Compensatory lengthening

Why does a word that loses a consonant so often keep its old length?

Germanic once had a word for the bird we call a goose that contained a nasal before the fricative — something close to gans. English lost the nasal. But it did not end up with a short gos. It ended up with Old English gos carrying a long vowel, which the later vowel shift turned into the modern diphthong. The same thing happened to tooth from a form with a nasal, and to five, and to us.

So something disappeared, and the word stayed the same size. That is a strange thing for a deletion to do. Deleting a letter from a written word makes it shorter. Why should deleting a sound from a spoken one so often leave the word measuring exactly what it did before?

The first move is to notice that we have been counting the wrong things. We have been counting segments — individual consonants and vowels. Ask instead whether speakers were ever counting segments at all.

There is good reason to think they count something else. Across many unrelated languages, stress placement, poetic metre and minimum-word requirements all care about a distinction between heavy and light syllables. In Latin, stress falls on the second-to-last syllable if it is heavy and retreats further if it is light. What makes a syllable heavy? Either a long vowel, or a short vowel followed by a consonant in the coda. Note what that lumps together: a long vowel and a vowel-plus-consonant count as the same thing.

That is a striking equivalence, and it invites a unit. Call it the mora — a unit of syllable weight. A short vowel is one mora. A long vowel is two. A coda consonant, in a language that counts it, contributes one. So a light syllable has one mora and a heavy syllable has two, whether the second mora came from the vowel or from the consonant.

Now return to gans. In moraic terms the syllable is: vowel, one mora; nasal coda, one mora; total two. Now delete the nasal. Ask the key question: does the mora go with it?

Here is the claim, and it is the whole mechanism. The segment is deleted but its mora is not. The weight of the syllable is what the grammar is tracking, and it persists as a slot that must be filled. The only remaining material that can spread into that empty slot is the vowel, so the vowel spreads into it — and a vowel occupying two morae is, by definition, a long vowel. Gans becomes gos. The length was not created out of nowhere. It was inherited from the consonant that vacated.

Test the idea by asking what it predicts. It predicts that lengthening should only happen when the lost segment was in a weight-bearing position. And that is largely what we see: consonants lost from the onset of a syllable, which does not contribute weight, generally do not trigger lengthening. It also predicts that in languages whose codas never count for weight, coda loss should be silent — and such languages do exist, and it is.

The historical record supplies further cases. Pre-Greek forms in which a sibilant was lost before a nasal show the same pattern — Attic Greek eimi 'I am' beside forms preserving the sibilant. Latin canus 'grey' descends from a form with a sibilant that was lost with lengthening. Old English lost a velar fricative in night and the vowel lengthened before the Great Vowel Shift got hold of it. Different families, same trade.

A caution about how settled this is. The moraic account, made explicit by Bruce Hayes in 1989, is the standard framework, but it is a formalisation of a pattern rather than a proven cause, and it does not explain every case. Some languages compensate by lengthening a following vowel, or a neighbouring consonant, rather than the preceding vowel. Some lose a coda with no compensation at all. And there is a live debate about how much of this is phonological representation and how much is the phonetic residue of a gradual weakening — a consonant that reduced to almost nothing while the tongue's movement into the vowel simply expanded to fill the time.

The name invites a misconception worth correcting: nothing is being compensated for in the sense of a repair carried out to preserve meaning or to keep words distinguishable. No speaker is protecting the length of goose. The lengthening is a mechanical consequence of a representation in which weight is a separate tier from segments, so a segment can go while its weight stays.

It is also worth being precise about what is conserved. Not duration in milliseconds — speech rate varies constantly and nobody is preserving absolute timing. What is conserved is a countable structural quantity, the syllable's weight, which is the thing the rest of the grammar — stress, metre, minimality — reads. This is a conservation law of the same kind as those in physics only in a limited sense: it holds because of how the system is built, not because of anything outside it, and it can be violated when the system is built differently.

That framing also explains why compensatory lengthening is so common across unrelated families. It is not a shared inheritance or a borrowing. It falls out automatically wherever two conditions happen to co-occur: codas count for weight, and a coda is lost.

• Double flop and other cases where the lengthening lands on a vowel that never neighboured the lost consonant.
• Why gemination — lengthening a consonant instead — is the compensation chosen in some languages.
• How metre in Greek and Sanskrit verse gives independent evidence that speakers really counted weight.

Every term the collection defines is gathered in the glossary.