[Top] [All Lists]

Re: [ontolog-forum] Using controlled natural languages for ontology

To: ontolog-forum@xxxxxxxxxxxxxxxx
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Thu, 10 Mar 2011 12:40:46 -0500
Message-id: <4D790D1E.3090702@xxxxxxxxxxx>
Adam, Ed, and Rich,    (01)

> Also, I should clarify that we did not use controlled English to
> define SUMO, but rather used SUMO as the definition for terms in the
> output of CELT.    (02)

Thanks for that clarification.    (03)

In any case, generating commentary from the implementation can be
very useful for solving a major problem in documentation:  keeping
it in synch with the implementation.    (04)

> Pease, A., and Li, J. (2010) Controlled English to Logic Translation. In
> Theory and Applications of Ontology, ed. Roberto Poli, Michael Healy,
> and Achilles Kameas, Springer, ISBN: 978-90-481-8846-8.
> http://www.adampease.org/professional/CELT.pdf    (05)

Thanks for the reference.    (06)

> The problem is that the average domain expert naturally _writes_
> a different language and takes some training to learn to write
> the controlled language.    (07)

Domain experts are often very highly paid people, and what they
write are usually unintelligible scribbles on a prescription or
autopsy report.    (08)

The dissertation by Tobias Kuhn is useful for a systematic study
of ACE (Attempto Controlled English) and MLL (a variant of the
Manchester notation for OWL).  He showed some interesting results:    (09)

  1. The test subjects learned ACE faster than MLL.    (010)

  2. They were significantly more accurate in writing ACE.    (011)

  3. They were significantly more accurate in understanding ACE.    (012)

  4. They required less time to read and write ACE.    (013)

  5. They preferred to use ACE.    (014)

And, by the way, most of the subjects were native German speakers.
It would be interesting to test them on a version of AC-Deutsch
as well as AC-English.  See Chapter 5 of    (015)

http://attempto.ifi.uzh.ch/site/pubs/papers/doctoral_thesis_kuhn.pdf    (016)

> My favorite recent example was:
>   The surface must be contained between two planes that are 0.25mm apart.
> The CNL interpreter understood the constraint to refer to two distinct
> instances of a class of object described as 'plane that is-apart by
> 0.25mm'!    (017)

What system generated that interpretation?  Did it have an echo that
showed how it interpreted the input?  Was there any comparison to the
error rates of people who used some formal language for the same kinds
of inputs?  How long did it take users to learn the formal language
compared to the CNL?  What were the users' preferences?    (018)

One anecdote means little.  And one anecdote about somebody's
research prototype is as useful as a study in 1970 that convinced
_Consumer Reports_ that digital watches would never sell.    (019)

> I have yet to see a domain expert working with a knowledge engineer who
> produces quality results.  The few good examples I have seen are where the
> knowledge engineer IS a domain expert.    (020)

I agree, but with some qualifications about the definition of 'expert'.    (021)

John    (022)

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
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (023)

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