Skip to content

ErgSemantics_OneAnaphora

EmilyBender edited this page May 14, 2014 · 5 revisions

ESD Test Suite Examples

Browne's barks.
The dog chased that one.
Three bark.
That's enough.

Linguistic Characterization

This phenomenon involves anaphoric noun phrases which either don't include a pro-form at all (Browne's, Three) or have the pro-form one, which is unusual among English pronouns in that it can take dependents. All of these expressions take their interpretation from the (linguistic or non-linguistic) context. Though one anaphora in particular are famous in the syntactic literature because Baker (1978) claimed that the antecedent could only be an N' (not a simple N), in fact Payne et al (2013) convincingly argue that the constraints on the possible antecedent of one in this use are semantic rather than syntactic. In particular, they show that the antecedent can be any pragmatically plausible expression of (in their terms) type <e,t>. In MRS terms, this means that the antecedent should be a subgraph rooted in a predication denoting an entity. Where Payne et al (2013) focus specifically on expressions involving the form one (e.g. The dog chased that one.), Nerbonne et al (1989) group these together with ‘null-head anaphora’ (e.g. Browne's barked) under the term ‘N-bar anaphora’. The ERG's semantic analysis further identifies expressions such as Three in Three bark and Those as in Those disappeared as members of this class. While at first glance the demonstratives in particular might seem like lexical NPs, examples like Those from the park disappeared motivate an analysis where the NP is built syntactically.

Motivating Examples

  • Sandy has a blue glass, but we couldn't find another one for Kim.

  • Sandy always likes the pictures of Kim better than the ones of Pat.

  • [Looking at a display of chocolates:] I want one!

Note that one of the testsuite examples actually involves two instances of this phenomenon:

  • That's enough.

In this example, that and enough are both associated with (separate) instances of generic_entity.

MRS Fingerprints

This phenomenon is characterized by the appearance of generic_entity in the MRS. This EP is meant to serve as a cue to further processing components that there is anaphora resolution to be carried out.

generic_entity(ARG0 x1)

Interactions

Reflections

  • One and null head anaphora are to be distinguished from partitive expressions (e.g. Some of the books arrived.), the latter being characterized by part_of predications.

Open Questions

  • In the current grammar, one introduces card analogously to other numeral adjectives. In examples like the following, this seems well-motivated:

    • Sandy saw three dogs, but Kim only saw one.

However, in other uses, it seems more surprising to see the card predication:

  • The dog chased the red one.

  • The dog chased the red ones.

  • I have the one I found yesterday.

Furthermore, the information provided by this predication (CARG 1 in the case of one and CARG 2+ in the case of ones) is redundant to the number information (sg v. pl) provided by the variable property NUM. It seems probable that there is actually a cline in usage here, where some examples clearly pattern with the other numeral adjectives, some clearly don't, and others are more muddled. One approach would be to ambiguate, creating two lexical entries for one, both of which would be able to participate in the analyses of the items above. (Presumably ones would be unambiguously non-card-introducing.) However, in the absence of any syntactic cues that could rule out one or the other of these ones in a given context, according to the design principle of minimizing unresolvable ambiguity, we opt to choose only one analysis to stand in for both. Given (a) the availability of the information that would be introduced by card (through CARG) in the NUM attribute and (b) the observation that an external source of MRSs for generation should not need to know to provide card in these cases, a likely future direction for the grammar is to drop card from one (except where it participates in larger number names, like one thousand one hundred and twenty one).

  • The example That's enough has two generic_entity EPs in it, one corresponding to that and one to enough. If generic_entity is supposed to be an instruction to find a relevant (linguistic?) antecedent, we are claiming that these instructions are the same for demonstrative pronouns as for one anaphora and partitives, modulo what we can get from the quantifier EP. On the surface, this claim is a bit surprising to me, but I'm not sure what kind of evidence could be used to support or refute it.

  • ‘subgraph rooted in a predication denoting an entity’ --- do we not have a more streamlined term for that yet? Also, I couldn't find ‘entity’ anywhere on the Basics page. Is it meant to be there?

  • ‘subgraph rooted in a predication denoting an entity’ --- do we expect these to always correspond to <e,t>?

Expert External Commentary

Grammar Version

1212

References

  • Baker, C. L. (1978). Introduction to generative-transformational syntax. Englewood Cliffs, NJ: Prentice-Hall.

  • Payne, J. & Pullum, G. K. & Scholz, B. C. & Berlage, E.(2013). Anaphoric one and its implications. Language 89(4), 794-829. Linguistic Society of America. Retrieved May 12, 2014, from Project MUSE database.

  • Nerbonne, J., Iida, M., & Ladusaw, W. (1989). Running on empty: Null heads in head-driven grammar. In Proceedings of the Eight West Coast Conference on Formal Linguistics. Stanford: CSLI.

More Information