Academic Genre Classification in Higher Education: A Systematic Review of Theoretical Traditions, Empirical Approaches, and Implications for Educational Assessment Genres

Abstract

Academic genre classification—the practice of distinguishing and labeling recurring forms of academic communication (e.g., essays, case studies, proposals, lab reports, literature reviews)—has become an increasingly consequential topic in higher education. It underpins curriculum design, writing pedagogy, educational assessment, learning analytics, and (more recently) computational tools that attempt to detect or evaluate student writing. Yet “genre” is theorized differently across research traditions, and classification methods vary widely in their assumptions, units of analysis, and validation practices.

This review synthesizes peer-reviewed and authoritative scholarship on academic genre and genre classification with a specific focus on academic assessment genres. Drawing on rhetorical genre studies, English for Specific Purposes (ESP), systemic functional linguistics (SFL), corpus/register studies, and computational genre identification, we examine (a) what is being classified (texts, assignments, tasks, or activity systems), (b) how classification is performed (qualitative typologies, corpus-driven dimensional analyses, machine-learning models, or mixed methods), and (c) how classifications are validated and used in educational contexts. We also introduce an integrative framework— Multi-Level Academic Genre Classification (ML-AGC)—that distinguishes genre labels across four analytically separable levels: (1) institutional activity/assessment function, (2) genre family and communicative purpose, (3) rhetorical organization (moves/steps), and (4) lexico-grammatical/register realization.

Across traditions, evidence converges that academic genres are socially situated and disciplinary, that boundaries are often fuzzy or hybrid, and that classification quality depends on explicit purpose statements and robust reliability/validity checks. For stakeholders, the review highlights how genre classification can (i) increase transparency and fairness in assessment design, (ii) support genre-aware pedagogy, and (iii) inform responsible uses of automated classification. The article concludes with a research agenda emphasizing multi-label/hierarchical classification, cross-disciplinary comparability, multilingual academic genres, and alignment between assessment goals and academic genre expectations.

Introduction

Universities rely on recurring forms of communication to teach, evaluate, and certify knowledge. Students write essays, lab reports, case studies, literature reviews, design specifications, reflective journals, and proposals; they also produce multimodal work such as posters, slide decks, annotated bibliographies, and portfolios. Professors use these academic genres not only to assess content knowledge but also to apprentice learners into disciplinary ways of reasoning, evidencing, and arguing (Hyland, 2004; Swales, 1990). As a result, academic genre classification—naming, grouping, and distinguishing genres—matters to curriculum design, educational assessment, and equity in student participation.

Research on genre classification spans several intellectual lineages. In rhetorical genre studies (RGS), genres are typified social actions rather than static forms (Miller, 1984), and classification attends to recurrent situations, motives, and institutional activity (Bazerman, 1994; Russell, 1997). In English for Specific Purposes (ESP), genre analysis often centers on communicative purposes and conventionalized rhetorical “moves” in professional and academic texts (Bhatia, 1993; Swales, 1990, 2004). In systemic functional linguistics (SFL), genres are staged, goal-oriented social processes realized through register choices shaped by field, tenor, and mode (Martin & Rose, 2008). Corpus linguistics and register studies quantify language variation and offer empirically grounded ways to distinguish registers and text types (Biber, 1988; Biber & Conrad, 2009). More recently, computational research has treated genre classification as a document classification task using structural, linguistic, and metadata features (Finn & Kushmerick, 2006; Kessler et al., 1997; Karlgren & Cutting, 1994).

Despite substantial scholarship, academic genre classification remains challenging in practice for at least four reasons. First, genre labels are often locally negotiated (e.g., what counts as a “report” in engineering vs. psychology). Second, genres frequently hybridize or embed other genres (e.g., a research proposal that includes a mini literature review). Third, the unit of classification varies: scholars may classify whole texts, sections (e.g., introductions), assignment prompts, or broader activity systems (Bazerman, 1994; Swales, 2004). Fourth, in educational assessment contexts, a single assignment may simultaneously pursue multiple learning purposes (e.g., conceptual understanding and professional communication), complicating a one-label taxonomy (Nesi & Gardner, 2012).

This review addresses the following questions:

• RQ1 (Conceptual): How do major genre traditions conceptualize “genre” and what implications follow for academic genre classification?
• RQ2 (Methodological): What empirical approaches have been used to classify academic genres, including academic assessment genres, and how are classifications validated?
• RQ3 (Applied): How can insights from genre classification research inform stakeholders in higher education (faculty, students, writing programs, assessment designers, and educational technology developers)?

We focus on higher education and prioritize academic assessment genres because classification affects what students believe is valued, how they allocate effort, and whether assessment expectations are transparent and fair (Boud & Falchikov, 2007; Wiggins, 1998). At the same time, we treat genre classification as an interdisciplinary problem spanning social action, discourse form, and measurable linguistic patterns.

Conceptual Foundations: What “Genre” Means in Genre Classification

Genre as Social Action (Rhetorical Genre Studies)

RGS foregrounds the idea that genres are typified responses to recurrent situations. Miller’s (1984) influential formulation defines genre as “social action,” shifting emphasis from formal features to the ways texts mediate activity, motives, and relationships. From this perspective, classification requires attention to what people are doing with texts in institutional settings, not simply how texts look. Bazerman (1994) extends this view by proposing “systems of genres,” where one genre (e.g., a call for proposals) triggers others (e.g., proposals, reviews, decisions), producing a coordinated sequence of actions. Russell (1997) similarly analyzes genre in activity systems, arguing that genres cannot be understood apart from the tools, roles, and objectives of a community.

Implication for academic genre classification: labels should be tied to recurrent educational situations and assessment functions (e.g., “demonstrate disciplinary argumentation under timed conditions,” “document empirical procedures,” “propose and justify a design”). Formal similarity alone is insufficient because two texts may share features but perform different actions in different disciplines or courses.

ESP Genre Analysis: Communicative Purpose and Rhetorical Moves

ESP genre analysis has been particularly influential in academic contexts because it provides operational tools for describing and teaching genre conventions. Swales (1990) defines genre as a class of communicative events with shared purposes recognized by a discourse community, while Bhatia (1993) emphasizes the relationship between genre, professional practice, and conventionalized rhetorical organization. A hallmark of ESP work is “move analysis,” which identifies functional rhetorical units (moves and steps) used to achieve a genre’s purposes (Swales, 1990, 2004).

Implication for academic genre classification: genres can be distinguished by their communicative purposes and by recurrent rhetorical structures (e.g., research article introductions vs. grant proposals). ESP approaches also support classification at multiple grains: the whole genre, its typical sections, and its moves.

SFL Perspectives: Genre, Register, and Staged Social Processes

In SFL-informed genre pedagogy, genres are staged, goal-oriented social processes, and differences among genres are realized through register configurations—patterns of meaning shaped by field (what is happening), tenor (roles/relationships), and mode (channel/organization of language) (Martin & Rose, 2008). SFL traditions often connect genre classification to pedagogy by making discourse stages explicit (e.g., “exposition,” “discussion,” “explanation”) and teaching how language choices realize those stages.

Implication for academic genre classification: classification can be grounded in stage models and linked to teachable language resources (e.g., nominalization, technicality, appraisal). However, educational assessment genres may not align neatly with a single stage model when assignments require multiple epistemic activities (describe, analyze, argue, reflect) in one product.

Corpus/Register Studies: Dimensional Variation and Empirical Text Types

Corpus linguistics and register studies approach classification by measuring linguistic variation across texts and interpreting systematic patterns. Biber’s (1988) multidimensional analysis demonstrates that texts vary along multiple co-occurring linguistic dimensions (e.g., involved vs. informational production), providing a quantitative basis for describing registers and “text types.” Register approaches often distinguish register (situationally defined varieties) from genre (conventional text categories), while acknowledging overlap and frequent conflation in everyday labeling (Biber & Conrad, 2009).

Implication for academic genre classification: some distinctions are better captured by continuous dimensions rather than discrete labels, and genres may cluster into families based on shared distributions of linguistic features. This is especially relevant for large-scale educational assessment genres, where discrete labels may oversimplify variation.

Academic Literacies and Educational Assessment Genres

Academic literacies research emphasizes that academic writing is embedded in power relations and institutional norms and that students must negotiate disciplinary epistemologies and identity positions through writing (Lea & Street, 1998; Lillis, 2001; Lillis & Scott, 2007). This tradition has been cautious about treating genre as merely a stable set of textual conventions, arguing instead that writing expectations are contested and often tacit, which can disadvantage students without access to implicit norms.

Implication for academic genre classification: classification is not neutral. The act of naming and stabilizing genres can increase transparency (a potential equity benefit) but can also reify norms, marginalize alternative ways of meaning-making, or obscure disciplinary disagreement. In assessment contexts, classification should therefore be paired with explicit discussion of purposes, criteria, and legitimate variation.

Method: Review Design and Evidence Identification

Review Approach

This article is a systematic, conceptually structured review that synthesizes authoritative and peer-reviewed work across applied linguistics, rhetoric/composition, educational assessment, and computational text classification. Because genre classification research is dispersed across disciplines and publication venues (journals, edited collections, conference proceedings, and books), the review follows systematic principles (transparent scope, explicit inclusion criteria, and structured synthesis) while also adopting interpretive integration appropriate to heterogeneous theoretical traditions (Petticrew & Roberts, 2006).

Guiding Reporting Standards

We used PRISMA guidance as a reporting and planning heuristic for documenting search and selection decisions (Moher et al., 2009; Page et al., 2021). Given that the present article also maps conceptual terrain and typologies (in addition to aggregating findings), it overlaps with scoping review logic (Tricco et al., 2018). Where the review makes illustrative claims (e.g., example workflows), these are labeled explicitly as author-generated representations rather than empirical counts.

Search Strategy (Conceptual and Reproducible Template)

A reproducible search strategy for academic genre classification typically spans databases that index applied linguistics, education, and information science (e.g., ERIC, Linguistics and Language Behavior Abstracts, Scopus, Web of Science, ACL Anthology). Search strings generally combine terms for genre and classification with educational contexts. Examples include:

• "academic genre" AND classification
• genre analysis AND (university OR higher education)
• (assignment OR assessment) AND genre
• automatic genre identification
• document genre classification AND academic
• student writing corpus AND genre

Because books and edited volumes are central in genre scholarship, backward and forward citation tracking is essential (Petticrew & Roberts, 2006). In practice, reviews that omit books risk systematically underrepresenting RGS, ESP, and SFL contributions.

Inclusion/Exclusion Criteria

Included sources met at least one of the following conditions:

• Theory-building on genre definitions with clear implications for classification in academic settings (e.g., Miller, 1984; Swales, 1990).
• Empirical classification of academic genres, including student writing or academic/professional academic texts (e.g., Gardner & Nesi, 2013).
• Computational approaches to genre identification with relevance to academic texts or transferable methods (e.g., Kessler et al., 1997; Finn & Kushmerick, 2006).
• Assessment-focused scholarship linking writing tasks/genres to learning purposes, transparency, or institutional practice (e.g., Boud & Falchikov, 2007; Wiggins, 1998).

Excluded sources were those that (a) addressed “genre” only metaphorically without operational implications for classification, (b) focused solely on K–12 genres without transferable higher education implications, or (c) treated automated essay scoring without meaningful attention to genre distinctions (though we refer to automated assessment sparingly when relevant).

Data Extraction and Synthesis Logic

To synthesize heterogeneous studies, we extracted (when available) the following elements: object of classification (text, prompt, section, activity), theoretical frame (RGS/ESP/SFL/register/computational), data type (student writing, expert writing, web documents, corpora), labeling scheme (single-label vs. multi-label; flat vs. hierarchical), features used (moves, structural cues, lexico-grammar, citations, metadata), and validation method (inter-rater reliability, triangulation, predictive performance metrics). We then integrated findings into a cross-tradition typology and derived implications for educational assessment genres.

PRISMA Flow (Illustrative)

Many published reviews provide a PRISMA flow diagram reporting records identified, screened, and included (Page et al., 2021). Because this article synthesizes across books and disciplinary traditions where comprehensive counting depends on database access and institutional holdings, we include an illustrative PRISMA-style flow that readers can adapt for future fully enumerated reviews.

Results: What Is Being Classified in Academic Genre Classification?

Objects of Classification: Texts, Tasks, Prompts, and Activity Systems

A key finding across traditions is that “academic genre classification” refers to multiple distinct objects. Confusion often arises when stakeholders assume that the genre of an assignment prompt is identical to the genre of the student submission , or that the genre of a text is identical to the genre of the broader activity system in which it circulates (Bazerman, 1994; Russell, 1997).

• Texts (products): Completed student papers, published research articles, theses, or grant proposals (Swales, 2004).
• Prompts/tasks (instructions): Assignment briefs that specify actions (analyze, compare, propose) and constraints (length, sources), sometimes functioning as “pre-genres” that shape textual outcomes.
• Sections/moves: Sub-genres or functional units such as introductions, methods sections, or literature reviews embedded within larger texts (Swales, 1990, 2004).
• Genre systems/ecologies: Sets of interrelated genres coordinating educational work (syllabi → assignments → drafts → feedback → revisions → grading rubrics) (Bazerman, 1994).

For educational assessment genres, classification benefits from explicitly stating which object is being classified. A “case study” may refer to (a) a teaching case provided by the instructor, (b) an assignment genre requiring application of theory to an authentic scenario, or (c) a student’s case-study report as a textual product.

Granularity: From Macro-Genres to Genre Families to Micro-Structures

Genre labels operate at different levels of abstraction. Some schemes classify at a macro level (e.g., “essay,” “report”), while others identify genre families with more specific communicative purposes and disciplinary patterns. A major contribution to academic assessment genres is the classification of student writing in university contexts into genre families and subtypes (Gardner & Nesi, 2013; Nesi & Gardner, 2012). These works show that university assignments often cluster into families such as critiques, explanations, narrative recounts, design specifications, and proposals, but that the same surface label (“report”) can mask substantively different rhetorical purposes across courses.

Disciplinary Context and the Problem of Label Equivalence

Disciplinary discourse communities stabilize expectations for evidence, argument, and stance (Hyland, 2004). Consequently, the same genre label can denote different communicative purposes across disciplines. For example, a “discussion” in a lab report may function to interpret results in relation to hypotheses and prior studies, while a “discussion” in a humanities essay may synthesize theoretical perspectives and contest interpretive claims. A central classification challenge is therefore label equivalence : determining whether two artifacts bearing the same name belong to the same genre or merely share a label.

Results: How Academic Genre Classification Is Conducted

Approach 1: Qualitative Typologies and Expert-Elicited Taxonomies

Qualitative classification commonly relies on expert judgment, discourse analysis, and iterative refinement of genre categories. In RGS and academic literacies, classification often emerges from ethnographic attention to institutional contexts and writing practices (Bazerman, 1994; Lea & Street, 1998). In ESP and SFL, analysts may develop teaching-oriented typologies that describe purposes and stages, then test them through classroom use and text analysis (Hyland, 2007; Martin & Rose, 2008).

Strengths: high interpretive validity for local contexts; sensitivity to purpose, audience, and institutional action.

Limitations: potential for limited generalizability; categories may remain underspecified for cross-institutional comparison unless operationalized with clear criteria and reliability checks (Krippendorff, 2018).

Approach 2: Corpus-Driven and Register-Based Quantitative Classification

Corpus approaches treat genre-related variation as measurable distributions of features (e.g., noun phrase complexity, stance markers, referential cohesion). Biber’s (1988) multidimensional framework demonstrates how texts can be mapped onto continuous dimensions rather than discrete bins, and later register/genre syntheses emphasize careful separation of situational characteristics, linguistic features, and functional interpretations (Biber & Conrad, 2009). Work on academic prose also shows that academic registers are not uniformly “complex” in the same way; instead, complexity is realized through specific constructions (e.g., compressed noun phrases) rather than clausal elaboration (Biber & Gray, 2010).

Strengths: empirically grounded; supports large-scale comparison across disciplines and tasks; can reveal latent groupings and dimensional continua.

Limitations: may underrepresent social action unless linked to situational analysis; dimensional similarity does not automatically imply genre equivalence.

Approach 3: Rhetorical Move/Step Analysis as a Classification Resource

Move analysis operationalizes genre knowledge by identifying functional rhetorical segments. Swales (1990) introduced move analysis as a method for analyzing research article introductions, and subsequent work has documented disciplinary variation in rhetorical organization (e.g., how fields justify research niches). For instance, Samraj (2002) shows discipline-linked variation in research article introductions, reinforcing that classification cannot assume a universal structure even within “the same” genre.

Strengths: connects classification to communicative purpose and pedagogy; supports fine-grained feedback.

Limitations: move boundaries can be ambiguous; reliability depends on strong coding manuals and training; some assessment genres mix multiple rhetorical sequences.

Approach 4: Computational Genre Identification and Machine Learning

Computational work often frames genre classification as a supervised learning problem: given labeled documents, learn a model that predicts genre for new documents (Finn & Kushmerick, 2006; Kessler et al., 1997). Early approaches used relatively simple surface metrics and discriminant analysis (Karlgren & Cutting, 1994), while later work expanded feature sets to include part-of-speech patterns, structural cues, and metadata (Finn & Kushmerick, 2006). Even when the target is not specifically “academic genre,” these methods are relevant because they operationalize genre signals in measurable ways.

For higher education stakeholders, computational classification is attractive for triage (sorting submissions), learning analytics (mapping task distributions), and writing support (genre-aware feedback). However, the educational validity of such systems depends on alignment with theoretically meaningful genre constructs and careful evaluation against human judgments and institutional definitions (Biber & Conrad, 2009; Miller, 1984).

Reliability and Performance Metrics Used in Classification

Genre classification studies that involve human labeling typically assess inter-rater agreement. A widely used coefficient is Cohen’s kappa (Cohen, 1960). When genre classification is automated, evaluation often reports precision, recall, and F1-score.

In Equation (1), is precision and is recall. For inter-rater reliability, Cohen’s kappa is commonly defined as:

In Equation (2), is observed agreement and is expected agreement by chance (Cohen, 1960). For genre classification in educational assessment contexts, reporting reliability is not merely a technical detail; it is part of validity evidence because unstable labels undermine downstream interpretations and fairness (Krippendorff, 2018).

Structured Synthesis: A Cross-Tradition Typology of Academic Genre Classification

A Four-Level Framework: Multi-Level Academic Genre Classification (ML-AGC)

To integrate disparate approaches, we propose Multi-Level Academic Genre Classification (ML-AGC) , a framework that distinguishes four levels at which “genre” claims are often made. The key premise is that disagreements about genre classification frequently stem from mixing levels (e.g., using a form label to stand in for an assessment function).

ML-AGC is designed for academic genre classification in educational assessment settings where stakeholders require both interpretability (Why is this a “proposal”?) and operationalizability (What features should be taught/assessed?). The four levels are:

• Level 1: Assessment/Activity Function (RGS/assessment emphasis). What institutional action is the text performing (e.g., demonstrating mastery, rehearsing disciplinary inquiry, simulating professional practice)? (Bazerman, 1994; Wiggins, 1998).
• Level 2: Genre Family & Communicative Purpose (ESP emphasis). What is the dominant purpose (argue, explain, critique, propose, design, reflect), and how is the audience positioned (Swales, 1990; Bhatia, 1993)?
• Level 3: Rhetorical Organization (ESP/SFL emphasis). What staged structure or move pattern typically realizes the purpose (Swales, 2004; Martin & Rose, 2008)?
• Level 4: Register/Linguistic Realization (corpus emphasis). What lexico-grammatical patterns and discourse features tend to cluster with the genre family in the given context (Biber, 1988; Biber & Gray, 2010)?

ML-AGC is not a claim that every genre has a single stable realization. Rather, it separates analytical questions so that classification schemes can specify: What exactly is being classified and for what downstream use?

Mapping Major Traditions to Classification Decisions

Table 1 summarizes how major traditions tend to operationalize genre and what that implies for academic genre classification, especially for academic assessment genres.

Academic Assessment Genres as “Institutionalized Hybrids”

University assessment tasks frequently blend multiple communicative purposes: explaining a concept while arguing a position, reporting results while reflecting on limitations, or proposing a project while reviewing literature. This hybridity is not merely an analytical inconvenience; it reflects real educational design choices. Nesi and Gardner (2012) and Gardner and Nesi (2013) demonstrate that student writing genres cluster into families yet still exhibit internal variation and cross-disciplinary adaptation. From an ML-AGC perspective, many academic assessment genres are best treated as multi-label at Level 2 (purposes) and sometimes at Level 3 (rhetorical sequences), even if instructors assign a single surface label.

Academic Genre Classification and Educational Assessment: What the Evidence Suggests

Transparency and Construct Representation

Assessment theory has long warned that what is assessed should align with intended learning outcomes and that criteria should be explicit enough to support learning (Biggs, 1996; Wiggins, 1998). Genre classification intersects with this principle because genre expectations effectively define part of the construct: what counts as competent performance is partly genre knowledge (Hyland, 2004; Swales, 1990). If an assignment requires “critical analysis” but students are not taught what a “critique” looks like in that discipline, assessment may inadvertently measure prior access to tacit genre knowledge rather than learning achieved during the course (Lea & Street, 1998; Lillis, 2001).

Genre Knowledge as Disciplinary Enculturation

Across applied linguistics and writing studies, a robust conclusion is that learning academic writing involves learning disciplinary genre knowledge—how arguments are warranted, what counts as evidence, how stance is expressed, and how writers position themselves relative to sources (Hyland, 2004; Swales, 2004). Tardy (2009) emphasizes that genre knowledge includes formal, rhetorical, process, and subject-matter dimensions, implying that classification cannot be reduced to textual form alone.

Genre, Identity, and Power in Student Writing

Academic literacies scholarship argues that classification and instruction should recognize power dynamics: institutional genre norms can regulate participation and may be experienced as gatekeeping (Lea & Street, 1998; Lillis & Scott, 2007). This does not imply that genre classification is undesirable; rather, it suggests that classification should be used reflexively: to surface expectations, legitimize variation where appropriate, and invite critical discussion of why particular genres dominate in particular disciplines.

Evidence of Linguistic Differentiation in Academic Registers

Register studies provide evidence that academic writing is not monolithic and that different academic genres/registers realize complexity differently. For example, Biber and Gray (2010) argue that stereotypes about academic writing being defined by clausal complexity overlook the role of phrasal compression in academic prose. This matters for educational assessment genres because instructors may conflate “mature academic writing” with vague notions of complexity, when in fact genre-appropriate complexity is patterned and teachable.

Practical Synthesis: Designing and Using Genre Classifications in Higher Education

A Stakeholder-Oriented View

Academic genre classification is most useful when designed for explicit stakeholders and decisions. The same classification may be inadequate for different uses: a writing center may need pedagogically meaningful genre families; a department may need an assessment map of tasks; an NLP team may need labels that are consistent enough for model training.

Implications for Faculty and Assessment Designers

For instructors, genre classification can function as a design and communication tool:

• Align genres to learning outcomes: Use constructive alignment to ensure the genre demanded matches the capability being taught (Biggs, 1996). For instance, if the outcome is “evaluate competing explanations,” an “explanation” genre may be insufficient; a “comparative critique” or “argument” genre may be more aligned.
• Reduce tacit expectations: Provide exemplars and explicit genre criteria (Hyland, 2007; Swales, 1990). A stable classification scheme can make it clearer what “counts” as analysis, synthesis, or reflection in that course.
• Differentiate labels from functions: Avoid relying solely on local labels like “report.” Instead, state the communicative purpose (e.g., “investigative report interpreting primary data”) and intended audience (e.g., “professional client,” “peer researcher”). This reduces label ambiguity.

Implications for Students and Writing Support Units

For students, classification supports “genre awareness”: recognizing that writing expectations shift across disciplines and tasks, and learning strategies for analyzing new assignments. Pedagogically, ESP and SFL traditions have developed explicit instructional approaches that teach how rhetorical purposes map onto staged structures and language choices (Hyland, 2007; Martin & Rose, 2008). Writing centers can use genre family classifications to triage support materials and workshops (e.g., sessions on proposals vs. critiques) while acknowledging local disciplinary variation (Nesi & Gardner, 2012).

Implications for Institutions: Assessment Ecology and Curriculum Mapping

At institutional scale, classifying academic assessment genres can support curriculum mapping: identifying whether programs overuse a narrow set of genres (e.g., timed essays only) or whether students experience a coherent progression of genres aligned with disciplinary enculturation. From an RGS view, institutions can also examine genre systems: how rubrics, feedback forms, and learning management system tools shape what students produce (Bazerman, 1994).

Implications for Educational Technology and Automated Classification

Automated genre classification can support discovery and organization (e.g., sorting resources, identifying task types). But applying computational genre identification in educational assessment raises validity and fairness concerns. If a model is trained on inconsistent labels or on historically biased patterns, it may perpetuate inequities. Therefore:

• Prefer interpretable features where possible: Structural cues, citation patterns, and rhetorical move indicators may be more instructionally actionable than opaque latent embeddings alone.
• Use multi-label and hierarchical schemes: Because assessment genres are often hybrid, forcing single-label prediction can misrepresent the task and produce misleading analytics (Gardner & Nesi, 2013).
• Validate against human judgment and institutional definitions: Predictive performance should not substitute for construct validity. Human-in-the-loop adjudication and transparent labeling manuals are essential (Krippendorff, 2018).

Tools for Implementation: Two Applied Frameworks

Framework 1: The Genre–Assessment Alignment Matrix (GAAM)

To translate classification into assessment design, we propose a simple planning tool: the Genre–Assessment Alignment Matrix (GAAM) . GAAM is not a psychometric model; it is a structured way to ensure that an assignment’s genre expectations are aligned with its learning purposes and that hybrid purposes are made explicit.

Framework 2: A Minimal Labeling Manual for Academic Genre Classification

Whether classification is conducted by faculty committees, researchers, or annotators labeling corpora, reliability depends on explicit definitions. Drawing on content analysis principles (Krippendorff, 2018), a minimal labeling manual for academic genre classification should include:

• Label name and alternative local names (e.g., “proposal,” “project pitch”).
• Inclusion criteria (what must be present).
• Exclusion criteria (boundary cases and what to label instead).
• Primary purpose and typical audience.
• Typical move structure (optional but useful).
• Examples (annotated exemplars and near-misses).
• Decision rules for hybrids (multi-label permitted? hierarchical tie-breakers?).

This manual supports both qualitative and computational classification by making categories auditable and teachable.

Discussion

Convergences Across Traditions

Despite theoretical differences, several convergent insights emerge:

• Genre is contextual: Classifications that ignore disciplinary and institutional context risk mislabeling or oversimplifying academic genres (Hyland, 2004; Russell, 1997).
• Genres are systems, not isolated forms: Especially in assessment, genres are linked to prompts, rubrics, feedback, and revision cycles (Bazerman, 1994; Wiggins, 1998).
• Boundaries are often fuzzy: Hybridization is common in academic assessment genres, suggesting that multi-label or hierarchical classification often better matches reality than single-label taxonomies (Gardner & Nesi, 2013; Nesi & Gardner, 2012).
• Operationalization matters: Whether through move coding, dimensional feature analysis, or machine learning, classification quality depends on explicit definitions and validation evidence (Biber & Conrad, 2009; Krippendorff, 2018).

Persistent Tensions and Open Problems

Genre vs. Register vs. Task

A recurring problem is the conflation of genre, register, and task. Register studies emphasize situational variables and linguistic distributions (Biber & Conrad, 2009), while RGS emphasizes social action (Miller, 1984). In educational settings, “task” may refer to what students are asked to do (compare, critique, propose) rather than the text produced. ML-AGC helps by separating assessment function, genre family, rhetorical organization, and register realization, but empirical research is still needed to validate how these levels interact in authentic curricula.

Local Stability vs. Cross-Institutional Portability

Genre labels are often locally stabilized: a department may have shared understandings of “case study” or “reflection,” but those labels may not transfer. Cross-institutional classification therefore requires either (a) very abstract categories (risking loss of pedagogical specificity) or (b) carefully mapped genre families with explicit definitions and examples. The BAWE-related work on genre families provides one model for balancing specificity with generalizability (Gardner & Nesi, 2013; Nesi & Gardner, 2012).

Validity and Fairness in Automated Genre Classification

Computational models can classify texts by detecting patterns that correlate with labels, but correlation is not validity. If labels reflect institutional bias, models may reproduce it. Moreover, models may rely on superficial cues (formatting, headings) that are not the intended construct of assessment. Responsible use therefore requires a validity argument that links automated outputs to defensible interpretations and decisions, with clear limitations stated (Krippendorff, 2018; Wiggins, 1998).

A Research Agenda for Academic Genre Classification

• Multi-label and hierarchical annotation: Develop and test schemes that represent hybrid academic assessment genres, and compare them to forced single-label schemes.
• Prompt–product relationships: Study how assignment prompts (as genres themselves) shape student texts and how misalignment contributes to evaluation disputes.
• Longitudinal genre development: Track how students acquire genre knowledge across a program and which sequences of assessment genres best support enculturation (Hyland, 2004; Tardy, 2009).
• Cross-disciplinary comparability: Build mappings that preserve disciplinary specificity while enabling institutional curriculum analytics.
• Multimodal academic genres: Extend classification beyond alphabetic text to posters, slide decks, and portfolios, where mode and medium reshape genre realization.
• Open datasets and reproducible methods: Encourage shared, well-documented corpora and labeling manuals to improve comparability and cumulative knowledge, especially for computational studies.

Conclusion

Academic genre classification is not merely a descriptive exercise; it is a consequential practice that shapes teaching, learning, and educational assessment in higher education. Across RGS, ESP, SFL, register studies, and computational approaches, the literature suggests that genres are socially situated, disciplinary, and often hybrid—especially in academic assessment genres. Classification efforts therefore require explicit statements about what is being classified (text, task, prompt, or activity), what level of granularity is intended, and what evidence supports reliability and validity.

This review contributed (a) a cross-tradition synthesis of classification approaches, (b) the ML-AGC framework to clarify levels of genre claims, and (c) practical tools (GAAM and a minimal labeling manual) to support transparent assessment design. Future work should prioritize multi-label/hierarchical models, stronger validation practices, and equity-aware uses of classification in pedagogy and educational technology. When thoughtfully designed, academic genre classification can serve as a bridge between disciplinary enculturation and fair, educative assessment.

References