[Attempto] A Note on Racer

[Norbert E. Fuchs]

On 14 Mar 2010, at 19:35, Adam Wyner wrote:

> I'm working with a set of 16 sentences for a policy discussion on
> garbage recycling.  I've got them all parsed and nicely semantically
> represented with ACE.
> 
> Then, I want to test which sets of sentences are consistent, using
> RACER.
> 
> Some observations.  There are sentences which ACE accepts, but RACER
> does not; in particular, RACER reports an error with sentences that
> include "should" or "say" -- sentential operators.  This makes sense
> since RACER is a first order reasoner, not a modal logical one, which
> would be needed for these sorts of constructions.  However, I didn't see
> documentation about what RACER can/can't do....  I must be missing an
> obvious link.
> 
> Also, just a reminder from our early lessons in FOL, just because
> something is intuitively inconsistent does *not* mean that it is
> logically inconsistent in FOL.  Be attentive to the meanings of the
> given logical forms.  For instance, Every man is happy together with No
> man is happy is reported as consistent.  But, obviously, this is so
> because where there are no men, both are true.  Assert that there is a
> man, and the inconsistency is clear.  While this is a simple example,
> the same issue might easily arise in more complex knowledge bases and be
> hard to root out.


Adam

Thanks for your considerations. Here are my responses.

1. The Attempto reasoner is called RACE, not RACER that is another animal.

2. The version of RACE available on the web (http://attempto.ifi.uzh.ch/race/) is not the latest version, and does not reflect the latest changes of ACE, for example the more compact representation of proper names. Thus the web version may appear buggy.

3. I am currently implementing a new version, and will make it public as soon as possible.

4. You are right, the language restrictions of RACE are not properly documented. I will document them in the near future when the new version of RACE will be made public.

5. Right, modality is not yet covered. There is a standard solution for possibility (can) and necessity (must) that is based on possible world semantics. However, I hesitate implementing this solution since it gives a large overhead for every sentence – even those not involving modality – which negatively affects RACE's performance. Thus I consider alternatives based on axioms like "p -> <> p" and "[] p -> p". These axioms can be implemented without affecting the rest of RACE. However, this approach has also disadvantages because the contents of p are not taken into account, as you may want, for instance, in "John eats a green apple." -> "John can eat an apple.". Possible world semantics would also cover sentence subordination, as in "John sees that Mary sleeps.".

6. Assuming that there will be a solution for possibility and necessity, I do not see a solution for your "should" and "may". Which deductions do you expect?

7. In your last paragraph starting with "Also, just a reminder from our early lessons in FOL, just because something is intuitively inconsistent does *not* mean that it is logically inconsistent in FOL. ... ", I do not see what you are getting at since RACE is behaving correctly in this case. Please elaborate.

Best regards from Palermo.

   --- nef