FeforDroppedArguments

Notes from the discussion on dropped arguments: To include or not include pronoun rels? Is it sufficient to constrain the u variable with e.g., png or definiteness information?

[http://faculty.washington.edu/ebender/prodrop.pdf Emily's slides (pdf)]

The 'eat' example can be misleading, since then we have to worry about whether there's two separate relations. 'earlier' as in "The earlier train" vs. "*The before train" is a clearer case. (Both are relation, take a 'than' complement, but with earlier it's optional.)

"Move!" shouldn't be able to mean "Nobody move!"

Current assumption is that the arity of a given predicate is fixed and known, so leaving off arg roles is not a legitimate means of underspecification.

This leaves:

[ _eat_v_1_rel

• ARG1 x ARG2 u ]

or

[ _eat_v_2_rel

• ARG1 x ]

But cf 'earlier' examples where it seems less plausibile to posit two separate relations.

What about dropped handle arguments? ("I persuaded Sandy") If that's just a u, but MRS wellformedness requires a handle to fill that hole, what to do? Possibility of u_ind and u_hand.

If we underspecify the dropped argument completely (leave off its arg_rel), then we potentially lose information coming from the morphology of the verb (e.g., png) or its lexical properties (definite/indefinite null instantiation).

Slippery slope: What about cognate objects? We have "The dog barked an enormous bark" "sleep a long sleep" "die a terrible death". Surely wedon't want to posit ARG positions for the cognate argument that are underspecified everywhere the verb is used without them. Similarly, what about denominal verbs ("to sunscreen someone") --- should the representation have an ARG role for the sunscreen?

Spent some time looking for cases where the verb or other constraints provides information about u that is 'genuinely semantic':

Bought a book, did he? Cut well, don't they? (scissors) Cut well, doesn't it? (scissors)

... relationship to other kinds of anaphora that seem sensitive to grammatical properties of other words for their referents, even if the other words haven't been used in the local discourse context. Also French "le machin/la machine" --- translations of 'thingamajig' with grammatical gender, often used to match the grammatical gender of words that the speaker can't retrieve.

It's clear from languages which show agreement for gender that we need a featural representation of gender to handle that agreement (e.g., French "Je l'ai cuite" where the past participle is agreeing with the preverbal direct object, in fact in this case, being the only exponence of that gender). It seems to be an open issue to what extent gender needs to be in the MRS (somewhat, for pronoun resolution, at least), and whether it should be represented in addition as a syntactic feature.