Discourse Processing and Coreference
Modeling meaning above the sentence: resolving what pronouns and noun phrases refer to, and analyzing how sentences cohere into structured, coherent discourse.
Definition
Discourse processing is the computational analysis of meaning relations that span multiple sentences, including reference resolution and the structure that makes a text coherent.
Scope
Covers discourse-level computational semantics — coreference and anaphora resolution, models of local coherence such as centering, discourse structure theories such as rhetorical structure theory, and discourse relation parsing. It addresses how reference and coherence are tracked across text. Sentence-internal meaning is covered in sibling topics.
Core questions
- How are pronouns and noun phrases linked to their referents?
- What makes a sequence of sentences a coherent discourse?
- How can discourse structure be represented and parsed?
- How do discourse models support summarization and question answering?
Key concepts
- coreference resolution
- anaphora
- centering theory
- discourse coherence
- rhetorical structure theory
- discourse relation
- salience
- discourse parsing
Key theories
- Centering theory
- A model of how attention to discourse entities shifts between utterances, predicting which referents are most salient and thus likely targets of pronouns.
- Rhetorical structure theory
- Analyzing text as a tree of nucleus-satellite relations such as elaboration and contrast, providing a structural account of coherence.
History
Discourse processing matured through theories of coherence and attention in the 1980s and 1990s, with centering theory and rhetorical structure theory offering structured accounts of how texts hang together. Coreference resolution became a standard shared task, and discourse parsing was later advanced by annotated corpora and neural models.
Debates
- Universality of discourse relations
- Whether there is a fixed, theory-neutral inventory of discourse relations or whether relations are framework-specific, a question that complicates annotation and cross-corpus comparison.
Key figures
- Barbara Grosz
- Aravind Joshi
- William Mann
- Sandra Thompson
Related topics
Seminal works
- grosz1995
- mann1988
Frequently asked questions
- What is coreference resolution?
- Coreference resolution is the task of grouping the expressions in a text that refer to the same entity, such as linking 'Marie Curie', 'she', and 'the physicist' to one person, which is essential for understanding connected text.