[developers] Subsumption check

Woodley Packard sweaglesw at sweaglesw.org
Mon Jun 10 19:20:45 CEST 2013

Hi Emily,

To clarify slightly, the hierarchy present in the SEM-I signature is not necessarily an exact duplicate of the hierarchy present in the grammar proper.  The SEM-I signature describes the types on the *righthand* side of the VPM, while the grammar proper describes the types on the *lefthand* side of the VPM.  In many cases these may look the same, but sometimes there are nontrivial differences, e.g. in the ERG the internal definition of the value of the PNG.PN feature is the `pn' type hierarchy, which is defined by 83 lines of relatively inscrutable multiple inheritance TDL describing something like the power set of the six terminal person/number combinations.  By contrast, in the SEM-I signature we just have:

 num := top.
  sg := num.
  pl := num.

  pers := top.
  1 := pers.
  2 := pers.
  3 := pers.

The SEM-I encapsulates all the information that you need to know about the properties used in the MRSes in order to be able to do some simple operations, such as check well-formedness and subsumption relations.  I think having that information packaged up in a relatively simple format could be important for enabling consumption of our MRSes by non-DELPH-IN-ites.  I would estimate that understanding and writing code to process the whole grammar type hierarchy is a much bigger proposition for someone who wants to come play in our sandbox than understanding just the SEM-I signature.  And if we present the SEM-I as the definition of the types that show up in MRSes, then it makes sense to use it for the generator subsumption test too.

It's true that it increases the maintenance chore, and may not be particularly relevant for smaller grammars that don't really care about presenting a self-contained description of their MRSes for external consumption.  For such situations, we either should have an automatic SEM-I signature generation system, or a way for the generator to fall back to using the grammar-internal hierarchy.

And yes, I absolutely agree that this is a great topic for the summit!


On Jun 10, 2013, at 9:41 AM, Emily M. Bender wrote:

> doing the subsumption check
> "outside" --- in the external, 'MRS' universe --- requires that the sem-i
> mirror all subsumption relations between types used in variable properties
> and between underspecified and fully specified PRED values that the
> grammar also has internally.  Is there some other reason that we want
> to be doing this?  (E.g., is there any reason an external MRS-producing
> component would need this info?)

More information about the developers mailing list