ADR 0022 Romanised Answers

ADR-0022: Answers typed in Latin letters are graded against the reading, in the language's scheme

Source of truth: this wiki page.

• Status: Accepted. Superseded in part by ADR-0025: readings are ISO 15919 letters as said; typed answers are graded as this ADR says.
• Date: 2026-10-03

Note, 2026-10-09. The readings ADR-0025 introduced are based on ISO 15919's letters, with deliberate deviations, not strictly ISO 15919; see the list of deviations.

Context

The owner asked for "transliteration support for answers, as no script cannot work without it", in production, listening and grammar. A learner who types kitnaa for कितना knows the word; one who types kitna does too. Comparing letters as typed would fail one of them, and every Indic language is typed several ways in chat.

Each language's readings now follow one popular, unmarked scheme, and decks/<code>/<code>-romanisation.yaml lists the spellings learners also type for the same sound (ee for i, w for v).

The owner decided the grade: "No-alphabet learners by default. Alphabet learners can toggle it", and a right romanised answer from an alphabet learner is graded 3.

Decision

• Where: a production, listening or grammar card with a reading, in a script that needs one, offers Script or Latin letters above the answer field. Number practice types digits and does not.

• How it is compared: RomanisedSpelling writes both the answer and each reading one way before AnswerGrader compares them:

  • lowercase, accents stripped, letters and digits only, so spaces, hyphens and apostrophes do not count;
  • each spelling in an equivalents group written as the group's first, read from the left, longest first, each written once.

  A grammar cell that lists several forms accepts each form's reading (Card.altReading).

• What counts: exact, a near miss to judge, or wrong, as for the script. A near miss that is another card's reading is that other word, and wrong. The script is still accepted in Latin mode, and counts in full.

• The grade: a right answer in Latin letters records 3 for a learner learning the alphabet who is past its script units: they recalled the word, not how it is written. A near miss they judge "I knew it" is 3 already. Before the script units, which a course reaches after its first six themes, answers start in Latin letters and count in full: the owner asked that "even for script learners, the initial writing exercises should be romanised". The no-alphabet path gives full credit throughout.

Consequences

• A language with no romanisation file still accepts Latin answers, folded only for case, accents and spacing.
• Equivalence groups that overlap the decks' own spellings, such as Hindi ch for छ, make grading more lenient, never stricter: both sides are written the same way.
• The scheme drops length and retroflex marks, so a few different words share a reading (दिन and दीन are both din). In Latin letters either is accepted for the other; the script still tells them apart.

See also: Contributing · ADR-Index · Home