Is RST sentence grammar?

No. It focuses on functional relationships among text spans, though those spans can sometimes be smaller than a sentence.

Does every relation have a nucleus and satellite?

No. Multinuclear relations connect spans without making one the supporting satellite, as in the neutral Contrast example.

Can two analysts disagree?

Yes. Context and inferred communicative purpose can support alternative analyses. Explicit definitions and annotation guidelines make those disagreements discussable.

AI Glossary / Natural language and speech / Intermediate

Rhetorical structure theory

Definition

Rhetorical structure theory, or RST, is a framework for describing how parts of a text contribute to its overall organization and coherence. It analyzes relationships between text spans, such as a claim supported by evidence or an instruction accompanied by information that helps someone carry it out.

What it means

RST asks what role a passage plays relative to another passage. Two sentences may discuss the same topic but serve different purposes: one may state a conclusion, and the other may explain why a reader should believe it. The relationship contributes meaning beyond the individual sentences.

In many relations, one span is the nucleus, central to the local communicative purpose, while another is a satellite that supports it. Other relations are multinuclear: the linked spans have comparable status. Nuclearity is therefore relational, not a permanent importance score attached to a sentence wherever it appears.

How it works

An analyst divides a text into suitable units, considers relations among them, and combines related spans into larger structures. A complete analysis can describe organization across a paragraph or document, rather than stopping at neighboring sentences.

Relation definitions impose constraints on the spans and their combination and describe an intended effect on the reader. For example, Evidence concerns increasing belief in a claim, while Background helps the reader understand material. A neutral Contrast relation links comparable alternatives without selecting one as the supporting satellite.

Analysis involves reasoned interpretation of the writer's purpose. An explicit connective can provide evidence, but relationships can also be implicit. More than one analysis may be defensible, and annotation projects must specify their relation inventory and segmentation conventions. RST describes textual organization; it is not itself a complete psychological model of how people read or produce language.

Three practical examples

1. Report review: An editor checks whether supporting observations are connected to the conclusion they are intended to justify, instead of appearing as unrelated facts.

2. Summarization research: A developer examines whether a summary preserves important discourse relationships, including the distinction between a claim and material qualifying or supporting it.

3. Help-content design: A writer distinguishes instructions from background explanation and from information needed to perform a step, then checks whether readers can find each function.

Think of it this way

Think of a guided museum exhibit. Some panels state the main message, others supply evidence or context, and two equal panels may present contrasting alternatives. Their arrangement explains how the visitor should connect them. The analogy captures communicative roles, but an RST analysis relies on defined relations rather than an editor's visual preference alone.

Worked example

Consider this original two-unit notice: (1) “The storage room is empty.” (2) “The inventory team checked every shelf this morning and found no items.” Suppose the writer expects readers to trust the inspection and uses it to increase belief in the first statement.

Under those assumptions, unit 1 is the nucleus and unit 2 is an Evidence satellite. The analysis contains two elementary units joined by one relation. Evidence identifies the communicative function of the inspection report; it does not independently verify that the inspection occurred or that the room remains empty.

Now consider another two-unit text: (1) “The east room has windows.” (2) “The west room has no windows.” If the purpose is a neutral comparison of the rooms, a multinuclear Contrast analysis is plausible. Neither sentence needs to be treated as a supporting satellite of the other.

These examples have the same number of units but different organization. Counting sentences cannot recover that difference. Changing the surrounding discussion or the writer's purpose might change the analysis, so the interpretation should be justified against the actual text rather than chosen solely from a familiar keyword.

Illustrative case study

Fictional case: A business analyst writes a proposal containing a recommendation, cost estimates, and an explanation of an operational problem. Reviewers find the proposal difficult to follow despite clear individual sentences.

The analyst labels what each paragraph is meant to do and draws a simple relationship map. Some cost details support the recommendation; another paragraph supplies background needed to understand the problem. The revision brings those connections into the prose. The exercise does not prove that the recommendation is sound, but it makes its structure easier to examine and challenge.

Common misconceptions and mistakes

A common mistake is reading “rhetorical” as a synonym for deceptive persuasion. RST also describes neutral exposition, instructions, and other coherent texts.

Another mistake is treating satellites as disposable. Supporting material may contain evidence or context essential to a responsible summary. Deleting it can distort what the text communicates even when the remaining nucleus is grammatical.

Do not identify every “because” with one fixed relation or assume every relation has an explicit marker. Also distinguish communicative Evidence from verified evidence: the analysis describes how information is presented to support belief, not whether the information is factually correct.

Business use

Businesses can use RST-informed review to make proposals, reports, and support materials easier to follow. For AI applications, discourse annotations can provide structured data for investigating summarization, generation, or document analysis.

Set a clear annotation guide before comparing analysts or training a model. Review disagreements in segmentation, relation choice, and nuclearity separately. Evaluate whether the resulting system preserves meaningful connections in realistic documents. A plausible diagram is useful analytical output, but it should not be sold as an objective measure of truth or writing quality without further validation.

Check your understanding

Does labeling the inspection statement as Evidence prove the room is empty?

Answer: No. It says that the statement is presented to increase belief in the claim. Checking the inspection and the room's condition is a separate factual task.

Related terms

Discourse relation
Discourse parsing
Natural language generation
text coherence
nucleus
satellite
summarization

Learn next

Select a short explanatory paragraph and label each unit's function. Try an analysis using Evidence, Background, Elaboration, or Contrast, then read the formal definitions and revise any unsupported choices. Explain why your analysis fits the writer's apparent purpose and where another interpretation remains possible.

Sources

RST introduction, created within the author-maintained RST resource: https://www.sfu.ca/rst/01intro/intro.html

RST relation definitions: https://www.sfu.ca/rst/01intro/definitions.html

Primary explanatory material was read. The room notices and proposal case are original teaching examples; their analyses are explicitly conditional on the stated communicative purposes.

Previous: Reward-to-go Next: Rich Club