Text Coherence

Text coherence includes what some may demarcate as:

• Appropriate use of coherence relations between elements of the discourse — called rhetorical structure.
• Appropriate sequencing of subparts of the discourse — discourse structure.
• Appropriate use of referring expressions.

Rhetorical structure

One theory of discourse structure, based on identifying relations between parts of the text, is Rhetorical Structure Theory [Mann, Matthiessen, and Thompson 89].

Aspects:

• How many and what types of rhetorical relations are there?
• [MMT] identify 23.
• Concept of nucleus/satellite encodes asymmetry in the relationships.
• Representative rhetorical (coherence) relations

  Relation	descriptive conjunction	Logical relevance
  cause-effect	because	satellite provides evidence for nucleus
  violated expectation	although; but
  condition	if?then; as long as	satellite provides precondition for nucleus
  similarity	(and) similarly	multinuclear or satellites viewed as elements of a class
  contrast	but; however	multinuclear relation
  elaboration	also, furthermore	specialize [class to instance] or generalize [instance to class]
  attribution	?said, according to?
  temporal sequence	before; afterwards	multinuclear relation

Some Problems with RST [Moore & Pollack 92]

• How many and what types of rhetorical relations really exist/are required?
• Extension of RST to dialogue from its use in monologue.
• Incorporation of speaker intentions into RST.
• How to model multiple relations between parts of a discourse?
• RST does not model overall structure of the discourse.

Identifying RS Automatically [Marcu 99]

• Train a parser on a discourse treebank.
  Example treebank characteristics

  - 90 RS trees, hand-annotated for rhetorical relations.

  - Elementary discourse units (edu's) linked by RR.

  - Parser learns to identify Nucleus and Satellites and their RR.

• Requires discourse segmenter to id edu's.

  - Trained to segment on hand-labeled corpus.

  - Features: 5-word POS window, presence of discourse markers, punctuation, occurrence of a verb.

  - 96-8% accuracy

Discourse Structure

A leading alternative theory of discourse structure is Structures of Discourse Structure [Grosz & Sidner 86].

• Provides for multiple levels of analysis:

  - Captures speaker's purpose as well as content of utterances

  - Captures speaker's and hearer's attentional states.

  - Identifies only a few, general relations that hold among intentions.

• Three components:

  - Linguistic structure

  - Intentional structure

  - Attentional structure

Linguistic Structure

• What is actually said/written
• Representation

  - discourse is segmented into Discourse Segments (DS)

  - determine segmentation agreement

Intentional Structure

• Discourse purpose (DP): basic purpose of the discourse
• Discourse segment purposes (DSPs): how this segment contributes to the overall DP
• Example intersegment relations:

  - Satisfaction-precedence: E.g., DSP₁ must be satisfied before DSP₂

  - Dominance: DSP₁ dominates DSP₂ if fulfilling DSP₂ constitutes part of fulfilling DSP₁

Attentional State

• Focus stack:

  - Stack of focus spaces, each containing objects, properties and relations salient during each DS, plus the DSP

  - State changes modeled by transition rules controlling the addition/deletion of focus spaces

  -- Information at lower levels may or may not be available at higher levels

  -- Focus spaces are pushed onto the stack when a new DS or an embedded DS (e.g. DS that are dominated by other DS) is begun; popped when they are completed