Definite Noun Phrases and the Semantics of Discourse

Manfred PinkalCOLING 1986

TL;DR

This paper explores how definite noun phrases (like "the book" or "he") are interpreted in discourse, proposing extensions to Discourse Representation Theory (DRT) that integrate insights from Natural Language Processing (NLP). It argues for a more nuanced understanding of anaphoric reference, particularly the locality constraint on pronouns and the role of focus in demonstratives, offering a more unified account of how we understand language in context.

What Problem Does This Paper Address?

Understanding how humans resolve references in language – figuring out what "he," "she," or "it" refers to – is a fundamental challenge in linguistics and AI. Traditional semantic theories often treated noun phrases as simple truth-conditional elements. However, recent work, including Discourse Representation Theory (DRT), has introduced a dynamic, context-sensitive perspective. DRT, along with similar theories, treats indefinite noun phrases as reference introducers and definite noun phrases as anaphoric expressions.

While DRT offers a unified account for these types of phrases, it faces a gap: the complexity of real-world reference resolution observed in NLP research. This paper aims to bridge this gap by refining DRT's treatment of definite noun phrases, specifically addressing:

  • The anaphoric range of pronouns vs. definite descriptions: Kamp's DRT suggests pronouns have a limited reach, while descriptions have a wider one. The author argues this is reversed in practice.
  • The locality constraint on pronouns: Pronouns seem to prefer recent antecedents, a phenomenon not fully captured by standard DRT.
  • The role of focus in demonstratives: How demonstratives like "this" or "that" direct attention and influence subsequent interpretations.

Methods

The paper builds upon the framework of Discourse Representation Theory (DRT). Key methodological contributions include:

  • Extensions to DRT: The author proposes two main extensions to Kamp's DRT:
  • The Locality Constraint (CU): A "current universe" (CU) is introduced, representing a distinguished subset of the discourse universe. This CU is divided into a current-sentence position (CU₀) and a last-sentence position (CU₁). This formalizes the intuition that pronouns have a limited "backward reach" and must find their antecedents within this local context.
  • Focus Indication (CF): A "current focus" (CF) set is introduced. This set tracks elements that are currently the focus of attention. Demonstratives are proposed to move referents into the CF, while pronouns confirm existing focus elements within the CF. This helps explain how demonstratives direct attention and how pronouns maintain focus.
  • Integration of NLP Insights: The author explicitly integrates concepts and findings from NLP research on anaphora, particularly the work of Grosz, Sidner, and Webber, to inform the theoretical framework.
  • Reinterpretation of Kamp's Concepts: The paper critically re-examines Kamp's notions of salience ranking and selection sets, proposing that a unified salience ranking might be more appropriate for definite descriptions, and that focus plays a more crucial role than Kamp's original model suggested.

Datasets

Not discussed in the paper. The paper focuses on theoretical linguistic analysis and does not present empirical evaluations on specific datasets.

Key Insights and Ideas

  • Pronoun Locality is Crucial: The paper strongly argues that pronouns, especially inanimate ones, are subject to a strict locality constraint. They must find their antecedents in the immediate linguistic context (current or preceding sentence), regardless of overall salience. This is a significant departure from theories that rely solely on salience.
  • Focus as a Driving Force: Demonstratives are not just about pointing; they actively direct the hearer's focus of attention to a new referent. This "focus-shifting" capability is a key function that distinguishes them from definite descriptions.
  • Reversing the Anaphoric Range: Contrary to Kamp's initial proposal, the author posits that definite descriptions have a wider anaphoric range than pronouns. This is because descriptions can carry more information (modifiers) allowing them to "reach further down" the salience scale, while pronouns, being "naked," must latch onto the most salient available antecedent within their local context.
  • Unified Treatment of Descriptions and Demonstratives: The paper suggests that both definite descriptions and demonstratives can be analyzed using a salience ranking mechanism, but with crucial differences in how they interact with focus and context.
  • Thematic Roles Matter for Pronouns: The paper hints that the thematic role of a referent (e.g., subject vs. object) can influence pronoun resolution, suggesting that parallelism plays a role in anaphoric relations.

Key Results

The paper does not present quantitative results. Its contributions are theoretical and conceptual:

  • Formalization of Pronoun Locality: The introduction of the "current universe" (CU) provides a formal mechanism within DRT to model the restricted anaphoric reach of pronouns.
  • Formalization of Focus: The "current focus" (CF) set offers a way to model the role of focus in directing attention and influencing reference resolution, particularly for demonstratives.
  • A Revised Taxonomy of Definite NPs: The paper proposes a more nuanced understanding of the anaphoric behavior of different types of definite noun phrases (pronouns, descriptions, demonstratives), highlighting their distinct constraints and functions.

Limitations and Future Work

  • Limitations of the Locality Constraint: The author acknowledges that the "current universe" (CU) is a "crude approximation" and that defining the "local linguistic context" more precisely remains an open question. Counter-examples to strict locality are mentioned, particularly for "it" referring to the topic of primary interest in task-oriented dialogues, suggesting a need for further refinement.
  • Complexity of Focus: The paper notes that the "current focus" (CF) concept is a first attempt and that a more elaborate syntactic foundation for discourse semantics might be required to fully capture the nuances of anaphoric text structure, including parallelism and multiple chains of anaphoric relations.
  • Unanswered Questions: The paper raises several open questions, including how the salience ranking extends to sub-DRSes, how additional reference objects are established, and the precise mechanism for constructing and modifying salience rankings.
  • Future Work: The author suggests further investigation into the "outer boundary" and "internal structure" of the current universe, as well as a deeper exploration of the relationship between focus, thematic roles, and anaphoric resolution. The paper also points to the need for a more elaborate syntactic foundation for discourse semantics.

Summary updated

BibTeX
@inproceedings{pinkal1986definite,
  author = {Manfred Pinkal},
  title = {{Definite Noun Phrases and the Semantics of Discourse}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1986},
}
Figure