On Context Dependence In Modal Constructions

Anette Frank, Hans KampProceedings From Semantics and Linguistic Theory 1997

Abstract: This paper investigates a new representation format for dynamic discourse in Discourse Representation Theory (DRT), where contextual dynamics is modeled in terms of update conditions. This new representation format is motivated by the study of context dependence in modal constructions, in particular by serious problems besetting earlier approaches to modality and modal subordination in DRT. We present an alternative DRT analysis that provides a unified analysis of relative modality and modal subordination, and which accounts for a wider range of data as regards modal subordination relative to negation and graded modal contexts.

Introduction

In natural language, the meaning of a sentence often depends on the context established by previous discourse. A classic challenge in formal semantics is modal subordination. Consider the sequence:

  1. A thief might break in.
  2. He would take the silver.

In sentence (2), the pronoun "he" refers to the thief, and the modal "would" is understood as being conditional on the hypothetical scenario introduced in sentence (1). Standard semantic theories often struggle to explain why the "thief"—introduced in a hypothetical "might" scenario—remains accessible to the second sentence, while in other contexts, hypothetical entities disappear.

The Problem with Traditional DRT

Previous attempts to model this phenomenon (notably by Roberts and Geurts) faced several representational gaps:

  • Representational Gap: Modality rules were often handled in the verification conditions (background logic) rather than within the Discourse Representation Structure (DRS) itself.
  • Negation: Existing models could not easily explain how speakers refer back to entities in negated contexts (e.g., "I don't have a microwave. I wouldn't know where to put it.").
  • Graded Modality: Theories struggled with graded operators like probably or unlikely, which provide a range of likelihood rather than a binary truth value.

Technical Innovation: Annotated DRT (ADRS)

The authors propose Annotated Discourse Representation Theory (ADRS). The core innovation is the introduction of Update Conditions, represented as $G :: F + K$. In this notation, context is treated as a first-class citizen within the logic.

Schematic representation of update conditions in ADRS, showing how context referents (G, H) are derived from previous states and new information (K).

Key features of ADRS include:

  • Context Referents: The system tracks context referents ($F, G, H$) alongside individuals.
  • Explicit Dynamics: An update condition $G :: F + K$ states that context $G$ results from updating input context $F$ with new information $K$.
  • Anaphoric Binding: Following Kratzer’s theory, modal operators are treated as relative to a "modal base." In ADRS, this base is simply a context referent that the modal verb points to anaphorically.
  • Wellfoundedness: To prevent logical paradoxes, the authors define a strict partial ordering on context referents, ensuring the discourse history remains non-circular.

Key Insights and Results

By making the update an explicit part of the representation, the authors move discourse dynamics from the meta-theory into the syntax itself. This allows for a more robust handling of complex scenarios:

  1. Negation as a Context Creator: Negation is argued to introduce a new counterfactual context referent. This explains why we can discuss the properties of a non-existent object in subsequent sentences—we are pointing to that counterfactual context.
  2. Unification: Modal subordination is revealed to be a special case of relative modality. Whether a modal refers to legal requirements, epistemic knowledge, or a previous hypothetical, the mechanism remains anaphoric binding to a context referent.
  3. Graded Modality: Using a "normalcy selection function," the authors show how probably restricts the set of worlds to the "most normal" ones, providing a more nuanced logic than simple possibility.
A complete ADRS representation for the 'thief' example, illustrating how nested update conditions and context referents (F, G, H) track the hypothetical scenario across multiple sentences.

Conclusion

The ADRS format successfully distinguishes between different types of modality at the representational level. While the explicit tracking of context updates adds complexity to the DRS, it provides the first robust DRT account for anaphora out of negated contexts and a unified framework for modal subordination. Future work remains to be done on the pragmatic constraints of "accommodation," where listeners must mentally fill in missing links in the discourse chain.

Summary updated

Abstract

This paper investigates a new representation format for dynamic discourse in DRT, where contextual dynamics is modeled in terms of update conditions. This new representation format is motivated by the study of context depen­dence in modal constructions, in particular by serious problems besetting ear­lier approaches to modality and modal subordination in DRT. We present an alternative DRT analysis that provides a unified analysis of relative modality and modal subordination, and which accounts for a wider range of data as regards modal subordination relative to negation and graded modal contexts.
BibTeX
@inproceedings{frank1997context,
  author = {Anette Frank and Hans Kamp},
  title = {{On Context Dependence In Modal Constructions}},
  booktitle = {Proceedings From Semantics and Linguistic Theory},
  year = {1997},
}
Figure