Punctuation checker: commas, marks and spelling

It works it out from the text as you type, and it has to, because the two languages disagree about almost every mark. Russian puts a comma before a subordinate clause where English usually does not; the quotation marks are different characters; the dash between subject and predicate is required in one and unknown in the other. Applying English rules to Russian text would flag every correct comma as an error, which is the failure mode of most general-purpose checkers. If a mixed passage is detected wrongly, the language can be set by hand — and each flagged issue links to the rule behind it, so a suggestion you disagree with can be checked rather than merely accepted or ignored.

It is a language model that reads the whole sentence instead of matching familiar word patterns. Rules are precise but see little: on an independent corpus they catch roughly one missing comma in five, because participial phrases, list items and parentheticals need the sentence parsed, not pattern-matched. The model roughly doubles that. Its 29 megabytes download once and stay in the browser cache; after that it computes on your device. Until you turn it on, the normal check works exactly as before — it never waits for the model.

Each step adds a different class of finding rather than more of the same. Mild shows only what the engine is sure of: categorical rules and the model’s highest-confidence hits. Standard adds the “worth a look” group — places where a comma is probably needed but the call is yours; “Fix all” deliberately leaves those alone. Strict adds style: repeated words, unknown words, optional marks. These used to be three confidence thresholds that barely differed — mild found nothing at all, and strict damaged one correct paragraph in eight.

Yes. Every fix is recorded in the history. Use the Undo button or Ctrl+Z to revert any change. You can also redo reverted changes.

Because part of Russian punctuation cannot be settled without parsing the sentence. The comma around a participial phrase depends on which noun the participle modifies, and word shapes alone do not tell you that; guessing does more harm than good — the rule we measured was wrong more often than right, so we removed it. The engine would rather stay silent than add a comma that does not belong. The deeper check finds noticeably more because it reads the whole sentence; the AI check finds more still. If it stays silent where you are sure there is an error, write to us — those letters are where new rules come from.