00 / project statement
is a curated editorial archive,
not a search engine.
Each entry begins with a word selected for its cultural weight: a word that has shifted meaning, accumulated associations, or traveled across registers over centuries. The archive traces that word through corpus frequency, lexical attestation, scanned evidence, and interpretive annotation, presenting them as distinct layers rather than a single authoritative answer.
This is not a dictionary. It does not define words. It does not claim that frequency reflects importance, that Gutenberg texts represent all historical usage, or that modern news snippets are comparable to historical corpora. What it does claim is that the available evidence is worth making visible, with its sources, limits, and gaps stated alongside the data.
Intended audience: researchers, writers, educators, and anyone curious about how language carries history.
01 / methodology
A selected-word archive for historical frequency, attestation, scanned evidence, and interpretation.
Not a general search engine. Each word is selected, structured, and annotated before it becomes a public entry.
design research
The visual structure draws on the Swiss International Typographic Style as developed between the 1950s and 1980s at the Basel School of Design and the HfG Ulm.
The six-column grid is not an aesthetic choice. It is a claim that the six categories of semantic evidence are commensurate and comparable. When a category is missing, the column is empty. The gap is not hidden.
the visual programme / six evidence columns
Signal
source-specific
Attestation
lexical proof
Variant
form policy
Context
snippet evidence
Boundary
claim limit
Rights
attribution
Faded columns = evidence not yet implemented. The absence is structural, not hidden.
Josef Müller-Brockmann
1914-1996
Grid Systems in Graphic Design
1961
The six-column grid that structures every word entry is a direct application of Müller-Brockmann's modular grid principle: a visible, auditable structure that makes the absence of data as legible as its presence.
Karl Gerstner
1930-2017
Designing Programmes
1964
The color token system (ink, wheat, blaze, fire, sun, nice, curious, wine, sail) is a programme in Gerstner's sense: each token is a rule, not a feeling. Orange marks emphasis and interactivity. Blue marks sources and data layers. Green marks confidence.
Emil Ruder
1914-1970
Typographie
1967
Helvetica Neue is used throughout as an information-neutral carrier. Ruder's argument that type must serve communication rather than express the typographer applies here: the typeface does not perform its own historicity.
HfG Ulm
1953-1968
Hochschule für Gestaltung
1953-1968
The Ulm model treated design as an epistemological practice: structure makes claims, not just appearances. The page modules and evidence diagrams are design research artifacts; they argue through their structure that evidence should be shown as a relational system.
This is not an application of Swiss design as historical style. It is an application of the underlying principle: that the structure of a design makes claims, and those claims should be as auditable as the data they present.
layered evidence
Two information layers share one structure. The source layer stays readable at rest; hover gently brings the output layer forward.
01 / source
long-run book / corpus frequency data
output / Corpus frequency
Frequency is read as a long-run signal, with corpus boundaries kept visible.
Corpus size, genre mix, and source breaks stay attached to the line.
02 / source
dictionary / lexical evidence
output / Lexical attestation
Dictionary evidence anchors claims about first known use without pretending it is corpus frequency.
Attestation can sit outside the archive corpus and still matter.
03 / source
verified scanned-book page
output / Scanned evidence
A page image or public-domain snippet gives inspectable context, still with uncertainty attached.
Scan quality, OCR noise, and public-domain limits remain visible.
04 / source
interpretive notes
output / Annotation
Notes separate spelling variants, semantic drift, licensing limits, and confidence.
Interpretive decisions are recorded instead of hidden behind the chart.
data sources
| Source | Role | Coverage | License |
|---|---|---|---|
| Google Books Ngram Viewer | Frequency time series | English corpora through 2022; queried at smoothing 0 before local transforms | Google Books Ngram terms; attribution required |
| Project Gutenberg | Public-domain context text | Selected public-domain books, mainly eighteenth to early twentieth century | Project Gutenberg License; public-domain status varies outside the US |
| Library of Congress / Chronicling America | Historical newspaper evidence | Digitized US newspapers, chiefly 1770s-1960s depending on collection availability | Library of Congress rights statements; item-level rights vary |
| Wikimedia / Wikinews / MediaWiki APIs | Modern context and attention signals | Contemporary page, article, and context metadata where relevant | CC BY / CC BY-SA family; project-specific terms apply |
| Lexical references | Attestation and sense-history checks | OED candidate checks, Online Etymology Dictionary, Wiktionary, Merriam-Webster, Cambridge | Publisher-specific; entries are not reproduced |
| Policy, clinical, and technical references | Domain context anchors | EU AI Act, GDPR/ICO, FTC/NIST/OECD, PubMed/MeSH, WHO/NIMH, APA/DSM-history pointers, Stanford HAI and related pages | Source-specific; used as citation targets and metadata, not republished text |
| Public law and human-rights repositories | Legal and rights anchors | Wikisource, CourtListener, Justia/Oyez, Cornell Wex, NY Senate, UN, ECHR, EUR-Lex, eCFR, govinfo, DOJ/HHS, FTC, OECD, CPPA and related public pages | Source-specific; legal text, summaries, and court opinions are cited or paraphrased, not redistributed as a corpus |
| Geographic and demographic context sources | Aggregate context signals | OpenAlex, GDELT, World Bank indicators, Our World in Data fallback values, Open-Elevation, and Google Trends availability checks | Source-specific; used as aggregate metrics, metadata, or unavailable-source audits only |
calculation methods
Source capture
Scripts fetch or ingest source-specific data into generated JSON files. Ngram queries are pulled as yearly series; Gutenberg and LOC material are stored as text or metadata extracts; policy, dictionary, clinical, and technical references are stored as source pointers or curated records when full-text reuse is restricted.
Frequency normalization
Google Ngram values are converted into comparable per-million visibility where needed. Most charts keep smoothing at 0, then apply local period aggregation, rank lookup, or max-normalization so that a chart compares terms within the same source family rather than across incompatible corpora.
Display transformation
Visual scales may use square-root, max-normalized, indexed, or ranked transforms. These transforms are display devices only: the interface labels them as visual intensity, visibility index, or relative signal rather than raw counts.
Phrase and variant policy
Each word page declares which forms belong together and which remain separate. Examples include spelling variants, compounds, X + word phrases, word + X phrases, singular/plural grammar, and domain phrases. Variant aggregation is treated as an editorial decision, not a default.
Semantic grouping
Semantic layers are built from curated phrase sets, keyword/collocate overlaps, source annotations, and domain-specific evidence records. They are not presented as machine-learned sense disambiguation; they are interpretive maps backed by visible source categories and caution language.
Branch and dependency scoring
For pages such as hub, form groups and dependency tiers are computed from curated examples: counts, object-type spread, phrase form, and modifier dependence are converted into visual branch maps. The score indicates how much the attached word specifies the object, not popularity or legal meaning.
Confidence and boundary labels
A claim can be source-supported, corpus-visible, manually attested, derived, pending, or cautionary. The page text must name that status instead of collapsing everything into proof. Absence, sparse data, OCR noise, rights limits, and genre bias remain attached to the claim.
epistemological position
The archive makes bounded claims. It can show that a selected form is visible in a named corpus, that a cited source supports an attestation, or that a curated semantic grouping organizes the evidence. It does not turn those signals into universal claims about all English usage, all communities, or all meanings of a word.
claims allowed
claims refused
public repository
The data pipeline and visualization components are public for inspection, reproducibility, and citation review.
github / dpan538
Words-Over-Time
Public repository / Data pipeline / All components
What is inspectable
What is curated (not generic)
citation note
Cite the archive and the upstream sources separately. A Words Over Time chart is an editorial synthesis of source retrieval, cleaning, transformation, semantic grouping, and visual design; it is not a replacement citation for Google Books Ngram, Project Gutenberg, Library of Congress, Wikimedia, dictionary publishers, policy pages, or clinical/technical references.
suggested website citation style
Words Over Time. "[Word page title]." Words Over Time, 2026, [page URL]. Accessed [day month year].
Example: Words Over Time. "Hub." Words Over Time, 2026, /words/hub. Accessed 27 May 2026.
Rights note: this page is a research and design archive, not legal advice. Public launch should keep source URLs visible, avoid full third-party reproductions, and remove or paraphrase any pending, restricted, or subscription-only excerpt.