ontolog-forum
[Top] [All Lists]

Re: [ontolog-forum] NLP Dealing with metaphor [was: RDF vs. EAR]

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: "doug foxvog" <doug@xxxxxxxxxx>
Date: Sat, 10 Dec 2011 09:03:56 -0500 (EST)
Message-id: <50816.72.83.246.198.1323525836.squirrel@xxxxxxxxxxxxxx>
On Fri, December 9, 2011 17:25, John F. Sowa said:
> On 12/9/2011 4:56 PM, doug foxvog wrote:
>> I do not know if the referenced phrase has been encoded in CycL, but it
>> could be encoded as follows:
>>
>>    (multiWordString Have-TheWord "it up to here" Verb
>>        (VeryHighAmountFn Frustration))
>>
>> The Cyc NLP has transformations that map
>>     "<Instance of Agent>  <HAVE>  <Emotion>"
>> to the predicate feelsEmotion, such that it would generate
>>
>>    (feelsEmotion ?SPEAKER (VeryHighAmountFn Frustration))    (01)

> That's a good example of how "frozen metaphors" can be
> represented.  As I said in my previous note, there are
> many special case techniques that work very well.    (02)

John,
Frozen metaphors are learned by people learning a language one-by-one.
If they haven't met the metaphor before, people have several options:
asking someone what it means, looking it up in a reference work, finding
other examples of its usage, and finally trying to intuit its meaning.    (03)

I'm not sure that i would call storing the meanings of words and
phrases a "special case technique".  It seems to me to be the main
technique which language understanding is based -- look up known
meanings of sentence components, combine them using standard rules,
use reasoning techniques to eliminate most of the combinations.  Score
the remaining results.  If all combinations are eliminated or the
remaining ones score low, then special case techniques, e.g. checking
for analogies and metaphor, would be applied.    (04)



> But developing a good, general, flexible method that can
> recognize, interpret, and represent all metaphors is still
> a major research problem.    (05)

I would agree to this with analogies, but often a metaphor could just
as reasonably have several different meanings.  A metaphor often derives
from several analogical steps.    (06)

Certainly, a reasonable understanding of the situation in which an
unknown metaphor is encountered could disparage some of the meanings.
But to expect a person, much less an NLP system to determine the meaning
of new (to it) metaphor seems unreasonable, imho.  One possibility would
be to search corpora for other usage of the metaphor, and look for
commonalities in the situations.    (07)

If a person met a metaphor anew and had no one else to ask, no reference
to check, and insufficient context to determine its meaning, s/he would
likely find multiple possible meanings.  "I've had it up to here" refers
to "having" something, but there are hundreds of ways to have something.
The word "up" indicates the existence of some scale of having, but that
would not limit the types of having that much.  The phrase could logically
mean:
      I'm full.
      I'm drunk.
      I'm extremely suffering from some disease.
and many other things.    (08)


You suggest below that NLP systems
> use reference material or human consultants when they get stuck.    (09)

I would suggest that the use of reference materials would be one of
the first techniques for an NLP system to use, not the penultimate
technique.  Looking up standard meanings of phrases -- whether they
be metaphors or not -- is easier than trying to determine what chains
of analogical reasoning could have produced a metaphor.    (010)

-- doug    (011)

> But we can still do a lot of NLP without solving every
> problem.  Even human beings vary widely in their ability
> to understand their native language.
>
> In fact, nobody can understand every document, even in their
> native language, without using a lot of reference material
> or asking help from teachers and experts in the field of
> application.
>
> A good goal is to develop computer systems to a level where
> they can interpret a lot of language, recognize what they
> don't understand, and use reference material or human
> consultants when they get stuck.
>
> For that legacy re-engineering example, the computer ran
> 24/7 for 3 weeks to download and process all the files.
> During that time, it logged all the problematical areas,
> stated it best guess and confidence factors, and let the
> two developers (Arun Majumdar and André LeClerc) decide
> what to do about them.  See slides (120 to 128) of
>
>     http://www.jfsowa.com/talks/goal.pdf
>
> That project was done several years ago, and the equivalent
> could be done much faster today.  The computers can process
> much larger files much more quickly, but they still need
> human teachers and advisers to help them.
>
> John
>
> _________________________________________________________________
> Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
> Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
> Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
> Shared Files: http://ontolog.cim3.net/file/
> Community Wiki: http://ontolog.cim3.net/wiki/
> To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
>
>    (012)


=============================================================
doug foxvog    doug@xxxxxxxxxx   http://ProgressiveAustin.org    (013)

"I speak as an American to the leaders of my own nation. The great
initiative in this war is ours. The initiative to stop it must be ours."
    - Dr. Martin Luther King Jr.
=============================================================    (014)


_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J    (015)

<Prev in Thread] Current Thread [Next in Thread>