word_roots

Browser-based tools built around Donald J. Borror’s Dictionary of Word Roots and Combining Forms: a searchable index of extracted entries, a species-name / compound builder from plain English, and an in-app link to the original PDF for full text and Borror’s introductory chapters.

The audience matches Borror’s—students, clinicians, and especially taxonomists who need conventional Latin and Greek combining forms when coining names.

Source material

The underlying work is Dictionary of Word Roots and Combining Forms (first edition 1960), Donald J. Borror, Mayfield Publishing Company: Greek, Latin, and other languages with emphasis on biological terminology and scientific naming.

The bundled PDF used for extraction and in the web UI lives at:

• web/public/dictionary_of_word_roots_and_combining_forms_borror.pdf

The printed book also contains sections on formulation of scientific names, transliteration of Greek, and common combining forms. Those sections are not duplicated in the app; open the PDF for the full treatment.

Web app (web/)

A static client-side app (Vite + React + TypeScript). At runtime it loads web/public/dictionary.json only; there is no server API.

Features

• Search + letter row — One screen: a search box and A–Z (plus #) buttons. With an empty search, choose a letter to list entries whose headword starts with that letter (book column order within the letter: each PDF page is read left column then right column, as in extract_dictionary.py). That order can differ slightly from strict A–Z when pdftotext mis-assigns a line to the wrong column; python3 scripts/lint_dictionary_alphabetical.py flags adjacent alphabetical inversions as a review aid (it does not change data). With text in the search box, up to 200 substring matches are shown across all letters; picking a letter clears the search and jumps to that letter’s browse list.
• Name builder — Enter a short phrase (e.g. “sea eagle” for a short field-note gloss). Stopwords (articles, “and”, etc.) are dropped; each remaining content word (up to a fixed cap) is matched to dictionary glosses, then stems are chained left-to-right with simplified rules from Borror’s Formulation of Scientific Names (Greek ο linking and omission before a vowel-initial next stem, o+o elision where applicable, Latin i link when a link step is not Greek–Greek). Output is a draft lowercase epithet and a genus-style capitalized stem; real names still need correct endings, grammar, and nomenclatural availability checks. Gloss-based notes compare the first and last words (or the two words when there are only two) for Borror’s usual “quality before principal” pattern.
• Entry cards — Show the root line, expand tags like (G) / (L) into full language names, and short notes on notation (commas, =, leading hyphens, etc.).
• Source book — The top of the page states what the app indexes, with Open book PDF (same file as web/public/…) on the next line. An Introduction panel (collapsed by default) contains Borror’s preface and How to use this dictionary in full (extracted from the PDF). The compound name builder sits below that so the dictionary search stays primary.

Requirements

• Node.js (for npm in web/).
• pdftotext from Poppler (e.g. poppler-utils on Debian) to regenerate dictionary.json from the PDF.

Commands

# Regenerate dictionary JSON from the bundled PDF (run from repo root)
python3 scripts/extract_dictionary.py

# Re-apply spurious filtering and overrides to the existing JSON (keeps book column order; no PDF)
python3 scripts/rebuild_dictionary_from_json.py

# Rebuild `spurious_alphabetical_inversion_entries.json` from the PDF (same rules as the linter; run after changing sort logic)
python3 scripts/regenerate_spurious_alphabetical_inversion_entries.py

# Optional: list browse-letter pairs that are out of strict A–Z order (column-OCR red flags)
python3 scripts/lint_dictionary_alphabetical.py

# Optional: full tab-separated report of every such pair (redirect to a file for review)
python3 scripts/report_alphabetical_inversions.py -o /tmp/alphabetical_inversions.txt

# `spurious_alphabetical_inversion_entries.json` drops the “higher sort key” row per pair from the web bundle;
# see `scripts/ALPHABETICAL_INVERSION_PAIR_ANALYSIS.md` for per-pair notes (many same-stem pairs are ASCII false positives).

# Develop
cd web && npm install && npm run dev

# Production build
cd web && npm run build

The extractor merges wrapped gloss lines in each column (continuation rows without a new (LANG). tag) and can carry an incomplete trailing line across page breaks when the gloss still ends with a comma. When two headwords appear in one OCR fragment (the column gutter was only 2–3 spaces, so the wide-space column split missed), it splits on repeated roots (LANG). patterns, and when a second headword is jammed onto the same line after the first gloss (e.g. =depas / =diazoma), it splits on roots (LANG). inside the English—including comma–hyphen stem forms (drom, -a, …, =us) before a second (G).. A short right-column line with no (LANG). may be merged backward onto a left-column entry whose gloss ends with ; a / , a (wrapped gloss continuations such as gift after doro). OCR reads Borror’s = as « in places; the extractor normalizes « to = everywhere. OCR_EQUALS_ROOT_FIXES is reserved for any remaining roots-level overrides after that (usually empty).

Serve web/dist/ with any static file server for production.

Subpath hosting — Production builds for https://hromp.com/word_roots/ use web/.env.production with VITE_BASE=/word_roots/ so JS, CSS, dictionary.json, and the PDF resolve under that path. For hosting at the site root, set VITE_BASE=/ (or remove the variable) before npm run build.

Roadmap

1. Richer search — Synonyms, stemming, or embeddings for English descriptions (beyond substring search).
2. Cleaner extraction — Improve wrapped lines and OCR glitches in the PDF-derived data.

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0).