Glossary

Centering Theory

A discourse coherence theory that models local focus by tracking a ranked set of forward-looking centers and a single backward-looking center to constrain pronoun interpretation.

Get in touch Learn more

ML engineer managing model training cluster on laptop, GPU utilization visible, technical deep learning setup.

DISCOURSE COHERENCE MODEL

What is Centering Theory?

A formal theory of local discourse coherence that explains how attentional state and entity salience constrain pronoun interpretation and text flow.

Centering Theory is a discourse coherence model that tracks the local attentional state by maintaining a ranked set of forward-looking centers (Cf) and a single backward-looking center (Cb) to constrain pronoun resolution and topic continuity. It posits that certain transitions between utterances—Continue, Retain, Smooth-Shift, and Rough-Shift—are preferred because they minimize the cognitive load required to integrate new information with the existing discourse focus.

The theory operates on the principle that each utterance contains a most prominent entity, the preferred center (Cp), which is likely to become the Cb of the following utterance. Pronoun interpretation is constrained by Rule 1: if any Cf is realized as a pronoun, the Cb must also be realized as a pronoun. This formal constraint explains why certain pronoun uses feel coherent while others create a sense of abrupt topic shifting, making Centering Theory foundational for modern coreference resolution and salience models.

DISCOURSE COHERENCE

Core Components of Centering Theory

The fundamental mechanisms of Centering Theory that model local discourse coherence by tracking a ranked set of forward-looking centers and a single backward-looking center to constrain pronoun interpretation and anaphora resolution.

Forward-Looking Centers (Cf)

A ranked set of discourse entities evoked by an utterance that serves as candidates for the backward-looking center of the subsequent utterance. The ranking reflects salience based on grammatical role, with subjects typically ranked highest.

Cf(Ui): The set of forward-looking centers for utterance Ui
Ranking hierarchy: Subject > Object > Indirect Object > Other
Entities are ordered by prominence rather than surface position
The highest-ranked member is the Preferred Center (Cp)

Backward-Looking Center (Cb)

The single most salient entity that connects the current utterance to the preceding discourse. The Cb must be a member of the previous utterance's forward-looking center set and represents the local topic of the utterance.

Cb(Ui): The backward-looking center for utterance Ui
Uniquely identifies the entity the utterance is about
Constraint: Cb(Ui) must be an element of Cf(Ui-1)
If no entity from Cf(Ui-1) is realized, the Cb is undefined

Center Transition States

Four coherence-defining transitions between utterances based on whether the Cb remains the same and whether it equals the Preferred Center (Cp). These states predict the perceived coherence of discourse.

Continue: Cb(Ui) = Cb(Ui-1) AND Cb(Ui) = Cp(Ui) — smoothest transition
Retain: Cb(Ui) = Cb(Ui-1) BUT Cb(Ui) ≠ Cp(Ui) — topic maintained but shifted focus
Smooth-Shift: Cb(Ui) ≠ Cb(Ui-1) AND Cb(Ui) = Cp(Ui) — acceptable topic change
Rough-Shift: Cb(Ui) ≠ Cb(Ui-1) AND Cb(Ui) ≠ Cp(Ui) — jarring, low coherence

Pronoun Rule (Rule 1)

A constraint on realization stating that if any entity in Cf(Ui) is realized as a pronoun, then the backward-looking center Cb(Ui) must also be realized as a pronoun. This rule explains why over-pronominalization of non-topic entities feels incoherent.

Enforces that the most central entity receives the most reduced form
Violation example: 'John met Bill. He smiled.' (He=John, not Bill)
Works in conjunction with grammatical role and salience
Provides a computational basis for pronominal resolution

Preferred Center (Cp)

The highest-ranked entity in the forward-looking center set Cf(Ui) that is predicted to become the Cb of the following utterance. The Cp serves as the default antecedent for subsequent pronouns.

Derived from the ranking function applied to Cf(Ui)
Typically the grammatical subject of the utterance
Guides anticipatory processing in discourse comprehension
Used by mention-ranking models to prune candidate antecedents

Centering Constraints

Three universal constraints that govern valid center assignments across utterances, forming the axiomatic foundation of the theory.

Constraint 1: Every utterance Ui has exactly one backward-looking center Cb(Ui)
Constraint 2: Every element of Cf(Ui) must be realized in Ui
Constraint 3: Cb(Ui) is the highest-ranked element of Cf(Ui-1) that is realized in Ui
These constraints enable deterministic algorithms for center tracking

DISCOURSE COHERENCE

Frequently Asked Questions

Explore the mechanics of Centering Theory, a foundational model for understanding how local focus and attention state guide pronoun interpretation and discourse coherence.

Centering Theory is a model of local discourse coherence that explains how the focus of attention shifts during a discourse and constrains the interpretation of referring expressions, particularly pronouns. It operates by tracking a set of discourse entities called forward-looking centers (Cf) , which are ranked by prominence, and a single backward-looking center (Cb) , which represents the most highly ranked entity from the previous utterance that is realized in the current utterance. The theory posits that coherent discourses exhibit smooth transitions between utterances, governed by rules that prefer the Cb to be the highest-ranked Cf (the preferred center (Cp) ) and to be realized as a pronoun. By modeling these transitions—Continue, Retain, Smooth-Shift, and Rough-Shift—the theory predicts the relative coherence of different pronominalization choices and discourse structures.

Enabling Efficiency, Speed & Accuracy

Intelligent Analysis, Decision & Execution

We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.

Talk to Us

Search across company data

Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.

Useful when people spend too long searching or get different answers from different systems.

Enterprise searchRAGPermissions

Automate internal workflows

Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.

Useful when repetitive work moves across multiple tools and teams.

AI agentsWorkflow automationGovernance

Add AI to products and internal tools

Build assistants, guided actions, or decision support into the software your team or customers already use.

Useful when AI needs to be part of the product, not a separate tool.

AI integrationDecision supportModel routing

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.

Limited slotsGet a Free AI Consultation

How We Work

Custom AI workflows for your Business

One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.

Talk to Us

Centering Theory

What is Centering Theory?