Closed arademaker closed 2 years ago
The compound
relation behaves like an underspecified preposition (see https://github.com/delph-in/docs/wiki/ErgSemantics_Compounding ), and so the labels are different.
But if udef_q
is interpreted as an existential, then it doesn't really matter in terms of evaluating the MRS -- an existential quantifier is classically interpreted as ∃x R(x)∧B(x). With that interpretation, your alternative suggestion is not the compositionally derived MRS, but it's logically equivalent.
With that interpretation, your alternative suggestion is not the compositionally derived MRS
Why? Indeed the generalized existential when converted to the FOL existential will make a conjunction from its RTSR and BODY. But if we have and universal quantifier here the split of the compound would matter. I will read the wiki page, but the splitting of a compound in different label is still confusing go me.
Consider prepositions, e.g. "Every box on some table is blue". With explicit quantifiers, we can see that the two nouns have different labels. The preposition takes the nouns as arguments, and shares a label with one of them.
The ERG analyses compounds so that compound
and _on_p
have the same kind of structure, e.g. "boxes on tables are blue", "boxes with tables are blue", "boxes for tables are blue", "table boxes are blue".
One analysis that LKB gives me:
The indexed MRS for this analysis:
One scoped MRS
Does it make sense to have a compound split in two different labels (e.g.
h21
andh14
)? Theudef_q
quantifier restriction shouldn't be_traffic_n_1(x16) /\ compound(e17, x10, x16) /\ _lane_n_1(x10))
?