Discourse analysis examines how language works across texts, conversations and social situations. It asks how people connect ideas, identify participants, organise actions, present knowledge and negotiate meaning. Its gold standard is an interpretation that another reader can trace back to the material: clear examples, an explicit method, relevant context and an honest account of alternative explanations.
This guide shows how to analyse a notice, a short exchange and a small collection of messages without confusing an attractive interpretation with demonstrated evidence. All examples and numerical datasets below are fictional teaching material. Begin with What Is Discourse? for the concept, then use this article for the investigative method. The Gold Standard master and Education parent provide the wider framework: understanding should remain useful when the text, speaker or situation changes.
Choose your discourse analysis route
Chapters 1–3: Frame the investigation
Chapters 4–6: Trace text and interaction
Chapters 7–9: Inspect records, claims and power
Chapters 10–12: Count, challenge and transfer
Chapter index
Frame the investigation · Chapters 1–3
1. Begin with a question that language evidence can answer
Trace text and interaction · Chapters 4–6
4. Trace reference before interpreting the message
Inspect records, claims and power · Chapters 7–9
7. Treat a transcript as a representation with choices
Count, challenge and transfer · Chapters 10–12
10. Count patterns with a denominator and a coding rule
11. Check agreement, alternatives and missing evidence
12. Transfer the analysis to a changed communication setting
Answer questions and continue learning · Chapters 13–14
CHAPTER 1 OF 14 · Frame the investigation
1. Begin with a question that language evidence can answer
A notice says, “Residents must leave the entrance clear.” A second says, “Please help us keep the entrance clear.” Both concern an entrance, but they organise the relationship with readers differently. One explicitly presents an obligation. The other frames the desired action as assistance to a collective project. That observation gives us a starting point, not a complete explanation of either writer’s motives.
A good question might be: how do these notices express the requested action and the reader’s responsibility? A weaker question would be: which writer secretly dislikes residents? The first can be investigated through wording and context. The second leaps from language to a psychological claim for which the text alone provides inadequate evidence.
Separate observation, interpretation and explanation
Observation identifies a feature: one notice uses “must”. Interpretation describes its likely function in context: it presents the action as required. Explanation proposes why that choice appeared: perhaps the notice follows repeated obstruction, belongs to a formal policy genre or was written by a different department. Each step requires additional support.
Write these levels separately in your notes. “The notice contains an obligation marker” is not interchangeable with “the organisation has become more authoritarian”. The latter might be a research hypothesis, but supporting it would require a broader comparison and information about practices, institutions and historical circumstances.
An efficient first pass asks four questions. What happens in the text? Which exact words or sequence make that happen? What surrounding information matters? What plausible alternative account should be tested? These questions keep analysis close to evidence while allowing larger social questions to emerge responsibly.
The subject is useful precisely because language does more than convey isolated facts. A message can allocate responsibility, soften a request, conceal an unclear referent or organise a response. Careful analysis makes those operations visible without pretending that every choice has one fixed meaning across all situations.
CHAPTER 2 OF 14 · Frame the investigation
2. Choose an approach that fits the material
Discourse analysis is a family of approaches rather than one universal procedure. Some investigations focus on how a text holds together. Others examine the organisation of interaction, patterns across a corpus, narratives, institutional categories or relationships between language and social power. These approaches can inform one another, but their claims and evidence requirements should remain clear.
A conversation recording can support questions about timing and turn transitions that a cleaned transcript cannot answer reliably. A collection of published notices can support comparisons of recurring wording, but it cannot directly show how every reader understood those notices. An interview about an event supplies an account of the event, not an unmediated replay of it.
Match the unit to the question
Suppose the question concerns how a request is accepted or declined. The useful unit may be a sequence containing the request, the response and any subsequent clarification. Extracting only the request would remove the evidence about how the other participant treated it. If the question concerns the organisation of an explanation, the unit may instead be several connected clauses or paragraphs.
For a project comparing announcements, define what counts as an announcement. Are repeated versions separate items? Do image captions belong to the text? Are comments included? Does a short update that links to a longer notice count as the same document or a new communicative event? These decisions influence the conclusions before analysis begins.
A method statement need not be complicated. It can say: “I compare twenty notices from two periods, treating each unique notice as one item, and code how the requested action is expressed.” That is more informative than saying only that a discourse approach was used.
Do not borrow impressive terminology to cover an undefined procedure. If you describe a study as conversation analysis, explain how sequential evidence supports the claims. If you use a thematic classification, explain how categories were defined and applied. Method names help readers understand the work only when the actual steps match the name.
Context includes the circumstances that make a text or interaction interpretable: who produced it, for whom, through which medium, for what activity and at what point in a larger sequence. Context is evidence to establish, not a story to fill in whenever the wording seems ambiguous.
Imagine the message “Fine.” It could accept a proposal, close an argument, signal annoyance or simply acknowledge receipt. Punctuation alone does not settle the matter. The preceding message, the relationship, the medium and any subsequent response may help. A responsible analysis can retain uncertainty when those materials are unavailable.
Build a compact context record
For a public notice, record the source, date, location or platform, intended audience, document type and any linked instructions. For an interaction, record the activity, relevant participant roles, recording conditions and the sequence boundaries. Keep observed information separate from assumptions about social identity, intention or emotion.
Consider a fictional exchange at an equipment desk. A visitor asks, “Can I take this?” The assistant replies, “Have you signed it out?” If the institution requires a checkout record, the reply can be understood as addressing a condition on taking the item. Without knowing the activity, an analyst might incorrectly treat it as an unrelated question.
Even with context, avoid assuming that an institutional rule determines every interpretation. The visitor’s next response matters. “Not yet—where do I sign?” supports one understanding; “I meant take a photograph of it” reveals a different problem. The evolving interaction can revise what initially looked obvious.
For a practical project, create a short “known, inferred, missing” note beside each extract. This prevents an inference from becoming a fact through repetition. If a missing detail is essential to the claim, either obtain it through an appropriate source or narrow the claim to what the available material can support.
CHAPTER 4 OF 14 · Trace text and interaction
4. Trace reference before interpreting the message
Texts refer to people, things, events and prior propositions. Names, pronouns, demonstratives and repeated descriptions help readers maintain those references. When the chain becomes unclear, every individual sentence may remain grammatical while the larger message becomes unreliable.
Take an original example: “The coordinator emailed the caretaker after the resident called. She said it was unsafe.” Who is “she”? What is “it”? The text may be clear to participants who share background knowledge, but an outside reader cannot confidently identify either referent from this extract alone.
A worked reference repair
Assume the intended meaning is that the resident reported an unsafe stair rail. A clearer version is: “After the resident reported that the stair rail was unsafe, the coordinator emailed the caretaker about the rail.” This rewrite names the report and the object. It does not establish whether the rail was actually unsafe; it preserves the source of that claim.
That last distinction matters. Rewriting “the resident reported that the rail was unsafe” as “the unsafe rail” can shift an attributed report into the writer’s own description. A cohesion repair should not silently strengthen the factual status of the statement.
Map each referring expression to its intended target. If more than one target remains plausible, record the ambiguity rather than choosing whichever supports your preferred reading. Check whether the full document resolves it. An extract may appear ambiguous because the analyst removed a sentence that introduced the referent clearly.
The cohesion across sentences guide develops these issues for translation. Its practical lesson also applies to analysis: preserving word-level meaning is insufficient if the connections among sentences have changed. Track the chain of reference before drawing conclusions about the larger argument or event.
CHAPTER 5 OF 14 · Trace text and interaction
5. Distinguish visible links from a coherent argument
Cohesion concerns linguistic connections such as reference, repetition and conjunctions. Coherence concerns how the discourse makes sense as a connected whole. A text can contain many linking words while still leaving its central reasoning unsupported. Conversely, a short exchange can be intelligible with few explicit connectors because participants share the relevant situation.
Consider: “The room is crowded. Therefore, the timetable was printed in blue.” The connector announces a consequence, but the relationship is unexplained. Replacing “therefore” with “however” does not repair the underlying problem. We need to know what relationship the writer intends and whether the surrounding evidence supports it.
Test the relationship, not just the connective
A stronger fictional sequence is: “The room is crowded. The organiser will move the next session to a larger hall.” Readers can infer a practical relationship without an explicit “therefore”. Adding the connector may make that relationship more visible, but it does not by itself establish that the move is feasible or sufficient.
Ask what each segment contributes. Does it supply a reason, an example, a contrast, a condition, an explanation or background needed for comprehension? These descriptions should be justified by the content and purpose, not assigned mechanically from a single word.
The Rhetorical Structure Theory relation definitions provide one explicit framework for analysing relations between text spans, including background, evidence, elaboration and contrast. Such a framework makes analytical choices inspectable. It does not remove interpretation or guarantee that two readers will always produce the same analysis.
For a worked exercise, take “The path is closed because a tree has fallen across it.” The second clause supplies a proposed cause of the closure. Change it to “The path is closed; the gate is locked and a closure sign is displayed.” The following material now supports a reader’s belief about the closure. Cause and evidence are related possibilities, but they answer different questions.
CHAPTER 6 OF 14 · Trace text and interaction
6. Analyse what a turn makes relevant next
Conversation unfolds through sequences. A question commonly makes an answer relevant; an invitation creates a place for acceptance, refusal or another response; a clarification request can suspend the main activity while a difficulty is addressed. Looking at the sequence helps us understand what participants are doing with their words.
Sacks, Schegloff and Jefferson’s 1974 study of conversational turn-taking treats the organisation of turns as a systematic feature of interaction. Its importance for a beginner is methodological: the position and organisation of a contribution matter, not merely its vocabulary. Detailed claims about timing require suitable recordings and transcription.
Follow the response rather than assuming the intention
Consider this fictional exchange. A says, “Are you using this chair?” B says, “You can take it.” B treats the question as connected to a possible request for the chair. That response is evidence about B’s understanding. It does not prove that every occurrence of the same question is a request or that A’s private intention is fully known.
Now change B’s response to “Only when the meeting starts.” The sequence remains open: the answer supplies information but leaves practical availability uncertain. A might then ask, “Could I use it for five minutes?” The request becomes explicit through the subsequent interaction.
Do not label every overlap an interruption or every silence a refusal. Those interpretations require evidence about timing, activity and what the participants do next. Two people may begin speaking at a possible transition point without either attempting to dominate the other. A pause may arise from searching for information rather than unwillingness to respond.
A useful analysis preserves enough preceding and following talk to show why a label fits. If you present only one line, explain the resulting limitation. The sequential environment often supplies the strongest evidence for the action being accomplished, and removing it can turn a defensible interpretation into speculation.
CHAPTER 7 OF 14 · Inspect records, claims and power
7. Treat a transcript as a representation with choices
A transcript transforms a time-based event into a readable record. It selects what to preserve: words, pauses, overlap, emphasis, laughter, gesture, speaker changes and other features. Different purposes require different detail. A readable meeting summary and an interactional transcript are not interchangeable forms of evidence.
The existing Transcript guide explains why transcription decisions matter. An analyst should identify which features were recorded and which were omitted. A cleaned transcript may be suitable for studying reported topics while being inadequate for a claim about hesitation or response timing.
Compare two representations of one fictional event
Version A reads: “A: Can you finish today? B: Yes, tomorrow morning.” Version B adds that B first said “yes”, paused, and then corrected the time to tomorrow morning. Without the temporal detail, a reader may see a contradiction. With it, the sequence may be better understood as a revision or clarification. The recording would be needed to decide exactly how the event unfolded.
Do not manufacture precision. If a recording does not allow reliable measurement of a pause, label the uncertainty rather than inventing a decimal duration. If speech is unclear, retain an uncertainty marker according to a stated convention. Guessing a word can change the analysis of responsibility, agreement or disagreement.
For a small project, keep a transcription key. Explain speaker labels, omitted passages and any notation for uncertainty. Maintain a secure link between the transcript and the source material where permission permits. Record editing decisions so another analyst can see why the readable version differs from the original event.
Automatic transcription can provide a starting draft, but proper names, accents, overlapping talk and quiet speech can require correction. Verify the features on which the argument depends. If your conclusion rests on whether someone said “can” or “can’t”, checking that word is essential rather than optional proofreading.
CHAPTER 8 OF 14 · Inspect records, claims and power
8. Examine modality and attribution without mind-reading
Language can present a claim as certain, possible, required, permitted or attributed to someone else. “The gate is closed”, “The gate may be closed” and “The caretaker says the gate is closed” differ in the writer’s commitment and the stated source of information. These distinctions are central to responsible reading.
The Modality guide explains how English expresses possibility, obligation, permission and confidence. In analysis, ask which interpretation fits the local context. The same modal form can perform different functions; counting it without examining examples may combine unlike uses.
A worked comparison of three notices
Notice A says, “Visitors must report to reception.” Notice B says, “The visitors must be at reception by now.” Notice C says, “Visitors may report to reception.” In A, “must” naturally expresses a requirement. In B, it can express an inference about location. In C, “may” could express permission, though a fuller context is needed to exclude other readings.
If a coding scheme labels every “must” as an obligation, B becomes a false positive for that category. The repair is not to abandon counting. It is to define the category by function, inspect context and retain an ambiguous category when the evidence does not permit a confident decision.
Attribution introduces another layer. “The manager confirmed the delay” differs from “A visitor suggested there might be a delay.” These expressions identify different sources and degrees of commitment. An analyst should preserve those differences when paraphrasing rather than flattening both into “there was a delay”.
Avoid turning linguistic caution into a verdict about honesty. A cautious expression can reflect genuine uncertainty, politeness, institutional convention or strategic ambiguity. To argue for one explanation, compare relevant contexts and seek additional evidence. The wording supports a description of how the claim is presented more directly than a conclusion about the speaker’s character.
CHAPTER 9 OF 14 · Inspect records, claims and power
9. Connect language and power through observable practices
Discourse can organise who gets to ask questions, whose account is treated as authoritative, which categories are available and what actions count as acceptable. These are important questions about power. They become stronger when grounded in specific practices rather than inferred from a single intimidating word.
Imagine a meeting in which one role allocates speaking turns, closes topics and decides which suggestions enter the written record. Those procedures are relevant evidence about participation. A transcript might show the turn allocation, while minutes reveal what was retained. Neither source alone provides the complete institutional picture.
Compare what the design permits people to do
A consultation form asks, “Which of these two options do you prefer?” It may allow useful input within a limited choice while excluding proposals outside those options. The analytical observation is about the available response space. Whether that limitation is appropriate depends on the purpose, constraints and wider decision process.
A different form asks, “What problems should the project address?” This invites a broader contribution, but the wording alone does not prove that responses will influence decisions. To study actual participation, examine what happens after people submit answers: how they are summarised, evaluated and connected to action.
Similarly, passive wording such as “The application was rejected” omits an explicit actor in that sentence. This can make responsibility less visible locally. It does not prove deliberate concealment; the actor may be established elsewhere or irrelevant to the immediate purpose. Inspect the document and compare alternatives before making a stronger claim.
A useful analysis can be critical without becoming accusatory. State the mechanism: a category excludes certain experiences, a question narrows possible answers, or a summary removes a distinction present in the source. Then explain the consequences supported by evidence and identify what remains unknown. Precision makes criticism more useful and more accountable.
CHAPTER 10 OF 14 · Count, challenge and transfer
10. Count patterns with a denominator and a coding rule
Quantitative summaries can help show whether an observation recurs. They are most useful when the unit, category and denominator are explicit. A raw count of a word does not automatically measure politeness, authority, uncertainty or another complex social feature.
Suppose a fictional collection contains thirty notices, eighteen of which include at least one explicit instruction. The proportion is eighteen divided by thirty, or 60%. This is a notice-level measure. If one notice contains ten instructions and another contains one, both contribute one to the numerator under this definition.
Change the unit and the result changes
Now suppose the same collection contains sixty instruction clauses in total, twenty-four of which use an explicit obligation modal. The clause-level proportion is twenty-four divided by sixty, or 40%. It would be incorrect to compare this directly with the earlier 60% as if the two percentages measured the same thing.
For another fictional comparison, collection A has twelve obligation expressions across 3,000 words and collection B has twenty across 10,000 words. Their rates per thousand words are four and two respectively. B has more expressions in total, while A has the higher rate. Neither measure automatically tells us which collection places greater practical demands on readers.
The Corpus Linguistics root develops sampling and frequency analysis. In discourse research, connect such measures to inspected examples. A numerical pattern can direct attention, but the function of a form still needs interpretation in context.
Record difficult cases and category changes. If you revise the definition halfway through, revisit earlier items rather than leaving two rules mixed together. Preserve the distinction between an exploratory observation and a pre-specified test. A transparent small study is more useful than a large table whose categories cannot be reconstructed.
CHAPTER 11 OF 14 · Count, challenge and transfer
11. Check agreement, alternatives and missing evidence
Two analysts can disagree because they noticed different evidence, used different definitions or interpreted the same context differently. Disagreement is not automatically a failure. It can expose a category that needs clarification or a text that genuinely supports more than one reading.
In a fictional exercise, two analysts independently classify twenty messages as either a request or not a request. They agree on sixteen and disagree on four. Their raw agreement is 16 divided by 20, or 80%. This number describes the matching decisions under that procedure; it is not an 80% probability that the interpretation is true.
Inspect what the agreement hides
If almost every message falls into one category, high raw agreement can occur even when the analysts struggle with the rarer category. Report the distribution and inspect the disagreements. If the question requires a more formal reliability measure, choose one suitable for the design and assumptions rather than treating all coefficients as interchangeable.
For a qualitative project, compare competing readings explicitly. A brief response may close a topic, acknowledge information or display uncertainty. What subsequent turn would favour one reading? Does the recording contain that evidence? If not, preserve the alternatives in the analysis instead of forcing a single answer.
Actively look for a case that does not fit the emerging pattern. If your provisional claim is that every request uses “please”, one unmarked request challenges it. The appropriate response may be to narrow the claim, refine the category or abandon the hypothesis. An exception is information about the analysis, not an inconvenience to hide.
Keep a decision log with the original label, the reason for revision and the examples affected. This lets another reader distinguish genuine refinement from selective adjustment. The aim is not to eliminate judgement; it is to make the path from material to judgement visible and open to challenge.
CHAPTER 12 OF 14 · Count, challenge and transfer
12. Transfer the analysis to a changed communication setting
A method that works for a printed notice may need adjustment for a group chat. Messages can arrive out of order, respond to different threads, use reactions instead of words and depend on shared images. The underlying questions remain useful, but the evidence is distributed differently.
Consider an original case. A group chat contains: “Can someone unlock the room?” followed by a thumbs-up reaction from one participant. Ten minutes later another person writes, “Still waiting outside.” Has the request been accepted and fulfilled? The reaction suggests some positive orientation, but it does not establish completion. The later message provides evidence that the practical problem remains.
Answer the changed-context question
A careful account would distinguish acknowledgement, commitment and completed action. It would inspect who reacted, whether that person had access, whether another message clarified the plan and what happened next. The same icon can serve different functions in different groups. Do not assign it a fixed meaning independently of the interaction.
Now imagine translating the exchange into formal minutes. “The room was unlocked” would add an unsupported outcome. “A participant acknowledged the request; a later message reported that people were still waiting outside” stays closer to the evidence. This is a practical transfer of discourse analysis into accurate documentation.
For a second challenge, move from chat to a public FAQ. A question may be written by the organisation rather than quoted from a reader. The question-and-answer format resembles interaction, but it does not prove that a real exchange occurred. Analyse how the format anticipates concerns while keeping its production context visible.
Successful transfer means adapting the method to the medium. Preserve references, sequence, attribution and uncertainty, but reconsider what counts as a turn, a response or a complete event. The analytical categories should help explain the material rather than force every communication channel into the same mould.
CHAPTER 13 OF 14 · Answer questions and continue learning
13. Frequently asked questions about responsible analysis
Is discourse analysis simply looking for hidden meanings?
No. Much of the work concerns visible organisation: references, sequence, argument relations, categories and the distribution of opportunities to respond. An implied meaning may be relevant, but it needs contextual support. The analyst’s task is to explain how an interpretation is supported, not to invent a secret message behind every expression.
Can one sentence be analysed?
Yes, but some questions require more context. A sentence can show a grammatical choice or a form of attribution. It may not establish the action accomplished in a conversation or the structure of a larger argument. State what the extract allows you to examine and what removing the surrounding material prevents you from knowing.
Does a passive sentence hide responsibility?
It can leave an actor unexpressed in that sentence, but the actor may be known, mentioned elsewhere or irrelevant to the immediate purpose. Begin with the local observation, then inspect the broader context. Deliberate concealment is a stronger claim that requires more than identifying passive grammar.
Is interpretation too subjective to be useful?
Interpretation involves judgement, but it can be disciplined. Define the question, preserve the material, show the relevant examples, explain the analytical steps and consider alternatives. Readers can then assess the reasoning. Pretending that a label assigns itself automatically is less rigorous than making the judgement explicit.
Can counts replace close reading?
Counts can reveal recurrence and guide comparison, but they depend on what was counted and how it was classified. The same word may perform different functions. Close reading helps establish those functions, while a broader collection helps test whether an interesting example is typical or exceptional.
Can analysis reveal what a speaker really believed?
Language provides evidence about what was said, how it was presented and how others responded. Private belief is not directly observable in a transcript. Claims about belief require appropriate additional evidence and should retain uncertainty. A defensible account of public communication can remain valuable without claiming access to another person’s mind.
CHAPTER 14 OF 14 · Answer questions and continue learning
14. Build a complete small project and report it clearly
Choose a manageable question, such as how a set of public notices expresses requested actions. Select a defined collection and record the inclusion rule. Preserve source information and relevant context. Read the material before settling every category, but distinguish exploratory decisions from any later formal comparison.
Prepare a small coding guide with definitions and examples. Identify the unit of analysis and a category for unresolved cases. Apply the guide consistently, inspect a sample with another reader where possible, and revisit disagreements. Keep the original material available under appropriate permissions so the analysis remains traceable.
A report structure that readers can inspect
Begin with the question and the bounded answer. Describe the collection and its limits. Explain the method in enough detail for another person to understand the decisions. Present a few carefully selected extracts, any relevant counts and at least one alternative or difficult case. End with what the evidence supports and what further material would be needed for a stronger claim.
For example, a defensible fictional conclusion could say: “In this thirty-notice collection, eighteen notices contained an explicit instruction under our definition. The examples show several ways of presenting that instruction, including obligation and cooperative requests. The collection does not establish how residents interpreted or followed the notices.” The conclusion is useful because its boundary is clear.
For further study, consult Sacks, Schegloff and Jefferson on turn-taking, the RST relation definitions and Mann and Thompson’s account of text organisation. These offer different analytical resources rather than a single interchangeable procedure.
Use the How X Works hub to connect this method with wider subjects, and the Language Acquisition root to explore how communicative abilities develop. A strong analysis leaves the reader able to examine a new text more carefully, explain the evidence more precisely and recognise where a conclusion has outrun what the material can show.