Corpus-Assisted Discourse Studies
- Introduced by
- Alan Partington and colleagues
- Year
- 2004
- Type
- Mixed-methods corpus-and-discourse analytic approach
- Family
- Corpus linguistics / discourse analysis
What does it do?
Corpus-Assisted Discourse Studies (CADS) is a mixed-methods approach that combines the quantitative power of corpus linguistics with the interpretive depth of discourse analysis to investigate how meanings, evaluations, and ideologies are constructed across large collections of text.
Pioneered by Alan Partington and colleagues, CADS uses corpus techniques such as keyness, collocation, and concordancing to identify patterns no analyst could find by reading alone, then 'shunts' back to close qualitative reading to interpret what those patterns mean in their discursive context.
The intuition behind the method
A discourse analyst reading a handful of newspaper articles can interpret them richly but cannot be sure the patterns they notice are typical rather than cherry-picked. A corpus tool can tell you that 'flood' and 'wave' systematically collocate with 'immigrants' across a million words, but it cannot tell you what that means for the construction of social fear. CADS marries the two: let the machine point to where the statistically salient patterns are, then read closely to understand them, then return to the corpus to test the new hunch. Partington calls this back-and-forth 'shunting' — the quantitative pass disciplines the analyst's intuition, and the qualitative pass gives the numbers meaning.
History
CADS emerged in the late 1990s and early 2000s as corpus linguists trained on lexicographic and collocational methods turned their tools toward discourse and ideology. Alan Partington named and theorised the approach, arguing in his 2004 essay 'Corpora and discourse, a most congruous beast' that corpus and discourse work were natural partners, and developing the programme fully with Alison Duguid and Charlotte Taylor in Patterns and Meanings in Discourse (2013), which set out the 'shunting' logic between statistics and close reading. In parallel, Paul Baker, Tony McEnery, Ruth Wodak and colleagues demonstrated the synergy of corpus linguistics and critical discourse analysis in their influential 2008 study of UK press discourse on refugees and asylum seekers, establishing CADS as a recognised methodology across corpus linguistics, media studies, and political discourse analysis.
When to use it
Use CADS when you have a discourse-analytic question — about representation, evaluation, ideology, or change over time — and a body of text large enough that reading alone would risk missing or over-claiming patterns. It is especially valuable for comparing discourses (across outlets, periods, or communities) and for adding empirical weight and falsifiability to critical discourse analysis. It is less suited to very small text sets where manual analysis suffices, to questions about interaction and multimodality that lie outside the textual co-text, or to projects without a clear comparison that would make corpus statistics meaningful.
Limitations and considerations
- Corpus measures locate salient lexis but cannot themselves interpret meaning; the qualitative shunt remains essential and analyst-dependent.
- Findings are bounded by corpus design: an unbalanced or unrepresentative corpus or a poorly chosen reference corpus distorts keyness and collocation.
- The lexical and co-text focus can miss discourse features that operate above the sentence, across modes, or in interaction.
- Integrating two epistemologies is demanding: naive number-crunching or impressionistic reading each undermines the synergy CADS depends on.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Disciplines analyst intuition: corpus statistics show whether a noticed pattern is typical or anecdotal across the whole dataset.
- Surfaces patterns invisible to manual reading, such as pervasive collocations and keyness, that signal where ideology is concentrated.
- Combines quantitative replicability with qualitative interpretive depth, strengthening the rigour and falsifiability of discourse claims.
- Scales discourse analysis to large, comparative datasets across outlets, communities, or historical periods.
Applications
- Media discourse: tracing how newspapers represent immigrants, minorities, or social groups across outlets and over time.
- Political discourse: analysing parliamentary debates, manifestos, or speeches for evaluative and ideological patterns.
- Diachronic studies: tracking how the discourse around a concept, value, or institution shifts across historical periods.
- Health and science communication: examining how risk, illness, or controversy is constructed in public and institutional texts.
- Institutional and promotional discourse: studying the language of corporations, NGOs, or governments for legitimation and self-presentation.
How it works
How it works
The researcher begins from a discourse-analytic question — how a group, topic, or value is represented — and assembles a principled corpus to answer it, often together with a reference or comparison corpus that makes difference measurable. Choices about sampling, time span, source, and balance are made so that statistical comparisons (for example, the language of one newspaper versus another, or one period versus another) are interpretable rather than artefacts of corpus design.
The analyst runs corpus measures over the data: keyword and keyness analysis to find words unusually frequent relative to the reference corpus, collocation to find the company those words keep, and frequency profiling to map salient lexis. As Baker and colleagues demonstrate in their study of UK press discourse on refugees, these measures surface the lexical and collocational regularities — such as quantifying and water metaphors — that signal where ideologically loaded representation is concentrated.
The analyst then shunts from statistics to close reading, returning to concordance lines and fuller co-text to interpret each salient pattern: what evaluative meaning a collocation carries, how a keyword functions rhetorically, whether a frequency reflects strategy or convention. This qualitative pass tests, refines, and sometimes overturns the quantitative leads, and it typically generates new questions that send the analyst back to the corpus, making the method genuinely iterative rather than a one-way pipeline.
Finally the converging quantitative and qualitative evidence is synthesised into claims about discourse: how the texts construct identities, legitimise or delegitimise actors, naturalise values, or shift over time. The corpus supplies the warrant that patterns are systematic and not anecdotal, while the close reading supplies the interpretation, so that conclusions about ideology and representation rest on both replicable measurement and contextual understanding, in the spirit of Partington's programme for CADS.
Common pitfalls
- Treating keyness or collocation lists as findings in themselves rather than as entry points that must be read back into context.
- Choosing an inappropriate or mismatched reference corpus, so that the keywords reflect register differences rather than the discourse of interest.
- Cherry-picking concordance lines that confirm a thesis instead of systematically reading the dominant pattern.
- Reporting frequencies and statistics without the close qualitative interpretation that turns them into a claim about discourse.
Frequently asked questions
How is CADS different from ordinary corpus linguistics or from critical discourse analysis?
Plain corpus linguistics often stops at describing linguistic patterns (frequencies, collocations, keywords) without a sustained interpretive account of social meaning, while traditional critical discourse analysis interprets meaning richly but on small text samples that risk cherry-picking. CADS deliberately integrates the two: it uses corpus statistics to find and warrant patterns systematically, then applies discourse-analytic interpretation to explain them, shunting between the two until the evidence converges.
What does Partington mean by 'shunting'?
Shunting is the iterative back-and-forth movement between quantitative corpus work and qualitative close reading. A keyness or collocation result points the analyst to a pattern; the analyst reads the concordance lines and co-text closely to interpret it; that interpretation raises a new question, which sends them back to the corpus for another measure. CADS is therefore a cycle, not a one-way pipeline, with each pass refining and constraining the other.
Why is a reference corpus so important in CADS?
Keyness — the core technique for finding what is distinctive about a text or set of texts — is always relative: a word is 'key' only by comparison with some norm. The reference (or comparison) corpus supplies that norm. If it is poorly matched to the study corpus in register or genre, the resulting keywords reflect those mismatches rather than the discourse of interest, so choosing and justifying the reference corpus is one of the most consequential design decisions in CADS.
Related methods
Sources
Partington, A., Duguid, A., & Taylor, C. (2013). Patterns and Meanings in Discourse: Theory and Practice in Corpus-Assisted Discourse Studies (CADS). John Benjamins.
ISBN: 9789027203885Copied.Baker, P., Gabrielatos, C., KhosraviNik, M., Krzyżanowski, M., McEnery, T., & Wodak, R. (2008). A useful methodological synergy? Combining critical discourse analysis and corpus linguistics to examine discourses of refugees and asylum seekers in the UK press. Discourse & Society, 19(3), 273–306.
Open DOI referenceCopied.Partington, A. (2004). Corpora and discourse, a most congruous beast. In A. Partington, J. Morley, & L. Haarman (Eds.), Corpora and Discourse (pp. 11–20). Peter Lang.
ISBN: 9783039102488Copied.