Our Methodology

Data Source

All word data comes from Wiktionary via kaikki.org, which provides structured JSON exports of over 1,000,000 Wiktionary entries. Wiktionary is the free, collaboratively edited dictionary maintained by millions of contributors worldwide under the Wikimedia Foundation. Pronunciations follow the International Phonetic Alphabet standard where contributor data is available.

Wiktionary content is licensed under the Creative Commons Attribution-ShareAlike 3.0 Unported License (CC BY-SA 3.0). We present this data in compliance with that license.

Word frequency rankings shown throughout the site (how common a word is, used to order results and flag rare terms) come from a second, separate source: the hermitdave/FrequencyWords open word-frequency dataset (MIT-licensed), derived from OpenSubtitles corpora. This is used only to order and rank words by how commonly they occur; it does not contribute any definitions, pronunciations, or etymologies, which remain sourced from Wiktionary alone.

Languages Covered

PlainSpell covers five major languages: English, Spanish, Portuguese, French, and German. English has the most comprehensive coverage in our database, with hundreds of thousands of word entries including definitions, pronunciations, etymologies, and usage examples. The four additional languages provide substantial vocabulary coverage but with varying levels of completeness for supplementary features like IPA pronunciation and etymology data, depending on contributor activity in each language's Wiktionary edition.

Processing Pipeline

We download Wiktionary JSON exports from kaikki.org and process them as follows. Each language is processed separately to account for language-specific parsing requirements:

  • Parses definitions: Extracts word senses, part-of-speech tags, and usage examples from structured Wiktionary entries.
  • Extracts pronunciations: Pulls IPA strings from pronunciation sections where available.
  • Extracts etymology: Parses etymology sections to show word origin information.
  • Builds confusable pairs: A curated list of commonly confused word pairs (e.g., affect/effect, principal/principle) with explanations of the distinction.
  • Groups homophones: Words that share identical pronunciation are grouped using IPA matching from Wiktionary pronunciation data.
  • Generates misspellings: Common misspelling variants for each headword are generated with an edit-distance algorithm (insertions, deletions, transpositions, and adjacent-key substitutions) and linked back to their correct forms.

No AI Generation

No content is fabricated or generated using artificial intelligence. All definitions, pronunciations, etymologies, and usage examples are presented as-is from the original Wiktionary contributor submissions, with light formatting adjustments for improved readability in our interface. This commitment to source fidelity ensures that users receive the same community-vetted linguistic information available in Wiktionary, presented in a more accessible and searchable format.

Data Coverage

Coverage varies by language. English has the most complete coverage in Wiktionary. Spanish, Portuguese, French, and German entries vary in completeness depending on contributor activity for each term. Many entries have definitions but may lack IPA pronunciation or etymology.

Data Collection Method

Wiktionary data is collected by millions of volunteer editors worldwide who document words, their definitions, pronunciations, etymologies, and usage examples following Wiktionary's editorial guidelines. The kaikki.org project extracts this collaboratively maintained content into structured JSON exports that preserve the original Wiktionary entry structure while making it machine-readable. We download these JSON exports and process them into our optimized database format. The Wiktionary community actively maintains and corrects entries, so the underlying data quality improves continuously.

Update Schedule

Wiktionary content is updated continuously by its community of editors. The kaikki.org JSON exports are generated periodically from Wiktionary database dumps. We refresh our database when new exports become available. Between updates, new words added to Wiktionary and corrections made by editors will not be reflected in our database until the next refresh cycle.

Limitations

  • Wiktionary is community-maintained, coverage and accuracy vary by word and language.
  • Rare, archaic, or highly specialized terms may have incomplete entries.
  • Confusable pairs and homophone groups are curated lists and may not cover all cases.
  • IPA pronunciations represent one or more dialect variants as contributed by Wiktionary editors.
  • Definition quality varies, some entries have detailed, well-sourced definitions while others have minimal stubs.
  • Frequency rankings are approximate, sourced from the FrequencyWords open corpus (see Data Source above), and may not reflect all registers of language use.

References and Standards

The algorithms and standards PlainSpell relies on are documented in established academic and government references. Our misspelling-variant generation follows the edit-distance approach described in Stanford University's Introduction to Information Retrieval (chapter 3, spelling correction), the canonical treatment of insertion, deletion, transposition, and substitution distance. Corpus and frequency conventions follow the data standards published by the Linguistic Data Consortium at the University of Pennsylvania. For American English spelling and style decisions surfaced in our style-guide section, we cross-reference the U.S. Government Publishing Office Style Manual, the official spelling and usage authority for federal documents. These references are cited for methodological transparency; PlainSpell is not affiliated with any of these institutions.

Not Affiliated

PlainSpell is not affiliated with the Wikimedia Foundation, Wiktionary, or kaikki.org. We are an independent portal that makes Wiktionary data more accessible.