Copy Link
Add to Bookmark
Report
AIList Digest Volume 5 Issue 095
AIList Digest Thursday, 2 Apr 1987 Volume 5 : Issue 95
Today's Topics:
Application - Text Critiquing
----------------------------------------------------------------------
Date: Wed 1 Apr 87 22:35:15-PST
From: Ken Laws <Laws@SRI-STRIPE.ARPA>
Reply-to: AIList-Request@SRI-STRIPE.ARPA
Subject: Policy - Text Critiquing
The following messages are about a system that critiques English
prose. It could be argued whether this particular system is
within the realm of AI, but the application area does seem to
be of interest. This discussion really should be moved to
the AI-ED@SUMEX-AIM list (which has just distributed the text
grading query that started all this), or perhaps to the
NL-KR list. For now, I will continue to distribute to the
Arpanet discussions on this topic that have circulated on Usenet.
-- Ken
------------------------------
Date: 27 Mar 87 19:07:40 GMT
From: ritcv!rocksvax!rocksanne!sunybcs!colonel@CS.ROCHESTER.EDU
(Col. G. L. Sicherman)
Subject: Re: automatic checking
) But there is no reason to grade automatically, just let the students
) ^<<* 23. SENTENCE BEGINS WITH BUT *>>
) work on their papers (with the automatic checker) until they are satisfied.
) <<* 21. PASSIVE VOICE: are satisfied. *>>^
) <<* 17. LONG SENTENCE: 24 WORDS *>>^
"They are satisfied" is in the passive voice? That's what comes of letting
computers run things....
"Hey, Rocky! Watch me pull a UNIX program outa m'
source directory!"
"AGAIN?"
"Nothin' up my sleeve ... PRESTO!"
IDENTIFICATION DIVISION.
PROGRAM-ID. PROCESS-DATA.
AUTHOR-NAME. B. J. MOOSE, FROSTBYTE DATA SYS.
SOURCE-COMPUTER. IBM-7044.
OBJECT-COMPUTER. IBM-7044.
. . .
"No doubt about it--I gotta get a new source directory!"
--
Col. G. L. Sicherman
UU: ...{rocksvax|decvax}!sunybcs!colonel
CS: colonel@buffalo-cs
BI: colonel@sunybcs, csdsiche@ubvms
------------------------------
Date: 30 Mar 87 00:24:00 GMT
From: kadie@b.cs.uiuc.edu
Subject: Re: AI Project Information Request
Several people have ask if the grammar checker I used was real. It
is. It is a commercial product for the IBM PC. Here is some more
information and an example.
I own a spelling checker that I always use. And a grammar and style checker
that I sometimes use. I have a lot of confidence in the spelling checker; I
take virtually all of its advice. The style checker is not as good. I always
consider it's suggestions, but I know that it has missed many grammar
and style errors and that not everything it flags is really wrong.
Enclosed find its critique of a draft report.
This gives a pretty good indication of how well the program works.
The program is RIGHTWRITER version 2.0, a Right Soft product by
Decisionware, Inc.
of 2033 Wood Street, Suite 218, Sarasota, Florida 33577. It runs on
IBM PC's and compatible computers. It costs about $100.00.
Carl Kadie
University of Illinois at Urbana-Champaign
UUCP: {ihnp4,pur-ee,convex}!uiucdcs!kadie
CSNET: kadie@UIUC.CSNET
ARPA: kadie@M.CS.UIUC.EDU (kadie@UIUC.ARPA)
(I disclaim any ulterior relationship to Decisionware.)
.+c "A Program To Compute Moore's Stable Expansions"
.pp
Moore has recently proposed a possible-world semantics for autoepistemic logic.
His method has the intriguing property of producing multiple expansions, that
<<* 16. UNNECESSARY COMMA *>>^
is it list the (finite) theories of what you believe about the world, given
the axioms.
^<<* 17. LONG SENTENCE: 27 WORDS *>>
For example, if your unbelief in proposition $P$ implies $Q$, and your unbelief
in proposition $Q$ implies $P$, then we can theorize that either
$P$ is true or alternatively $Q$ true.
<<* 17. LONG SENTENCE: 31 WORDS *>>^
<<* 31. COMPLEX SENTENCE *>>^
.pp
In Lisp notation the axioms are expressed:
.(L
(and (imp (not (l 'p)) q) (imp (not (l 'q)) p))
.)L
and the conclusion is expressed:
.(L
(Q) (P)
.)L
.pp
I have written a program that finds the stable expansions of
formula in Moore's autoepistemic logic. As might be expected
<<* 21. PASSIVE VOICE: be expected *>>^
the program run in time exponential to the number of variables.
<<* 32. INCOMPLETE SENTENCE OR MISSING COMMA *>>^
Let's look at some runs:
.(L
A non-autoepistemic sentence:
(expand
'(and p (imp p (not q)) (imp (not q) r)) ;; axioms
'(p q r) ;; propositions
0) ;; trace level
returns:
((P (NOT Q) R))
.)L
In other words, the axioms entail that $P$ is true, $Q$ is false, and $R$ is
true. This is of course just what we expect for this propositional sentence.
.pp
Here is a trace of the run of the example we saw before:
.(L
[Figure goes here. -- CMK]
.)L
.pp
The program also identifies cases where no stable expansion exists:
.(L
[Figure goes here. -- CMK]
.)L
.pp
At higher trace levels, the program provides counter-models
to non-grounded theories. For example:
.(L
(expand
'(and (imp (not (l 'p1)) p2)
(imp (not (l 'p2)) p3)
(imp (not (l 'p3)) p4)
(imp (not (l 'p4)) p1))
'(p1 p2 p3 p4)
2)
...
(P1 P2 P3 P4) in theory is stable w.r.t. the axioms.
S5 is ((P1 P2 P3 P4))
(s5:((P1 P2 P3 P4)) , V:((NOT P1) P2 (NOT P3) P4)) is a model of A
Counter-model: (s5:((P1 P2 P3 P4)), V:((NOT P1) P2 (NOT P3) P4))
Theory (P1 P2 P3 P4) is NOT a stable expansion of the axioms
...
((P2 P4) (P1 P3))
.)L
.pp
In fact it is just this test of groundness that makes Moore's logic
different from the logic of Shoham that we will see later.
<<* 17. LONG SENTENCE: 24 WORDS *>>^
For example when we give Shoham's gun
example to the program it replies that there are no stable
<<* 1. REPLACE: that there BY there *>>^
expansions. This is because it does not have Shoham's
chronological ignorance criteria with which to choose ungrounded
theories. Here is the trace:
.(L
[Figure goes here. -- CMK]
.)L
.pp
Having no stable expansion and believing nothing are two separate case.
Here is a case where the only stable expansion is the theory
where nothing is believed.
^<<* 21. PASSIVE VOICE: is believed. *>>
.(L
[Figure goes here. -- CMK]
.)L
.pp
The program works by enumerating every theory, then constructing
the corresponding S5 structure. Next, it tests every world
of the S5, if any world fails to support the axioms then
it is unstable and the theory is removed from consideration.
<<* 21. PASSIVE VOICE: is removed *>>^
<<* 17. LONG SENTENCE: 27 WORDS *>>^
Stable theories are next tested for groundness. This is done
<<* 21. PASSIVE VOICE: are next tested *>>
<<* 21. PASSIVE VOICE: is done *>>^
by trying every variable assignment $V$. If an assignment
makes the axioms true then $V$ must correspond to a world
in the S5, or else the theory is not grounded. A theory
<<* 21. PASSIVE VOICE: is not grounded. *>>^
<<* 17. LONG SENTENCE: 33 WORDS *>>^
that is both stable and grounded is added to the stable
<<* 21. PASSIVE VOICE: is added *>>^
expansion list to be returned at the end of the program.
<<* 21. PASSIVE VOICE: be returned *>>
<<* 17. LONG SENTENCE: 24 WORDS *>>^
.pp
Overall, the program works very well on small problems (four
variable problems take only seconds on a SUN). The program
accepts any formula that Lisp can evaluate;
so very complex formula may be input. However, since
the program relies on enumeration, it can not be expanded
<<* 21. PASSIVE VOICE: be expanded *>>^
to first-order logic, nor can it be considered practical
<<* 21. PASSIVE VOICE: be considered *>>^
unless the problems can be guaranteed to be small.
<<* 21. PASSIVE VOICE: be guaranteed *>>
<<* 17. LONG SENTENCE: 31 WORDS *>>^
<<* 31. COMPLEX SENTENCE *>>^
<<** SUMMARY **>>
READABILITY INDEX: 7.63
Readers need an 8th grade level of education to understand.
STRENGTH INDEX: 0.19
The writing can be made more direct by using:
- the active voice
- shorter sentences
- more common words
- fewer abbreviations
DESCRIPTIVE INDEX: 0.74
The use of adjectives and adverbs is within the normal range.
JARGON INDEX: 0.25
SENTENCE STRUCTURE RECOMMENDATIONS:
15. No Recommendations.
<< UNCOMMON WORD LIST >>
The following words are not widely understood.
Will any of these words confuse the intended audience?
AUTOEPISTEMIC 3 AXIOM 2 AXIOMS 40
CHRONOLOGICAL 1 CRITERIA 1 DRIBBLE 1
ENTAIL 1 ENUMERATING 1 ENUMERATION 1
EXPONENTIAL 1 FINITE 1 FIRE4 20
GROUNDNESS 2 IMP 24 INTRIGUING 1
LISP 2 LOAD1 20 MOORE 1
MOORE'S 3 NIL 5 NOISE6 17
P 4 PROPOSITION 2 PROPOSITIONAL 1
PROPOSITIONS 2 Q 4 R 1
SEMANTICS 1 SHOHAM 1 SHOHAM'S 2
THEORIZE 1 UNBELIEF 2 UNGROUNDED 1
V 2 VACUUM5 17 WRT 19
<< END OF UNCOMMON WORD LIST >>
<<** WORD FREQUENCY LIST **>>
A 31 ABOUT 1 ACCEPTS 1
ADD 1 ALSO 1 ALTERNATIVELY 1
AN 1 AND 18 ANY 2
ARE 4 AS 1 ASSIGNMENT 3
AT 3 AUTOEPISTEMIC 3 AXIOM 2
AXIOMS 40 BE 7 BECAUSE 1
BEFORE 1 BELIEVE 3 BOTH 1
[Rest of word frequency list goes here -- CMK]
<<END OF WORD FREQUENCY LIST>>
------------------------------
Date: Mon, 30 Mar 87 11:31 EST
From: "Linda G. Means" <MEANS%gmr.com@RELAY.CS.NET>
Subject: Re: AI Project Information Request
means%gmr.com@relay.cs.net >Date: 25 Mar 87 01:44:00 GMT
>From: kadie@b.cs.uiuc.edu
>Subject: Re: AI Project Information Request
>
>
>Automatic checking and automatic grading are different things. I think
> <<* 3. WEAK: I think *>>^
>automatic computer checking is a good thing, especially for spelling
>and simpler grammar.
>
>But there is no reason to grade automatically, just let the students
> ^<<* 23. SENTENCE BEGINS WITH BUT *>>
>work on their papers (with the automatic checker) until they are satisfied.
> <<* 21. PASSIVE VOICE: are satisfied. *>>^
> <<* 17. LONG SENTENCE: 24 WORDS *>>^
>Then have them turn in their work and the final computer critique to a human
>grader.
>
>The situation is similar to programming, where the compiler
>automatically checks the syntax. It would be unthinkable to make people turn
>in programs without letting them compile the programs first. On
>the other hand it would unthinkable to leave a syntax error in
>when the compiler tells you right were it is.
>
>
> <<** SUMMARY **>>
>
> READABILITY INDEX: 10.42
> Readers need a 10th grade level of education to understand.
>
> STRENGTH INDEX: 0.41
> The writing can be made more direct by using:
> - the active voice
> - shorter sentences
>
> DESCRIPTIVE INDEX: 0.65
> The use of adjectives and adverbs is within the normal range.
>
> JARGON INDEX: 0.00
>
> SENTENCE STRUCTURE RECOMMENDATIONS:
> 1. Most sentences contain multiple clauses.
> Try to use more simple sentences.
>
> << UNCOMMON WORD LIST >>
>The following words are not widely understood.
>Will any of these words confuse the intended audience?
> CRITIQUE 1 SYNTAX 2 UNTHINKABLE 2
> << END OF UNCOMMON WORD LIST >>
>
>
>Carl Kadie
>University of Illinois at Urbana-Champaign
>UUCP: {ihnp4,pur-ee,convex}!uiucdcs!kadie
>CSNET: kadie@UIUC.CSNET
>ARPA: kadie@M.CS.UIUC.EDU (kadie@UIUC.ARPA)
Carl,
Your submission regarding grammar/style checkers sends a mixed
message to me. The content appears to advocate the use of such
systems as tutoring tools. The automatic critique interspersed
throughout the text, however, seems to belie your intention.
First of all, it failed to point out three blatant errors in the text:
> But there is no reason to grade automatically, just let the students work
on their papers (with the automatic checker) until they are satisfied.
- a comma is an inappropriate conjuntion for the two independent clauses
here; a semi-colon would be more appropriate.
> On the other hand it would unthinkable to leave a syntax error in when
the compiler tells you right were it is.
- 'where' is misspelled as "were".
- 'be' was omitted before 'unthinkable'.
Second, I have a number of objections to the types of criticisms the
program does make. You characterize automatic style checking as
"... a good thing, especially for spelling and simpler grammar".
I would call it simple-minded, not simple. The complaint about
the use of passive voice in "until they are satisfied" is ridiculous.
This is not an example of passive voice at all; it's a predicate
adjective. And even if passive voice had been used there, so what?
The strength index in the summary gives the text a low grade on the
basis of two supposed weaknesses: one occurrance of passive voice,
and one sentence which is overly long (24 words). This evaluation
is quite misleading to a student, who will subsequently comb his
papers for constructions like "they are satisfied" to be purged, and
will frantically count words in sentences. No machine or human
critic should object to the use of "they are satisfied" in this
context (or probably any other). And if you want to evaluate
sentence length as an index of readability, number of words is
too superficial an index to use. The readability of a sentence
is better judged by the embedding of clauses, or syntactic complexity.
I think it would be dangerous to have students become obsessed with
counting words in sentences while ignoring sentence structure, just
because the teacher requires them to use an inadequate computer program
as a teaching aid.
I could go on and on. And I think I will, because I get so angered
by the commercial crap which is passed off to gullible teachers and
parents as computer-aided instruction! I don't want my child to
learn how to write with a program that scolds him every time he
begins a sentence with the word 'but', and tells him that he should
"try to use more simple sentences" because "most [of his] sentences
contain multiple clauses". There's nothing wrong with multiple clauses,
even in most of your sentences. Try writing most sentences with
single clauses. The technique will not enhance your writing style,
I assure you. Granted, you don't want to embed clauses in your
sentences to the depths of Hell. But the difficulty in comprehending
sentences with a lot of embedding stems from the syntactic structure,
and not simply the number of clauses. (Nyaaa nyaaa, I just started a
sentence with 'but'. Did it make your skin crawl as you read it? No,
of course not. Some sentences just cry out to begin with 'but', although
not ALL your sentences should.) Which sentence do you find more
readable: the sentence criticized for its length in your text (number 1
below), or my utterly grammatical and very short sentence number 2?
1. But there is no reason to grade automatically, just let the
students work on their papers (with the automatic checker) until
they are satisfied.
2. The man the girl the boy loved kissed died.
I don't want my child's creativity and personality in his writing to
be stifled by a computer program which performs such rigid and
superficial analysis. Nor do I want his writing to be limited to
an average 10th grade level when he's in 10th grade if his writing
ability goes beyond the level of the average 10th grader. Thanks
for including your style checker's criticisms in your message. It
serves as evidence that those programs may do more harm than good
to a child's developing literary talents.
Linda Means
means%gmr.com@relay.cs.net
------------------------------
End of AIList Digest
********************
