Re: OWL "Sydney Syntax", structured english

Still not weighing in properly :) Cold laying me up. But....

On Nov 29, 2006, at 10:57 AM, Kaarel Kaljurand wrote:

[snip]
>> I agree that rendering to a constrained dialect of English is an
>> interesting user-interface technique; I made a note about that
>> during the OWL-ED workshop
>>  http://swig.xmlhack.com/2006/11/10/2006-11-10.html#1163171978.746560
>>
>> "(Tom? stanford guy) says the English format spit out by SWOOP is  
>> very
>> valuable"
>
> Thanks, I'll check out the latest version of SWOOP.

It's not in the latest version of Swoop, but here's the paper  
describing our pilot, focused, study:
	http://image.ntua.gr/swamm2006/resources/paper11.pdf

The key bit (which was a surprise to me):

	"""Using a Wilcoxon matched-pairs signed-ranks test, we found that  
the NL format
gnificantly outperformed the Concise Format (ranked second on  
average) for both
Anjou and Beaujolais with p<0.05. There was not a significant benefit  
over the Con-
cise format for the simplest class, AlsatianWine, but NL did  
significantly outperform
the Abstract Syntax, which was the average third ranked format. This  
allows us to
conclude that in the pilot study the NL format offers significant  
benefits to users when
they are trying to understand the meaning of classes, particularly  
complex classes. """

Earlier paper:
	http://www.mindswap.org/papers/nlpowl.pdf

This was *not* testing authoring, but preference of the user for the  
sake of *understanding* the class, where, I at least as a subject,  
understood "understanding" to be roughly being able to explain it to  
someone who might not be an OWL or Logic person, or over the phone,  
or later, from memory.

I'll see what I can do about getting the NL tab back in by default.  
There were issues about licensing the POS tagger, but we could make  
all that a separate download. It is in the 2.3 beta release:
	http://www.mindswap.org/2004/SWOOP/downloads/SWOOP-2.3beta3.zip

(But it is neither editable nor does it have hyperlinking.)

My understanding of the report at OWLED was that they generate a  
first draft using "normal" tools, then refine the class description  
in interaction with domain experts via the NL rendering. I'm not  
exactly sure how they translate back.

Cheers,
Bijan.

Received on Wednesday, 29 November 2006 11:29:33 UTC