How the word frequency counter handles text

What counts as a word

Words mode recognizes sequences of Latin-script letters, combining marks and numbers. A straight apostrophe, curly apostrophe or ordinary hyphen can join adjacent parts of a token. Spaces and most other punctuation separate tokens. This is a deterministic text rule shared with the layout editor, rather than a linguistic model or a dictionary lookup.

The number option applies to entries made only of Unicode number characters. It is not a numeric-value parser. A mixed token such as B2B is kept even when number-only tokens are excluded. Decimal punctuation does not remain part of a single number. Read the table before using numbers as labels in a design. Inspect your text in the counter.

Case and accents

Text is normalized to Unicode NFC before counting. Canonically equivalent composed and decomposed accented spellings can therefore merge. Accents are not stripped: cafe and café remain different. Merge letter case uses English locale lowercasing; it is not a language detector or a locale-specific search engine.

With Merge letter case off, Cloud and cloud stay separate. Normalization does not rewrite straight apostrophes into curly ones, replace unusual dashes or collapse multiple internal spaces in phrase mode. Words such as learn, learns and learning remain separate because this counter does not stem or lemmatize text. Check the editor’s language and font support.

Whole lines are literal phrases

Thank you
Thank you
thank you!

Whole lines as phrases trims the ends of each non-empty line and counts repeated complete lines. With case merging enabled, Thank you appears twice while thank you! remains separate because of the exclamation mark. Internal spacing and punctuation matter. The mode does not find recurring two-word expressions inside a paragraph.

Each complete phrase can appear at most once on its own line, so occurrence and line weights are equal in this mode. Stop-word and custom exclusions match entire phrases. The number-only setting applies only to Words mode; a phrase line containing 2026 is retained. Other scripts can be preserved as reviewed lines, although the editor’s bundled fonts may not cover their glyphs.

Counting and filtering order

1. Validate the total input size and normalize text to NFC.
2. Tokenize words or trim whole-line phrases, then apply the selected case rule.
3. Exclude stop words, number-only tokens if disabled, over-length entries and unsupported script groups.
4. Count occurrences and lines containing each kept entry. Select the weight unit.
5. Apply minimum count and character length; retain omitted entries in the report.

Every non-empty input line contributes to the line denominator, even when all its entries are excluded. Occurrence shares use all counted occurrences before minimum filters. The report records the settings and both count columns. The on-screen exclusion list is shortened for large inputs; the report includes the complete recorded exclusions.

Resource and interpretation limits

Input is limited to 100,000 Unicode code points, UTF-8 TXT files of 512 KiB and 1,000 distinct parser entries before minimum filters. Ordinary word entries can contain at most 64 code points and complete phrases at most 100. These are counting limits; the layout editor has a separate 250-word placement limit and reports words it cannot place.

Words mode does not segment non-Latin scripts and reports unsupported letter groups. Punctuation and symbols outside the token rule are separators, not counted entries. A literal frequency table does not measure sentiment, identify themes, verify respondents or prove representativeness. Use the source text and a stated method to interpret the result. Choose an appropriate survey counting unit. Understand the separate layout limits.