[Attempto] RACE thinks if Chicago and Detroit are cities, then every city is Chicago
Norbert E. Fuchs
fuchs at ifi.uzh.ch
Sat Nov 16 10:43:27 CET 2019
> On 16 Nov 2019, at 24:14 , Tanner Swett <tannerswett at gmail.com> wrote:
>
> Hi everyone,
>
> I've noticed that RACE can make some questionable inferences given axioms containing proper names.
>
> I tried the following using the RACE client at http://attempto.ifi.uzh.ch/race/:
>
> Input:
> Axioms: Chicago is a city. Detroit is a city.
> Theorems: Every city is Chicago.
>
> Output: True.
>
> Both axioms need to be present in order for RACE to conclude that the theorems follow; "Chicago is a city." alone is not enough for RACE to conclude that every city is Chicago. The axioms "Chicago is a city. Chicago is a city." also don't suffice, but if I give it "Chicago is a city. A city is Chicago.", that *is* enough for it to conclude that every city is Chicago.
>
> A similar problem occurs if I mention multiple different types of objects:
>
> Input:
> Axioms: Chicago is a city. France is a country
> Theorems: France is a city.
>
> Output: True.
>
> All of these counterintuitive theorems are said to be true because of "Prolog Axiom transcompadj2: Transitivity of comparative adjectives."
>
> Am I misunderstanding what proper names mean to RACE, or is this a bug?
>
> Tanner Swett
> _______________________________________________
> attempto mailing list
> attempto at lists.ifi.uzh.ch
> https://lists.ifi.uzh.ch/listinfo/attempto
Dear Tanner Swett
Thank you for testing RACE.
You are right, this is a bug in the auxiliary axiom transcompadj2 that I recently introduced. I am working on this and will let you know once I will have removed the bug.
Best regards.
Norbert E. Fuchs
Institute of Computational Linguistics
University of Zurich
More information about the attempto
mailing list