The gold standard of digital humanities is research that uses computational and digital methods to illuminate human history, culture, language and expression without losing contact with original sources, interpretive context or uncertainty.
Digitised archives, encoded editions, large text corpora, maps, networks and visualisations create new ways to search and compare evidence. They also create new ways to misread it: missing pages, imperfect transcripts, disputed place names and unrepresentative collections can produce confident charts about an incomplete record.
This eduKate root guide shows how to choose a humanities question, build reliable collections, link analysis back to sources and preserve work for future readers. Its numerical illustration is fictional and does not describe any real archive.
Use digital methods without losing the source.
Chapter index
- 1. Begin with a humanities question
- 2. Respect archives, context and provenance
- 3. Document selection before scanning begins
- 4. Separate the scanned image from an interpreted transcription
- 5. Use metadata and text encoding to expose structure
- 6. Build a corpus with a defensible denominator
- 7. Combine close reading and computational comparison
- 8. Treat maps and networks as models, not photographs
- 9. Worked case: a search result is not a population estimate
- 10. Respect rights, communities and language
- 11. Plan digital preservation and ongoing access
- 12. Make interpretation auditable and teachable
Chapter 1. Begin with a humanities question
Digital humanities brings computational and digital methods into conversation with questions about history, literature, language, culture, art, philosophy and human expression. Its goal is not to replace close reading with software or to treat a visualisation as proof. The gold standard is an interpretation whose sources, transformations and methodological choices remain visible. A researcher may use a searchable edition, text corpus, map, database, network or digital exhibition because it helps answer a question that matters.
Start with that question and articulate what different evidence could change the answer. Are you studying how a word changed meaning, which archives preserve a particular community, how a text circulated or how a public narrative developed over time? Each needs a different unit of analysis and different historical or cultural knowledge. Choosing a tool before identifying the source problem can make a project technically elaborate but intellectually empty.
Back to chapter index · Related root guides
Chapter 2. Respect archives, context and provenance
A digital item is not the event or object from which it came. An archival photograph may lack a precise date; a letter may be a copy; a printed transcription may silently modernise spelling; a catalogue description may reflect an institution’s inherited categories. Preserve creator, collection, date range, repository, original identifier, source language and the relationship between original and surrogate. Maintain the uncertainty rather than forcing every item into neat modern categories.
Context includes why a source survived and what did not. Archives reflect collection priorities, censorship, loss, funding and power. Digitisation adds another layer of selection: even when a catalogue lists thousands of items, only a portion may be scanned or transcribed. A claim about “what people believed” needs care if the surviving record mainly represents officials, wealthy publishers or a single language. Return to History for source criticism beyond any one interface.
Back to chapter index · Related root guides
Chapter 3. Document selection before scanning begins
A digital collection is shaped by inclusion rules: date, language, geography, genre, legal status, fragile condition and available funding. An honest project states what it scanned, what it did not scan, and why. A visually attractive online gallery may be a curated subset rather than a representative sample. Count the accessible objects alongside catalogue totals and make clear when catalogues themselves are incomplete.
Digitisation requires technical decisions about image capture, colour, resolution, page order and handling of damaged or folded materials. Different originals may need different approaches. Version files so that researchers can distinguish master capture, access copy, corrected transcription and derivative analysis. Access interfaces should allow citation back to the holding institution or stable object identifier rather than forcing readers to rely on screenshots without provenance.
Back to chapter index · Related root guides
Chapter 4. Separate the scanned image from an interpreted transcription
Optical character recognition and handwriting recognition can help create searchable text, but recognition accuracy varies by script, typography, page damage, layout and training material. Words omitted from a search result may be missing because software failed to recognise them rather than because they never appeared in the historical source. A project must disclose the transcription method, correction status, error characteristics and which material was never converted into reliable text.
Preserve links from a transcribed passage to its image and physical source description. Human corrections should be attributable, and contested readings should not be silently flattened into one answer. For complex scripts, mixed languages, marginal notes or unusual typography, transparent uncertainty may be more valuable than falsely clean text. Do not call a transcription authoritative merely because it can be copied into a search box.
Back to chapter index · Related root guides
Chapter 5. Use metadata and text encoding to expose structure
An unstructured text file can conceal divisions essential to interpretation: speakers, verse, quotations, deletions, notes, editorial insertions, dates and place names. Encoding may express these features so that readers and software can distinguish them. The Text Encoding Initiative P5 Guidelines provide a framework for representing textual resources, documenting editorial choices and supporting processing and exchange.
Encoding is interpretation. Two scholars may reasonably disagree about whether a mark is a correction, an abbreviation or a different author’s hand. Describe the editorial policy, rules and version rather than disguising decisions as automatically discovered facts. Consistent identifiers, metadata and controlled vocabularies improve retrieval, but a standard schema still needs careful application to the specific language and collection. See Information Science for information structure and indexing.
Back to chapter index · Related root guides
Chapter 6. Build a corpus with a defensible denominator
A corpus is a defined set of documents, not a neutral window into every text from a period. Language, genre, publication place, survival, copyright and digitisation policy all affect its coverage. Specify what counts as one document: an issue of a newspaper, an article, an advertisement, a page or an author. Counting each newspaper page as an independent expression of one person’s opinion would distort interpretation.
Text comparisons also require choices about spelling modernisation, tokenisation, compound words, punctuation, OCR corrections and duplicate editions. A word-frequency result is meaningful only under those conventions. Compare periods with transparent denominators, and avoid interpreting a rising raw count when the collection’s number of words or documents also increased. The Linguistics root provides complementary foundations for language analysis.
Back to chapter index · Related root guides
Chapter 7. Combine close reading and computational comparison
Search, concordances, topic models, embeddings and frequency charts can reveal promising patterns across large collections. They are tools for locating questions, not independent readers of historical intention. A clustering algorithm may group texts based on print artefacts, spelling or publisher conventions rather than the concept a scholar hoped to study. Human interpretation should examine representative original passages and counterexamples before a thematic label becomes a claim.
Close reading can also improve computational analysis. If the word “liberty” shifts between legal, religious and political senses, reading a sample of its contexts helps determine whether a numerical trend supports a historical argument. Record what changed when preprocessing or modelling assumptions changed. Invite another reader to inspect evidence without requiring blind trust in a black-box chart. Data Science develops the related disciplines of sampling, evaluation and reproducibility.
Back to chapter index · Related root guides
Chapter 8. Treat maps and networks as models, not photographs
Historical maps can display the movement of people, ideas, books and objects, but an uncertain place name or approximate route cannot simply be converted into a precise coordinate without warning. Include date ranges, georeferencing uncertainty and ambiguous locations where applicable. Administrative boundaries may have changed and modern place names may introduce anachronisms. A plotted line between two cities may indicate a documented connection, not the exact path a traveller used.
Network diagrams likewise depend on what counts as a node and edge: correspondence, publication, shared membership, citation or physical meeting. Removing unrecorded or hard-to-identify actors can make well-documented institutions appear disproportionately central. Weighted links may mix multiple sources of unequal reliability. Explain the source rules behind each edge and allow readers to inspect the underlying evidence before offering an interpretation of influence.
Back to chapter index · Related root guides
Chapter 9. Worked case: a search result is not a population estimate
Imagine a fictional archive catalogue listing 120 pamphlets for a period. Ninety have been digitised, and only 75 currently have text of sufficient quality for the project’s keyword search. Thus 90/120 = 75% have images; 75/120 = 62.5% of the catalogue population can be reliably keyword-searched under the stated rule. Fifteen of the scanned items still require transcription correction, while 30 have no scan at all.
If a search finds a slogan in 12 of those 75 searchable pamphlets, the observed share is 12/75 = 16% of the searchable subset, not proof that 16% of all 120 pamphlets—or all surviving pamphlets in the country—contained it. The missing 45 items may differ systematically. To improve the claim, inspect some of the missing records, establish selection reasons, correct transcription errors and explain how conclusions change. This is an example of archival criticism meeting numerical discipline.
Back to chapter index · Related root guides
Chapter 10. Respect rights, communities and language
Public accessibility, legal permission and ethical acceptability are different questions. An item may be in an archive but carry copyright, donor conditions or privacy risks. Historical records about vulnerable communities can include identifying details and language that causes harm when presented without context. Consider consultation, access restrictions and explanatory notes where appropriate, and distinguish faithfully quoting a source from endorsing its viewpoint.
Digitisation priorities affect whose cultural knowledge becomes visible to online users and search engines. Language coverage matters: tools trained for dominant languages often perform poorly on historical spelling, local scripts and mixed-language collections. Label translations and editorial interpretations rather than replacing original voices. Metadata can include contested names and identities without assigning unsupported certainty. Good digital humanities improves the conditions for responsible reading, not simply the volume of indexed text.
Back to chapter index · Related root guides
Chapter 11. Plan digital preservation and ongoing access
A website launch does not guarantee that a scholarly collection will remain usable. Files, software dependencies, domains, rights and storage systems change. Create an inventory of master objects, formats, metadata, identifiers, versions, checksums where suitable, backup arrangements and migration responsibilities. Plan for the possibility that an interactive visualisation disappears while the underlying source, code and methodological explanation should remain accessible.
The Library of Congress Recommended Formats Statement provides preservation-oriented format preferences, including structured textual resources and datasets. It is a reference for planning, not a universal mandate for every project. A responsible project provides stable citations, exportable data where legally possible, usable documentation and a named steward for corrections and future maintenance.
Back to chapter index · Related root guides
Chapter 12. Make interpretation auditable and teachable
A strong digital-humanities publication provides a research question, collection description, selection criteria, original references, transcription and encoding policies, dataset versions, analysis methods, limitations and a route to corrections. A reader should be able to move from a summary graphic to individual items, understand what was excluded and identify reasonable alternative readings. The publication is strongest when it invites criticism of its sources and model rather than presenting a digital interface as a neutral authority.
Students can begin by comparing an image of a historical page with its searchable transcription, tracing a claim to the original and documenting one correction. From there they can explore text counts, metadata, maps and network evidence while resisting the temptation to treat more data as better evidence in every case. The Education root explains why transferable research habits matter beyond a particular assignment.
Back to chapter index · Related root guides
Digital-humanities evidence checklist
- Question and population: state which cultural or historical claim is being investigated and what survives.
- Provenance: identify original sources, institutions, identifiers, dates, languages and degrees of uncertainty.
- Selection: disclose catalogue totals, digitised subsets, searchable subsets and their exclusions.
- Methods: document scanning, transcription, encoding, versioning, counting and mapping decisions.
- Interpretation: test patterns through source-level close reading and counterexamples.
- Stewardship: respect permissions, record corrections and preserve reusable documentation.
The eduKate Gold Standard subject directory
Digital humanities is a bridge. These published roots allow readers to move into the underlying scholarly and technical questions rather than treating all digital scholarship as one method.
- History — evidence and context.
- Library Science — collections, catalogues and access.
- Linguistics — language patterns and interpretation.
- Information Science — metadata and retrieval.
- Data — definitions and provenance.
- Computing — software and reproducible analysis.
- Data Science — sampling, evaluation and limits.
- Design — readable and accessible interfaces.
- Education — learning that transfers beyond one case.
Frequently asked questions
Is digital humanities just scanning old books?
No. Scanning can make sources accessible, but digital humanities also studies how computational models, editions, databases, maps and digital interfaces shape questions and interpretation.
Can keyword counts prove a cultural trend?
Only under defensible source selection, transcription, language and denominator assumptions. Frequencies in a searchable subset should not be presented as evidence about all historical texts without qualification.
What is the point of TEI encoding?
TEI offers methods of representing relevant features of texts and documenting editorial choices so that readers and software can inspect and process structured evidence.
What makes a digital collection trustworthy?
Stable links to sources, transparent selection, documented transcription and metadata, accurate rights statements, preserved versions, interpretive caution and an accountable corrections process.
Primary sources and scope
- Text Encoding Initiative: TEI P5 Guidelines (2026 edition).
- Library of Congress: Recommended Formats Statement 2025–2026.
- NIST: FAIR Data Principles.
Editorial policies, archival permissions and applicable law govern real collections. A digital surrogate should not be presented as an unquestionable substitute for the original object or its historical context.