Dialogue Acts, Synchronizing Units, and Anaphora Resolution
What it is about
Not read: this card is written from what is reliably known about the paper, and every idea is marked to check against the text. It is the journal version of the EACL 1999 paper, which has its own card (Resolving Discourse Deictic Anaphora in Dialogues). The authors resolve pronouns and demonstratives in two-party spoken dialogue (Switchboard). Dialogue acts are grouped into initiations and acknowledgements, and an initiation with its acknowledgement forms a synchronizing unit: only acknowledged material enters the common ground and becomes available as an antecedent. Anaphors are classified as pointing to individuals, to abstract objects (facts, events, propositions: discourse-deictic reference) or as vague, using the predicate context. Individuals are resolved with a salience list in the style of Strube's S-list, and abstract objects with a separate list of recent clauses and sentences. The journal version adds more detail on the classification and an annotation reliability study.
Ideas
- I1. Inter-annotator agreement before rules: have two independent teacher passes label anaphor type (individual, abstract, vague) on the same dialogue sentences, and report kappa per type; a type whose agreement is low gets no hard rule in
coref. (to check: the kappa values the paper reports) - I2. Predicate-context incompatibility lists: predicates that only take individuals and predicates that only take abstract objects decide the anaphor type, with the list entry named in the explanation. Grow the lists from our own lexicon and measure their coverage on dialogue lessons. (to check: the paper's list of contexts; the 1999 card's I4 covers the same idea for the short version)
- I3. Vague anaphors as an explicit outcome:
corefmay answer "no antecedent, vague" instead of forcing a link, and the eval counts these separately; measure whether precision rises without a large recall loss. (to check: the share of vague cases in the paper's data)
For Mova
The chapters are 35. Reference and coreference and 38. Dialogue, discourse markers and reported speech. The code is coref/src/sieve.rs, coref/src/mention.rs and the speech-act labeller in prag. Switchboard is LDC-licensed: tests only.
Cite
@article{eckert2000dialogue,
title = {{Dialogue Acts, Synchronizing Units, and Anaphora Resolution}},
author = {Eckert, Miriam and Strube, Michael},
journal = {Journal of Semantics 17(1):51–89},
year = {2000},
doi = {10.1093/jos/17.1.51},
url = {https://doi.org/10.1093/jos/17.1.51}
}This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/enprag-2000-eckert-dialogue-acts-synchronizing-units-anaphora/