SaarlandLtopDiscussion

Discussion on constraining LTOP on NPs in MRSs

Leader: Emily

Participants: Joshua, Stephan, Rebecca, Lars, Dan

Potential examples where we might want LTOP available after composing an NP:

• English appositives, if we want the appos_rel's LBL to be identified with the head rel of the left NP.
• Turkish N-Det-Adj
• Yiddish Intens-Det-Adj-N

Obstacles to making the LTOP available: Semantic algebra prohibits binding LTOP of a quantified NP, since the LTOP of an argument should either be shared with that of the head, or be a scopal argument of the head. This sem-alg constraint is useful if we dispense with quantifiers for pronouns, since the pronoun's non-q relation's LTOP needs to be identified with that of its functor. Similarly for our treatment of apposition in the 1212 ERG.

Options:

• (A) Add a unary rule that empties the LTOP of a quantified NP (which starts out with the LTOP bound to the head noun's LBL, to sustain further intersective modification), and liberates it to be an argument. Currently implemented in the ERG to enable e.g. |everybody in Paris who dances|, which would be problematic since |everybody|. (B) Give up the semantic algebra constraint (abandon strong hypothesis), and allow LTOP of NP to be identified with the head noun's LBL. Potentially problematic if we give up quantifiers for pronouns, as has been contemplated for awhile for the ERG. We could go further and make pronouns introduce no rels, and then coerce them to add a nominal rel for |you in the corner|. But without a quantifier, we would have the LBL of the preposition identified with that of the head verb in |you in the corner sing|. Not appealing, since it would be a first instance of an intersective modifier whose ARG1's source relation does not share its label. (C) Add another attribute in HOOK for NPs to keep track of the head noun's LBL. Weakest, and potentially problematic downstream.