On the Relative Expressiveness of Description Logics and Predicate Logics
Borgida
first order logic description complexity decidability two-variable
@article{borgida:aij-1996,
title={On the Relative Expressiveness of Description Logics and Predicate Logics},
author={Borgida, A.},
journal={Artificial Intelligence},
volume={82},
number={1-2},
pages={353--367},
year={1996}
}
Even for more expressive DLs, only encompasses fragment of first order logic expressible in three variables
- Providing a clear boundary on the limits of their expressiveness---they cannot express all things
Example of non-atomic role: grandsons \equiv compose[children, restrict[children, MALE]]
Core expressiveness results:
- The description logic language DL consists of top, nothing, and, or, not, all, some, at-least, at-most, at-least-c, at-most-c, same-as, subset, not-same-as, fills, one-of, top-role, identity, role-and, role-or, role-not, inverse, restrict, compose, product, trans
- The description logic language DL-{trans, compos,; at-least, at-most} and L^2 are equivalent
- The description logic language DL-{trans, at-least, at-most,} and L^3 are equivalent
Corollaries
- Cannot express 4-clique
- compose cannot be constructed from other language elements
- Subsumption is decidable for DL without trans, compos, at-least, at-most
- Undecidable for DL without trans, at-least, at-most
- DL-{trans,compose} is equivalent to L^2_{CNT}
- Which is decidable, correct?
- DL-{trans} is equivalent to L^3_{CNT}
Alternate proof path: ALC is equivalent to a modal logic, which in turn can be expressed in L2, which can in turn be expressed in a modal logic
- These are not necessarily the same modal logics
There are applications requiring terminological reasoning which DL constructors simply cannot represent
DLs are likely insufficient as query languages over a database, etc. since they have limited expressiveness
- Even the least powerful query languages are more expressive and decidability
- Due to operating over finite model?