Word Frequency Counter

How It Works

Every time you type, change a control or load a file, the tool re-tokenizes the text with a Unicode-aware regular expression, applies your filters (minimum length, case handling, stopword list), counts the surviving tokens in a map and records each word's first position in the stream. The results feed four places at once: the summary stat cards, the top-15 bar chart on canvas, the ranked table (capped at the first 50 rows for readability) and the CSV/clipboard exports, which always reflect the current filters rather than the raw text.

The tokenizer regex

Words are found with /p{L}+'?p{L}*/gu: one or more Unicode letters of any script, optionally with one apostrophe inside so English contractions ("don't", "l'été") survive as single tokens. Because the class is letters-only, digits, hyphens, underscores and punctuation act as separators — "2026", "e-mail" and "$1,000" contribute nothing whole. The u flag makes the pattern surrogate-pair safe for emoji and rare CJK code points.

Why stopword filtering matters

Zipf's law guarantees the top of any raw English list is occupied by the, of, and, to — words that describe grammar, not topic. For the usual jobs this counter gets (checking whether an article actually covers its promised keywords, auditing page copy, spotting filler), you want the content-word profile, so the ~200-word English stopword list is on by default. Turn it off to see the true token distribution, which is also the honest way to compare two texts of different lengths.

Reading the table

The % column is the word's share of all tokens that passed the current filters (the percentages sum to 100 only if you show every unique word). A 3% share is genuinely high for a content word — English prose rarely gives a topic noun more than 1–2%. The first-position column is the word's index in the token stream, useful for spotting which terms a piece introduces early (a real signal in search-intent analysis).

Frequently Asked Questions