Repeated words with punctuation
“Tea, tea; coffee.” counts tea twice and coffee once with case combining enabled. Their shares are 66.67% and 33.33% of the three counted tokens.
Counting rules
0 counted tokens · 0 distinct tokens · 0 tokens excluded by rules
| Word | Count | Share of counted tokens |
|---|
Maximum 100,000 UTF-16 code units. Tokens are Unicode letters/numbers with combining marks and internal apostrophes; hyphens split words. This is a lexical count, not language-aware segmentation or keyword relevance analysis. Processing stays in this browser; text is not submitted to a server.
Text is processed locally in your browser. No tool input is sent to a remote processing service.
How to use the Word Frequency Counter
- 01
Paste the text to analyze.
- 02
Choose whether to combine case variants, include number-only tokens, and impose a minimum token length.
- 03
Enter any words you want to exclude. There is no hidden stop-word list.
- 04
Filter the report for inspection, then copy or download the full counted-word CSV.
Examples
Excluding words changes the denominator
In “the river and the bridge”, excluding the and and leaves river and bridge with one occurrence each. Both then account for 50% of counted tokens.
Apostrophes and hyphens
“don’t” is one token because its apostrophe is internal. “well-known” splits into well and known. Straight and curly apostrophes remain distinct token spellings.
Know the token definition
Tokens begin with a Unicode letter or number and continue through letters, combining marks, and numbers. An internal straight or curly apostrophe joins parts of a token. Most other punctuation separates tokens. Hyphenated phrases therefore split, while alphanumeric IDs remain together. Decimal punctuation splits values: 12.5 produces 12 and 5 when number-only tokens are enabled. Emoji and punctuation are not words in this report.
Apply visible counting rules
Case combining uses English lowercasing after NFC normalization. With it disabled, Tea and tea receive separate rows. The number option controls tokens containing no letters, not IDs that mix letters and digits. Minimum length counts Unicode code points after normalization, rather than bytes or visual width. The exclusion field follows the same tokenization and case rules as the source. Excluded words are removed completely from both counts and percentages.
Read frequencies as descriptive evidence
The total is the sum of all retained token occurrences. Distinct tokens count the number of rows, and each share divides its row count by the retained total. Equal counts are ordered by token text, making ties deterministic. The table displays up to 100 matching rows; the filter changes which rows you inspect but does not change counts or the export. The CSV contains all retained words with counts and percentages to four decimal places.
Where lexical counts stop being enough
This is not stemming, lemmatization, topic modeling, or search ranking analysis. Run, runs, and running remain separate words, as do unrelated spellings that sound alike. Scripts without spaces may form long runs rather than dictionary words because this tool uses an explicit lexical pattern instead of a language-specific tokenizer. Compare drafts using the same settings and treat large changes as prompts for editorial review rather than automatic evidence that writing is good or bad.
What to know
- Maximum input: 100,000 UTF-16 code units; exclusion list: 2,000 code units. Table displays 100 rows.
- Lexical tokens only: no language-specific segmentation, stemming, semantic relevance, or keyword ranking score.
Common questions
Is there an automatic stop-word list?
No. The exclusion field starts empty so no words disappear without your choice. Add the words that matter for your particular review.
Why are my Chinese words not separated individually?
The lexical rule groups adjacent Unicode letters. It does not perform dictionary segmentation for Chinese, Japanese, Thai, or other languages that need language-aware word boundaries.
Does filtering the table change the CSV?
No. Filtering only narrows the visible rows. The export includes every word retained by the counting rules.