Why I Don't Love the International Phonetic Alphabet

[ Guest post by Sally Thomason, from Language Log; some links added by EB. ]

Don’t get me wrong: I love the idea of an international phonetic alphabet, and most of the IPA symbols are the same as the phonetic symbols used by linguists all over the world, including me. But some of them are different, and some of those differences make the IPA non-ideal for me, and I suspect for a lot of other fieldworkers out there too. Some of my reasons are fairly trivial, but one of them is a serious problem. I’ll save that one for last.

The IPA doesn’t go in for diacritics much, notably hacheks. So, for instance, the “sh” sound is an elongated letter [ʃ], as opposed to an ordinary [s]. For linguists who got A’s in penmanship in grade school (if there’s anyone still alive who ever got grades in penmanship), this might work just fine when they’re transcribing data from speakers or from tape recordings. But I’m not one of those people, and there’d be a real risk that my [ʃ]’s would turn out looking like [s]’s and vice versa, and that’s a bad thing when you’re trying to figure out a language’s phonological system. If you use a hachek for “sh”, it’s [š], much harder to confuse with [s]. So I use hacheks, and so do most other fieldworkers I know.

For a few sounds, different transcription practices have developed for the use of ordinary roman letters that aren’t needed for the sounds they spell in Western alphabets. The most common of these letters is c: we have [s] for transcribing the c in cell and [k] for the c as in cat, so c itself is available for use as a phonetic symbol for another sound. In the IPA, [c] is a voiceless palatal stop. This is actually a pretty common sound in the world’s languages, but it’s rare as a separate phoneme — often, where it occurs, it’s an allophone of /k/, and it’s commonly transcribed with k plus some diacritic indicating palatalization. When I studied Sanskrit in graduate school, the letter c was used to transliterate a “ch” sound, and as a phonemic symbol for the “ch” — an affricate that many other linguists (including me) transcribe as [č], using the same hachek diacritic as for [š]. No need for a palatal stop in transliterating (or phonemicizing) Sanskrit; lots of need for a [č], and why bother with the diacritic when everyone understands the c as [č]? Years later, when I morphed into a specialist in American Indian languages, I adopted the typical Americanist practice of using c to transcribe a “ts” affricate: lots of need for that in New World languages, little or no need for a palatal stop. (Actually, in the Pacific Northwest, /k/ often represents a prevelar or even a palatal stop, contrasting with /q/, which is a “back k” or uvular stop. But there too c is reserved for the affricate.) In transliterating or transcribing Sanskrit, it’s very convenient to have an easy way to represent the common phoneme “ch”, and ditto for the common affricate “ts” in American Indian languages. Using the letter c is a handy solution in each case. But strict adherence to the IPA closes off this convenient use of c, which seems a shame, because as long as linguists are always careful to say what symbols stand for what, no confusion can result.

But if the IPA uses c for a palatal stop, how do you transcribe affricates in IPA symbols? Answer: you use digraphs. So [tʃ] represents “ch” and [ts] represents “ts”. Because the IPA isn’t enthusiastic about diacritics, you don’t get any ligature in standard IPA (though the TexTipa phonetic fonts that I use in LaTeX documents do have the t and the fricative symbol sort of smooshed together so that you can tell it’s an affricate and not a consonant cluster). When I wrote a paper on Montana Salish phonetics with the late great Peter Ladefoged and Edward Flemming — a somewhat misleading author listing; Flemming did almost all the analysis and writing, Ladefoged collected the data when I took him to the Flathead Reservation in the mid-1990s, and I was third author — we had a problem with the IPA system: Montana Salish has a contrast between syllable-initial affricates and syllable-initial stop + fricative clusters. The word for ‘be soft’, for instance, is /čep/ (in Americanist transcription for “ch”), and the word for ‘bull elk’ is /tšec’/ (also using the Americanist transcription for [ts’]). There’s no doubt about the contrast, because stops in stop + fricative clusters are released before the fricative and are thus clearly differentiated phonetically from a corresponding affricate. So, as we worked on the paper, I kept insisting that using IPA [tʃ] for the affricate was unacceptable, because it concealed the phonetic and phonemic distinction between affricate and cluster. Peter was a dedicated proponent of strict IPA transcription, though, and I won the battle (the distinction had to be made) but lost the war: the paper — to appear, possibly even fairly soon, in the Journal of Phonetics — ended up transcribing ‘be soft’ as [tʃep] and ‘bull elk’ as [t.ʃets’].

So the IPA is problematic for people transcribing linguistic data in the field — where you need to write fast to avoid wasting your consultant’s time, so that super-careful writing isn’t an option — and it doesn’t allow for variation in symbol choices to facilitate the transcription of sounds that are especially common in different parts of the world. And it also doesn’t make it easy to distinguish affricates in the admittedly rare languages where they contrast with stop + fricative clusters. But the worst problem I’ve come across is in the IPA vowel chart. I only realized this last month, when a student in my historical linguistics class objected to my complaint, in a problem set, that he hadn’t generalized enough in stating a certain sound change. I wanted the students to say that the change happened “before all front vowels”, but he protested that his “before non-low front vowels” had a necessary constraint because the change did not happen before /a/. I tend to deviate from IPA norms more in vowels than in consonants — some of the IPA vowel symbols are hard to draw, and life is short, so I use diacritics quite a lot in transcribing vowel distinctions. But I keep forgetting that the IPA insists that the letter “a” represents a low front vowel. Like many other linguists, I use the symbol known as ash, [æ], for the low front unrounded vowel. (In the IPA this symbol represents a not-quite-so-low front unrounded vowel; I won’t bore you with the details of how I make that distinction in transcribing.) Also like many other linguists, I use the letter “a” for a low central vowel.

The IPA does of course have a symbol for a low(ish) central vowel: it is an upside-down “a”, namely, [ɐ]. And for a low back unrounded vowel, the IPA uses a symbol that looks just like the letter a that I, and a whole lot of other linguists and non-linguists, use in printing by hand. There are two problems here with the IPA system. First, the IPA symbol for the low central vowel is hard to draw, and it’s all too easy to draw in a way that makes it very hard to interpret afterward. This vowel sound is so common as to be near-universal in the world’s languages, so it’s particularly unfortunate that the IPA transcribes it with such a user-unfriendly symbol. The lower-than-æ front vowel, by contrast, is absent from many, many languages. Second, and worse, everyone who hand-prints an a with something that looks like the IPA low back unrounded vowel symbol [a] has a life-long habit of equating that hand-printed vowel with the printed letter “a”. The habit is hard to break, which makes confusion all too predictable when (for instance) one is typing up fieldnotes.

Geoff Pullum and Bill Ladusaw, in their wonderful little fat book Phonetic Symbol Guide, observe that ‘The IPA’s effort to establish [a] and [a] as separate symbols “has not met with the success originally hoped for” (Principles, p. 19).’ Surprise, surprise. But it’s worse than that plaintive comment by the International Phonetic Association suggests. After my student pointed out that my use of the symbol “a” for the low central unrounded vowel is not sanctioned by the IPA, I did a quick check of the language grammars closest to my desk and found an overwhelming majority of authors using the symbol “a” for a low central vowel, not for a low front vowel. This was a rough count of a non-random set of languages, and I omitted grammars whose authors did not describe the language’s sounds or show the symbols in a chart or otherwise indicate what sounds the symbols represented (a sign of extreme wickedness in any linguist-author of a grammar of any language). Here are the results. All the authors who clearly indicated the pronunciation of the vowels treated [a] as a low vowel. In the Americas, 28 grammars had [a] as a central (or at least non-front) vowel and not one grammar had it as a front vowel. In Australia, 35 grammars had [a] as non-front and 4 had [a] as front, and in New Guinea, 18 grammars had [a] as non-front and 2 had it as front. A spot-check of a few languages elsewhere in the world turned up seven languages with [a] as non-front (Finnish, Hungarian, Hausa, Gimira, Yemsa, Turkish, and Uighur) and one (Evenki) with [a] as a front vowel. All of these grammars were written by modern linguists, so I think it’s reasonable to conclude that on this point it’s the IPA, not me, that’s out of step with the world.

In any case, being a stubborn type, I’ll stick to my Americanist practice of using hacheks and a user-friendly low central vowel symbol, and even the less obviously defensible non-IPA transcriptions that I’m used to and happy with. I do routinely warn my students that there is no sacred phonetic alphabet and that there is a lot of variation among linguists who do phonetic transcription, and I tell them they can use any system they want as long as it’s reasonably standard, and as long as they stick to a single symbol for each sound they’re transcribing. I freely acknowledge that the IPA may be absolutely ideal for people who do their phonetics in laboratories rather than in the field, and even that there are probably quite a few fieldworkers who don’t share my reservations about it. I also realize that no single transcription system is likely to be ideal for all linguists and/or all purposes. So I’m not dogmatic (at least not about this). I hope, though, that the IPA enthusiasts out there won’t try too hard to convert us IPA dissidents.

16 thoughts on “Why I Don't Love the International Phonetic Alphabet

  1. Pingback: Pages tagged "Strict"

  2. Michael Covarrubias

    And of course there is the problem not only with handwriting but with the rendering by computer fonts. On my computer the following sentence shows identical symbols in this posting.

    ‘The IPA’s effort to establish [a] and [a] as separate symbols “has not met with the success originally hoped for” (Principles, p. 19).’

    (Tho they are differentiated by italics. And in the original LL post they reflect the two IPA forms).

  3. Darin Flynn

    Arial aside, most fonts render [a] as (low back) [ɑ] in italics, so (1) the graphemic neutralization that Sally describes w.r.t. hand printing is also pervasive in books and journals (e.g., IJAL) that italicize data, and (2) bloggers should avoid relying on italics to distinguish the low back vowel, and perhaps avoid italicizing data in general, at least when discussing vowels.

  4. Philip Spaelti

    Michael Covarrubias wrote:
    On my computer the following sentence shows identical symbols in this posting.
    ‘The IPA’s effort to establish [a] and [a] as separate symbols “

    Well that is going to appear that way for everybody. But that has nothing to do with the IPA or the font. That has to do with the fact that Sally didn’t enter the character correctly as Unicode.

  5. Eric Armstrong

    It seems to me that many many people have their own private IPA that they use for speed and ease of use. They need to know the “official” version so that they can read documents that use it, and publish documents that others will “get”. But for your own private use, who cares what version you use? Hachek it up, I say! Different people have different needs. In comparing the sound associated with /a/ in the Handbook of the IPA, (listening to the samples provided on the web), I was shocked to see what variety there was in the usage of that symbol. But the reality is that, when you have a language that has only got 5 vowels in it, the sound that is CLOSEST to /a/ is going to get the symbol. It’s just easier that way. Is it accurate? As long as you describe what you mean by that sound, sure it is.

    As someone who teaches actors accents for theatre and film, my students get frustrated that the IPA isn’t all they might imagine it to be. They hope that one set of symbols will be a universal code that will allow them to describe and learn the wealth of sounds of the worlds languages and, in particular, the great variety of English accents. It just isn’t possible. Even WITH diacritics, one just can’t be accurate enough with symbols. So we use sound samples to give students a sense of “When I say /a/, I mean ….”

    I do feel though that poor handwriting seems an odd reason why one has it in for the IPA. But I bet that many of us have frustrations with symbols that we find look quite similar in our own handwriting style. My students often struggle to differentiate between /t/ and /ɫ/ because their /t/ symbols lack the hook-tail that the typographic version has. As I teach them phonetics, I actually take the time to teach them the letter shapes, so that they’re more easily distinguished. The barred i /ɨ/ is also hard to distinguish for some. I suggest that students add serifs to their handwritten /ɪ/ symbols, so there is less chance that they will be confused with /i/. But making the effort to learn how to lengthen your /ʃ/ so it won’t be confused with /s/ seems like very little work to me. Again I say though, in your private papers, who cares?

    In the example that ends ” ‘be soft’ as [tʃep] and ‘bull elk’ as [t.ʃets’] ” I must say that what you finally agreed upon is much clearer to me, an outsider, because of that period (syllable break) in the second word. THAT reveals the difference between affricate and cluster very clearly. Why reinvent the wheel when the symbols are already there?

    “First, the IPA symbol for the low central vowel is hard to draw, and it’s all too easy to draw in a way that makes it very hard to interpret afterward.” This is the “upside-down “a”, namely, [ɐ].” Yes, it is unfamiliar to draw, but it needn’t be DIFFICULT to draw in a distinct way. It’s essentially a handwritten /e/ with a little hook on the top left. This ISN’T hard to draw, it’s just not something you’ve taken the time to PRACTICE. I find, with my students (and in myself when I first learned IPA, to be honest), that they spend quite a lot of time to learn the sounds, but aren’t willing to spend a few minutes drilling a the shape of a new symbol so that their hand remembers the shape. I guess we figure WE KNOW HOW TO WRITE! Most of us don’t remember learning to write/print as something that was fun. It was a chore. Unfortunately it still is. But it is a chore that was worth doing, and is still worth some effort.

  6. Paul Carter

    I can appreciate some of the comments in this post, and the IPA alphabet is a long way from being perfect but I think some of the ideas here arise more from starting from a position of familiarity with an alternative system (ie the American tradition). The IPA only seems harder or more perverse if it’s not what you’re used to. The American system seems weird to me but that’s only because I’m a British phonetician who was taught the IPA way of doing things. Don’t get me wrong: I’m no IPA evangelist. One of my own pet hates with the IPA is the way superscript consonantal symbols sometimes imply temporal sequence (eg aspiration) and sometimes don’t (eg secondary articulation). But it’s a toolkit. I just learnt how to use it and then bored my own students with my grumblings in class. But to suggest that there’s no way in IPA to distinguish between plosive+fricative clusters and affricates is mistaken: it says quite clearly on the chart that “affricates […] can be represented by two symbols joined by a tie bar if necessary”. So we can use a tie bar for the affricate and no tie bar for the plosive+fricative. I’m reading a paper right now which uses exactly that graphic distinction for another language which has both plosive+fricative and affricate.

    My students seem to manage to distinguish esh from s in their handwriting with a bit of practice, but perhaps it’s different with the pressures of the field. Still, I can’t see anything wrong with using alternative symbols as long as their value is explicit in the conventions which accompany the text of the transcription. “Strict adherence to the IPA” doesn’t close off the use of “c” suggested because normal roman letter shapes can be used where they’re useful in particular languages, as long as their value is made clear. I suspect the issue at hand isn’t so much how appropriate the IPA alphabet is, but rather confusion between narrow, impressionistic phonetic transcription and the phonological value of particular sounds in particular languages. We find this in the common use of “r” in transcriptions of English: this thing varies quite widely in different Englishes but it’s relatively uncommon for it to be an alveolar trill, the “strict” IPA meaning of the symbol. But it’s useful to be able to use the symbol “r” as a cover-all for what might (in English) be a post-alveolar approximant or an alveolar tap or a labiodental approximant or whatever. As long as we make clear that’s what we’re doing.

    I’m surprised that the two symbols which look like typographical variants of the first letter of the alphabet are criticised on the grounds that one of them will be hard to write, seeing that I find much more difficulty in handwriting the ash symbol (a-e digraph) which is used in the American tradition for a low open vowel. That’s why I think this whole thing is a problem of habit (ie if we’re used to IPA it’s easy; if we’re not used to it it’s hard). A proper low central vowel symbol might indeed be very useful, mind (I tend to use diacritics to solve that problem). Writing on computer is indeed a tricky issue here, especially with italics (though TIPA is pretty good in this respect). Doulos seems to do it right, though Charis doesn’t (even though it’s supposed to have a proper italic character set). Nor do Gentium or Junicode, among the free fonts with decent phonetic character sets. Lucida Sans unicode copes, but don’t get me started on what the phonetic symbols look like in that font!

    Far be it from me to try and convert a non-IPA enthusiast, especially since I see the IPA alphabet as no more than a useful toolkit (useful despite its theoretical problems). But please, let’s all just agree to make clear what our transcriptions mean. Let’s spell out what our use of particular symbols represents. This sounds basic, but it’s not always followed. I have read published texts which explicitly state they’re using IPA transcription and then do things like use “j” for a palatal approximant on one page (ie as suggested by the IPA) and then use “y” for the same sound on the next page. It did take me a moment to work out what on earth that rounded front vowel was doing there. Nothing wrong with using one symbol or the other (though one tradition or the other would find it annoying to have to translate all the time), but everything wrong with causing uncertainty by not making the conventions clear.

    If we don’t make this sort of thing clear, how can one person comment on or reanalyse someone else’s data? How on earth would we know what potentially interesting bit of phonetics has been glossed over in the transcription for ease of writing? If it’s crucial that open vowels could be front or central, we’d never get that information if the author of the transcription didn’t make it clear which vowel their symbol referred to. So we’d never know whether some theoretical phonological accounts were based on phonetic fact or on some vague almost-phonetics. We’d never know whether things transcribed with the same symbol in two different languages were actually the same sound and it would be nigh-on impossible to draw robust cross-linguistic phonological conclusions.

    Just a few comments from a grumpy old phonetician, sorry. :-)

  7. Pavel Iosad

    Re italics, the IPA wasnæt meant to be italicized at all, precisely because italics is a different typeface, not a mechanical operation on letter shapes. This is why the tipa fonts that Sally mentions substitute slanted for italic, since a slanted font keeps all the contrasts intact. So this is the IJAL’s fault, rather than the IPA’s. This is also one of the reasons I don’t like using italics for this sort of data and tend to use the square brackets for transcriptions and italics for orthography.

  8. Darin Flynn

    I mentioned that the webiquitous typeface Arial is contrast preserving in that [a] ≠ [ɑ] across std, ital, bold, etc., but should add that it is graphemically neutralizing in other ways. Notably, because it is sans-serif, IPA/American [small capital i] merges with [i] under superscripts (accents, etc.).

  9. Darin Flynn

    A terminological hair-split: Pavel mentions that “italics is a different typeface”, but my (very possibly incorrect) understanding is that italics, bold, etc. are different fonts within a given typeface.
    So for instance, one might say that the Times typeface generally provides distinct serif idiographs for the IPA graphemes ɑ and a, but it provides a single allograph for both in italics; or that the Arial typeface provides the same sans-serif idiograph for both í and high-tone small capital i, such that these IPA graphemes are not distinguished in any Arial font –std, ital, or other.
    Btw, my students tend to be confused rather than helped by relating phonemes/allophones/phones to graphemes/allographs/graphs, so I think I’ll avoid that comparison in my teaching :)

  10. Pingback: From Under My Sculpt

  11. Pavel Iosad

    Darin, you are of course right re the typeface/font, sorry for the confusion.

    Good comments about the Arial too, One more reason to use TeX & Friends…

  12. Bruce Hayes

    The IPA is standardized and backed by a clearly written, widely distributed handbook. Therefore, if you use IPA symbols in your publications, you will maximize the chance that people reading your symbols will understand what they mean.

    The amount of confusion created by non-standardized transcriptions in linguistics is huge. People who aren’t experienced in the art of interpreting ad hoc transcriptions–for instance, my undergraduate students–tend to be resentful of the scholars who use them.

    It took me a long time to realize this (my own phonetic training was non-IPA). But now that I fully understand the advantages of standardization, I use strict IPA in my own publications.

  13. Michael

    The international phonetic alphabet is pretty bad but it would be useful for high school esl if there an actual guide to writing the symbols. Children aren’t good at listening and don’t have enough exposure in non-English speaking country- they need to be able to write and read the symbols rather what they end up doing is writing in local language that can’t do English sounds well and leads to katakana English and other bad English versions. IPA symbols are so poorly chosen that the IPA association admits that you can’t make it with one font and that some symbols are near impossible to tell apart while not even listing a handwriting guide.

  14. Patrick Lauser

    It is good to have a standardised system people can easily look up and reference, and likelihood that they will be familiar with it is also nice. But it seems that if there was another system that was also standardised and which people could easily look up it would be equally useful, even if fewer people were familiar with it.

    One thing that bothers me is when it seems that people are using different IPAs. I remember seeing an IPA symbol and not finding it on Wikipedia, until I found out that it was an “obsolete” symbol that had been replaced by a new one. There are symbols in this chart:
    which I also did not find on Wikipedia, and the sound he described was represented with a similar but different symbol (but perhaps it is a font difference?). I can see where updates would be nice, but when the main benefit of the system is standardisation it seems that it would be better to keep it warts and all (or save up the fixes and start a whole new system when they reach critical mass, if that ever happens). It’s one of the casualties of creating a system: before it is well-used the errors are not clear, after it is well-used the errors can’t be removed without hurting the system and what it was meant to be used for.

    I wonder about aspiration. I can see that “p” in “pin” is different in “spin”, but then how are non-aspirated plosives distinguished by voice? That is, does any language actually distinguish between “this bin” and “the spin”? Then when I actually hear languages which differentiate based on aspiration, like Hindi, it does not sound like “pin” but like an affricate or even a cluster, like “tap hard”. In the same way Tolkien describes aspirates as “followed by an h, more or less as in backhand, outhouse”, not as in “car” or “two” (or “pin”). If aspirates are thus more distinct than some make them out to be, almost being consonant clusters, that would explain to me how there is room in some languages to distinguish between voiced and unvoiced as well as distinguishing aspirates. But this case shows how there are apparently differences of opinion on even how descriptive terms like “aspiration” are used, which damages standardisation.

    One other thing about the IPA though is that it is merely international. As a conlanger creating languages experimentally and artistically, you use every sound that can be used, whereas the IPA only transcribes sounds already in use. For example, sounds made with the tongue against the upper lip (labio-glossal I suppose?) are easily pronounced and completely distinct, but they are used by no languages that I have heard of, and are not represented in the IPA. Of course conlangs can also use sounds like clapping or whistling or imaginary sounds from non-human voices, but there are many sounds (like the whole labio-glossal group) for which there is no reasonable cause that they should not be represented in a standardised system.
    Actually, I remembered hearing of extensions to the IPA, and looked it up while writing this, and it does have labio-linguals (what I called labio-glossals). Though, the main purpose they give for the extensions is for transcribing “disordered speech”, haha.

  15. Nick

    One of my pet peeves is how writers omit diacritics and lowering marks from their vowel symbols. How is this justifiable?
    Perhaps they are concerned that a diacritic might give a false impression of a phoneme’s invariability, in other words, over-accuracy. But when the actual vowel space is concerned, why should one vowel symbol be given more importance simply by virtue of *not* having a diacritic or lowering mark? Why should one vowel symbol be preferred because of orthographic simplicity?
    It is clear to me that the IPA is being misused when one symbol is substituted for another for no good reason, and when phonetic accuracy is not seen as important.
    For a long time I would be cross whenever a writer would write that the low-front vowel in ‘path’ in Northern English was [æ], because to me it was clearly [a], as per IPA. Only after reading your post here have I realised that these writers might not in fact be phonetically tone-deaf but simply unconcerned about accuracy or adherence to standardisation. But why is this an acceptable situation? It’s worse when you think about how the ‘trap’ vowel in American English is also transcribed with this same symbol æ (which, in this case, is actually phonetically accurate).
    I have some serious suggestions: first, let us always be concerned with phonetic accuracy, and when there is any chance of ambiguity, we should describe things in full, e.g. ‘low central vowel’. Secondly, let us all use the same symbol for the same sounds, across all languages. (And different symbols for different sounds, so that the German central A [ä] is not transcribed with the same A as ‘trap’ or ‘bath’ in BE or AmE.) Thirdly, could we please use new and better symbols for those ‘forsaken’ IPA vowels? (I mean those vowels whose symbols use diacritics or lowering marks, and are avoided as a result.) For instance, we could use a capital A for the central vowel, instead of ‘a’ or ‘ä’. And we could use capital E for the mid-front vowel, or the Greek letter omega for the mid-back vowel. All this would make it easier for us to compare languages and to study the phonetics of a language. The current situation is madness.

Leave a Reply

Your email address will not be published. Required fields are marked *