Review

This file reads the literature on typed units of writing. Existing reviews come first, then the seminal theory, Information Mapping read first-hand. DITA, reader studies, transfer, cost, models as readers, and practice follow.

How numbers and scores are marked: the evidence README.

0. Existing reviews

Review	Scope	Finding for this review	Read
Hebert 2016	meta-analysis, 45 studies, grades 2 to 12, expository text structure instruction	effective on proximal measures, and across three levels of transfer: maintenance, untaught text structures, general comprehension; far transfer small and inconsistent	abstract (secondary)
Pyle 2017	meta-analysis, 19 of 21 studies, kindergarten to grade 12, 1970-2013	mean effect 0.95; larger with one or two structures taught, 11 to 20 hours, researcher implementers	abstract (secondary)
Bogaerds-Hazenberg 2020	meta-analysis, 44 studies, grades 4 to 6, informational and narrative	immediate g = 0.25 (questions), 0.57 (summary), 0.37 (recall), 0.38 (structure knowledge); no advantage at delayed posttest	abstract (secondary); the thesis chapter read for transfer (Bogaerds-Hazenberg 2023)
Eiriksdottir 2011	review of instructions: procedures, principles, examples	specific instructions help first performance; general ones help learning and transfer	abstract
Andersen 2015; Batova 2014; Batova 2017	integrative and systematic reviews of component content management	none extracted here	title and venue only here; read in document-model

• No review was found of DITA, Information Mapping or topic types measured on readers. That is a statement about this search.
• The three meta-analyses test teaching readers the types, not writing to a type. Their effect is a lower bound on what a schema in the reader can do, not a test of a template.

1. Seminal theory: schematic structures and story grammars

• Kintsch 1978 (read first-hand, section "Schematic structures of discourse"): a conventional text type is a set of categories with rules for their order.
  • Its examples are a story, an argument and a psychological research report.
  • The rules of formation are "(sometimes recursive)", in the paper's words.
  • The schema controls which propositions reach the gist; it is shared by a cultural group (section "Role of the schema").
• Rumelhart 1975: the first schema for stories (title and venue only).
• Mandler 1977: an analysis of simple stories; children and adults recall them differently, so schemata develop (ERIC abstract).
• Thorndyke 1977: two experiments vary the amount of structure in passages and separate structure from content (ERIC abstract, secondary).
• Stein 1975: a story schema tested on 48 first and fifth graders; recall differed by age and delay, but which categories were recalled was stable across stories (ERIC abstract, secondary).
• Mandler 1978: the story schema used in retrieval (title and venue only).
• Black 1979 evaluated the grammars: one formally adequate, many stories not generated, one class of non-story generated, nothing added over content models (abstract, secondary).
• Trabasso 1985: causal networks of events, the content-based account that followed (title and venue only).

2. Information Mapping, read first-hand

Horn 1969 (the 1969 Air Force report, read in full scan) replaces the encyclopedia page that reader-ergonomics cited under the same key.

• Units: separable information blocks, each with a marginal label, grouped into information maps; each map type has its own format (chapter 1).
• Eight map types in the appendix chart: concept, structure, process, procedure, classification, decision, fact, proof (pp. 122-123).
• Derivation from a base already appears: a structure map takes every block of the concept map; a process map takes every concept and structure block; a procedure map takes every concept block.
• The concept map is the most frequent type (chapter 2).
• The one controlled comparison (chapter 5, Harvard series):
  • same content, same questions, either as information maps (IM) or with the labels, boxes and unit pages removed (prose, P); random assignment;
  • adjusted posttest means 19.56 (IM, n = 20) against 17.65 (P, n = 21), t = 1.17, 38 df, not significant;
  • the authors offer two readings and favour the second: the one-session class test was not sensitive enough to show a difference.
• Chapter 7 warns that such evaluations test one book, not the method, and have no external reference point.
• Jonassen 1980 (journal abstract, secondary): information mapping against programmed instruction; no difference in recall, but retrieving information from the text benefited significantly from mapping.
• Later formal work, mostly not opened: Horn 1981 (structured writing for software documentation), Horn 1999 (STOP against structured writing), Fields 1981 (a survey, ERIC abstract). Also later formal work, mostly not opened: Fields 1982 (an evaluation, title only), Hartley 1981 (a critique).
• Kools 2008 varied one health text by the method: headings sped search, the number of headings did not change speed, and the method's information order had no effect (abstract).

3. DITA and topic-based authoring

• OASIS 2015 (DITA 1.3, read first-hand):
  • every topic has the same base: title, short description or abstract, prolog, body, related links, nested topics (2.2.1.6);
  • concept, task and reference derive from topic by specialization, a subtype hierarchy in which a derived type inherits meaning and default processing (2.5.3.1);
  • the @class attribute lists each element's ancestry, so any processor can treat a derived element as its ancestor (2.5.3.6); content can be generalized back to any ancestor (2.5.4.1);
  • information typing is said to rest on Horn's Information Mapping and Hughes Aircraft's STOP; no study is cited (2.2.1.4);
  • limits: sections cannot nest, every topic needs a title, and a type that breaks these is no longer DITA (B.11).
• Priestley 2001 and Priestley 2002: the IBM architecture papers on reuse and on specialization (title and venue only).
• Evia 2015: teaching DITA through rhetorical and computational thinking (title and venue only).
• Swarts 2022: a corpus of topic-based documentation; relative clauses signal context to readers who arrive at one topic alone (abstract).
• Heinonen 2022: DITA layered by skill level, tested with maintenance staff in three countries by observation, interview and questionnaire; well received (abstract, secondary; no control).

4. Does a typed structure help the reader (Q2)

Study	Manipulation	Result	Grade
Kintsch 1982	essays in good rhetorical form with cues, against bad form, no cues, deviant order	better comprehension with good form and cues on questions about the topic and main points (global); the cloze test (local) did not differ; 2 experiments, 72 students; no numbers in the abstracts	measured (ERIC and APA abstracts)
Horn 1969	typed maps against the same text as prose	no significant difference in learning	measured: one controlled test, 41 students, in a report not peer-reviewed
Kools 2008	headings and order of an information-mapped text	headings help search; amount of headings and order do not	measured (abstract)
Zwaan 1994	one text framed as literature or as news	literary framing: longer reading, better surface memory, weaker situation memory	measured (secondary)
Triantafyllopoulos 2021	the same framing, three online experiments	no difference in comprehension by genre expectation	measured (abstract, secondary)
Markevich 2025	narrative, expository and poetic texts, 44 adolescents, eye tracking	forward reading for narrative, more lookbacks for expository and poetry	measured, descriptive (abstract)
Carroll 1987; Lazonder 1993	task-oriented minimal manual against a standard manual	not extracted	title and venue only

5. Does knowing one type transfer (Q3)

• Hebert 2016: instruction transferred to untaught text structures (near transfer); far transfer was small and inconsistent (abstract).
• Bogaerds-Hazenberg 2023: some delayed posttests in the pooled studies used untaught structures; too few to test transfer (thesis, chapter on the meta-analysis).
• Meyer 2001: nine hours of structure strategy training raised adults' total recall from a variety of texts and from an informative video; interest training did not (ERIC abstract, secondary).
• Eiriksdottir 2011: abstract instructions favour transfer, specific ones favour first performance.
• Writing studies treat transfer of genre knowledge as recontextualisation (Cheng 2007; Artemeva 2008; Brent 2011; titles and venues only).
• No study was found that teaches readers a base type and tests them on a type derived from it.

6. What typing costs (Q5)

• Batova 2018, on writers' motivation in a move to component content management, is reviewed in document-model.
• Bailie 2015 and Coggio 2015: case studies of content management changing how writers write (titles and venues only).
• Black 1979: a type grammar can accept non-stories and reject stories, so form alone does not define the kind.

7. Models as readers (Q6)

• Jimeno Yepes 2024: chunking financial reports by element type improved retrieval-augmented question answering over paragraph chunks (preprint, abstract).
• Mortezaagha 2026: structure-aware retrieval for models reading a tree, reviewed in progressive-disclosure; its trade-off is in against.
• No study was found that gives a model DITA topic types or Information Mapping blocks and measures the answer. General findings on layout and models: llm-readers.

8. Practice

Who	Since	Base and derived types	Evidence of use
Information Mapping (Horn 1969; Information Mapping 2026)	1969	blocks in maps; eight map types in 1969	company page: nine gains, such as +83% initial learning and -51% reading time, credited to Horn's studies without a reference (example only)
DITA (OASIS 2015)	OASIS Standard 1.0 in 2005; 1.3 on 2015-12-17	topic, then concept, task, reference	802 firms listed by DITAWriter 2026; DITA-OT 2026 repository, 458 stars
Kubernetes 2026b	page dated 2026-05-03	concept, task, tutorial, reference	the project's contributor style guide; no use count
Red Hat 2026b; GitLab 2026	undated	concept, procedure, reference modules; topic types	reviewed in documentation
SchemaOrg 2026	2026-08 counts	CreativeWork, then HowTo, then Recipe; Article, then TechArticle	HowTo on 100K to 1M domains; TechArticle and Recipe on 10K to 100K
Baker 2013	2013	eight traits of a topic, one of them that it conforms to a type	publisher page

Q/A sites (Writing, Stack Overflow, User Experience; searched 2026-10-29; 12 threads, 2 under 10 of 20; strings in protocol). Practitioner opinion, no data.

• Q1: DITA topics and S1000D data modules are both modular units; DITA lets a team derive types, S1000D only constrains a fixed schema (WSE 2018b).
• Q1: code must act on @class values, not tag names, because DITA has no single grammar (SO 2016b).
• Q1: the base topic is the most forgiving type; deriving a type needs a developer, says the asker, and is trivially easy, says the accepted answer (SO 2013a).
• Q2: readers expect a genre's general rules, and familiar elements make the new ones stand out (WSE 2018d; WSE 2014b).
• Q2: a topic should solve one user problem, so length is the wrong measure (WSE 2018c, an answer by the author of Baker 2013 by its own account).
• Q3: moving a story from the hero's journey to the four parts of kishotenketsu means retelling it, not relabelling it (WSE 2022a).
• Q5: DITA tooling ranges from free editors to integrated systems; the hidden cost is set-up time (WSE 2014a); drop a story template when it forces content (WSE 2020a, example only).
• Q4: DITA is little used by software library developers, says one asker (SO 2019a, example only); name document types from user research (UXSE 2014a).