Перемикання кодів і матрична мова

Three things that look the same from outside

By this level you already replace a Russianism when you catch one. What you have not been given yet is a way to say what kind of mixing you just produced, and the kinds are not equally serious.

• A borrowing is a single word that has settled in and changes its endings like a Belarusian word: камп'ютар, кава, цукар. Note that in Belarusian the dictionary test alone does not settle the question - a long-established international loan and an ad-hoc Russian word pushed into a Belarusian sentence can both be found written down, which is why Русізмы, калькі і трасянка at B2 exists at all.
• Code-switching (пераключэнне кодаў) is a speaker moving between two whole languages inside one conversation, sometimes inside one sentence, with each language intact. It is a normal bilingual skill, not a defect.
• Belarusian-Russian mixed speech is neither. Millions of people in Belarus speak it, usually in addition to Russian rather than instead of it. Its everyday name, трасянка, is a disparaging one - researchers write of Belarusian-Russian mixed speech as the technical term and keep Trasjanka in scare quotes, and so should you when you are describing how somebody talks.

Which language is holding the frame

The tool that separates them comes from Carol Myers-Scotton's matrix language frame model. In codeswitching inside a sentence, one language - the matrix language - supplies the grammatical structure: word order and the elements that hold the sentence together. The other - the embedded language - supplies content words that are slotted into that frame.

The Russian sentence Я пошерил эту ссылку shows it in one line. Russian supplies the pronoun, the prefix по-, the past-tense -л and the accusative on эту ссылку; English supplies one stem, share, wrapped in all of it. Nobody would call that sentence English. Russian holds the frame, and one English word is a guest inside it.

Two piles, and a third

Run the same test on your Belarusian and your mistakes sort themselves by depth.

The middle row is the one worth drilling, because nothing in those sentences looks Russian. The grammar underneath them is. Most of these are the government patterns the B2 article lists, seen from the other side; the last pair is new here:

Switching looks different again. Я табе патэлефаную, как только освобожусь is one clause of intact Belarusian followed by one of intact Russian: two frames end to end, neither damaged. Mixed speech is the case where that seam disappears - a Russian stem carrying a Belarusian ending inside a clause whose connectives are Russian too, so there is no point at which one language stops. This article prints no specimen of it, deliberately: a mixed lect is a property of a corpus of real speech, not of a sentence somebody composes for a textbook.

Where mixed speech sits

Now turn the model on mixed speech itself. Corpus work reports Russian dominating the lexicon and much of the morphosyntax, inflection behaving as a hybrid, and pronunciation leaning Belarusian where the difference between the two languages is purely one of sound.

Feed that to the model's two tests and neither returns a single language. Word order is near enough identical in Belarusian and Russian that it decides nothing here. And the elements that hold a clause together - the endings - come from both languages at once, which is what hybrid inflection means. A frame supplied by both is not a Russian frame with Belarusian words in it, nor the reverse: it is what later work on the model calls a composite matrix language. Note that this is a later addition to the model, worked out in Myers-Scotton's Contact Linguistics (2002); the 1993 book assumes one matrix language per utterance.

A separate caveat sits on top of even that verdict. The model reads the frame off morphosyntax and never looks at pronunciation - and work on Russian in contact with Estonian, a genetically unrelated language, shows that a word can be adapted in its sound and not in its endings, or the reverse. So the one level the model consults is not the only level doing the work.

What mixed speech is not is a self-contained third language. The researchers who built the corpus stop short of calling it one stable lect across Belarus, and describe a bundle of regional patterns instead. The variation in it is structured rather than random, but it is not a system you could sit down and learn.

The question to ask

Of any sentence you have just produced: is Belarusian setting the grammar here, or only supplying the words? If the answer is the second, no amount of vocabulary work will fix it.