[Attempto] or is interpreted as inclusive or , or xor?

[George Herson]

Thanks for the feedback.

It is agreed that FOL is limited.  And that machine understanding of 
unconstrained natural language without human assistance is impossible for the 
foreseeable (ie, is AI-complete).

But I'm not suggesting a tool for capture of arbitrary meaning in Attempto, it's 
only first order logic.  We know, however, that Attempto is enough to represent 
real world English texts, such as guideline recommendations "Diagnosis, 
Treatment, and Evaluation of the Initial Urinary Tract Infection in Febrile 
Infants and Young Children," [1] for potential use in a medical decision support 
system.  Which is actual utility.  Which is the kind of usage I'm talking about.

The medical doctors desirous of making this translation were fully aware of the 
guideline's intent, yet called on the Attempto team's help to make said 
translation.  This shows the need I've identified.

Back to feasibility.  A translation tool that starts with the guideline as input 
then queries the doctors as needed to disambiguate it is definitely possible to 
some degree because
1. the guideline is well understood by the doctors and 
2. is representable in ACE (entirely in this case).  
But: how much work does the doctor have to do versus automated by the tool?  How 
much querying of the experts would be necessary for the translation tasks?  For 
support of a new domain?

I don't know the answers but hopefully we'll find out.

George

[1]  http://attempto.ifi.uzh.ch/site/pubs/papers/cnl2009main_shiffman.pdf 



----- Original Message ----
From: Norbert E. Fuchs <fuchs at ifi.uzh.ch>
To: George Herson <gherson at snet.net>
Cc: Gabriele Kahlout <gabriele at mysimpatico.com>; attempto at lists.ifi.uzh.ch
Sent: Sun, July 11, 2010 12:46:10 PM
Subject: Re: [Attempto] or is interpreted as inclusive or , or xor?


On 11 Jul 2010, at 18:29, George Herson wrote:

> A brilliant solution, agreed, but something that requires brilliance is only 
> usable by the consistently brilliant.  I would like to see a tool that allows 
> users to paste in normal English and stepwise translate it into ACE.  Unless 
> that's known to be too difficult it would also be really interesting (and 
>thesis 
>
> worthy) to try.  

George

Your proposal does not fall into the category "too difficult", but plainly into 
the category "impossible". Normal English and controlled English, like ACE, are 
completely different beasts. 


Let me give you four reasons for my statement.

1. ACE is a first-order logical language that uses part of the English syntax. 
Being first-order means that the language is only on one level. English, 
however, is multi-level. It allows for instance, that a sentence talks about 
itself.

2. ACE does not use any context besides preceding noun phrases to resolve 
anaphors. English cannot be correctly understood without referring to context. 


3. ACE is not ambiguous. Any ACE sentence – that read as English sentence might 
appear ambiguous – has but one parsing. English is highly ambiguous and you need 
context to resolve ambiguity.

4. In ACE words are processed as uninterpreted symbols, while in English words 
carry a meaning.

> Attempto, its authors will agree, is an existance proof that 
> audacious experiments can resolve in the affirmative.

ACE is not a replacement for English but a formal knowledge representation 
language that can be processed by a computer and read by anybody.

Regards.

   --- nef