Am I right that this is suitable only for case marking adpositions and not for adpositions of the kind in [the forest]:
basic-adposition-lex := single-rel-lex-item & basic-icons-lex-item &
[ SYNSEM [ LOCAL [ CAT [ HEAD adp,
VAL.COMPS < [ LOCAL [ CAT cat-sat,
CONT.HOOK [ INDEX #ind,
ICONS-KEY.IARG1 #clause ] ] ] > ],
CONT.HOOK.CLAUSE-KEY #clause ],
LKEYS.KEYREL arg12-ev-relation & [ ARG2 #ind ] ] ].
This is for contentful adpositions, not the case marking ones. single-rel-lex-item gives a non-empty (length 1) RELS list.
What does icons-key .IARG1 means? Why is it a clause for the complement which is presumably a noun?
That has to do with Sanghoun’s analysis of information structure, and I don’t know it off the top of my head, but roughly: information structure is analyzed as binary ICONS relations between individuals (x or e) and the clause they are a part of, as represented by the main event variable of that clause.
So, this is for “information structure-marking adpositions”, then. For “just adpositions”, do I want a yet another supertype which doesn’t say anything about ICONS? (I mostly need them to model things like in which city did you stay, and incidentally in which city will be focused however this probably isn’t what is meant by an “information structure-marking” adposition).
Here’s an MRS for Ivan lies in the book (sorry, that’s just the vocab I currently have ):
I think it matches what I get from the ERG except the name of the relation:
In the ERG, it is just in_p, without rel even. What should I do for the Grammar Matrix?
I’m pretty sure this is just an effect of the display — if you look in ERG’s lexicon.tdl, you should see _in_p_rel. That particular display strips “rel” everywhere.
In fact the rel-less predicates are now the expected form for all grammar-external MRSs, not just this display. The
_rel suffix is just for establishing a namespace in the grammar so predicate types don’t clash with other grammar types. Even the raw MRS output of ACE strips the _rel suffixes. It’s possible that some display modes of the LKB have not been updated to reflect this.