Discourse Referents

Lauri KarttunenCOLING 1969

TL;DR

This paper explores how language manages to refer back to previously mentioned entities, a phenomenon known as "discourse referents." It argues that existing linguistic theories, which rely on simple "referential indices," are insufficient. Instead, it proposes a more abstract analysis where noun phrases are treated as variables bound by quantifiers, allowing for a more robust explanation of how discourse referents are established.

What Problem Does This Paper Address?

When we speak or write, we constantly refer back to things we've already introduced. For instance, after saying "I have a car," we can then say "It is black" or "The car is black." But this isn't always possible. If we say "I don't have a car," we can't then say "It is black" because "it" doesn't refer to any car introduced in the discourse. This paper tackles the fundamental question of what makes an entity in a sentence available to be referred back to later in a conversation or text. Linguists have grappled with this, and the paper critiques existing models that assign simple "referential indices" to noun phrases, finding them inadequate to capture the nuances of how "discourse referents" are established and maintained.

Methods

The paper critiques the existing linguistic framework, particularly Noam Chomsky's "Aspects of the Theory of Syntax," which associates noun phrases with "referential indices" and a "specific" feature. It argues that this approach fails to adequately explain the problem of discourse referents. Instead, it advocates for a more abstract analysis, drawing on work by Emmon Bach and James D. McCawley. This proposed framework treats noun phrases as underlying variables that are bound by quantifiers, similar to how variables are handled in predicate calculus. This shift allows for a more principled explanation of when an indefinite noun phrase successfully establishes a discourse referent.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Beyond Simple Indices: The paper's core insight is that simply marking a noun phrase with a "referential index" isn't enough to explain its ability to become a discourse referent. The relationship is more complex.
  • Quantifier Binding is Key: The proposed framework, which views noun phrases as variables bound by quantifiers, offers a more powerful way to model this phenomenon. It suggests that the "establishment" of a discourse referent is tied to whether the proposition containing the noun phrase is asserted or presupposed by the speaker.
  • Negation is Tricky: The examples clearly illustrate how negation prevents the establishment of discourse referents. If you don't have a car, you can't refer to "it" later.
  • Ambiguity and Context: The paper highlights that even seemingly straightforward sentences can be ambiguous, and the ability to establish discourse referents depends heavily on the context and the speaker's intent.
  • Special Cases Exist: The authors acknowledge that the proposed framework, while powerful, will need to account for special cases like counterfactual conditionals, modals, and certain adverbs (e.g., "always," "often") that can influence discourse referent establishment.

Key Results

The paper doesn't present quantitative results in the traditional sense, as it's primarily a theoretical argument. However, its key "result" is the proposal of a new theoretical framework for understanding discourse referents. This framework, based on quantifier binding, is presented as a more adequate explanation than existing models that rely on referential indices and specificity features. The paper demonstrates this by showing how its proposed analysis can account for the examples discussed, particularly the contrast between sentences that establish discourse referents and those that do not.

Limitations and Future Work

The authors acknowledge that their proposed framework needs to be extended to handle specific linguistic phenomena. They explicitly mention the need to account for:

  • Counterfactual conditionals: Situations that describe what would have happened if something else were true.
  • Modals: Words that express possibility, necessity, or obligation (e.g., "might," "must").
  • Quantifier-like adverbs: Adverbs that indicate frequency or scope, such as "always" and "often."

These are identified as areas requiring further investigation and refinement of the proposed theory.

Summary updated

BibTeX
@inproceedings{karttunen1969discourse,
  author = {Lauri Karttunen},
  title = {{Discourse Referents}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1969},
}

Loading related content...

Figure