Six errors vs two: benchmarking Arabic checkers

Take one sentence with a misspelled name, a case error after a preposition, a semantic misuse, and assorted style slips, and run it through the tools your writers already have.

The result

Microsoft Word flags two errors, both spelling-shaped. WriteX flags six, and classifies them: the entity (ايلون مسك ← إيلون ماسك), the genitive after the preposition (المشاركون ← المشاركين), the semantic misuse, the style issues, plus language-compliance findings against the style guide.

Why the gap is structural

Word matches against lists; lists cannot see grammar. Arabic case endings, hamza seats, and plural patterns require morphological analysis: the word must be decomposed into root and pattern before correctness is even a well-formed question. That analysis is what our engine's 32 components do, which is also why every finding carries its rule.

Run your own text: the web editor is free, and the full comparison shows the benchmark side by side.

Update, September 2026: the full evaluation

One sentence and two tools made the point; a published evaluation now makes it properly. In September 2026 we compared L3, the Lisan language engine behind WriteX, with 26 large language model configurations on a linguist-reviewed benchmark of real Arabic sentences, scored on exact spans with one output per system. L3 scored 80.29% edit F0.5 against 44.68% for Claude Fable 5.1, the strongest model in the field. The results page has the charts, four benchmark rows and the method; the announcement has the short version.