself,
that I didn't mention Wittgenstein himself (on page 130) when
discussiong the definition of "game". Apologies to his ghost.
------------------------------
Date: Fri, 24 Jul 87 09:29 EDT
From: sriram@ATHENA.MIT.EDU
Subject: Journal Issue - Planning (Int. J. for AI in Engineering)
[Excerpted from AIList]
INTERNATIONAL JOURNAL FOR AI IN ENGINEERING
SPECIAL ISSUE ON PLANNING
APRIL 1988
The April 1988 issue of the International Journal for AI in
Engineering will be dedicated to Planning. The guest editors for this
issue are: Prof. Chris Hendrickson, Dept. of Civil Engineering, C-MU,
Pittsburgh, PA 15213 (hendrickson@cive.ri.cmu.edu) and Mrs Julie
Gadsden, Admiralty Research Establishment, Procurement Executive,
XCC5.2, Portsdown, Portsmouth, Hants PO6 4AA, UK. Papers in all areas
of engineering, as related to planning, are solicited. Each paper
should not exceed 10,000 words (roughly 30 doubly spaced pages),
including figures. The deadline for submission is September 1, 1987.
Please send the papers to either of the guest editors.
Sriram & McCallum (Editors)
------------------------------
Date: Mon, 20 Jul 87 15:30 EDT
From: mchalem@radc-lonex.arpa
Subject: CHAT
[Excerpted from PROLOG@SUSHI.STANDFORD.EDU]
Chat-80 is a PROLOG-based front-end natural language interface to
a worldwide database. It was written in partial fulfillment of
Fernando Pereira's PhD. thesis at the University of Edinburgh for the
degree of Doctor of Philosophy in Artificial Intelligence. Michael
McHale, John Crowter, and Mary Ann Huntley of RADC are working in
conjunction with Dr. Richard Kittredge, a linguist from the University
of Montreal and the Odyssey Research Corporation in Ithaca, to extend
Chat-80.
The worldwide database has been replaced with the database
Michael Hilton is using for his simulation, and includes parts of West
Germany, East Germany, and Czechoslovakia. Scott Gregory wrote LISP
routines which convert the database from ASCII to PROLOG. It is hoped
that the modified Chat will serve as the natural language interface to
the aforementioned simulation.
Chat-80 takes as input a natural language query, such as, What is
the capital of the United States. The query is first syntactically
parsed, and categorized as being one of four basic types of queries:
declarative, yes/no, wh-type, or imperative. This parse is then sent
off to semantics, where it encounters query planning and then query
evaluation. In this (not so well understood) stage of processing, the
query is mathematically optimized such that the most efficient access
to the database is performed. The database is then searched for the
answer, which is printed.
Modifications to CHAT-80
Many people suggested that Chat-80's grammar should be changed as
little as possible due to its extreme complexity and nice
organization. We therefore have been extensively modifying the
lexicon. In addition to addding new words, we have added a clever
feature to handle verbs. Previously, to add a verb and all its
tenses, six new entries had to be added to the lexicon. Now we have a
rule whereby we add the verb in only two places, and the tenses of the
verb are automatically created and added to the lexicon. It is also
necessary to modify the semantic dictionary, as well as the domain
dependent rule base when adding a new item to the syntactic
dictionary. It is interesting to note that although a complete
syntactic parse is made before entering semantic interpretation of the
input, some key words are coded directly into the grammar, i.e.,
wh-terms: where and what.
Preprocessing
In an effort to utilize as much of Chat-80 as possible without
changing the grammar, we decided to preprocess the input query. In
this way, we massage the query into one which Chat-80 (hopefully)
understands and can answer. The preprocessor contains several steps
in accordance with several problems we have encountered:
Imperatives
We want to be able to handle imperatives, such as Show me road
7, and Please show me the railroads. (This is with the intent that
Jean Carletta's graphics routines will be utilized. When the user
asks to see the railroads, for instance, the railroads on the screen
will be highlighted.) We thus look at the input query, and try to
match the beginning of the query with the three types of imperatives
our system can handle: show, show me, and please show me. Upon
finding that the input starts with one of these three phrases, the
query is transformed into one of the form: Where is ..., and this new
query is handed to Chat, as if it were the original query.
Compound words
Compound words such as Erfurt_North (anything with an underscore in
it) and type one, type two, and type three (words which have a special
meaning when used together) posed another problem. We have made it
possible to type a compound word as either having an underscore or not.
Also, type one, type two, and type three now are treated as a single
entity. Compound nouns were also problematic; i.e., road 7, where both
road and 7 are considered nouns. We have enabled the grammar to handle
double nouns by classifying them as a special type of compound word.
Synonyms
We wanted equivalent words to have equivalent meanings. Ergo, we
have made the following synonyms in our system:
waterway = river = stream = brook = creek
town = city
road = roadway = route
Spelling Checker
Due to the complexity of the spelling of the proper nouns in the
database, as well as the human error of misspellings and typographical
errors, a pattern matching spelling checker is implemented. If a word
in the input is not in the dictionary, then the user is informed, the
closest match replaces the unknown word, and the query proceeds as
usual.
"What is proper noun" query
It is helpful for the user to ask the system, for example,
What is Apolda. The system should then reply a town. However, the
response supplied by Chat was Apolda. Similarly, when asked what 7
was, the system replied 7, when what was intended was a road. We thus
needed a way to solve this problem. Our solution was to turn to the
preprocessor, which will look at the input query, and determine if it
is of the form "What is proper noun." If it is, then we check to see
what kind of thing the proper noun is during the logic phase of the
query processing.
Single noun phrase queries
It is desirous to have the ability to input a single noun
phrase such as Merseburg or road 7, and have Jean's system highlight
the appropriate area on the map. Thus, we needed to be able to parse
single noun phrases. Again we turned to the preprocessor to solve
this problem. If the input query contains a single noun phrase, then
we transform it to the form: "Where is np," and the process proceedes
as usual.
semantic hierarchy
We found it necessary to change the semantic hierarchy in
accordance with the information in the specialized database with which
we are working. The new hierarchy is as follows:
feature
area point line
block town land water
country heliport
terrain bridge
wetland dam road river
mountain obstruction railroad
airstrip power line
border
Test file
A test file was created which contains several queries which
currently work on our system. This was added so that whenever we add
a new feature to the system, we can execute this file to see if the
modification creates errors elsewhere.
Help file
David Warhoftig has created a help file to orient new users to the
system. It is composed of three parts:
Welcome for new users
This introduces the user to Chat; i.e., what it is, what it does, and
how to use it.
Sample queries and responses
Similar to the module in Chat-80, this facility aids the user in
formulating a query which the system can understand.
Helpful hints and reminders
Suggestions are offered on how to represent the input; i.e.,
the user is informed that the query need not start with a capital
letter, nor does it need to end with a question mark. Also, commas in
the input are ignored, and a spelling checker is implemented.
Current problems
Distance
We are currently examining distance queries, such as 'How far is Goslar
from Merseburg', and 'What is the distance between Apolda and Artern'.
Jean Carletta has written LISP routines to compute distances, and we will
either use them or modify them to suit our needs. In either case, they
must be translated to Prolog.
Pronomial Reference
We would like for the user to be able to use pronouns, and have them
refer to the response from the previous query.
Difficulty with adjectives
We have found particular difficulty in adding adjectives to Chat-80. We
feel that the implementation of adjectives may be somewhat weak, as when
we asked the original Chat-80 if Tokyo is European, the reply was 'yes'. We
would like to have the capacity to ask queries involving the adjective
mountainous.
Remarks
In light of the fact that a great deal of effort was expended to gain an
overall feel for the code, we feel that we have been making steady progress
in modifying Chat-80.
------------------------------
End of NL-KR Digest
*******************
