Critical Discourse Analysis: language, power, ideology

CDA’s core move, clearest in Fairclough’s framework, is to analyse every text at three connected levels. Text: the linguistic detail — lexical choices, transitivity (who is made an agent, who a patient), nominalisation (processes frozen into things: “the closure” instead of “we closed it”), modality (what is asserted as certain, possible, obligatory), presupposition (what the text treats as already agreed).

Discursive practice: how the text is produced, distributed, and consumed — whose voices are quoted directly and whose are paraphrased, which genres it borrows authority from, how it will be read and re-used.

Social practice: how the text connects to wider structures — the institutional arrangements and ideologies it draws on and reinforces. The discipline is traffic between levels: every social claim cashed in textual evidence, every textual observation asked what social work it does.

The field has several schools sharing the critical stance. Fairclough’s dialectical-relational approach is set out in Language and Power and Critical Discourse Analysis (Routledge). Van Dijk’s socio-cognitive approach — how discourse shapes the mental models through which domination is accepted — is stated programmatically in Principles of critical discourse analysis. Wodak’s discourse-historical approach (SAGE) adds systematic attention to context and to how discourses evolve across time and genre chains.

Van Dijk’s principles include the field’s most quoted discipline: CDA addresses social problems through discourse, makes its stance explicit, and is judged by the adequacy of its analysis, not the righteousness of its conclusions.

A practical entry sequence: read van Dijk’s principles for the stance, Fairclough for the toolkit, and one exemplary empirical study in your domain for calibration of how much linguistic detail a publishable analysis carries.

• Use it on consequential public texts: policy documents, media coverage, political speech, institutional communications — texts that organise how issues may be thought about.
• Use it when the question is naturalisation: how did this framing become common sense? who is systematically agent, victim, threat, burden?
• Use it comparatively where possible — across outlets, periods, or genres — because ideological work is most visible as patterned difference.
• Do not use it for interaction mechanics (conversation analysis owns that) or for content frequency questions (content analysis is cheaper and sufficient).
• Do not use it as a costume for a position paper. If the conclusion would survive the deletion of all textual evidence, it was not an analysis.

A study examined how two national newspapers covered disability-benefit reform across a budget cycle: 214 articles, sampled by rule, analysed in full. The standpoint — concern with how claimants are discursively positioned — was declared up front.

Text-level analysis was tabulated. Claimants appeared as grammatical agents in 9% of clauses mentioning them, overwhelmingly in verbs of receiving and claiming; officials were agents of verbs of deciding and reforming. Nominalisation did consistent work: “the tightening of eligibility” appeared 41 times; a named actor tightening eligibility, 3 times. Modality split by voice: claimant hardship was reported as allegation (“claims that…”), fiscal necessity as fact. Direct quotation ran two-to-one toward officials; claimants were typically paraphrased, their words arriving pre-summarised.

Discursive-practice tracing found a phrase — “trapped in dependency” — moving intact from a policy paper through ministerial interviews into unattributed editorial voice within five weeks: naturalisation made visible. The social-practice interpretation connected these patterns to the reform’s legitimation, with the limits section stating plainly that reception effects were not studied. The analysis was contestable — that is the genre — but contestable at the level of tabulated evidence, which is what separates CDA from commentary.

• Politics first, text optional. Conclusions preceding analysis, with quotations as decoration — the failure that gives the field its critics.
• Illustration instead of pattern. Three vivid examples cannot carry a claim about a discourse; corpus-wide tabulation can.
• Ignoring counter-discourse. Texts that resist the dominant framing belong in the analysis; their absence reads as suppression.
• Over-claiming effects. Positioning readers is what texts do; changing minds is an empirical claim needing reception data.
• Standpoint smuggled rather than declared. The critical stance is legitimate only when visible.
• Linguistics without the social level — the inverse failure: careful grammar cataloguing that never says what work the patterns do.

CDA’s declared standpoint means its findings persuade differentially: audiences sharing the stance credit them, others discount them. The only defence is analysis strong enough to compel across the disagreement, which most studies do not achieve.

Interpretation of ideological function is underdetermined by text: the same nominalisation can be evasion or economy. Comparative design and production evidence narrow the readings; they cannot close them.

And the method reads texts, not minds or outcomes. What audiences do with discourse — resist, ignore, repurpose — is outside the design, and the strongest CDA writes that boundary into its claims.

Discourse analysis is the broader family — language as social action — without the committed critical standpoint.

Conversation analysis studies interaction’s mechanics and resists importing context CDA insists on; the contrast is instructive.

Document and textual analysis treats texts as records and traces; CDA treats them as acts.

Narrative analysis examines how stories are built — a complementary lens on the same public texts.

The tabulation burden — agency patterns, quotation structure, phrase genealogies across 200 articles — is exactly what corpus-scale coding removes. Evidano codes a document set against a CDA-derived codebook (agent roles, modality types, voice attribution) with every application linked to its passage, and its cross-segment comparison shows the outlet-by-outlet and period-by-period patterns the argument needs.

The reading of ideological function stays with the analyst, as it must — a tool can show that officials own the deciding verbs; what that does to the reader’s common sense is the researcher’s case to make.