There’s no shortage of proposed etymologies for she, the third person singular feminine pronoun, but its origins remain uncertain. It appears to have arisen in the 12th century, but how it did so has proved difficult to establish. The Shorter Oxford English Dictionary says it is
probably a phonetic development of Old English hīo, hēo hoo pronoun feminine of he pronoun. Other suggested etymologies include derivation from Old English sēo, sīo feminine adjective . . . or from hypothesized forms in West Germanic.*
This array of possibilities corresponds, more or less, with the “probably”s of other authorities. According to the American Heritage Dictionary, she is “probably [an] alteration of Old English sēo, feminine demonstrative pronoun”. Eric Partridge, in Origins: A Short Etymological Dictionary of Modern English, says it comes
through Middle English she (earlier scae), variants sche, scheo, scho:? from Old English sēo, variant sīo, feminine of the article ‘the’, originally a demonstrative pronoun: cf Old High German siu, Middle High German siu, sie, sī, German sie, Old Saxon siu; cf also Sanskrit syā, this one.
This tallies with the route outlined by Walter Skeat in his Concise Etymological Dictionary of the English Language:
Middle English sche. Anglo-Saxon seó, used as feminine of definite article, but in the Northumbrian dialect as demonstrative pronoun. Feminine of se originally ‘he;’ cognate with Mœso-Gothic sa, that.
Merriam-Webster dates its first known use to the 12th century, and suggests its origin is
probably [an] alteration of hye, alteration of Old English hēo she
Douglas Harper’s Online Etymology Dictionary has a comparatively detailed entry:
mid-12c., probably evolved from Old English seo, sio (accusative sie), feminine of demonstrative pronoun se “the.” The O.E. word for “she” was heo, hio, however by 13c. the pronunciation of this had converged by phonetic evolution with he “he,” so the feminine demonstrative pronoun probably was used in its place (cf. similar development in Dutch zij, German sie, Greek he, etc.). The original h- survives in her. A relic of the O.E. pronoun is in Manchester-area dialectal oo “she.”
An alternative that’s generally overlooked, as it is by each of the above, is the possible connection with the Irish word sí /ʃi:/. It means “she”, it is pronounced identically to she, and it can be traced back to Old Celtic. Its roots appear, inevitably, in MacBain’s Etymological Dictionary of the Gaelic Language:
she, Irish í, sí, Old Irish í, hí, sí, Welsh, Breton hi: *sî; Gothic si, ea, German sie, they; Sanskrit syá: Indo-European sjo-, sjā- (Brug.).
Loreto Todd, in Green English: Ireland’s influence on the English language, makes the case for sí:
in the tenth century, there were parts of England where the same pronoun he could mean ‘he’, ‘she’ and ‘they’. Such a high degree of ambiguity was not allowed to continue. Speakers, especially in the north of England, began to adopt the Norse forms þai, þeʒʒm and þeʒʒre, which developed into modern ‘they’, ‘them’ and ‘their’. The change from he(o) to ‘she’ is much less easy to account for. No dialect of English or Norse had a personal pronoun that would or could have developed directly into ‘she’, although many etymologists have struggled to explain it by invoking combinations of Old English and Old Norse personal pronouns and by suggesting that the demonstrative pronoun seo, probably pronounced like ‘say + o’, can help in explaining the shift from ‘he’ to ‘she’.
Proposing sí as a conceivable inspiration, Todd notes the presence of Irish clerics and scribes in many communities in England at the time, and quotes from Martyn Wakelin’s English Dialects: An Introduction:
the early Scandinavian settlements (ninth century and earlier) in this country were mainly Danish and were on the Eastern side of England. Norwegian settlements occurring somewhat later (mainly in the first half of the tenth century by men who had been living in Ireland) were in the northwestern counties and the North and West Ridings of Yorkshire.
The emphasis is Todd’s. She acknowledges that this etymology is controversial, but I don’t know how controversial it is. Maybe it’s simply dismissed as unlikely, though at least one scholar finds Todd’s contention “satisfactory” (Radoslava Pekarová, The Influence of the Irish Language on Irish English Grammar; PDF, 308 KB).
But the change from he(o) to she might not be so unlikely or unusual. A. H. Smith’s Some Place-Names and the Etymology of “She”** (1925) showed that there are
certain place-names in the north of England and in Scotland which illustrate a peculiar sound development in English. Old English initial he—in these cases shows a tendency to become late Middle English sh-[ʃ].
the evidence of the place-names . . . shows that a development of O.E. he- to M.E. sch- did take place, which could explain the derivation of modern English she from O.E. hēo, especially as the periods when M.E. ʒhe, ʒhe and M.E. sche, scho were prevalent agree more or less with the periods when He-, Hy, Yh- and Sch-, Sh- were prevalent in the place-names
and summarises as follows:
Smith’s research forms part of the evidence to which M. L. Samuels refers in his analysis in Linguistic Evolution: with special reference to English (1975):
The ME reflex of OE heo was he, so that large areas of the country were left without a formal distinction between ‘he’ and ‘she’, while even in the remaining areas the other surviving forms (hy, heo) were not ideal for the purpose. This systemic gap, which can be shown from the ambiguities in many surviving texts, was filled by a typical drag-chain process – the selection of originally rare variants, the stress-shifted forms /hjo/ and /hje/. These then changed, via the intermediate stage /ço, çe/ to /ʃo, ʃe/, perhaps first in the heavily Norse-influenced Cumberland-Yorkshire belt which provides numerous parallels for the change. (pp. 114–116)
Some of this material is based on Samuels’ earlier paper, The Role of Functional Selection in the History of English*** (1965), which includes the following image and accompanying text:
The intermediate stage /ç/ would, as pointed out by Vachek, survive for a time as a marginal phoneme (spelt ʒ or ʒh), but would naturally give way to /ʃ/, which was equally distinctive for the purpose of functional differentiation, yet far better integrated phonemically.
Convincing proof for this theory (as against the older derivations from Old English sēo or Old Norse sjá) is now available from study of the Middle English distributions of forms. The later ME distribution, schematized in Fig. 3, shows ʒ(h)-forms in border areas which divide the newer s(c)he, s(c)ho in the Midlands from the older he, hy, heo in the south. . . . In other words, a belt of ʒ(h)-forms started in the northern belt and moved southwards across the country, always followed, after what appears to have been a comparatively uniform lapse of time, by s(c)h-forms. . . . we may justifiably conclude that the modern form she arose from an originally unusual phonetic variant in the spoken chain, and that it spread to large areas in which, failing such a form, the pronominal system was wholly lacking in balance.
The argument is developed in detail in Samuels’ book and paper. Although I found it fascinating, I’m not qualified to assess it, and I don’t know if there are more recent findings that substantially confirm or contradict it. Certainly there is nothing like a consensus on the matter, and it’s easy to see why the OED describes the etymology of she as “difficult”!
There’s a short and helpful discussion of the various possibilities in the comments of this post at Language Hat, where linguist Marie-Lucie says that “the Old English and Old Irish s- forms could have a common ancestor in Proto-Indo-European, or at least a branch of it”.
I love a good mystery, but this is one I wouldn’t mind seeing resolved. And much as I’m taken by Todd’s sí-hypothesis, I wouldn’t bet on it.
* In this and other quoted text, I’ve written some abbreviations in full.
** Review of English Studies, 1(4), Oct. 1925, pp. 437–440.
*** Transactions of the Philological Society, 64(1), Nov. 1965, pp. 15–40.