Can a text be coherent while discussing many entities?

Yes. The important issue is whether their introduction and relationships are understandable, not whether only one entity appears.

Are pronouns harmful to coherence?

No. Clear pronouns can support continuity and avoid unnecessary repetition; ambiguous ones can cause problems.

Is entity-based coherence the same as discourse coherence?

It is one perspective within the broader topic, which also includes relations between propositions and overall document organization.

AI Glossary / Natural language and speech / Intermediate

Entity-based coherence

Definition

Entity-based coherence is the continuity a text achieves through its references to people, things, organizations, and other discourse entities. It concerns how these entities are introduced, maintained, and shifted across sentences, helping readers follow what a passage is about and how its participants connect.

What it means

A paragraph can contain individually grammatical sentences yet be difficult to follow as a whole. One reason is that its references jump between participants without enough context. Entity-based coherence focuses on that part of the problem.
Consider a report that introduces a device, describes “it,” and later discusses “the unit.” If these expressions clearly refer to the same device, readers can maintain a connected interpretation.
This is one aspect of coherence, not a complete definition of good writing. Readers also need meaningful relationships between claims, a sensible overall purpose, and enough background information. A paragraph that consistently discusses one device can still contradict itself or leave an important causal step unexplained.

How it works

An analysis identifies entity mentions and determines which mentions refer to the same referent. It then examines how those entities continue or change across sentences.
Different approaches emphasize different information. A simple analysis may track presence and absence. A role-based entity grid additionally records whether an entity is a subject, object, or other participant, making patterns of continuity available to a scoring model.
The interpretation depends on the text and the method. A deliberate change of subject may be perfectly coherent when the writer signals the connection. Conversely, repeated pronouns may create the appearance of continuity while leaving readers unsure which entity is intended.
Reliable analysis therefore requires attention to actual reference links, not just repeated strings.

Three practical examples

An editor reviews a product explanation to check whether “it” consistently refers to the same component.
A summarization researcher studies whether selecting sentences has removed the introduction of an entity mentioned later.
A writing tutor asks learners to trace the main participants through a paragraph and explain each change of focus.

Think of it this way

Think of following several colored threads through a piece of fabric. A continuing thread helps you see connections, while an unexplained break can make the pattern harder to follow. The analogy is limited because readers also interpret events, intentions, and background knowledge; coherent writing is more than continuous threads.

Worked example

Compare two invented versions of a short maintenance update.
Version A reads: “The technician inspected the pump. It showed a cracked seal. The technician replaced the seal.” The pump and technician are introduced before later references depend on them. The seal is introduced as a part associated with the pump and then referred to again.
Version B reads: “It showed a cracked seal. The technician replaced the seal. The technician inspected the pump.” If read without earlier context, the opening “it” lacks a clear antecedent. Reordering has also made the sequence of information harder to interpret.
This comparison supports an editorial diagnosis, not a numerical proof that Version A is universally better. In a larger document, a previous sentence could already have introduced the pump, changing how Version B is understood.
Now replace Version A’s middle sentence with “It had no damaged parts.” The references can still be followed, but replacing the seal now needs further explanation; an undamaged seal could, for example, be replaced preventively. Entity continuity alone therefore cannot establish consistency of the described events.

Illustrative case study

In a fictional project, Cedar Manuals generates short equipment summaries from longer service notes. A draft occasionally begins with “this component” after the sentence naming the component has been omitted.
An editor traces the entity chain back to the original note and restores an explicit introduction. The team also checks event relationships because a well-linked reference chain does not guarantee that the service sequence is correct.
The imagined workflow uses entity continuity as a specific review lens. It does not claim that every pronoun should be replaced by a full noun phrase or that a single automated coherence score can approve a manual for publication.

Common misconceptions and mistakes

Entity-based coherence does not mean repeating the same name as often as possible. Excessive repetition can make a passage awkward without improving understanding.
It is not identical to coreference resolution. Coreference identifies shared referents; coherence concerns how reference patterns contribute to a connected text.
Nor does every new entity disrupt coherence. Introducing a related participant or a relevant part can advance an explanation naturally.
Finally, an automated entity-continuity score is not a factuality check. Correct links between mentions cannot verify whether the statements about those entities are true.

Business use

This perspective is useful when reviewing generated reports, summaries, instructions, and customer-facing explanations. It can help identify missing introductions, ambiguous pronouns, abrupt focus changes, and references made unclear by sentence removal.
A practical review records the specific reader problem and the proposed edit, rather than treating a numerical coherence score as the final product. For example, replacing an ambiguous “it” with “the backup pump” may resolve a concrete uncertainty.
Teams should evaluate their analysis on representative documents, including technical names and short fragments. Errors in entity detection or reference resolution can otherwise be mistaken for weaknesses in the writing itself.

Check your understanding

A summary repeatedly mentions the same company but gives incompatible dates for its founding. Does strong entity continuity resolve the problem?

Answer: No. The company reference may remain clear while the claims conflict. Entity-based coherence and factual consistency need separate assessment.

Related terms

Entity grid
Coreference resolution
Anaphora
Discourse relation
discourse coherence
local coherence
discourse entity
information status
discourse relations
text summarization

Learn next

Study coreference resolution to understand how reference chains are identified. Next, construct an entity grid for a short passage, then examine discourse relations such as cause and contrast. Comparing these views shows which continuity problems each can reveal and which it leaves unaddressed.

Sources

Primary research: Barzilay and Lapata, “Modeling Local Coherence: An Entity-Based Approach” (2008), https://aclanthology.org/J08-1001.pdf. The original entity representation and local-transition model were consulted. Textbook context: Jurafsky and Martin, Speech and Language Processing, discourse coherence chapter, https://web.stanford.edu/~jurafsky/slp3/25.pdf. Entity-based and relation-based coherence discussions were consulted. Independent review checked the original paper’s entity-based coherence discussion and its explicit limitation of local coherence for sentence ordering.

Previous: Enterprise information integration Next: Entity grid