Centering in Dynamic Semantics

Daniel HardtCOLING 1996

TL;DR

This paper proposes a novel integration of Centering Theory into a Dynamic Semantics framework to provide a unified account of "sloppy identity" phenomena, like VP ellipsis and paycheck pronouns. By treating the discourse center as a dynamic entity that can shift, the authors elegantly explain how pronouns and elided phrases can refer to different antecedents in context without complex recovery mechanisms.

What Problem Does This Paper Address?

Traditional semantic theories often struggle to explain how pronouns and elided phrases (like in VP ellipsis) can refer to different entities or meanings than their explicit antecedents. This phenomenon, known as "sloppy identity," presents a puzzle: if meaning is compositional and fixed, how can these anaphoric expressions seem to "change" their reference? Existing approaches often resort to complex, ad-hoc mechanisms to resolve these ambiguities. The paper aims to bridge the gap between Dynamic Semantics, which models discourse as a growing context, and Centering Theory, which tracks the discourse's focus (the "center"), to offer a more principled and unified explanation for these phenomena.

Methods

The core of the paper is the extension of a dynamic logic system (based on Muskens, 1996) to incorporate the concept of a "discourse center." This center is a distinguished discourse entity that can change its reference over time, a mechanism termed "center shift."

Key methodological contributions include:

  • Discourse Center: Introducing a dedicated position (position 0) in the discourse context to represent the current "center."
  • Center Shift Mechanism: Allowing the discourse center to be reassigned to a new entity, which is crucial for explaining sloppy readings.
  • VP Ellipsis Integration: Modifying the system to handle VP ellipsis by treating verb inflections (INFL) as properties that can be accessed from the context, and allowing these properties to be the discourse center.
  • Paycheck Pronoun Handling: Extending the framework to account for "paycheck pronouns" by introducing "dynamic individuals" – entities that represent a relationship between contexts, enabling anaphoric reference to contextually dependent entities like "his paycheck."
  • Unified Framework: Demonstrating how these extensions allow for a single, uniform treatment of sloppy identity across VP ellipsis, NP anaphora (like "his"), and paycheck pronouns, whenever the antecedent contains a proform.

Datasets

Not discussed in the paper. The paper uses constructed examples to illustrate its theoretical points.

Key Insights and Ideas

  • Sloppy Identity as Contextual Shift: The most significant insight is that sloppy identity is not about complex syntactic or semantic recovery mechanisms, but rather a natural consequence of how discourse context evolves, specifically through the reassignment of the discourse center.
  • Dynamic Semantics + Centering = Elegance: Combining the dynamic nature of discourse representation with the focus-tracking of centering theory provides a powerful and elegant solution to a long-standing problem in semantics.
  • Uniformity is Key: The paper argues for a unified treatment of various anaphoric phenomena (VP ellipsis, NP pronouns, paycheck pronouns) under the umbrella of dynamic context shifts, rather than separate, specialized accounts.
  • The Discourse Center as a Dynamic Variable: Treating the discourse center not as a static topic but as a dynamic variable that can be updated is central to the proposed solution.

Key Results

The paper doesn't present quantitative results in the traditional sense (e.g., accuracy scores on a dataset). Instead, its key results are theoretical:

  • Successful Modeling of Sloppy Identity: The framework successfully derives the correct "sloppy" readings for examples involving VP ellipsis (e.g., "Tom loves his cat. John does too.") and paycheck pronouns (e.g., "Smith spent his paycheck. Jones saved it.").
  • Expanded Paradigm: The proposed approach accounts for a broader range of sloppy identity phenomena, including cases where the antecedent contains a sloppy VP ellipsis or NP anaphora, and where the controllers are themselves complex.
  • Theoretical Simplicity: The paper demonstrates that a dynamic account, augmented with centering, is more general, empirically adequate, and simpler than alternative approaches that rely on complex matching or dependency mechanisms.

Limitations and Future Work

  • Simplified Centering: The paper acknowledges that it uses a simplified version of Centering Theory, omitting elements like the "forward-looking centers list" and preference orderings on transitions.
  • Future Work: The authors suggest incorporating these more complex aspects of Centering Theory into the dynamic system. They also propose investigating how to handle more complex linguistic phenomena, such as "backwards conditionals," within this framework. The paper also notes that the current system omits the semantic contribution of verb inflections (INFL) beyond their interaction with the discourse context.

Summary updated

BibTeX
@inproceedings{hardt1996centering,
  author = {Daniel Hardt},
  title = {{Centering in Dynamic Semantics}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1996},
}
Figure