Three things decide what a Russian word sounds like and only one of them is written down. The consonants and vowels are on the page. Where the stress falls is not, and it moves within a single word's own forms. And whether each consonant is hard or soft is marked indirectly, by the vowel letter that follows it or by a sign that makes no sound of its own.
Two of the three are therefore invisible, which is why a Russian word can be read correctly and said wrongly, and why the repair has to happen at the moment a word is first stored rather than later.
Six things that decide whether you are understood
- unstressed о is said a. This is the single biggest gap between Russian spelling and Russian sound, and it applies everywhere.
- the stress is unpredictable and is not written outside textbooks. It also moves between forms of the same word, and it changes meaning on real pairs.
- молоко, milk, has three written o and only the last is stressed, so it arrives as ma-la-KO. One word demonstrates both rules above.
- soft consonants are a separate set, made with the tongue raised towards the palate. Nearly every consonant has a hard and a soft version and the difference is a difference of word.
- final voiced consonants harden, so хлеб ends in p and город ends in t.
- что is said shto, one of a small set of words whose spelling and sound simply disagree, and it is among the commonest words in the language.
Stress moves, and it takes the vowels with it
Stress in Russian is not fixed to a position and it is not fixed within a word. Голова stresses the last syllable, головы stresses the first, and the pattern belongs to the word rather than to a rule you could apply. Nothing in ordinary printing marks it.
What makes that expensive is the reduction that follows it. An unstressed о is not a weakened o, it is pronounced close to an a, so молоко comes out with two a-like vowels and one o. Unstressed е and я move towards an i sound. Move the stress and the vowels of the whole word change, which means the same word in two of its own cases can sound like two different words to an ear that learned only the spelling.
The third dimension is palatalisation and English has nothing corresponding to it. Almost every consonant comes in a hard and a soft pair, and the pair is a genuine phonemic distinction: брат is a brother, брать is to take, and the only difference is whether the final t is soft. It is marked by the soft sign or by which of two vowel letters follows, so it is present in the spelling and it is not present in any sound an English speaker already makes.
Realising every word in the deck needs its stress again
Here is the stall and it usually arrives around month six. A learner has built a deck of several hundred words from reading, none of them carrying a stress mark, and discovers that they cannot say any of them with confidence. The words are known and unusable, and re-learning the stress for all of them is a much larger job than learning it once would have been.
The response most people choose is to ignore it and keep reading, which stabilises into a reader who cannot speak. The response that works is two rules and a triage. Never store another word without its stress and its soft signs. Then take the two hundred most frequent items already in the deck, fix those, and let the rest be corrected as they resurface. Doing that on the whole deck is a month nobody has, and doing it on the frequent tail is a week, and the frequent tail is what you actually say.
One word, four forms
голова, and what happens as the stress moves
The same word in its own paradigm. The stressed syllable is the only one holding its full vowel in any frame.
- 01головаStress on the last, and the two o sounds before it go towards a
- 02головыStress moves to the first syllable, and the vowels rearrange behind it
- 03головуStress on the first again, with a different ending reduced almost away
- 04головойBack to the end, and the word sounds different from frame two
Nothing in the printed forms shows any of this, which is why a learner reading a declension table aloud will get most of it wrong and will not know. The table teaches the endings and hides the sound.
Two things that mark the stress for you
Russian books printed for children and for learners mark the stress on every word, and that is a genuine and underused resource: a page of children's Russian is a page where every word can be read aloud correctly on the first attempt. It replaced guessing from the spelling, which cannot be done at all, since nothing in the letters carries the information. Free in any library, and the vocabulary is simple enough that the reading is not the point.
A Russian text to speech voice is the second, and in 2026 the built-in voices are good enough to use as a check. Paste a sentence and listen to where the stress lands. It replaced needing a native speaker present to model a sentence you invented rather than a word somebody recorded. Free in the operating system and the browser. The limitation is precise and worth remembering: these voices get common words right and mis-stress rare ones and homographs, so treat it as a check on words you half know rather than as an authority on words you do not.
The mistake: hearing the soft consonant as a consonant plus a y
An English speaker told about soft consonants produces the hard one and then adds a small y sound after it, so брать comes out as brat plus a y. It is understood and it is not the distinction, which lives in the shape of the consonant itself rather than in anything added afterwards.
The way in is the vowel that follows. Say the hard consonant with the tongue already in the position for an ee sound, and the softness happens as a property of the consonant rather than as an extra segment. Minimal pairs are the training, listening before producing, because this is a category the ear does not yet have and the mouth cannot copy what the ear has not split. That ordering is the general principle for a sound that does not exist in your first language, and it is the argument set out in the stalls every learner hits, and the way through each.
What the sounds decide next
Every ending in the language is an unstressed syllable, so the case system and the reduction are the same problem heard from two sides, which is what each ending does, and which preposition demands it. A word stored from the page arrives with none of this attached, which is why the card design in roots and prefixes that build whole families of words has to carry the stress along with the meaning. Reading fast enough for any of it to be automatic depends on the alphabet being invisible, in thirty three letters, and the false friends among them. And what all this reduction does to a sentence at speed is real speed, and the sounds that disappear in it.