A vanished i pulled the vowel forward - then deleted itself, leaving the grammar fossilised in the vowel.
English keeps a strange little family of pairs where a changed vowel keeps the trace of a vanished conditioning sound: foot / feet, mouse / mice, full / fill, long / length, old / elder. There is one law behind all of them. About fifteen centuries ago, an i, ī or j standing in the next syllable reached back and dragged the stressed vowel forward in the mouth - and then, its work done, that little vowel weakened to nothing and dropped off. The conditioning sound is gone; other parts of the ending can survive, as -th does in length. So the rule reads backwards: you see the moved vowel and infer the lost trigger that must once have stood behind it.
Each pair below is one word in origin. The form on the right carries a fronted vowel because, long ago, it had an i or j after it that the form on the left never did - a plural, a causative, an abstract noun. Touch a pair to see the Old English forms and the ending that did it.
- the same vowel that fronts foot into feet fronts every pair here. Pick one.
This is the part that makes it a law and not a list. The outcomes depended on environment and dialect as well as the starting vowel. This map selects correspondences, mainly at the early West Saxon stage; it is not an exhaustive rule for every context. Pick a correspondence: the vowel slides forward to its reflex, and every doublet on the page that runs that rule lights up below. (These are the vowels at the Old English stage, where the change happened - modern spellings are these vowels carried on through later changes, the Great Vowel Shift among them.)
Pick a correspondence above to drive the rule.
Here is the whole strange move in one word, foot and feet, step by step. The plural ending arrives, conditions the vowel, then weakens and vanishes - and the plural now lives entirely in the vowel. Step it.
The cause deletes itself. This is why the rule reads backwards: you see the changed vowel (ē) and infer the lost trigger (-i) that must once have stood after it.
Given the vowel at the mutation stage and a reconstructed trigger, what does the vowel become? The starred word forms show earlier stages; the results are Old English. You now have the map: run it.
The interactive instruments need JavaScript. The word pairs on the wall and the sources are readable without it.
i-mutation is conditioned, not free. The vowel changes ONLY because an i, ī or j stood in the next syllable. That is why foot keeps its ō in the nominative/accusative singular: those forms lacked the trigger. The Old English dative singular fēt did have it. No trigger, no mutation - which is exactly what makes the surviving doublets readable as fossils of lost endings.
In standard Modern English the clean mutation plurals are just feet, geese, teeth, men, women, mice, lice. They are not a living pattern: by Middle English most former mutation plurals had been re-formed on the regular -s model. Book once had the OE umlaut plural bēc - now it is books, not *beech (and beech the tree, though related to book far back, is NOT that lost plural). Oak, goat, friend, borough all had mutation plurals too, now levelled to -s. Frequency is one proposed contributor to their survival, not a complete explanation for the set.
The visible woman → women alternation is in the FIRST vowel, and its history is tangled (OE wīfmann → wimman, with a later vowel split), not a clean i-umlaut of the kind in foot → feet. We name it rather than use it as a clean example.
The archaic or dialectal plural kine (cattle) keeps the umlaut of OE cū → cȳ. It adds -en to the already mutated plural ky: a double plural, like brethren in combining umlaut with a later ending.
The same Germanic change gives German Fuß/Füße, Maus/Mäuse, Mann/Männer, lang/Länge - and German still SPELLS the front vowel with the umlaut dots ä ö ü. English merged the mutated vowels into ordinary spellings, then ran them through the Great Vowel Shift (so OE ē in fēt became the modern /iː/ of feet). The dots are gone; the change is the same.
The verifier checks the dataset is internally consistent with the fronting map and that the asterisks fall only on reconstructions. It does not re-prove Old English philology, and the starred trigger forms are schematic reconstructions at differing historical stages, checked against the cited grammars and etymologies rather than quoted verbatim from them.
i-mutation is usually dated to an onset around 450–500 CE, working earliest and most completely in Old English and Old Norse, and essentially complete before the earliest Old English texts (7th–8th c.). It is shared in broad outline across West and North Germanic (English, German, Dutch, the Scandinavian languages) but absent from Gothic, the one well-recorded East Germanic language - which is why it is dated separately per language rather than as one event. The exact relative chronology, and how much was a single change versus parallel developments, is debated; we keep the dating an honest range.
The only new thing here is the form: a standard, textbook sound change made into something you can drive - apply the fronting map and watch the vowel move, step the trigger as it conditions the vowel and then deletes itself, and read a wall of Modern English doublets back to the lost endings that shaped them.
Nothing here is a new claim in historical linguistics. The sound change and doublets are established material; the map selects correspondences, and the starred triggers schematise the histories given in the cited grammars and etymologies. Every pair is checked by an offline verifier (research/i-mutation/verify.mjs): that each doublet's fronted vowel really is the map's image of its base vowel, that every mutated form carries a stated i/j-bearing trigger, that the asterisks fall only on reconstructions, and that the data drawn in this page is byte-identical to the source of truth.
Corrected 2026-10-14: The page said whole endings disappeared and left only a changed vowel. The conditioning i/j was lost or altered; other suffix material survives, including -th in length and health and -an in OE settan and fyllan; found by the assay line, 2026-10-14. Length etymology.
Corrected 2026-10-14: The page said every vowel outcome depended only on its starting position in the mouth. Historical outcomes also depend on environment and dialect; this map shows selected correspondences; found by the assay line, 2026-10-14. Old English correspondences.
Corrected 2026-10-14: The mouse/mice gloss put /iː/ after unrounding and the Great Vowel Shift. Unrounding gave /iː/; the later shift led to modern /aɪ/; found by the assay line, 2026-10-14. Mice vowel stages.
Corrected 2026-10-14: The page excluded kine from i-mutation and called lice the only other surviving ū → ȳ plural besides mice. Archaic or dialectal kine adds -en to the already mutated plural ky, continuing OE cȳ; found by the assay line, 2026-10-14. Kine etymology.
Corrected 2026-10-14: The breadth entry used *braidiþu and said -th was re-added. OE brǣdu is an īn-stem formation, represented by Proto-Germanic *braidį̄; -th was added later by analogy; found by the assay line, 2026-10-14. Brædu etymology and declension.
Corrected 2026-10-14: The page said the singular of foot never had *-i. The nominative/accusative singular lacks that trigger, but the OE dative singular fēt also reflects an i-bearing ending; found by the assay line, 2026-10-14. Wright §408.
Corrected 2026-10-14: The page attributed every displayed trigger reconstruction to the reference grammars. Wright §382 places breadth in the īn class, not the displayed *-iþu formation. That entry is corrected, and the sourcing note distinguishes schematic triggers from quotations of the grammars; found by the assay line, 2026-10-14. Wright §382.
Corrected 2026-10-14: The page said the map table worked without JavaScript. Its rows are created by JavaScript; the static wall pairs and source list remain readable without it; found by the assay line, 2026-10-14. The page.
Corrected 2026-10-14: The Campbell citation identified §§115–204 as i-mutation and §656 as comparatives. The contents give §§190–204 for i-mutation and §§657–660 for adjective comparison; found by the assay line, 2026-10-14. Campbell contents.
Written 2026-06-21. Claims re-read against their sources on 2026-10-14: 138 checked, 109 confirmed, 29 wrong, 0 unverifiable. By the assay line (research/assay-line/cited-1014/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex.
This pass re-read the evidence for all 30 directed triage items for i-mutation at repository read 01f4ee4471, corrected 22 WRONG and seven MINOR items through this layer’s sources and generator, and rejected X29 with evidence. It carries forward the 108 CONFIRMED claims and their qualifications from the Codex source/checker record in research/assay-line/cited-1014/i-mutation/N1-sources.json and N2-codex.md; those are the earlier checker’s readings, not a claim that this fixer repeated all 108 external checks. MINOR lines count in wrong here, with a separate finding for each. Nine page corrections cover the repeated WRONG items and are mirrored in the catalogue body. MINOR changes have no dated page correction. The full edit and command report is research/assay-line/cited-1014/i-mutation/fix-report.md. Literal \u2014 in carried quotations represents the original punctuation.
This record is long (138 claim lines and 29 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.
The assay office: what a record is, and every layer re-read so far.