ESD Test Suite Examples
Did you? There were. Abrams tried to.
English allows the complement of an auxiliary or of the infinitival marker to to be omitted, if it can be reconstructed from context. Sometimes called VP ellipsis, this construction is better identified as post-auxiliary ellipsis, since the missing constituent need not be VP if the auxiliary in question can take a non-VP complement (e.g. Who is at the door?/Kim is.). The interpretation of elided constituents is an interesting and complex area of English grammar, especially as it appears to be constrained by the linguistic context, though that can be from a prior sentence, including one uttered by a different speaker. [FIXME: Citation Needed]
A sample context that would support each of the test suite examples is given below:
I saw the movie. Did you? I thought there weren't any dogs, but there were. Browne didn't want to go, but Abrams tried to.
From the point of view of the sentence-by-sentence processing done by the ERG, the interpretation is not given by the sentence itself. Accordingly, the elided element(s) are represented via one of two EPs. These characteristic EPs can be used to trigger the search for an appropriate antecedent, which will typically be not just a single EP but rather a subgraph of some MRS from some utterance in context.
The first, used when the subject is referential (non-expletive) has an ARG1 role.
This serves two purposes: First, it reflects the fact that the subject in such cases will be interpreted as playing some role within the reconstructed subgraph after the antecedent expression has been found. Second, it ensures that the MRS produced by the grammar in cases of ellipsis with contentful subjects will be connected graphs. Without that ARG1 position, the semantic contribution of the subject (of the head licensing the ellipsis) would not be integrated.
The second characteristic EP appears when the subject is an expletive:
This EP also has an ARG1 role, but it is not linked to anything.
When the auxiliary verb licensing the ellipsis is negated, as in They didn't., the ellipsis_ref or ellipsis_expl is represented as the argument of neg (via a qeq constraint).
The grammar provides two analyses for Were there?, one with two predications (_be_v_there and ellipsis_expl, and one with just ellipsis_expl). On first glance, it might seem that the analysis with _be_v_there is the only plausible one, since something must be licensing the expletive there. However, the second one, with just ellipsis_expl is motivated by examples like I thought there were going to be dogs, but there weren't. where one possible resolution of the ellided constituent is (the semantic representation of) going to be dogs. On that reading, the _be_v_there predication is provided by the resolution.
Note that the predicative copula and then identity copula are not distinguished in ellipsis. The claim is that ellipsis_ref could resolve to either antecedent.
I'm hungry. Are you?
Who is Macbeth? Kim is.
- ARG1 is meant as an underspecification over several possible roles (at least ARG1, ARG2, ARG3; others?). Should we choose here a different symbol for that underspecification (ARGn, ARG)? [Decision on 4/15/14: go with ARG.]
Abrams wasn't hired, but Browne was.
Abrams wasn't given an assignment, but Browne was.
I can't get ellipsis_expl to show up with it, even in the example from the testsuite (1212). [Identified as grammar bug.]
Why would ellipsis_expl need an ARG1? [No. To fix in the grammar.]
Possible case of conflicting desire for symmetry: VP ellipsis and NP-taking auxiliaries (identity copula, there copula, British possessive have).
External Expert Commentary
[To be solicited]
Dalrymple, M., Shieber, S. M., & Pereira, F. C. (1991). Ellipsis and higher-order unification. Linguistics and philosophy, 14(4), 399-452.