vowel inventories

It is good from time to time to stand back and consider what the phonetics of familiar languages look like from the perspective of language universals, and how they might appear to those who don’t know them.

Vowel systems of different languages vary widely in size.

The ever-fascinating World Atlas of Language Structures Online devotes the second of its 142 “feature” articles to Vowel Quality Inventories. There is (i) a narrative discussion and (ii) a customizable map of the world plotting vowel systems with unusually small (2–4) or unusually large (7–14) inventories. Various theoretical or procedural issues, such as the status of diphthongs, have to be settled before you start counting. That done, though, the average number of vowels in a language turns out to be just fractionally below 6. The smallest vowel quality inventory recorded is 2 and the largest 14.(Click here for the full-size map.)

And which language would hold that record number of fourteen vowels? Standard German! (Count them: iː ɪ eː ɛ aː a yː ʏ øː œ uː ʊ oː ɔ, to which you could of course add ɛː for some speakers. The weak vowels ə ɐ are not counted, nor the nasalized vowels used in borrowed French words.) The “variety of British English included here” is reckoned to come in equal second place, with thirteen. Its posited inventory presumably comprises iː ɪ e æ ɑː ɒ ɔː ʊ uː ʌ ɜː eɪ əʊ, the last two being regarded as unitary. Not only the remaining diphthongs but also schwa are excluded.

Considerably more languages have an inventory of five vowels than any other number — just over a third of the sample. Familiar examples would be Spanish, Greek, Japanese and Swahili, all with just i e a o u.

Polish adds one more, ɨ, giving six vowels (ignoring the nasalized ones, which we can arguably analyse away). Korean adds two, ʌ ɯ, giving seven.

Four languages in the sample have only two contrasting vowel qualities.

You can see why many foreign learners of English (but not speakers of other Germanic languages) would find the English vowel contrasts so difficult to master. Not to mention foreign learners of German.

_ _ _

No posting tomorrow. Next: 2 December.

Irish parties

On Friday, as I watched live television coverage of the result of the by-election in the Irish Republic, I was astounded to hear the returning officer for Donegal South-West pronounce the name of the party in third position, Fine Gael, as faɪn ɡeɪl. The results in English, in which I heard this, were preceded by the results in Irish, but since the Sky News commentator saw fit to talk over the Irish-language results I could not hear what she may have said when pronouncing this name in an Irish-language context.

We English know that we are supposed to pronounce the first word of Fine Gael as ˈfɪnə. How come this Irish lady doesn’t know that? Or was it just a slip due to nervousness?

In Irish, I believe, both consonants in Fine are palatalized, making the name of the party in Irish ˈfʲɪnʲə ˈɡeːɫ̪.

Compare the name of the other major party, Fianna Fáil ˈfiənə ˈfɑːlʲ, usually anglicized as ˈfiːənə ˈfɔɪl. The pronunciation of the first word is confusingly similar to Fine. The second word, Fáil, is alternatively anglicized as fɔːl, though this is appropriate really only for speakers who make all ls clear, as most Irish people do. In Irish I think it must be the genitive of a noun with the nominative form fál: the lateral would be velarized in the nominative and palatalized in the genitive. We map this palatalization onto the second part of the English ɔɪ diphthong.

The winning party in Donegal SW was Sinn Féin. We all know how to pronounce that in English: ˈʃɪn ˈfeɪn. Looking at LPD, I find I give the Irish as ˌʃin̪ʲ ˈheːnʲ. Although this has remained unchanged in LPD since the first edition twenty years ago, I think it must be wrong. Wikipedia says it should be ʃɪnʲ fʲeːnʲ, which I am inclined to believe. Perhaps someone who knows Irish can set us straight. Where did I get the idea that the f- is lenited to h here? Anyhow, surely the lenition product of f is zero? And if there were lenition here, surely it would be signalled in spelling? How come no one has complained about the entry, or reported the mistake (if that is what it is)?

intrusive r in LPD

And so finally to LPD. Rather than omit reference in the entries to possible intrusive r like EPD, or use an ad-hoc notational convention like ODP, for LPD I applied a notational convention not restricted to the matter of intrusive r. Namely, LPD divides “optional sounds” into two categories. Those shown in italics are “sounds which the foreign learner is recommended to include (although native speakers sometimes omit them)”. Those shown as ^{raised letters} are “sounds which the foreign learner is recommended to ignore (although native speakers sometimes include them)”. Put another way, sounds shown in italics may optionally be elided; those shown raised may optionally be inserted.

Since both categories are optional sounds, what is the basis for distinguishing them? You may well ask. It is covertly orthographic, though not only orthographic. The t segment that some but not all speakers use in mints mɪnts is reflected not only in the spelling but also morphologically, since the t of the singular mint is not a candidate for elision. In mince mɪn^ts, on the other hand, there is no such support for the plosive used by some but not all speakers.
Here is what LPD says about linking and intrusive r.

Since LPD, unlike the other dictionaries, transcribes in full the inflected forms of monosyllabic verbs, its entries for saw, -ing and soar, -ing are longer, more explicit, and therefore more complicated.
Here is the entry for withdrawal.A word on my own personal usage. I use intrusive r freely after ə, ɑː and the centring diphthongs, even word-internally, but never after ɔː. So I would readily put an r in china and glass, Grandma again and even in semi-nonce forms such as concertinaing, magentaish; but not in thaw out, sawing, withdrawal. Historically, I think this constitutes an intermediate step between an earlier generation (DJ) that still distinguished ɔə (soar) from ɔː (saw), with consistent r links after the first but not after the second, and later generations for whom the two categories are entirely merged as ɔː. You could say that those like me have the phonetic merger of the two categories but not yet the phonological (lexical) merger.

You may think that the theoretical underpinning of the distinction between optional elision and optional insertion is weak. In its defence I would say that it does nicely cater for (i) usage such as my own, consistently distinguishing mints – mince and sawing – soaring, and (ii) statistics such as those furnished by Hannisdal, who showed that in newsreaders’ usage linking r was found significantly more often than intrusive r, even though both categories are frequent.

intrusive r in ODP

The Oxford Dictionary of Pronunciation for Current English, to give it its full title, was published by OUP in 2001, edited by Clive Upton, William A. Kretschmar Jr. and Rafal Konopka. It has a longer discussion of intrusive r than either of the other two pronunciation dictionaries.No two ways about that: intrusive r is a “very significant feature judged worthy of inclusion in the model”, and “a firmly established feature of today’s mainstream RP”.

Accordingly we find the possibility of intrusive r indicated through the entries by a parenthesized italic (r). Historically justified linking r, on the other hand, is shown by a parenthesized plain roman (r). As far as I can tell, this is the only use of italicization in the ODP transcriptions. Our potential homophones sawing and soaring are shown like this.The word withdrawal is transcribed in four different ways per variety, with the possibility of intrusive r shown throughout for BrE.This places ODP firmly in the camp of objective description (yesterday’s blog).

Tomorrow: LPD.

intrusive r in EPD

Some of Monday’s commentators will be pleased, some annoyed, if I now take up the question of whether the use of intrusive r is to be regarded as a characteristic of RP.

As I see it, the answer depends on how you choose to define RP. Do you consider it

• the implicitly agreed model of good BrE speech?
• a codification intended mainly for EFL pedagogical purposes?
• or an objective description of how people at the top of BrE social stratification actually speak?

Let us examine the three specialist pronunciation dictionaries from this point of view.

Here is what the Cambridge English Pronunciation Dictionary, says (17th edition, 2006, Daniel Jones, edited by Peter Roach, James Hartman and Jane Setter) on the subject of r links.
So as far as intrusive r is concerned you could say that EPD prioritizes “safe” advice to learners over the documentation of reality.

The raised ^r symbol replaces the asterisk used by Daniel Jones in earlier editions of EPD. While recommending foreign learners to use linking r only in such cases, Jones also (12th edition, 1963) carefully and in great detail explains how intrusive r in words ending in ə, ɑː, ɔː but not transcribed with the asterisk “is a very noteworthy feature of south-eastern English”. In the current EPD, Jones’s page-long discussion has been reduced to the two sentences you see here.

The environment in which intrusive r is perhaps most disfavoured is in word-internal position, between a stem and a suffix beginning with a vowel sound. Thus the pair sawing and soaring are distinguished by some, but certainly not by all, real-life RP speakers. EPD, in line with its policy stated above, entirely ignores the possibility of an r link in the former.

Tomorrow we’ll look at ODP.

contrastive customers

Transport for London is the body responsible for running London’s public transport services. Like the railway companies, to general derision, it now calls those who travel on its trains and buses not “passengers” but “customers”. (You can even argue that a child or a pensioner, who in London can travel free of charge, is not actually a customer at all. A customer is “someone who buys goods or services from a shop or company”. But that is by the way.)

Here is a recorded announcement I heard at a tube station recently.

\/Customers are re°quested | to ˈtake their ˈlitter \home with them.

As you can tell if you say this aloud, the lady making the recording spoke the right words but used the wrong intonation for them. The intonation pattern that she used bears an implication that the request is aimed at customers, but not at others. It implies a contrast between customers and some other possible subject. We might gloss this implied meaning as ‘although those who are not customers can leave their litter behind’.

What she ought to have said was

ˈCustomers are re\/quested | to ˈtake their ˈlitter \home with them.

or just

ˈCustomers are reˈquested to ˈtake their ˈlitter \home with them.

If she had placed the intonation break differently and said

\/Customers | are reˈquested to ˈtake their ˈlitter \home with them.

then the fall-rise could have been interpreted as signalling merely non-finality, rather than the inappropriate contrastivity.

Dwight Bolinger has a number of further examples of people using the wrong accentuation (wrong tonicity) when reading scripted material aloud. Taken from broadcasts over San Francisco radio stations, they are to be found in Chapter 16 of his Intonation and its Uses (Edward Arnold, 1989). Here is one.

Some restrictions apply.

Bolinger comments “There is nothing that a restriction can do but apply.” It ought to be

Some restrictions apply.

Bolinger concludes

Maybe all this should be excused simply as the ingredients of a professional style. But for the constant listener it would be restful if occasionally the newscasters and their associates would just COOL IT. A daily exercise in taming the wild accents and toning down the ends of sentences might help.

Nannying station announcements are different from the compelling urgency of newscasters. But their use of inappropriate intonation can be just as annoying.

Ma and Marr

The reason that I was unable to post on Thursday and Friday of last week was I was attending funerals, one on each day.

West Indian funerals are different from English ones. In particular, they are attended by several hundred people, whereas for English funerals there are often no more than a handful of mourners, or at most a few dozen. (Funerals of public figures or particularly well known people are an exception.) For a West Indian funeral, everyone who knew the deceased is expected to attend; and those who did not know him or her personally, but know one or more members of the family, are under an obligation to attend in order to show respect. For funerals in London, special coaches are often laid on to bring mourners from Birmingham or Leicester. Other mourners fly in specially from the United States or the West Indies.

For Friday’s funeral I did not know the deceased, a man in early middle age, but I do know his mother and so was happy to be there in her support.

So now, at last, to phonetics — or rather to spelling and what it implies about pronunciation. Here is an excerpt from a tribute printed on the back cover of the service booklet.
I will never hear you call me Marr again. In non-rhotic English English, Marr is a homophone not only of mar but also of Ma. All three are mɑː. In the context of the following ə in again an r-link is to be expected: if the preceding word were indeed Marr we would call it linking r, but since it is properly Ma we would call it intrusive r. Misspelling Ma as Marr is entirely understandable.

But that is not the end of the story. I do not know who typed up the service sheet, but I surmise that it was someone who grew up in England rather than in Montserrat. Since the deceased’s mother did indeed grow up on the island, I surmise that she herself did not pen the words in question.

The reason is that in Montserratian local pronunciation Ma ‘mother’ is not actually a homophone of Marr and mar (insofar as this name and this word are known on the island). Montserrat English has a rule of Final Shortening. Although long vowels and short vowels are well distinguished, in final position historically long vowels become short. (Details here.) Crucially, however, this rule evidently operated historically before the deletion of final r.

Thus jaw dʒa has a short vowel, but jar dʒaː a long one. Similarly we have snow sno (short) but snore snuo ~ snoː (long), and bay be (short) but beer-bare-bear bia ~ beː (long).

In Jamaican, on the other hand, (i) there is no across-the-board final vowel shortening and (ii) final r is retained.

For someone who grew up on Montserrat it would not make sense to spell Ma as Marr. For a Londoner, it would.

depressors

Yesterday I mentioned the ‘depressor’ consonants of Zulu and other Nguni languages. I have not read any recent scholarly literature on this subject, but the account in Wikipedia at least is very weak. So I thought it might be worth summarizing what I learnt of this topic in the “Introduction to Zulu” course I followed at SOAS some thirty-five years ago. It was taught by the late David Rycroft and A.B. Ngcobo.

Zulu has two (or three) phonological tones: High tone (H), shown in the examples by an acute accent (´), Falling (F) (ˆ), and unmarked or Low. However the actual pitch contour of a syllable is affected by several additional factors, notably tone assimilation, depression, and of course intonation, downdrift etc.

1. Assimilation: unmarked syllables standing between two marked ones adopt the same pitch as the previous H (subject to certain exceptions).
   
2. Depression: a vowel following a depressor consonant, if H or F, begins with a rising-pitch onset and reaches a lower high point than would otherwise apply. If it is unmarked (low), it receives low pitch, overriding the Assimilation that would otherwise apply.

Among the depressor consonants are the voiced obstruents, including the voiced clicks, but excluding the consonants written b and k, which are implosives ɓ, ɠ. There are paired depressor and non-depressor semivowels, glottal fricatives, nasals, and nasalized clicks. There is also a free-floating depressor effect characterizing certain vowel-only syllables. Rycroft claims the depressors all have breathy voice, and writes them d̤, z̤ etc. Auditorily, as already discussed, the greatest difference between the implosives ɓ, ɠ and the plain bh, g b̤, g̈ is the depressor nature of the latter but not the former.

Here is a nice example of the depressor effect. Zulu has borrowed the English word spoon, but has modified it so as to conform to the usual Bantu noun pattern of “classifier” prefix plus stem: isi-punu. (The s is taken to be the classifier isi-. The English p, unaspirated in this position, is duly mapped onto Zulu p̕ rather than onto aspirated pʰ.) Nouns that have isi- in the singular regularly form their plural by changing the prefix to izi-. Thus the plural prefix contains a depressor consonant, although the singular prefix does not. The English stress on a monosyllable is mapped onto Zulu H tone, with the result you see in the graphics, which are scanned from Rycroft’s duplicated teaching materials. Notice how the -si- syllable of the singular is assimilated to high pitch, while the depressor consonant in -zi- overrides that effect and causes low pitch.

Here’s another of his examples. The words abantwana and amadada have the same tone pattern (H on the second syllable, otherwise unmarked). The pitch patterns differ in the third syllable: partial assimilation in abantwana aɓántwaːna but depression in amadada amád̤aːd̤a.

In the word-by-word sound file of Thula Sizwe (blog, 15 Nov.), listen again to the first two words, thúla sízwe. Notice the assimilation of the pitch of unmarked -la to highish, but the abrupt drop to low pitch on -zwe caused by the depressor z̤. In úngabókhâla the g of the second syllable is a depressor, but not the implosive ɓ of the third. In úJehóva you hear the depressor effect of the dʒ and the v. And so on.

I was impressed by Rycroft’s analysis of Zulu tone: a simple system of lexical tones, but complicated realization rules. His description, and the output of his rules, agreed exactly with what our native-speaker language consultant pronounced, although Mr Ngcobo, as seems often to happen with speakers of tone languages, had great difficulty in analysing the tones he deployed so effortlessly. (The first consonant in his name is a depressor nasalized click, the second a non-depressor implosive.)

For Rycroft’s analysis see his 1980 monograph ‘The Depression Feature in Nguni languages and its interaction with tone’, Communication No. 8. Department of African Languages, Rhodes University, Grahamstown.

To bid farewell in Zulu you say salani kahle saláːni ɠaːɬé. Literally, this means ‘stay (pl.) well’, and is used when saying goodbye to people who are staying where they are as the speaker leaves. Cf hambani kahle hamb̤áːni ɠaːɬé ‘go well’, used if the addressees are leaving and the speaker is staying. See here. You will now be able to work out how the pitch patterns of the two expressions differ.

* * *
Over the coming days I shall be busy elsewhere. Next blog posting: 22 Nov.

Salani kahle!

switching airflow

The graphic you see alongside is taken from Ladefoged and Maddieson's The Sounds of the World's Languages and shows aerodynamic data for a Sindhi word with ɓ.

The comments on yesterday’s blog raise several interesting questions (ignoring for the moment those relating to Arabic emphatics).

VP asked

How does one pronounce a vowel that immediately follows an implosive? Is the airstream for the vowel also ingressive, or does it suddenly change direction to be egressive?

The answer is that in every case — as far as I know — the vowel following an implosive consonant has an unremarkable pulmonic egressive airstream. Note though, that this means that not only does the direction of airflow change, but so in principle does the identity of the initiating cavity. Since voiced implosives (the usual kind) actually involve a combination of airstreams, we can say that in the sequence ɓa there is a constant pulmonic egressive airstream, but that in the first segment it is accompanied by a glottalic ingressive component.

In a vowel-ejective-vowel sequence such as ap̕a the air flow is egressive throughout, but the initiating cavity for the consonant is glottalic — and takes place during a glottal closure, ʔ, that interrupts the egressive pulmonic airstream used for the surrounding vowels.

Clicks have a velaric ingressive airstream, but this is always combined with a velar articulation that interrupts what is normally an egressive pulmonic airstream. So in aǀa (= old aʇa) we have an ordinary pulmonic-air sequence aka with the velaric ingressive operation taking place entirely during the hold phase of the velar plosive.

Ejectives never cluster with implosives. The ingressive pulmonic airstream mechanism appears always to characterize entire utterances rather than individual segments. Reverse clicks are only paralinguistic: they certainly never cluster with ordinary clicks. So we can formulate the universal that we never get an abrupt reversal of the direction of airflow within the same initiating cavity as we pass from one segment to the next in speech.

Glen Gordon asked rhetorically

Ejectives can evolve into implosives or pharyngeals and back again, can't they?

The most striking related instance of which I am aware is in the Nguni languages of southern Africa. There we find a regular correspondence between the z of (isi)Zulu and the ejective t̕ of the closely related (si)Swati. We see this in the name of the latter language, which is also known by its Zulu name of Swazi.

The Wikipedia page on Nguni languages gives these example sentences meaning “I love your new sticks”:

Zulu	Ngi-ya-zi-thanda izi-ntonga z-akho ezin-sha
Swati	Ngi-ya-ti-tsandza ti-ntfonga t-akho letin-sha

in which all four instances of Zulu z correspond to Swati t (= t’).

So how did this correspondence come about? What kind of consonant could historically have given rise on the one hand to a pulmonic-air voiced fricative and on the other hand to a glottalic-air voiceless plosive? (I can confirm from my own observation that this Swati consonant is indeed ejective.)

It may be relevant that Zulu z is a ‘depressor’ consonant, one of the set that cause the pitch of the following vowel to start lower than it would otherwise do. That perhaps provides a link to the change at the glottis, but the possible pathways of change still seem pretty obscure. Perhaps someone knowledgeable about proto-Nguni can comment.

Or, mutandis mutatis, of some other relevant language family.
(Map taken from here)

implosives

Implosives are sounds made with a glottalic ingressive airstream mechanism. Along with ejectives (glottalic egressive) and clicks (velaric ingressive), they are a sound-type taught to all serious students of phonetics. Every year at UCL I trained budding phoneticians not only to recognize them but also to produce them to order.

Most of the implosives found in the world’s languages are voiced. That means that the rarefaction in the pharynx and mouth (the glottalic ingressive mechanism) is combined with a simultaneous vibration of the vocal folds dependent on a pulmonic egressive airstream. Interestingly, in my experience learners generally find this combined airstream easier to produce than the purely glottalic one that is needed for voiceless implosives.

But does what we teach in the classroom agree with what we find in real languages?
Just over a year ago I wrote about the Zulu song Thula sizwe (blog, 8 Sep 2009).

Now Catherine Paver, in a very late comment on that posting, has drawn our attention to a website called “Sing Africa!”. Thank you! This site contains the words and music to a number of South African songs — Zulu, Xhosa, and Sotho — along with sound clips comprising not only the complete songs but also slow demonstrations of the authentic pronunciation of each word in turn.

Here you can confirm what I wrote last year about the Zulu implosives:

The implosives are only very mildly implosive and the main auditory difference between them and the plain voiced plosives is that plain [b] and [ɡ] are depressors, while weak-implosive [ɓ] and [ɠ] are not. In singing, though, such tonal subtleties are naturally lost.

In the word-by-word clip of Thula sizwe, listen to uŋgabokhala uŋɡaɓɔˈkʰala, the third word in the song. That’s what a bilabial implosive ɓ in this real language sounds like. The “implosiveness” is much less than in the exaggerated versions we tend to get in the classroom. Mea culpa, perhaps. We see the same thing in the last word, uzokuŋqobela uzɔɠuˈŋǃɔɓɛla, with both velar ɠ and bilabial ɓ.

The implosives of Sindhi seem to be similarly weak.

I wonder if there are any languages with really strong, noisy implosives.

facilis descensus Averno: agreed?

Yes, there are plenty of words such as cigar
sɪ ˈɡɑː(r), guitar ɡɪˈtɑː(r), and divide dɪˈvaɪd that have a syllable-final weak vowel ɪ with no tenser variant.

A particularly interesting candidate minimal pair is dissent vs. descent. The first is morphologically dis + sent (compare as+sent, con+sent), with regular lexical simplification of underlying double s. The prefix dis- is pronounced with ɪ (except in Australian-type accents where the only weak vowel allowed in this environment is schwa).
dissent dɪ ˈsent də-
The second is morphologically de + scent (compare as+cent). It has the regular weak prefix de-, which for the reasons discussed this week I write in the third edition with i.
descent di ˈsent də-
These entries are based on my supposition that there are people for whom the two words are not homophones.

It is possible, though, that I am wrong in this supposition. Perhaps I ought to test it out in the next preference survey I conduct.

If I am indeed wrong, and they are homophonous for everyone, then clearly I ought to revert to writing bɪˈliːv etc rather than the 3rd-edition biˈliːv. (Perhaps I ought to do that in any case if the proportion of people claiming non-homophony is very small.)

If on the other hand I retain bi-, di-, pri-, ri-, there are certain cases which still need discussion. Those that worry me sometimes are se- and e-. In secure, seduce etc. I give only sɪ- (or sə-). Likewise in event. On the other hand Michael Ashby persuaded me that I ought to transcribe eleven with i-, though there e- is not a prefix.

The pair diffuse (v.) — defuse is interesting. Although I use strong ˌdiː- in the second myself, so that they are not homophones, many other people weaken the de- and not only pronounce the two words identically, dɪˈfjuːz, but get the meanings confused. While traditionally to diffuse is to spread and to defuse is to render (a bomb) safe, you do sometimes come across references to “diffusing” a tense situation. I would write “defusing”.

further study

I have always tried to make sure that the entries in LPD are based on objective observation of usage as well as on my native-speaker intuitions (and, in some cases and to a minor extent, on what other reference books assert). But I too am human, and it is possible that I have sometimes got things wrong.

I am aware, for example, that many people use a laxer, ɪ-like quality when a stem ending in “i” is combined with a consonantal inflectional ending, as in studied, studies. (Personally, I have no such alternation, since like DJ I have phonetic ɪ throughout.) Yet (in my experience) such speakers often maintain, nevertheless, that studied does not thereby become a homophone of studded, even if the latter does not rhyme with (nonrhotic) juddered. So that’s why I allow for a three-way contrast: studied ˈstʌdid — studded ˈstʌdɪd — juddered ˈdʒʌdəd. (There are also people who do rhyme the latter two, and have ˈstʌdid — ˈstʌdəd — ˈdʒʌdəd, while rhotic speakers obviously distinguish juddered from the others. For me personally, on the other hand, studied and studded are homophones.)

If I had introduced an extra symbol ᵻ, giving study ˈstʌdi, studied ˈstʌdɪd, studded ˈstʌdᵻd, juddered ˈdʒʌdəd, I am sure people would have regarded that as too complicated. There are never four such weak vowels in contrast in the same environment. (Notice, too, how ODP, which does use the symbol ᵻ, sometimes gets it wrong, as when armistice is transcribed ˈɑːmᵻstɪs instead of the correct ˈɑːmᵻstᵻs, as demonstrated by Australian ˈaːməstəs.)

The same arguments apply, mutatis mutandis, to the -s ending: taxi — taxis, compare taxes, taxers.

I also leave the i transcription unchanged when suffixes such as -ness (shabbiness, holiness) are added. The suffixes which for many speakers — it seems to me — clearly do affect final i by changing it either to ɪ or all the way to ə are those involving the lateral: -ful, -less, -ly. So beautiful has -əf- or more conservatively -ɪf-; merciless likewise; and angrily ends in -əli or -ɪli. This is seen most dramatically when steadily and readily are pronounced with lateral release plus syllabic l̩, for which the rule of syllabic consonant formation requires a schwa input. As I see it, the synchronic phonological derivation of this variant has to be ˈstedili → ˈstedəli → ˈstedl̩i. Likewise penniless → ˈpenl̩əs.

Why do I write -ɪ-, -ə- in polythene but -i- in polytheism? Not because of some general principle, but because it is my impression that people treat them differently. Am I wrong? Similarly with various other supposed “inconsistencies”. Ultimately, every word has its own pronunciation.

Still more on this tomorrow.

believing descriptions

I promised I would write about the phonetic representation of the prefixes be-, de- (etc.). Let’s start with some background.

1. English, as we know, has contrastive phonemes iː and ɪ, as in the minimal pair green ɡriːn vs. grin ɡrɪn. 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 iː. We might transcribe happy in this newly respectable pronunciation as ˈhæpiː.

In 1978, for the first edition of LDOCE (JWL will correct me if I am wrong), its then pronunciation editor Gordon Walsh introduced the additional symbol i, to cover such cases where the contrast between iː and ɪ was irrelevant. He transcribed happy as ˈhæpi.

This notational innovation was widely welcomed and has since been adopted by most phoneticians dealing with English. EFL students were advised that in these cases they could use whichever they preferred, iː or ɪ. For the very many EFL learners in whose L1 there was no such contrast, this meant they could just use their undifferentiated native-language i. (In stressed syllables, of course, the contrast remains important and should be learnt.)

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 etc.

2. In the first two editions of LPD, most words with the one of prefixes be-, de-, pre-, re-, in cases where the prefix is not stressed and not used productively, were shown with a main pron involving ɪ and variant prons with ə and iː, the last of these being marked with a symbol to show that it was non-RP. The entry for believe looked like this:
No one has ever raised any queries about the inclusion of the iː variants.

As adumbrated in my blog of 29 Jan 2007, I decided for the third edition to save space by abbreviating the entries for these prefixes in the same way as I already had for the -y ending. So the entry for believe became
The logic is the same as with happy. It is an abbreviatory notational convention. I am surprised, therefore, when people say they find it ‘counterintuitive’. (Is it relevant to mention the common txtng spelling b4 for ‘before’?)

You could say that for my own speech I find it counterintuitive to write happy with anything other than ɪ. But I am happy (!) to go along with the i notation, given that so many people have a tenser vowel. By adapting the notation we make it more inclusive.

There is a lot more that could be said about the happY vowel. Whole articles (JWL, Susan Ramsaran) have been written about it. Likewise, there is plenty to say about its extension to the be-, de- etc prefixes. Here are some quick points.

• Quite often, particularly in rapid speech, the quality of the happY vowel is indeterminate as between iː and ɪ. Or the quality used may be intermediate between the two canonical qualities. Using a special symbol arguably helps to draw attention to this good news for the EFL learner.
  
• Some speakers have a lax quality finally but a tense quality prevocalically, thus ˈhæpɪ but ˈhæpiːə. We certainly don’t want to burden the EFL learner with this sort of irrelevant detail. We avoid it by agreeing to write ˈhæpi, ˈhæpiə.
  
• I think there is a good case for saying that English has not just one vowel system, but two: a ‘strong’ system and a ‘weak’ system. Like ə, the happY vowel i is part of the weak system. (Notice how the strong aɪ of variety alternates lexically with weak i in vary.) Trubetzkoy would have spoken of the iː ~ ɪ archiphoneme, reflecting a vowel neutralization in weak positions. Unfortunately, polysystemic phonology is something non-specialists find it difficult to get their heads round.

Lastly, you might like to know that the conversion for the third edition was done automatically from the existing files of the second edition using an algorithm devised for me by a programmer. I checked the results, of course, but this may have led to certain inconsistencies. But it is not an inconsistency that various words that lacked an iː variant in the second edition have ɪ rather than i in the third.

More tomorrow.

what do raised letters mean?

What does it mean if certain phonetic symbols are printed small and raised above the line? How does the IPA define their meaning? Here’s what it said in the 1949 booklet The Principles of the International Phonetic Association. I’m not sure how far back this convention went, but I think it was valid for most of the twentieth century.At the Kiel Convention of the IPA in 1989, however, this was revised. You see the outcome of the Kiel deliberations in the current IPA Chart and the 1999 Handbook of the International Phonetic Association. Now there is no general rule such as previously existed, but just a listing of the following diacritics (and only these):

ʰ	Aspirated	tʰ dʰ
ʷ	Labialized	tʷ dʷ
ʲ	Palatalized	tʲ dʲ
ˠ	Velarized	tˠ dˠ
ˤ	Pharyngealized	tˤ dˤ
n	Nasal release	dn
l	Lateral release	dl

Thus in the current official IPA raised symbols may denote

• phonation features, more specifically VOT (ʰ),
• secondary articulations (ʷ, ʲ, ˠ,ˤ), or
• types of plosive release (ⁿ, ^l).

The old convention, by which ʃ^s denoted an s-like variety of ʃ, is no longer recognized. Nor is the use of ʳ (or ʴ or ʵ or ʶ) for an r-coloured vowel.

It is worth mentioning that at no time has the IPA used raised symbols to denote quick transitional sounds, although that is what people often wrongly infer as the meaning of this convention. Nor does ^ə denote a “weak” schwa, as one of my recent correspondents assumed. (In English every ə is weak, anyway, in my terminology.)

All uses of raised alphabetic symbols other than those listed above are ad hoc conventions that must be defined by the author who uses them. (That’s why I defined what I meant by ˀ the other day when I used it to show the Danish stød.)

In LPD I use raised symbols to denote optional sounds that the EFL learner is advised to ignore (“although native speakers sometimes include them” — see the panel on Optional Sounds, p. 567 in the third edition).
I use them in particular for

• the əC alternative to a syllabic consonant, thus hidden ˈhɪd ^ən
• possible epenthetic plosives, thus emphasis ˈem^pf əs ɪs;
• possible epenthetic schwa before a liquid, thus fail feɪ^əl.

David Attenborough

The only well-known public figures with whom I have a personal family connection are the Attenborough brothers, the actor Richard (Lord Attenborough) and the naturalist and television presenter Sir David. Their late aunt, née Clegg, was married to my late uncle Gilbert. So both they and I called her Aunt Margaret.

I have just been watching the first part of David Attenborough’s new television series, First Life. Brilliant, like all his work. (And what an example to us all, still to be so productive at over 80 years of age!)

As you might expect, Sir David speaks RP. (If you didn’t already know, you surely wouldn’t be able to place him as coming from the east Midlands.) He also has an excellent clarity of diction. I find him a real pleasure to listen to. Given this, here are some notes on his pronunciation that I jotted down as I viewed the programme.

fungi	ˈfʌŋɡaɪ
glacier	ˈɡlæsiə
paleontologist	ˌpæliənˈtɒlədʒɪst (not -ɒn-)
stromatolite	strəˈmætəlaɪt
kilometre	ˈkɪləˌmiːtə
believes	bəˈliːvz (not bi-)
virtually	ˈvɜːtʃəli (not -tʃuə-)
sexually	ˈseksjuəli
manufacture	ˌmænəˈfæktʃə (not -nju-)
circulatory	ˌsɜːkjəˈleɪtəri
possible	ˈpɒsɪbl̩ (not -səb-)
Newfoundland	ˈnjuːfəndlənd, ˌnjuːfəndˈlænd (inconsistently)
Ediacara	ˌiːdiˈækrə (not -kərə)

Plus one very unusual pronunciation: unless my ears deceived me, he pronounced for the first time as fɒ ðə ˈfɜːst ˈtaɪm. I know that in for it, for us we sometimes get fɒr (see LPD), but I hadn’t previously registered the possibility of fɒ rather than fɔː, fə in for the.

Danish

Occasional comments on this blog made by a Danish reader remind me of my impression that of all the languages of Europe the one with the most difficult phonetics (for the outsider) is Danish.

It is the only (standard literary) European language whose vowels I feel despairingly uncertain about identifying, and even the consonants are not without problems.

Why should this be?

• No two published sources seem to agree on the makeup of the vowel system or how to write the vowels in IPA. I try to follow the system used by Peter Molbæk Hansen in his Udtaleordbog (Gyldendal, 1990) — but he recognizes thirty distinct vowels (16 short, 14 long) as against the 20 [sic] or 17 [count them] listed in the Wikipedia article on Danish language or the 27 listed in its article on Danish phonology. (At least the latter article acknowledges that the vowel system is “unstable”, with various potential ongoing mergers.)
  
• Vowels undergo striking lowering when adjacent to /r/, which is uvular. After a vowel, (historical) /r/ gets vocalized and sometimes absorbed by the vowel, leading to further difficulties in identification.
  
• Several sets of adjacent vowels are very similar to one another.
  

  ˈmiːlə	mile
  ˈmeːlə	mele
  ˈmɛːlə	mæle
  ˈmaːlə	male

  That looks fine, but then you find that the eː is considerably closer than cardinal 2, the “ɛː” is about cardinal 2, and the “aː” is about cardinal 3. Everything is pushed forward and up. (I also get the impression that the “iː” is slightly centralized — a push-chain?)
  
• All the plosives are voiceless, but with a contrast of aspiration, often verging on affrication. Postvocalically they are strikingly lenited, sometimes disappearing completely. There’s a suburb of Copenhagen called Amager which is ˈɑmɑːˀ. It must once have had a velar consonant corresponding to the written g. Someone who comes from Amager is an amager (uncapitalized), which is ˈɑmɑːˀɑ. Clear? Furthermore, the word mad ‘meal’ is supposed to be /mað/ (i.e. [mɛð̞]. But that final approximant sounds to me awfully like a lateral.
  
• There is a phonation phenomenon known as stød, represented in the examples just given by the ad hoc symbol ˀ. This is the phonologically contrastive use of creaky voice on a vowel or vowel plus sonorant. There are plenty of minimal pairs, such as mor moːɐ̯ ‘mother’ vs mord moːˀɐ̯ ‘murder’.
  
• As with English, the spelling does not enable you to predict the pronunciation with any certainty.
  

Any comments comparing Danish to a throat disease will be deleted.

gnot again!

Intentionally wayward spelling pronunciation (blog, yesterday) can also be exploited for comic effect. Nothing illustrates this better than Flanders & Swann’s famous song about the gnu. (I apologize, I’ve discussed this before.)

ˈaɪm ə ɡəˈnuː …

ju ˈrɪəli ˈɔːt tə kəˈnəʊ wəˈhuːz wəˈhuː

əʊ ɡəˈnəʊ ɡəˈnəʊ ɡəˈnəʊ, ˈaɪm ə ɡəˈnuː

We know that really in words spelt gn- the g is silent (= there is no associated sound g): gnaw, gnarled, gnat, gnash, gnome. So I suppose we really should be saying nuː, or possibly njuː. But in serious discourse this animal is usually called a wildebeest (which we can debate whether to pronounce with w or with v). In even more serious discourse we can use its scientific name, Connochaetes (though the OED for some reason thinks it is Catoblepas).

The etymology of gnu is given in the OED as “Hottentot word”. But in the Concise Oxford this is replaced by the slightly more precise “Bushman nqu, probably via Dutch gnoe”. Neither “Hottentot” nor “Bushman” is the name of a language in polite circles these days: the first we call Khoe, Khoekhoe or Nama, while the second term refers to any of a number of languages, including !Kung (!Xũũ) and !Xóõ. I wonder from which of the Khoisan languages, then, the word derives.
_ _ _

The sports reporter on the Today programme (BBC R4) this morning was having trouble with the name of a racehorse, Abriachan. This came up at 06:25 and was followed by a brief impromptu studio discussion of what might be the correct pronunciation. An hour later, it was time to repeat the report, and I’m glad to say that by then the producer had tracked down the trainer of the racehorse, who was able to tell us all over the phone that it was named after a village in the Scottish Highlands, pronounced əˈbriːəxən.

They could just have consulted Wikipedia.

non-spelling pronunciation

The inconsistencies of English spelling and the indeterminacies of its relationship to pronunciation mean that in English we frequently get the phenomenon of spelling pronunciation.

Spelling is, with rather few exceptions, fixed, while pronunciation varies. (That is why a high proportion of words in a pronunciation dictionary have more than one pronunciation shown.)

A speaker who is familiar with the written form of a word but not with its spoken form may, on the basis of the spelling, infer a pronunciation different from the traditional or generally used one. This is spelling pronunciation. Well-known examples include often with a t-sound and clothes with -ðz. In the case of backwards the spelling pronunciation with w has entirely displaced the earlier ˈbækədz, ˈbækɚdz. In the case of falcon, formerly ˈfɔːkən, my own pronunciation ˈfɔːlkən and the newer ˈfælkən represent successive stages of spelling pronunciation, as first the letter l and then the letter a receive their usual ‘value’.

There is an extensive discussion, with many other examples, here.

I particularly relish the Italian spelling pronunciation of Colgate toothpaste as kolˈɡaːte.

There are two related phenomena. One is pronunciation spelling, in which a new spelling is applied, reflecting the pronunciation better than the traditional spelling does. Popularly this is sometimes called phonetic spelling. An example would be the proper name Leicester ˈlestə, ˈlestɚ respelt as Lester. Another is though respelt as tho. Jack Windsor Lewis uses nonce respellings such as he’rd for heard and dou’t for doubt. Spelling reform projects typically involve systematic application of the principle of writing as we speak.

The other related phenomenon has no generally agreed name, but we could perhaps call it ’non-spelling pronunciation’. This is the adoption of a new pronunciation that does not match the traditional spelling. An example is the mɪsˈtʃiːviəs variant of mischievous that we discussed last week and that so upset the Daily Telegraph’s feature writer. Another is the widespread pronunciation of Westminster with -ˈmɪnɪstə instead of -ˈmɪnstə. Non-spelling pronunciation does particularly upset purists, and even from me tends to receive a warning triangle in LPD.

I suppose that if such pronunciations become established they are likely to lead to correspondingly changed spellings, thus mischievious and Westminister.

Non-spelling pronunciation followed by respelling is readily seen in the case of the word pronunciation, traditionally prəˌnʌnsiˈeɪʃn̩. Morphological regularization, ɡiven the base form pronounce prəˈnaʊns, produces prəˌnaʊnsiˈeɪʃn̩, which in turn gives rise to the unorthodox spelling pronounciation.

I have frequently had on-air conversations with radio presenters asking me about “pronounciation”, and I must confess that in reply I tend to pronounce pronunciation with extra clarity and care.

trahison des clercs?

An article in Friday’s Telegraph mentioned that

some listeners to Today yesterday morning detected a certain trahison des clercs in the moderate opinions of Professor John Wells, the successor of Daniel Jones (alias Henry Higgins) at University College, London. He wouldn't say kil'ometre himself, he admitted, but that was because he was getting on a bit. (He is 71.) He knew better than to say mischievious, but he breathed no word of criticism of those who did.

What had I said to give rise to this comment? I had done no more than make an objective report on some findings from the pronunciation preference polls I had conducted as part of the research behind LPD. As I truthfully stated, of the (British) people I asked about kilometre, only 37% had expressed a preference for ˈkɪləmiːtə, while 63% had preferred kɪˈlɒmɪtə. (By the way, the corresponding figures for Americans were 16% and 84%.) For mischievous, I reported that 65% of British respondents to my questionnaire had said that they preferred the pronunciation ˈmɪstʃɪvəs, 20% had voted for mɪsˈtʃiːvəs, and 15% had expressed a preference for mɪsˈtʃiːviəs.

By reporting objectively on the views of a representative sample of native speakers of English, how could I incur the charge of trahison des clercs, of betraying the educated? True betrayal would be to suppress the honest and accurate reporting of people’s opinions.

Unfortunately there are those who expect expert commentators to do no more than reinforce their own prejudices. But I report what I find.

I’m glad to say that the article did finish on a more realistic, if patronizing, note.

Professor Wells also knows enough to realise that if all the world says … kil'ometre, there is nothing that can be done about it.

speech and writing

Speech is sounds, vibrations in the air arising from movements of the human organs of speech under the control of the brain/mind (or of an electrical device simulating this). Sounds can be heard, but not seen directly.

Writing is marks on paper or some other surface, or patterns of light and dark on a screen. Letters can be seen, but not heard directly.

It is incredible how difficult people find it to grasp the difference.

Several times last week, in our discussions of pronunciation, interviewers or commentators raised the matter of text messaging and the innovative spellings associated with it. But that is not speech! Other than occasional initialisms such as lol spoken aloud as lɒl, txtng has nothing at all to do with pronunciation. It is a matter of writing, not speech.

Other commentators, under the heading of pronunciation, complained not only about “text talk” but also about misplaced apostrophes.

Apostrophes, whether misplaced or not, are not part of speech. They are part of our writing system. Why don’t people get it? Even highly educated journalists?

Perhaps one reason for the confusion is the common sense of ‘pronounce’ in the sense of ‘say letters aloud’ (er... you can’t actually do that), i.e. ‘say the sounds corresponding to written letters’.

The letters ng are pronounced ŋ, or sometimes ŋɡ or ndʒ.

Spanish has a letter ñ, which is pronounced as a palatal nasal.

But this rests on the fallacy that writing is primary, speech secondary. It implies that when we speak we are merely supplying sounds appropriate to the written form of the words we use.

If that were so, how could the illiterate ever speak at all? How is it that children learn to talk before they learn to read and write?