Classify text overall tone from a custom word lexicon

This overall tone classifier measures a piece of writing against a lexicon that you control. Supply word lists for formal, casual, urgent, and friendly language, and the classifier counts whole-word matches without using a model or making a network request. It returns the dominant tone, per-tone counts, the words that matched, a confidence share, and whether a requested default was used. Because the same input and lexicon always produce the same result, the output is suitable for repeatable editorial checks, routing rules, and tests.

Build a lexicon that represents your writing rules

A useful classification begins with a lexicon designed for the material being reviewed. Add words such as “therefore,” “regarding,” or “hereby” to formal when they genuinely signal formality in your organization. Casual might contain “hey,” “cool,” or “yeah,” while urgent could include “immediately,” “deadline,” and “critical.”

Friendly language might be represented by “welcome,” “thanks,” and “glad.” Every category is explicit, so reviewers can inspect and revise the rule instead of guessing why a hidden model chose a label. Entries are matched as complete words without regard to letter case.

A word may appear in more than one category; when it does, each occurrence contributes to every category that contains it. Duplicate entries within one category are collapsed, preventing an accidental repeated lexicon item from multiplying its influence. Keep entries specific enough to indicate tone in your domain.

Common words with little tonal meaning can overwhelm a more informative but less frequent signal.

Understand the score and deterministic tie rule

The classifier tokenizes the supplied text and looks up every token in the four category lists. Repeated words in the text are counted repeatedly because repetition can strengthen the apparent tone. The result includes counts for formal, casual, urgent, and friendly, plus distinct matched words for each category, so the conclusion remains auditable.

The dominant tone is the category with the greatest match count. Confidence is that winning count divided by all category matches; it describes the share of observed indicators supporting the selected category, not a statistical probability about the author’s intent. If categories tie for the highest count, the stable priority is formal, then casual, then urgent, then friendly.

This fixed rule ensures identical output in automation and browser use, though callers can inspect the counts and apply a different tie policy downstream. Punctuation and capitalization do not affect ordinary matches. Lexicon entries must each be one word, so phrases are intentionally outside this capability’s contract.

For a phrase-level or semantic judgment, a model-based analyzer is more appropriate; this tool is designed for transparent word evidence.

Choose whether unmatched text should fail or use a default

An absence of matches is meaningful. By default, the classifier returns an invalid-input error when none of the supplied tone indicators appears in the text. That behavior prevents a workflow from silently labeling neutral or out-of-vocabulary writing as though evidence supported the decision.

A practical workflow might omit the default during lexicon development so missing vocabulary is visible, then add a default only after the dictionary has been evaluated on representative content. The function does not infer sarcasm, sentence structure, audience, or context. Its strength is reproducibility: the same bounded text and the same lexicon always yield the same counts, label, and fallback status, making changes easy to test before editorial rules are deployed.