Documentation

Linglib.Core.Discourse.CoherenceRelation

Discourse Coherence Relations #

@cite{hobbs-1979} @cite{kehler-2002} @cite{umbach-2004}

Coherence relations classify how adjacent discourse segments connect. Each relation belongs to one of three classes (resemblance, cause–effect, contiguity) and has a directionality that determines which segment provides the cause/explanation.

Key insight for IC bias #

@cite{kehler-2002} argues that coherence relations determine which participant listeners seek as a cause/explanation in sentence continuations:

This interacts with verb semantics to produce implicit causality (IC) bias. @cite{solstad-bott-2022} @cite{solstad-bott-2024}

Contrast vs Correction #

@cite{umbach-2004} argues that both CONTRAST and CORRECTION are resemblance relations requiring alternatives that are similar (common integrator) and dissimilar (semantically independent). They differ in their type of exclusion:

Both are distinct from PARALLEL/SEQUENCE ("and"-coordination), which requires similarity+dissimilarity but no exclusion.

Kehler's three coherence classes, corresponding to Hume's three associative connections between ideas.

Instances For
    Equations
    • One or more equations did not get rendered due to their size.
    Instances For

      Individual discourse coherence relations. Each relation specifies how the current segment connects to the prior one.

      @cite{umbach-2004} §3: CONTRAST and CORRECTION are distinct resemblance relations that both require similarity+dissimilarity in their alternatives but differ in their exclusion type.

      Instances For
        Equations
        • One or more equations did not get rendered due to their size.
        Instances For

          Causal direction: does the relation seek a cause in the prior segment?

          Instances For
            Equations
            • One or more equations did not get rendered due to their size.
            Instances For

              German/English connective forms used as experimental stimuli (@cite{solstad-bott-2022}, Exps 1–4).

              Instances For
                Equations
                • One or more equations did not get rendered due to their size.
                Instances For

                  "and so" selects for effects (forward causal / I-Cons).

                  CONTRAST and CORRECTION are both resemblance relations (@cite{umbach-2004} §3.2, @cite{kehler-2002}).

                  CONTRAST and CORRECTION are distinct relations despite sharing a class. @cite{umbach-2004} §3.2: they differ in exclusion type (additional vs substitution) and in the implicit question they respond to. German lexicalizes the difference: aber (contrast) vs sondern (correction).