Why people fail the A2 language check

If you are preparing for the Latvian state language proficiency check at A2, you have probably heard that "most people fail it". There is one published data set behind that claim, and it says less than people think — but the structure of the check itself explains a great deal. This article separates the two.

The only published numbers

The government open-data set on state language proficiency check candidates covers March to September 2023. In those seven months, 17 656 people sat the A2 level and 6 421 passed — roughly 36 %. Across all levels in the same data set, 19 503 people sat a check.

Two conclusions follow, and only one of them is safe.

What the 36 % does not mean

The published data covers only those seven months. And they were precisely the months when a very large number of people were sitting the check against a deadline set in law. Both the number of candidates and the makeup of that group were therefore unusual: many people sat not because they were ready, but because a date was approaching.

So 36 % should not be read as an ordinary-year pass rate, and this data set cannot be used to work out how many checks take place in a year — its own coverage note does not allow it.

What does hold: A2 is overwhelmingly the level people sit — 17 656 sittings out of 19 503, i.e. more than 90 % of all state language proficiency checks — and the share of candidates who fail it is high. It is the check almost everyone takes, and it is not a check almost everyone passes.

Reason one: 60 % is required in each skill separately

The check has four parts — listening, reading, writing and speaking. Each is worth 15 points, 60 in total.

The total, however, decides nothing. Paragraph 47 of Cabinet Regulation No. 157 states that the check is passed only if the result in each skill separately is not below 60 % of the maximum — that is, at least 9 points out of 15 in every one of the four parts.

In practice:

• 45 out of 60 spread evenly (say 11 + 12 + 11 + 11) is a pass;
• 45 out of 60 with an 8 in writing is a fail, no matter how strong the other three parts were.

This is the first and most common planning mistake. People revise what they are already good at, because it feels better, and the weak skill stays under 9 points. One skill sinks the whole thing. The arithmetic is set out in detail in how many points you need at A2.

Reason two: writing task 2 tests endings, not vocabulary

The writing part (15 points, 35 minutes, on paper) has three tasks, and the middle one catches people out.

Task 2 gives five internet news headlines, each with one gap, and the base form of a word in brackets. You have to write the correctly inflected form. Where the brackets say vasara (summer), the headline needs vasarai; where they say notikt (to happen), it needs notiks.

That is five points out of fifteen, and they are not awarded for knowing words. They are awarded for grammar and endings — cases, gender, number, verb tense. Someone who speaks comfortable everyday Latvian but has never studied the case system can lose this task completely, and must then find all 9 required writing points in the remaining two tasks, which are worth 10 between them.

Those other two tasks are text production: a guided description of a person from pictures and a business card (minimum 30 words), and a functional text written to a five-point plan — an e-mail, a letter, an invitation or a small ad (volume 35 words). These are judged on how fully the prompt is covered, whether the required genre is used, whether the required language functions appear, whether the text hangs together, whether the vocabulary is sufficient, and how far the mistakes made interfere with understanding the text. An error is not automatically a penalty; an error that stops the reader understanding you is.

Reason three: in speaking task 3 you have to ask, not answer

The speaking part lasts 10–15 minutes and is taken individually, either in person or by videoconference. The first two tasks are the ones candidates expect: an interview of 9–10 questions drawn from the published topic list, and a description of two pictures. Those two are scored together and carry 12 of the 15 points.

The third task is different. You are given three short advertisements, each missing one piece of information, and you must ask a full-sentence question to obtain it — if an ad does not say when a match takes place, you have to produce the question that asks for the date.

That is 3 points, and it is the hardest part to carry on conversational experience alone. In daily life many learners signal a question with intonation or a single word. Here the grammatical form of the question in a full sentence is what is being assessed. Lose those three points and you need 9 out of the remaining 12 — possible, but a much narrower margin.

Reason four: if you skip a part, nothing at all is graded

This is the rule that costs the most. Results are assessed only if the candidate sits all parts of the check. If a part is skipped or abandoned, it is not that one part that goes unmarked — nothing is graded.

The day is long. The written parts run in sequence (listening and reading on a computer, then writing on paper), and only afterwards does each candidate take the speaking part individually. That is exactly the moment when someone who feels the written part went badly is tempted to go home. It is the worst available decision: walking out throws away the points already earned and leaves no result at all.

Failing does not, by itself, cancel a confirmation of a level you had already passed: the rules provide that an earlier decision remains valid where the person sat a higher level or degree and did not pass it.

What you cannot find out about marking

This is worth stating plainly: the point bands for writing and speaking are not published. The official programme lists what is assessed, but no band descriptors are published — there is no table saying how many points a particular shortcoming costs. For speaking, the only published split is 12 + 3, with no further breakdown across the first two tasks.

So nobody — including any course provider — can legitimately promise you "this mistake costs exactly this many points". Listening and reading are machine-scored in the information system; writing is marked by two independent raters, and where they differ by more than three points a third rater is brought in. Results appear in the e-service on the latvija.gov.lv portal from the seventh working day.

What to do with all this

The practical conclusion is simple and unglamorous: plan around your weakest skill rather than your total, drill word forms separately for writing, drill question formation separately for speaking, and sit all four parts to the end whatever happens.

On this site you can try all four parts free of charge in the check's own format — the same task types and the same point split — and see immediately which skill falls below 9 points. Common questions are collected in the FAQ.

This article is not legal advice. Requirements, deadlines and procedures can change — before you register, confirm the current position in the in-force text of Cabinet Regulation No. 157 and in the official state information sources, including the latvija.gov.lv portal.