Corpus Linguistics: Corpus-Based Analysis and Move Analysis

Collocates analysis identifies words that frequently appear near target words, revealing linguistic associations. AntConc's Collocates tool uses adjustable parameters including window span (words to search left/right) and minimum frequency thresholds. The default ranking uses Mutual Information (MI) score, which measures exclusivity - words strongly associated with the target despite low absolute frequency. MI prioritizes statistical significance over mere frequency, meaning a word appearing only 3 times with strong exclusivity may rank higher than one appearing 22 times with weak exclusivity. Researchers can alternatively sort by raw frequency for different analytical perspectives.

Corpus linguistics provides association measures combining frequency with other statistical properties like exclusivity or dispersion. Mutual Information (MI) expresses the ratio between collocation features and random co-occurrence frequency, treating corpus like a box with words randomly selected. Log Dice uses harmonic mean of two proportions expressing tendency of words to co-occur relative to their individual frequencies, removing the random co-occurrence element and highlighting exclusivity.

This section introduces collocation analysis, which quantifies the strength of word co-occurrence by comparing observed frequencies to expected frequencies under statistical independence. The expected frequency is calculated by multiplying individual word frequencies and dividing by total corpus size. T-score measures collocation strength by calculating the difference between observed and expected frequencies, normalized by the square root of observed frequency. A T-score of 2-3 indicates a strong collocation. MI Score (Mutual Information) uses logarithmic scaling to measure association strength, where a score of 3 indicates words co-occur approximately 8 times more frequently than expected. Log-likelihood provides an alternative measure less affected by corpus size.

Different statistical methods calculate keyness differently. Log likelihood measures frequency differences between corpora. Text dispersion measures how many files a word appears in across both corpora—words appearing in unusually many files in the target versus reference get high keyness values. Researchers can choose different probability measures (log likelihood, text dispersion) and set p-value thresholds (default 0.05) to control statistical significance. Bonferroni correction adjusts for multiple comparisons.

Collocation analysis uses statistical measures to identify strong associations between words. Mutual Information (MI) is commonly used to identify collocations with strong lexical associations. Different statistical measures (like chi-square) will produce different lists of collocations. Researchers should choose measures appropriate for their analysis goals, as some measures favor content words (nouns, verbs, adjectives) while others may highlight function words.

Genre analysis is a research methodology that systematically examines all aspects of texts within a specific genre to understand social contexts, reveal typical organizational patterns and language features, and explain how social factors influence textual realizations; the CARS model (Create a Research Space), developed by linguist Jean Swales, provides a three-move framework for analyzing research introductions: Move 1 (Establishing Territory) involves claiming centrality, making topic generalizations, and reviewing previous research; Move 2 (Establishing a Niche) identifies research gaps through counter-claiming, gap indication, or continuing traditions; and Move 3 (Occupying the Niche) outlines research purposes, announces principal findings, and indicates article structure.

This segment presents the foundational framework for analyzing research article introductions. It establishes that introductions play a pivotal role in exhibiting research novelty and measuring contributions to international publications (Swales 1987, Samage 2002). The genre approach move analysis, developed since Swales' seminal work 'Creating a Research Space' (1981), provides a systematic framework for examining how writers organize their introductions through specific rhetorical moves. The study employs a qualitative comparative design, analyzing a corpus of eight introductions from experienced Indonesian authors across soft and hard sciences. Data collection involved obtaining permissions, selecting best papers via Google Scholar, extracting introductions, and conducting collaborative online analysis sessions to ensure inter-rater reliability.

This presentation critically examines the Swalesian genre approach in English for Academic Purposes, highlighting key issues including the need for clearer definitions of communicative versus social purpose, the potential overemphasis on linear move structures while underestimating cyclic patterns in texts, and the limited empirical evidence supporting the effectiveness of genre-based teaching in EAP classrooms, calling for more systematic research on how genre awareness impacts student writing performance.

Genres are institutionalized categories of language use that shape how we think and communicate, serving as frameworks for social action rather than mere text types; they are dynamic and evolve through repeated performances by different people in different contexts, exhibiting prototypicality and structural patterns such as general-to-specific or specific-to-general organization, which EAP instruction should teach students to recognize and apply through rhetorical consciousness raising and corpus-based analysis.

Research article introductions follow identifiable rhetorical structures known as 'moves' that convey specific meanings. Based on corpus linguistics research by John Swales, introductions typically contain three main moves: establishing territory (defining the research area), identifying a niche (demonstrating a gap in existing research), and addressing the niche (proposing how the research fills this gap). Within each move, there are specific steps such as indicating gaps, proposing hypotheses, highlighting problems, presenting justifications, and raising research questions. These patterns vary across disciplines, requiring discipline-specific corpus analysis for effective teaching.

Advances in internet technology and increased availability of electronic newspaper formats have enabled more quantitative methods in media discourse analysis. The combination of Critical Discourse Analysis with Corpus Linguistics creates what O'Keef describes as a sharp analytical tool. By using corpora in their studies, CDA specialists can identify how ideology and cultural mindset are communicated across different times and mediums, enabling systematic pattern identification that was previously difficult to achieve.

CDA reveals ideology by examining how it is encoded in the structures and strategies of discourse—in what is said and how it is said and what is left unsaid. Fairclough emphasizes that ideology is often embedded in linguistic form in word choice, grammatical patterns, and discursive organization. At the textual level, CDA investigates specific linguistic features: lexicalization (word choice that encodes value judgments—for instance, calling a group 'freedom fighters' versus 'rebels' signals opposing ideological positions); transitivity (which presents actors and recipients of actions—for example, a headline like 'teachers strike over pay dispute' positions teachers as active agents while 'students miss classes due to strike actions' positions students as passive victims); modality (expressions of certainty or obligation like 'must,' 'should,' or 'may' indicate authority, evaluation, or bias in the speaker's stance).

Critical discourse analysis reveals how language perpetuates social inequalities through ideological framing and presupposition. News articles about social movements often give excessive attention to secondary details while relegating core messages to background, reinforcing existing power structures. Questions like 'Does your boyfriend help you with the house?' presuppose that women bear primary household responsibility while men have optional collaboration. To analyze discourse critically, one must ask counterfactual questions: How else could this message be expressed? What would be different? This reveals what discourse explicitly states versus what it omits. Language users bear responsibility for creating realities through their speech and writing, requiring critical awareness of linguistic choices for social progress.

Corpus-based CDA uses computer-assisted techniques to reduce research bias through quantitative analysis. A corpus is a large electronic text sample representing specific language varieties or topics. Tools enable statistical measures like keywords, collocations, and n-grams to identify significant linguistic patterns. Collocation analysis examines which words statistically co-occur with targets, following the principle that 'a word is known by the company it keeps.' In OCD research, this might reveal connections between OCD and anxiety/depression, identify first-person accounts, analyze how sufferers are constructed linguistically, and detect colloquial uses. While corpus methods decrease cherry-picking risk, they still require qualitative researcher decisions about which findings are most significant.

Corpus-based discourse analysis employs computational methods from corpus linguistics—including collocates, collocational networks, concordances, and keywords—to systematically identify and analyze discourses in large collections of naturally occurring texts, enabling researchers to uncover non-obvious patterns of ideological meaning that might be missed through traditional qualitative analysis alone.

Corpus linguistics offers two pedagogical approaches: indirect application uses corpus data to inform syllabus design, materials development, and curriculum planning before instruction; direct application integrates corpus tools into classroom activities where students actively explore language patterns. Data-driven learning (DDL) extends this by having learners base their discoveries on empirical corpus findings rather than intuition. Teachers can build custom corpora from learner texts to identify common errors and develop targeted interventions, creating personalized evidence-based teaching materials.

This section introduces corpus linguistics as a principled collection of natural texts stored on computers for qualitative and quantitative analysis. It covers three main teaching applications: highlighting student work issues, suggesting alternative language, and illustrating typical language use. The presentation defines corpus characteristics (representative of specific discourse varieties, computer-stored), types (spoken, written, combined), and classifications (general vs. specialized). It introduces Data-Driven Learning (DDL) as an approach where teaching decisions are based on data analysis, tracing its history from Tim Johns' 1990-1991 papers through the 1992 predictions of revolutionizing language teaching. Scientific evidence from a meta-analysis of 64 studies by Bon and Cob demonstrates significant language gains in DDL classrooms.

Data-Driven Learning (DDL) is an educational approach that involves the direct, hands-on use of language corpora by learners to discover linguistic patterns such as collocations, frequency distributions, and grammatical constructions, enabling them to acquire language through inductive learning based on authentic language data rather than rule-based instruction; this approach operates through two main modes—direct (where learners consult corpora themselves using tools like CorpusMate) and indirect (where teachers prepare corpus-derived materials)—and follows a four-stage framework of illustration, interaction, induction, and intervention to facilitate meaningful language learning experiences.

This extended section presents the complete framework for implementing data-driven learning (DDL) in language education. It begins with the theoretical foundation established by Tim Johns in 1991, positioning DDL as an educational approach where learning decisions emerge from data analysis rather than teacher intuition. The section presents research evidence from Bon and Cobb's meta-analysis of 64 studies, demonstrating DDL's effectiveness in improving language acquisition. Multiple pedagogical benefits are articulated: authentic data presentation replacing textbook contrivances, pattern identification over rule memorization enabling informed stylistic choices, discourse community conventions for specialized fields, learner autonomy through self-directed corpus exploration, and research-informed curriculum design. The section addresses three implementation challenges: lack of teacher training, misunderstanding that personal corpora compilation is necessary, and perception that creating materials requires complex research projects. Solutions are provided through basic terminology instruction and starting with existing online corpora. The section concludes with detailed demonstrations of core corpus functionalities—concordance lines revealing authentic usage patterns like 'nonetheless' at sentence beginnings, and KWIC analysis showing 'between' vastly outnumbers 'from' in academic writing for preposition selection. These tools transform abstract grammar instruction into concrete pattern recognition, helping students develop metalinguistic awareness and self-editing skills through exposure to authentic language data.

Data-driven learning (DDL) combines corpus linguistics applications with language teaching and learning. At the individual level, it follows a constructivist approach where learners engage with multiple sources of information including language data, statistical information about language use, and frequency information. In classroom contexts, DDL incorporates social-cognitive approaches where peers and teachers collaborate using scaffolded instructional materials. Research supports cognitive benefits beyond language learning, though specific areas require further investigation. Traditional DDL involves reading and manipulating concordance lines—the most typical corpus output—where users query terms and receive results showing statistical regularities of word usage, including collocations and contextual patterns.