DemoImplementationGo to paper →

Centering: A Framework for Modeling the Local Coherence of Discourse

Barbara J. Grosz - Harvard University

Aravind K. Joshi - University of Pennsylvania

Scott Weinstein - University of Pennsylvania

Abstract: This paper concerns relationships among focus of attention, choice of referring expression, and perceived coherence of utterances within a discourse segment. It presents a framework and initial theory of centering intended to model the local component of attentional state. The paper examines interactions between local coherence and choice of referring expressions; it argues that differences in coherence correspond in part to the inference demands made by different types of referring expressions, given a particular attentional state. It demonstrates that the attentional state properties modeled by centering can account for these differences.

Guide

Key Concepts

  • Utterance (U): A spoken or written statement, e.g., "John went to the store."
  • Utterance Sequence (Un, Un+1): A pair of utterances in a discourse. For example, Un: "John went to the store." Un+1: "He bought some milk."
  • Discourse Segment (DS): A coherent segment of connected utterances within a conversation or text.
  • Backward-Looking Center (Cb): The primary entity from Un that continues as the focus in Un+1. Example: In "John went to the store. He bought milk," the Cb is "John."
  • Forward-Looking Centers (Cf): Potential entities in Un that could become the focus in Un+1. Example: In "John went to the store," Cf might include "John" and "the store."
  • Preferred Center (CB): The most likely Cf to transition to Cb in Un+1.

Centering Rules

Rule 1: Each utterance has exactly one Cb. This links the discourse back to the previous context.

Rule 2: If there are multiple Cfs, the Cb in Un+1 should be the highest-ranked Cf from Un.

Rule 3: Avoid unnecessary shifts of Cb. Coherence is improved when transitions between utterances minimize shifts.

Preferences for Transitions:

  • Continue: Cb remains the same, and CB is maintained.
  • Retain: Cb remains the same, but CB differs.
  • Shift: Cb changes, marking a transition to a different focus.

Understanding Transitions

Transitions measure how well the discourse maintains coherence. The most coherent discourse occurs with Continue transitions, while Shift transitions are less preferred.

Examples:

  • Continue: "John went to the store. He bought milk."
  • Retain: "John went to the store. The store was closed."
  • Shift: "John went to the store. Mary stayed home."

Notice

This demo uses Natural Language Processing (NLP) techniques to extract backward-looking (Cb) and forward-looking (Cf) centers from text. While it handles a variety of inputs, NLP models may not always interpret abstract or ambiguous references perfectly. For best results, input clear and concise sentences.

Default Examples

John has been acting quite odd. He called up Mike yesterday. Mike was studying for his driver's test. He was annoyed by John's call.

Jill caught a ball. She tossed it to John. He caught it while in the air.

Try Your Own Sentences

Purpose

This demo serves as a bridge between theory and practice, aiming to make a foundational concept in linguistics accessible to a wider audience.

  • Offer an interactive way to explore the principles of Centering Theory.
  • Bridge the gap between theoretical linguistics and practical applications.
  • Assist students and researchers in better understanding discourse coherence.
  • Highlight the historical and ongoing influence of Centering Theory in linguistics and AI.
  • Inspire further exploration and application of these concepts in modern contexts.
ACL Logo

ACL 2020 Test-of-Time Award (25 years)

With well over 3000 citations, this paper has had a profound impact on the study of linguistics, AI, and discourse analysis. Its framework has been instrumental in advancing how systems understand and process language, bridging gaps between human and machine communication. The theories presented have been foundational in applications ranging from machine translation and chatbots to text summarization, dialogue systems, anaphora resolution, and sentiment analysis, underscoring its relevance and influence in both academic and practical fields.