Monday, 10 October 2011

the full gamut

A correspondent writes that his wife
seems to think that the IPA cannot represent the full gamut of human sounds in terms of pronunciation in languages. Not wanting to argue on a point about which I cannot claim to know, I decided to email you.

If it is the case that any language can be represented by the IPA, is there a good book which one can use as a pronunciation reference when looking at IPA? And an unrelated question, can anyone actually really use IPA to speak like a native speaker if they don't have reference sounds (given they could reproduce them if they tried)? I would guess you would need some tutelage in order to properly use IPA...?

I replied along the following lines.

1. There is no universal enumerable list of discrete “sounds”, and therefore there can be no set of symbols in a one-to-one relationship with them. Rather, we are faced with a multi-dimensional continuum of possibilities. Putting it another way, there is no super multilingual phoneme system in the sky, of which the sounds of each particular language are a subset.

In principle, the IPA contains all the symbols needed to represent the pronunciation of any human language so far described — that is, it is adequate to cover the contrastive sounds (phonemes) of any language. Not all finer shades can be represented except by ad hoc symbols. For example the English ʃ sound in sheep is somewhat different from that in sharp and that in short. But it is not necessary to symbolize these finer nuances. The French ʃ (orthographic ch) is not identical with any of these, being usually “darker” than in English ʃ, nor in the other direction is the Japanese ʃ (orthographic し, romanized sh). However the same IPA symbol will serve for all. For languages that do make phonemic contrasts among sounds that we English speakers would regard as varieties of ʃ, the additional symbols ʂ (less palatal) and ɕ (more palatal) are available. As far as is known, these are sufficient to cater for the ʃ-type sounds of all languages.

2. It is of course true that learning a set of symbols does not equip you to pronounce any language perfectly. Phonetic symbols are a reminder of what you should be aiming for when pronouncing a word in a given language. They still need to be interpreted in the light of detailed phonetic information about that language (point 1 above). And even possession of this knowledge does not automatically give you the articulatory motor skills to perform the sounds accurately.

3. Yes, study phonetics! Or at least consult the IPA Handbook (CUP 1999) and a textbook or two.

More generally, it does seem to be the case that for many people practical phonetic skills are best learned through face-to-face tuition given by an expert. Partly, this is to do with the “hearing bias” imposed on us by our native language. We just don’t automatically hear all the details of speech sounds if those details are irrelevant in our L1. A tutor can draw our attention to aspects we may have overlooked, and perhaps offer ear-training to improve our perception of “foreign” sounds and sound contrasts. It ought to be possible for experts to deliver this kind of tuition through distance learning rather than face-to-face, but that’s something we’re still working on (though we made a start at UCL with PhonLine).

For original investigation into the phonetics of a language, there is no substitute for dealing with a live native-speaker “language consultant” (informant).


  1. As John suggests IPA is at best a phonemic, not a phonetic alphabet: describing finer (and in cases even less fine) articulatory details is clearly impossible with IPA (consider flap vs. tap). However, I don't agree with him that it's not necessary to have unique symbols. Phonemically one could say that using /ʃ/ is a reasonable simplification, but phonetically there must be a way to show the exact articulations.
    The number of distinct speech sounds any human being is — in theory — able to produce and recognize goes into the thousands. Any speaker uses just a small subset of these sounds that can be grouped into the phonemes of a specific language. To describe what a speaker pronounces it is necessary to have a transcription system that dedicates a symbol to each distinguished speech sound regardless of language or what phonemes they realize, or if they are phonemic in any language. A super multilingual phoneme system John mentions simply cannot exist, as that would violate the definition that phonemes exist only within the scope of a single language. A language-independent phonetic transcription system listing all the human speech sounds is, however, not impossible. I don't think the number of discrete sounds is unlimited, because our speech organs, our ears, and our brain put a limit on that, and if so, it is theoretically possible to create a system with an adequately high number (2,000+) of symbols, giving us the possibility to accurately transcribe what's actually pronounced, not just some vague approximations.
    In short, the correspondent's wife is quite right: IPA cannot represent all the human speech sounds.

  2. I suppose the diacritics would be pertinent to mention as well. I had trouble with this too when I was first studying, but no writing system could possibly convey the huge array of phonetic nuances of every language, indeed, every person's idiolect. We must make do with what we have.

  3. Maybe off-topic...but there are two symbols in your image I don't recognise: the gamma and the lambda. The standard IPA symbols that these are often confused with - the Latin gamma (voiced velar fricative) and the upside-down Y (palatal lateral) - are there too.

    Am I working from an old version of the IPA?

  4. Not according to this, Pete:

    (The heading Phonetic Symbols looks like a link, but this is a masquerade. The subheadings link to pdfs.)

    Last updated: - 22 September 2011

    This site, updated on 2007-09-27 has the following puzzling legend for λ as opposed to ʎ:
    IPA #295 Lambda: Voiceless dental or alveolar lateral fricative (Not IPA usage)

    I can't find any site explicitly saying anything like that about γ as opposed to ɣ, however

    The β here looks almost as if it may be an attempt to give the Greek β a bit of a serif, which it was supposed to have in serifed IPA fonts before Unicode failed so miserably to disunify it into Greek and IPA, as done with γ. Perhaps this β is from some specialized phonetic font, but I doubt it.

  5. Pete: The illustration is just a pretty graphic that I took from an article discussing whether IPA symbols such as β and θ are typographically identical with letters of the Greek alphabet or not, and more generally how to harmonize Latin and Greek letters within the same typeface.
    You're right that γ and λ are not IPA symbols.

  6. This is really interesting as I'm just learning the IPA in my introdutory linguistics course. My professor had given me the impression that the IPA was an actual phonetic system but your explanation gives me almost the opposite impression. I guess I'll have to resolve all this in my head now.

    On the point of learning the IPA though: is something like IPA Help ( actually useful for this task? I've learned to recognize musical intervals and chords quite well using a simple ear trainer such as this so it seems like a similar method would work for the IPA. I can't even imagine trying to learn this system without auditory guidance, though. My textbook already gives me examples for vowels that don't work for my speech (like using both pot and bar to explain the sound of a symbol when I pronounce those vowels differently).

  7. The modified IPA system made by Luciano Canepari (University of Venice) is much better than 'plain IPA'. Even Spanish or Croatian O's and E's lack the proper simple representation in IPA (you have use symbols of '''lower than close O/E'' or ''higer than open O/E's'''

    canIPA Natural Phonetics

    Plain IPA is terrible for comparing (between) different languages, at best it is useful for describing sounds of one single language (without trying to compare them with similar sounds present in other languages.).

    Portuguese bom and French bon sound exactly the same, yet, different symbols IPA used.

  8. @ Urban: That's a very interesting website. It's big so it'll take me a while to form an opinion on it. I've looked at his PDF on English pronunciation and he appears to have identified 34 different accents in England alone. Only a few are available for free online. I wonder how an Italian managed to map English in such detail.

  9. John,
    «an article discussing whether IPA symbols such as β and θ are typographically identical with letters of the Greek alphabet or not, and more generally how to harmonize Latin and Greek letters within the same typeface.
    You're right that γ and λ are not IPA symbols.»

    So I was right in my comment above about this being a disunification issue. But there is no whether about it! The IPA symbols *are* typographically identical with letters of the Greek alphabet. So what was this article and what did it conclude?

    Some time ago you discussed disunification on your blog, giving jpgs of Daniel Jones's typographic specifications for the IPA in the 1949 Principles booklet (The Principles of the International Phonetic Association, pages 1-2).

    I will give them here again in text so that I can add emphasis:

    "The non-roman letters of the International Phonetic Alphabet have been designed as far as possible to harmonise well with the roman letters. The Association does not recognise makeshift letters; It recognises only letters which have been carefully cut so as to be in harmony with the other letters. For instance, the Greek letters included in the International Alphabet are cut in roman adaptations. Thus, since the ordinary shape of the Greek letter β does not harmonise with roman type, in the International Phonetic Alphabet it is given the form… .

    The form in question, which I can't specify a font for in Comments, is an upright beta with a serifed descender as opposed to the slanted form in the jpg of the original. I use the Charis SIL font a lot because it's best for phonetics in so may respects, including this one: the beta is serifed.

    I would add that the situation is worse than that: the variant form ϐ (U+03d0, and bloody well not in Charis SIL, but I see it's in Times new Roman and Caslon, for a start), which certainly will not do for IPA, is not distinguished from the "ordinary shape", whether serifed or not, by search engines etc, and is quite likely to be thrown up by Greek fonts, especially in modern Greek.

    This also applies to the variant form ϑ (U+03d1, also bloody well not in Charis SIL, but in Times new Roman and Caslon) of θ, and here again is Jones on that:

    "And of the two form of Greek theta, θ and ϑ, it has been necessary to choose the first (in vertical form), since the second cannot be made to harmonise with roman letters."

  10. I feel like I've seen the article John is referring to and have made the same sort of comment before, but here goes...

    The Greek letters used as IPA symbols are meant to be harmonized typographically with the Roman letters. Greek letters, especially those designed with a 'traditional ductus' as in Garamond Premier Pro or SBL Greek will look out of place next to Roman letters. Also, as Mallamb mentioned, Greek fonts may well use variant forms of letters such as ϐ or ϑ in place of the 'regular' forms—some fonts use one variant for the upright and another for the italic letter. They are perfectly acceptable for writing Greek, but not for IPA. So the typographical form of the Greek letters used as IPA symbols is subject to a tighter constraint than those of the Greek letters used for writing Greek.

    It would have been better from my point of view if Unicode had separate 'Latin small letter theta', 'Latin small letter beta', etc. to be used as IPA symbols.

  11. It does seem somewhat arbitrary for which sounds IPA provides dedicated symbols. Seeing a sound written as /β/ doesn't tell me whether it is a fricative (as it is supposed to be on the IPA chart) or an approximant, because there is no dedicated symbol for a bilabial approximant. I get the impression that most uses of /β/ in transcribing languages are for the approximant.

    Meanwhile, the labiodental nasal, which appears as a phoneme in a tiny minority of languages if at all, gets its own symbol, /ɱ/.

  12. outside of the typographical discussion, I find it interesting that when I attempt to pronounce a French word such as "chat" with a lighter ʃ, I can't get the vowels right, yet when I try it with a darker ʃ , it sounds fine. I'd never noticed this distinction before. Thank-you!

  13. I've learned to recognize musical intervals and chords quite well using a simple ear trainer such as this so it seems like a similar method would work for the IPA.

    Golden Ears or...?

    The illustration is just a pretty graphic that I took from an article...

    Does it say which typeface the image shows?

  14. There have been several modifications to the IPA system, but I wonder why IPA has failed to dedicate a symbol to one of the most frequent vowels of the world: the open central unrounded sound found in e.g. Italian or Spanish. There's been a tendency to use plain [a] for this sound, despite it's supposed to represent an open front vowel (causing ambiguity). Alternatively [ä] could represent this sound, but why use diacritics for such a frequent sound?
    On the other hand, [ɘ] for example is found in just a couple of languages, yet it is not indicated by means of diacritics, but it has its own symbol.
    The symbol [ð] is commonly used to transcribe a sound of English, Spanish, and even Danish, but they're actually rather different sounds.
    What real scientific value can any description have with such inconsistencies and ambiguities in the standard transcription system? Science is to describe reality as accurately as possible, but IPA effectively prevents us from doing so.

    Canepari's alphabet seems to be a great alternative, it could provide a lot more accuracy, without ambiguity, so phoneticians that want to do real scientific work should consider using it instead of IPA.

    I think our host, being a well-known phonetician, could do a lot to help abandon the unscientific IPA, and promote Canepari's system instead to improve overall phonetic quality. Are you going to help, John?

  15. I've been a casual browser of Canepari's free online materials for some time, and I do find his system appealing, if a bit intimidating. I'll have to chime in with the others in criticizing the IPA vowel chart: I think it was a great mistake not to include simple symbols for central [a], mid [e]/[ɛ] and mid [o]/[ɔ], three of the five components of the extremely common five-vowel system.