The fallac[ɪj]of schwee

hap-peaIn a recent blog post, John Wells returned to the topic of the happY vowel and the symbol i, or “schwee” as they’re often nicknamed. He began:

It seemed like a good idea at the time, but it clearly confuses a lot of people.

Though widely adopted in dictionaries and textbooks, schwee certainly has been widely misinterpreted, and I’ll try to show here that the common misinterpretations are not a good idea. Further, I think that even the original, authentic schwee has been rather undermined by changes in standard BrE over the last generation or two, so that it no longer really serves a useful purpose.

John went into greater detail in an earlier post:

At the end of words like happy speakers of traditional RP, as represented by Daniel Jones (and for that matter by me) have a vowel that is clearly to be identified with the ɪ of grin, so ˈhæpɪ.

However many other accents, including what you might call today’s neo-RP, have a tenser vowel that speakers identify instead with their . We might transcribe happy in this newly respectable pronunciation as ˈhæpiː.

In 1978, John tells us, “pronunciation editor Gordon Walsh introduced the additional symbol i, to cover such cases”. John emphasizes that this additional symbol is an “abbreviatory notational convention”, standing for two equally recommendable alternative possibilities – and not a third vowel in addition to FLEECE and KIT:

This convention saves space. It means that we do not need to give two separate pronunciations for the thousands of words involved: instead of
coffee ˈkɒfɪ, ˈkɒfiː
happy ˈhæpɪ, ˈhæpiː
valley ˈvælɪ, ˈvæliː
etc. we just put
coffee ˈkɒfi
happy ˈhæpi
valley ˈvæli

As John points out, the history of phonological theory has a more technical term for this kind of thing: “Trubetzkoy would have spoken of the ~ ɪ archiphoneme”. We could say that schwee as originally intended is not a vowel at all, but a disjunction of two vowels, FLEECE or KIT. It doesn’t have a pronunciation – it stands for two alternative pronunciations.

Hindsight is a wonderful thing. Today it seems perhaps unsurprising that an archiphoneme would be misunderstood. In fact we can find two flavours of misunderstanding, or two False Schwees. Whereas True Schwee is archiphonemic, one False Schwee is phonemic, and the other False Schwee is allophonic.

The False Schwees seem to have displaced True Schwee in many people’s minds at various academic levels. I recently gave a talk at Middlesex University, where I found that excellent students of English & TEFL were under the impression that English has three close front unrounded vowels, namely FLEECE, KIT and (whether they know the nickname or not) schwee. Unfortunately, this is not merely untrue of English but is, I think, a violation of a phonological universal. Unless I’m mistaken, human language cannot distinguish three close front unrounded vowels.

A three-vowel picture is presented at the highest level. Here, from a textbook aimed at Dutch teachers of English (English pronunciation for student teachers, Noordhoff Uitgevers, 1997):

RP /iː/, as in piece, sea, is a close, front unrounded vowel. It is typically somewhat diphthongised, much like [ɪi]…

RP /ɪ/, as in sit, is a mid-close, centralised front, unrounded vowel…

RP /i/ is used word-finally, as in city /ˈsɪti/… and before vowels, as in mediocre /ˈmiːdiˈəʊkə/… Older RP speakers may have a vowel quality approaching that of [ɪ], but otherwise a close, short vowel is used…

Note the parallel use of brackets, giving the impression that /i/ is a phoneme like the others, with various allophonic realizations, such as [ɪ]. Let me point out that the authors, Carlos Gussenhoven and Ton Broeders, are not only scholars of the first rank and awesomely native-sounding speakers of English, they’re also my erstwhile employers (I’m one of the native speakers on their accompanying CD). I’m certain there are no better teachers of English pronunciation for Dutch speakers; but “/i/” as they present it is not an “abbreviatory notational convention”, and rather implies that English exemplifies what I suspect is a phonological impossibility.

Let’s turn now to the allophonic kind of False Schwee. It’s perfectly true that many speakers often realize the underlined vowels of happy and mediocre as a short close front vowel which we can show in narrow transcription as [i]. But such context-specific realizations, or “allophones”, are not generally represented in broad, dictionary-type transcription. Compare for example the vowel of the word they.  In citation form, with a nuclear accent, it’s long and clearly diphthongal, [ɛj]. However, most of the time this word is unstressed, as in When are they coming?, in which case the vowel will be much shorter and less diphthongal, something like [e]. But this narrow detail isn’t something we’d include in a broad transcription or list in our dictionaries, eg

they ðɛj, ðe

The same thing can be said of the stressed and unstressed forms of GOAT in eg low and yellow.  The narrow difference is not the kind of thing we include in broad transcriptions, though it’s audible enough in, for instance, the Macmillan online dictionary:

So although [i] is a common realization of the final vowel in happy, this is not the intended meaning of John’s “abbreviatory notational convention” i which, to repeat, stands for the alternatives “/iː/” or “/ɪ/”.

Another objection to allophonic False Schwee is that [i] is certainly not the only realization of the final vowel in happy: we frequently hear pronunciations with exactly the same quality and quantity as stressed FLEECE. Here’s BBC newsreader Kasia Madera saying party with a final vowel which I would transcribe narrowly as [ɪj] rather than [i]:

And here is Prince William making a speech in Nottingham last week:

During the lifetime of Prince William, born the same year as John Wells’s masterpiece Accents of English, BrE has shifted considerably. Some older RP speakers (including John himself) still have KIT in happy and coffee, but this has turned out to be highly recessive in the face of happy=FLEECE, whether realized as [i], [iː] or [ɪj]. There is no good reason to transcribe William’s opportunity, charity etc with anything but the FLEECE vowel.

It might be worth pointing out to non-native readers how dated native Brits now find RP happY vowels like those in the following quality, vitality and family, from old information films and the Queen’s 1957 Christmas broadcast:

(In the second clip there we hear happY encroaching on the RP DRESS vowel [e̞]. Schwee hobbyists shouldn’t miss Jack Windsor Lewis’s discussion here of a variety of old happYs.)

Of course, final [ɪ] is still used in some regional accents.  Here is Lancashire comedian Peter Kay saying there were a guy in t’toilets collared me and very weird:

There we hear KIT in both the unstressed pronoun me and finally in very.  But in standard BrE, happy=FLEECE now predominates.

And this brings us back to True Schwee, the abbreviatory archiphoneme which stands for unstressed FLEECE or KIT. When it was first used in the 1970s, there was at least some justification for abbreviating ˈkɒfɪ, ˈkɒfiː to coffee ˈkɒfi, because the two alternatives were equally recommendable. But this is simply no longer the case. In any given phonetic environment, either FLEECE and KIT are contrastive, or else one is predominant and recommendable, while the other is dated or a regionalism. And with the loss of contexts in which FLEECE and KIT are equally acceptable, the raison d’etre for the abbreviation i is also lost.

On the contrary, the FLEECE-KIT distinction carries an immense functional load; one has only to think of pairs like sheet-shit, beach-bitch and peace-piss. And for most learners of English, this contrast is very hard to perceive and to produce: in teaching, therefore, the distinction should be prioritized and insisted upon. Encouraging learners to think of FLEECE and KIT as neutralizable, while burdening them with a non-existent third phoneme, is pedagogical nonsense. If BrE were being analyzed and transcribed for the first time today, I doubt that any linguist would use schwee: as John Wells made clear in his blog, it was only invented as a sop to the obsolescent happy=KIT pronunciation, preserving it alongside the new standard happy=FLEECE.

In stressed syllables, the FLEECE and KIT vowels ɪj and ɪ are contrastive and have to be learned word by word.  (The orthography provides guidance, as in fleece and kit, although there are some tricky cases like mini with ɪ v. linguine with ɪj.)  For unstressed syllables, the basic rule is simple:

Unstressed FLEECE/KIT Rule
• if a consonant follows, ɪ (ə for some speakers/accents)
• if not, ɪj

Let’s go through some examples. First, inital unstressed syllables:

• if C follows, ɪ (ə for some):
retain, menagerie, beneath, fidelity, prefer, divine, enough
• if not, ɪj:
create, meander, beatify, fiesta, preoccupy, viola

Distinct from these unstressed syllables are the stressed “word-level” prefixes, re-, pre– and de-. These have transparent meanings, can be added to words rather freely and are sometimes written with a hyphen.  They have FLEECE even if a consonant follows, unlike the unstressed “root-level” prefixes with the same spelling which do not have transparent meanings and cannot be freely added:

re- ‘again’ eg ˌrɪjˈeducate, ˌrɪjˈheat cf rɪˈduce
pre– ‘previously’ eg ˌprɪjˈowned, ˌprɪjˈcooked, cf prɪˈfer
de– ‘undo’ eg ˌdɪjˈactivate, ˌdɪjˈpressurize, cf dɪˈcide

Moving on to medial syllables, here are examples before the primary stress:

• if C follows, ɪ (ə for some):
irrigation, elevation, animosity, California
• if not, ɪj:
variation, permeation, grandiosity, Riviera

And medial syllables after the primary stress:

• if C follows, ɪ (ə for some):
telescope, caffeinated, aniseed, audible
• if not, ɪj:
deviate, radiator, lineage, audio

Lastly, we come to final syllables.

• if C follows, ɪ (ə for some):
coffin, trumpet, minute, porridge, ragged
• if not, ɪj:
coffee, mini, chimney, twenty, rally

Traditional RP’s use of KIT in absolute final position was rather anomalous, since English generally allows short-lax vowels only before consonants. RP made exceptions to this generalization in allowing unstressed KIT and FOOT to occur finally, as in happy ˈhæpɪ and thank you ˈθæŋkjʊ – neither of them extinct, but both of them now old-fashioned and not very recommendable to learners, who tend to be young. We can now see that modern BrE has regularized the system by shedding these forms, so that none of the English short vowels, except final unstressed schwa, may occur without a following consonant.

Where KIT occurs in unstressed syllables, some speakers/accents use schwa, or an intermediate quality like [ɨ].  This is true of many Americans, but also some BrE speakers.   In the following samples, former Prime Minister Tony Blair exhibits such schwa-like qualities in regarded, activated and minutes:

Note that this is not possible in absolute final position: whereas happɪ is merely old-fashioned, happə is out of the question. I take this as confirmation that, when no consonant follows, the contemporary language chooses the FLEECE option, not the KIT/schwa option.

Finally, there are some apparent exceptions to the Unstressed FLEECE/KIT Rule, but these fall into two systematic categories. One category arises when word-formation processes attach a further consonant to a word-final unstressed ɪj. For example, if we add the inflectional endings –s or –ed to pity or rally, we get pities, rallies, pitied, rallied. In these words many speakers keep the FLEECE vowel even though it’s now preconsonantal. Here, for example, is the BBC’s Jeremy Paxman saying our most elite universities:

So for many speakers, there’s a difference between pitied pɪtɪj+d and pitted pɪt+ɪd, between rallied ralɪj+d and valid valɪd.  Other speakers, however, conform more fully to the Unstressed FLEECE/KIT Rule by switching from FLEECE to KIT when a consonant is attached to the end of the word. Both kinds of speaker work for Sky Movies:

(A similar variability applies to the -(e)y in compounds like bellydancer and chimneysweep, the portmanteau gerrymander, and also poly when it’s used as a productive word-level prefix transparently meaning ‘many’, eg polysyllabic, polymath. Cf polythene, which definitely has KIT and not FLEECE.)

Another category of apparent exception is illustrated by words like chlorine, athlete, phoneme, species, colleague, caffeine. The underlined vowels contain FLEECE despite a following consonant, within the same morpheme. However, these can be analyzed as containing a secondary stress: ˈchloˌrine, ˈathˌlete, etc. Such an analysis is supported by words such as ˌchloriˈnation and ˌdeˌcaffeiˈnation, where the underlined vowels find themselves between two stressed syllables and become fully unstressed; as a result, they switch from FLEECE to KIT and thus conform to the Unstressed FLEECE/KIT Rule.

To summarize. A phonemic misinterpretation of schwee is wrong because schwee isn’t contrastive, and because it implies a linguistically odd triple set of high front unrounded vowels. An allophonic misinterpretation of schwee is inappropriate for broad transcription, and makes no allowance for realizations other than [i]. Schwee in its authentic, archiphonemic interpretation was justified to the extent that unstressed FLEECE and KIT were equal alternatives in certain contexts; in contemporary BrE, however, the distribution of unstressed FLEECE and KIT has been regularized – KIT before a consonant, FLEECE elsewhere – so that archiphonemic schwee obscures valid generalizations and recommends inadvisable pronunciations.

All of these interpretations involve an addition to the list of English vowel symbols, with schwee generally accompanied by its even weirder sister schwoo, u. (In a nutshell, schwoo gives a common description to the middle vowel of occupy and the first vowel of fruition, the implication being that both can equally be GOOSE or FOOT.  Whether or not this was true of RP, it doesn’t match what I tend to hear in BrE today, namely schwa in preconsonantal forms like occupy and GOOSE elsewhere.)

Some readers will feel that schwee is a familiar symbol which does no harm. But the English vowel system is large and difficult enough without adding extra symbols, and we should keep future learners in mind. Carlos and Ton’s Dutch readers at least have the advantage of coming from a first language with a large and difficult vowel system – Dutch has almost as many vowels as RP. But think of, say, Spanish speakers. For every one of their five native vowels, the RP system confronts them with four. (If you’ve studied the piano, imagine encountering a new keyboard where each octave is divided not into twelve keys but forty-eight.) English vowels are a huge challenge, and I think we should take every opportunity to economize and prioritize.

Here is a chart of vowel frequency (in AmE) from Ladefoged and Disner’s Vowels and Consonants:

The prevalence of schwa, KIT and FLEECE is attributable to their being the three vowels which occur in unstressed as well as stressed syllables (ə here also includes stressed STRUT vowels; a proportion of unstressed ə tokens here would probably come under ɪ in BrE). To me, this chart suggests that anyone who wants to sound like a native English speaker needs to prioritize these three vowels – their perception, their production and their distribution. How much are learners helped by relegating “weak vowels” to a secondary group containing additional symbols which actually obscure and/or confuse the differences between them?

Of course a “polysystemic” view of vowels in stress languages is well motivated.  Many languages have diminished vowel possibilities in unstressed syllables.  Russian, for example, has both e and i in stressed syllables, but only i in unstressed syllables.  In BrE, ɪj and ɪ freely contrast in stressed syllables, while in each unstressed environment one is pr[ɪ]ferred over th[ɪj]other.  The question is whether it makes sense to describe a decrease in phonetic possibilities by means of an increase in the vowel inventory.  I tend to think twent[ɪj]is more than [ɪ]nough.