Mini-discussion prompted by quiz at the end of Mathieu's presentation on (D)MRS comparison: [Self-appointed volunteer scribe: Emily]
From the ERG's semi:
"_about_a_1_rel": ARG0e, ARG1 i, ARG2 h "_about_a_1_rel": ARG0e, ARG1 h "_acceptable_a_for_rel": ARG0 e, ARG1 p, ARG2 i "_acceptable_a_for_rel": ARG0 e, ARG1 e "_acceptable_a_for_rel": ARG0 e, ARG1 h
How many different predicates are these? (It's important for (D)MRS comparison, transfer, conversion from DMRS to MRS)
Ann: That depends on what you mean by predicate. At some point there was a discussion between the more logically oriented and others, where it was pointed out that you can't swap arity and call it the same predicate. But by that time a lot of that happened in the grammar. They have to correspond to logically different predicates in probably all cases. However, in an MRS they are taken to be a naming convention which is convenient for some reason; use the general get-out clause that MRS is a description of a set of semantic objects (not really semantics itself) and therefore you can get away with having only one rel name there.
Probably the right way of doing it is to treat them all as different --- invent a naming convention.
Emily: If you're referring to the same discussion I remember, I recall that this was cast in terms of underspecification, but I don't see it just now.
oe: That was at a sub-group gathering last year (at Hankø). We said that our MRSs (as produced by the ERG, say) can be viewed as descriptions of underspecified representations, i.e. these predicate symbol are underspecified… (but AlexLascarides was unfortunately not present at that meeting).