🔬 Exam format

NAATI CCL Marking Criteria: How Marks Are Actually Lost, Segment by Segment

Beyond the 63/90 pass mark: how idea units are counted inside a segment, what tips a minor error into a major one, the language-quality component, and the specific things that reliably cost marks.

Our scoring guide covers the 63/90 pass mark and the 29-per-dialogue floor — the rules of the game. This page goes underneath those rules, into how marks are actually lost inside a single 35-word segment: the three deduction families, what separates a shrug-worthy minor error from an expensive major one, and the specific habits that quietly bleed marks across a whole dialogue. Worked Punjabi–English examples throughout, and one honesty note held to all the way down: NAATI does not publish these deduction values, and neither does anyone else with a straight face.

Key takeaways
  • Errors fall into three families: distortions (meaning changed), omissions (meaning dropped), and insertions/unnatural language (meaning added or expressed unidiomatically).
  • The same error type can be minor or major — what tips it over is whether a real listener would act differently because of it.
  • Segments (~35 words) are marked by idea unit, not as one all-or-nothing block — so one dropped detail costs less than the whole segment.
  • A separate language-quality component (fluency, grammar, vocabulary, register) decides the gap between two candidates who conveyed identical content.
  • Negations, conditions, third-person reporting and long silences are specific, nameable habits that cost marks beyond the well-known 'numbers and dates' warning.
  • NAATI does not publish per-error point values. Any numeric model here — including ours — is a self-marking approximation, not NAATI's rubric.
TL;DR
Marks are lost in three families — distortion, omission, and insertion/unnatural language — applied idea unit by idea unit inside each segment, with a minor/major severity split that depends on whether the change would actually mislead a listener. A separate language-quality score covers fluency, grammar and register. Negations, conditions and third-person reporting are specific, under-warned traps. NAATI never publishes exact deduction values, so treat any numeric model — including the one further down this page — as a training aid, not a leaked rubric.

Deeper than the pass mark

Knowing you need 63 out of 90, with at least 29 on each of the 2dialogues, tells you the target. It doesn't tell you how a single segment actually gets marked — and that gap is where most self-directed preparation stalls, because people practise "for accuracy" in the abstract without knowing what an examiner is actually listening for inside 35 seconds of speech.

35
Words per segment
roughly, per chunk
3
Deduction families
distortion, omission, insertion
2
Severity tiers
minor and major

The three deduction families

Every content error in a CCL segment sorts into one of three families. Learning to hear your own errors this way — in practice, not just in theory — is the fastest route to fixing them.

FamilyWhat happensListener impact
DistortionMeaning is changed — a number, name, date or condition rendered incorrectly.The listener acts on wrong information.
OmissionMeaning is dropped entirely — a clause, detail or whole idea never makes it across.The listener never receives that idea at all.
Insertion / unnatural languageMeaning is added that wasn't said, or the rendering is technically complete but expressed unidiomatically.The listener is misled, or has to work harder to understand a message that should have been clear.

Worked example: a distortion

Worked example: an omission

Worked example: an insertion

What tips a minor error into a major one

The same error type can land as minor or major depending on one question: would a real listener, in that room, act differently because of this change? That single test does more work than any list of examples.

  1. 1
    Does it change a decision or an action?
    A wrong dosage, a wrong date, a dropped condition — these change what someone does next. That's the signature of a major error.
  2. 2
    Does it change a fact that matters, or a fact that doesn't?
    Dropping 'please' is a courtesy loss. Dropping 'not' in 'do not drive after taking this' is a safety loss. Same grammatical category (a small word), completely different weight.
  3. 3
    Could the listener still infer it correctly from context?
    If the rest of the segment makes the missing piece obvious or low-stakes, severity drops. If the missing piece was the only carrier of that information, severity rises.
  4. 4
    Is it isolated, or does it compound with another error nearby?
    Two minor errors in the same idea unit — say, a slightly wrong number AND a dropped condition on the same instruction — often compound into something closer to major, because together they leave the listener with a materially wrong picture.
✕ Myth
All omissions are roughly equal, so I should just try to say 'something' for every part of the segment.
✓ Fact
Omissions vary hugely in weight. Dropping a softener costs a sliver; dropping a negation, a condition, or a safety instruction costs far more. Prioritise carrying the high-stakes idea units intact over saying something for everything.

How idea units are counted inside a segment

A useful way to prepare — and to self-mark practice recordings — is to break each 35-word segment into its discrete idea units before you listen for errors, rather than judging the segment as one pass/fail blob.

Broken into idea units: (1) two tablets, (2) in the morning, (3) the condition — if pain returns, (4) the time marker — after 3pm, (5) take one more, (6) the ceiling — never more than four in a day. Six units, one segment. A candidate who renders five of six but drops the ceiling has kept most of the segment intact while dropping the one unit most likely to matter if someone actually followed the instruction.

This is a practical framework for self-assessment, not a claim about NAATI's internal marking software — but breaking segments into idea units this way is exactly the skill that makes our guide to avoiding omissions actionable rather than just cautionary.

Practise counting idea units in a real dialogue
Free Punjabi mock dialogues, scored the way NAATI scores the CCL — see which idea units you're actually dropping.
Try a free mock test

The language-quality component

Content accuracy is the majority of the mark, but it isn't all of it. A separate, real component assesses how you said it: fluency, grammar, vocabulary choice and register. This is the component that decides the gap between two candidates who both technically conveyed every idea unit — one smoothly and naturally, the other haltingly or in stiff, translated-sounding sentences.

Language-quality factorWhat it actually assesses
FluencySmooth, continuous delivery without excessive hesitation or false starts.
GrammarCorrectly formed sentences in both languages — not word-for-word transfer of source grammar.
VocabularyNatural, idiomatic word choice a real community member would use, not a stilted or overly literal one.
RegisterA tone appropriate to the professional scenario, held consistently across the whole dialogue.
Two candidates can carry identical content and land on different marks. The difference is never the meaning — it's how it arrived.

First person, not third

Community interpreting convention — and what CCL dialogues expect — is direct, first-person rendering. The patient says “I have a headache” and you say it in first person too, not “she says she has a headache.” Switching to third-person reporting under pressure is a common habit, especially for candidates more used to summarising than interpreting, and it's treated as more than a style slip: it inserts a layer of reported speech the original speaker never used, and it often signals other fidelity problems in the same breath (softened tone, dropped urgency, paraphrase creeping in). Our guide to direct-speech interpreting covers the habit and the fix in full.

Negations and conditions: the two most-missed idea units

Numbers, dates and names get warned about constantly — rightly. Two categories get warned about far less, and they fail candidates just as often: negations and conditions.

TypeExampleWhat goes wrong
Negation“Do not take this on an empty stomach.”Under pressure, the 'not' vanishes and the instruction inverts into its opposite — a major distortion, not a minor slip.
Condition“Only come back if the fever is still there after two days.”The condition drops and an optional, contingent instruction becomes a flat command — exactly the error in the worked distortion example above.
Frequency“Every other day, not every day.”The contrastive correction gets lost and the wrong frequency survives into the rendering.

Long silences and what they actually cost

Silence itself isn't a named deduction category the way a distortion is — but a long pause is rarely free. It damages the fluency dimension of the language-quality component directly, and in practice it's usually a leading indicator of a bigger problem: candidates who freeze for two or three seconds then often rush the rest of the segment, which is exactly when omissions and distortions cluster. Treat a long silence in your own practice recordings as a flag to investigate, not a harmless pause.

A self-marking model — clearly labelled as ours

Error, roughlyIllustrative weightWhy (our reasoning, not NAATI's)
Minor omission (a softener, a filler phrase)Very smallLow-stakes, rarely changes what a listener would do.
Major omission (a condition, a negation, a safety instruction)SubstantialChanges what a listener understands or does next.
Minor distortion (a slightly off but harmless rewording)SmallMeaning drifts but the practical outcome is unchanged.
Major distortion (wrong number, date, name or inverted meaning)HeavyA listener would act on incorrect information.
Insertion of invented contentModerate to heavyMisleads the listener about what was actually said.
Unnatural language, content intactSmall, cumulativeCosts language-quality marks, not content-accuracy ones — but adds up across a dialogue.

Use this the way you'd use a practice rubric for any skill: consistently, so your self-scoring means the same thing from one practice session to the next — not as a substitute for the actual score calculator, which estimates your likely pass position against the real 63/90 and 29-per-dialogue thresholds.

After every practice segment, check for each family:0/5

If you're running this checklist against a real result you already have rather than a practice recording — because you sat the test and it didn't go the way you hoped — our guide to what to do after a NAATI CCL fail picks up from exactly here: how to diagnose which of these families cost you the most marks, and what that means for your next attempt.

Quick check

🧠 Quick check
1. A candidate drops 'only if the pain returns' from a conditional instruction. What family of error is this, and how severe?
2. Two candidates convey identical content. What decides which one scores higher?
3. Does NAATI publish the exact number of marks each error type costs?

Frequently asked questions

No. NAATI does not publish a per-error deduction table, and neither does anyone else with legitimate access to it — that detail lives inside examiner training. Any specific numeric weighting you see, including the illustrative one on this page, is a self-marking approximation for practice, not NAATI's actual rubric.

Sources

The pass mark, segment length and dialogue structure referenced above are NAATI's published figures, checked against naati.com.au on 14 September 2026. The deduction weightings are explicitly our own approximation, not NAATI's — see the callout above.

Part of: The full NAATI CCL mock test

Ready to put this into practice?

NaatiGuru turns everything you just read into real, scored practice — 112 exam-format dialogues, mock tests marked like NAATI, and model answers in Gurmukhi. Start free; go Unlimited from $29/mo — or a one-time $59.

Try the 90-second challenge or create a free account →