fast-is-english-word

Blazingly fast English word lookup. ~42M ops/s, 555 KB on npm, zero runtime dependencies.

Works in Node.js, Bun, Deno, browsers, edge runtimes, and Cloudflare Workers — same import everywhere.

Uses an xor filter — a space-efficient probabilistic data structure that's smaller and faster than bloom filters. The entire 234k-word dictionary compresses into a 282 KB binary with only 3 array accesses per lookup.

Install

npm install fast-is-english-word

Usage

import { isWord } from "fast-is-english-word";

isWord("hello");     // true
isWord("world");     // true
isWord("asdfgh");    // false
isWord("Hello");     // true  (case-insensitive)
isWord("it's");      // false (alpha-only)

Works identically in Node.js, browser bundlers (webpack, vite, rollup), and edge runtimes. The right entry point is selected automatically via the package exports map.

Performance

| Benchmark | ops/s | ns/op | |-----------|------:|------:| | Known word ("algorithm") | 42,000,000 | 24 | | Non-word ("xjqkv") | 76,000,000 | 13 | | Short word ("a") | 140,000,000 | 7 | | Long word ("extraordinary") | 30,000,000 | 34 |

npm run bench

How it works

An xor filter stores 8-bit fingerprints in a compact array. For each query, 3 positions are computed and the XOR of those 3 entries is compared against the word's fingerprint.

• Zero false negatives — if a word is in the dictionary, isWord always returns true
• False positive rate ≈ 0.39% (xor8) — ~1 in 256 non-words may return true
• O(1) lookup — exactly 3 array accesses, independent of dictionary size
• 282 KB — vs ~2.5 MB for a raw word list, or ~36 MB for a Set

Platform support

| Environment | Entry | How it loads | |-------------|-------|-------------| | Node.js (ESM) | dist/index.js | readFileSync of binary filter | | Node.js (CJS) | dist/index.cjs | readFileSync of binary filter | | Browser / bundler | dist/browser.js | Self-contained, filter embedded as base64 | | Deno / edge / workers | dist/browser.js | Self-contained, no filesystem needed |

The correct entry is resolved automatically by the exports map — you always just write:

import { isWord } from "fast-is-english-word";

API

isWord(word: string): boolean

Returns true if word is a recognized English word.

• Case-insensitive (lowercased internally)
• Returns false for empty strings, numbers, hyphens, apostrophes, spaces
• Only ASCII a–z characters are supported

Configuration

The filter is pre-built with xor8 by default (smallest size). You can rebuild with different options:

Fingerprint size

| Mode | FP rate | Filter size | |------|---------|------------| | xor8 (default) | ≈ 0.39% | 282 KB | | xor16 | ≈ 0.0015% | 563 KB |

# Rebuild with 16-bit fingerprints (higher accuracy, 2x size)
npm run build:filter:xor16
npm run build

Custom word list

# Use any newline-separated word list
node scripts/build-filter.mjs --words=./my-words.txt

# Combine options
node scripts/build-filter.mjs --bits=16 --words=./my-words.txt

npm run build

Comparison

| Package | Approach | Size | Lookup | |---------|----------|------|--------| | fast-is-english-word | Xor filter | 555 KB | 24 ns | | an-array-of-english-words | JSON array | 3.4 MB | ~ms (linear scan) | | word-list | Text file | 2.8 MB | user builds own | | check-if-word | Giant regex | 39.6 MB | slow |

License

MIT