Voicing Assimilation and Final Devoicing

Polish spelling is famously phonemic — what you see is, almost always, what you say. But there are two systematic exceptions, and they are exactly the ones English speakers walk straight past. Voiced consonants quietly turn voiceless at the end of a word, and inside a consonant cluster every obstruent surrenders its own voicing to whatever obstruent comes last. The letters do not change; the sounds do. Once you know these two rules, dozens of "surprising" pronunciations stop being surprising — and you understand why the same stem looks voiced in one form and voiceless in another.

Rule 1: Final devoicing

At the end of a word, a voiced obstruent (b, d, g, w, z, ź, ż/rz, dz, dź, dż) is pronounced as its voiceless partner. The spelling keeps the historically voiced letter, but the sound is voiceless.

The crucial point is that the spelling is etymological, not phonetic here: chleb is written with b because the stem really is voiced — it just cannot show its true colours at the end of a word. Add an ending that puts a vowel after it, and the voiced consonant resurfaces immediately.

Listen to those two: chleb ends in a clear "p" sound, but chleba (genitive) has an unmistakable b, because now a vowel follows it. This is the single most useful diagnostic in Polish spelling.

Rule 2: Regressive voicing assimilation in clusters

When two or more obstruents sit next to each other, the last one decides the voicing of the whole cluster. The earlier consonants assimilate backwards (hence "regressive") to match it. Again, the spelling stays put while the sound shifts.

Notice that assimilation runs both ways: a voiced consonant devoices before a voiceless one (wódka), and a voiceless consonant voices before a voiced one (także, prośba). The direction is always backward — from the last obstruent to the earlier ones.

In prośbę you say a clear ź even though it's written ś, because the following b is voiced and drags the ś into voice. In także the written k comes out as g before ż.

The special case of w and rz

Two consonants behave asymmetrically, and this trips up nearly everyone. The letters w and rz are voiced, so you'd expect them to obey the cluster rule both ways. They half-do: they devoice after a voiceless consonant, but they do not voice a preceding voiceless consonant. They are, in a sense, "weak" — they give in to a neighbour but cannot impose on one.

In twój, the w comes after voiceless t, so it devoices: you say "tfuj," not "tvuj." But the t stays t — the w could not pull it into voice.

Here przez begins "psh-" (the rz devoices after p) and trzy begins "tsh-" (the rz devoices after t). The p and t stay voiceless. This single quirk fixes the pronunciation of hundreds of common words starting tw-, kw-, św-, przy-, trz-, krz-; it has its own dedicated page, the devoicing of w and rz, because it is that important.

Why the spelling stays voiced

The logic is morphological. Polish spells a stem the way it appears when it is "protected" — when an ending shields the consonant from word-final position or from a devoicing neighbour. The voiced stem of chleb shows itself in chleba, chlebem, chlebowi; the orchard sad shows its d in sadu, sadem, sady. The spelling commits to the underlying form and lets the two automatic rules handle the surface. This is also why the genitive (and other vowel-final case forms) is the great revealer: it is the form where the consonant is least disturbed.

You hear "sat" in the first sentence and "sadzie" (with dz, the softened d) in the second — same stem, two surfaces. This interlocks with the rz-versus-ż spelling rules: because rz devoices to "sh" after p/t/k, learners constantly mishear it and want to write sz — but the spelling stays rz for the same etymological reason the b stays in chleb.

For English speakers

English actually has final devoicing too — but partial and unreliable, so you've never had to notice it. The end of English bed is a weak, half-devoiced d, and crucially English uses vowel length to carry the contrast: the vowel in bead is long, in beat short. Polish does it cleanly and completely: a final voiced obstruent becomes fully voiceless, with no vowel-length compensation. So you must devoice harder than feels natural. Bóg is a clean "Buk," not a soft "Bug."

The bigger trap is cluster assimilation. English does not freely voice a consonant to match the next one — English keeps footpath with a "t," not a "d." So when an English speaker meets prośba, the instinct is to keep the ś voiceless ("proshba"). You must let it voice to ź ("proźba"). Train yourself to scan a cluster for its last obstruent and set the whole cluster to that voicing.

Common Mistakes

Key Takeaways

• Final devoicing: a voiced obstruent at word-end is pronounced voiceless — chleb = "chlep," Bóg = "Buk," wóz = "wus." The spelling stays voiced.
• Cluster assimilation: the last obstruent sets the voicing for the whole cluster, backwards — wódka = "wótka," prośba = "proźba," także = "tagże."
• Sonorants (l, ł, m, n, ń, r, j) and vowels never trigger assimilation; only obstruents do.
• w and rz devoice after a voiceless consonant but never voice a preceding one — twój = "tfuj," przez = "pszez."
• The declension test reveals a stem's true final consonant: add a vowel ending (genitive) and listen — sadu shows the d of sad.