Posted by & filed under Podcast.

George gives a short talk about how phonology affects phonetic transcriptions and why the narrow “phonetic” transcription of your language does not have to be overly specific (especially with vowels!). We should have a regular episode again next month.


There is a tendency in the conlanging community to hew toward more narrow, standardized definitions of terms than among linguists. There have been a few times when someone has criticized me on a use of a term that I defined in the text where they felt another term fit. This is most obvious in morphology, which is pretty notorious for having different traditions among different languages for naming different categories or morphs, but it extends to other places as well. I understand the impulse of conlangers to hew to standardized systems.

Conlangers, by our nature, are describing our languages to an audience that is unfamiliar with it and wants to understand quickly. We want to standardize our description so that other people can quickly comprehend it and give feedback. That is somewhat the case in linguistics, but as a linguist you are not only talking to the broader linguistics community, but also to an audience of linguists who are particularly interested in a specific language, family, or areal group, and will be familiar with the peculiar traditions in describing that language.

While I understand the interest in hewing to the standard, it’s important to know that there is some flexibility, and all sorts of descriptive terms are used differently among different languages. That’s important for reading linguistic research, but it’s also important for writing descriptions, because the labels you give things will be influenced by an analysis of the language at hand. I want to give an example through the lens of a part of linguistics I know fairly well: phonetic transcription.

A note: yes, I do, in fact, mean phonetic transcription. It is obvious that the labels you use for phonemes in a phonemic transcription will be influenced by the structure of the language. It’s also pretty clear that those labels have a degree of arbitrariness, we choose which symbol is the phoneme based on an analysis of how it behaves and on standard conventions of the field. What is less obvious is that the symbols we use in phonetic transcriptions are informed this way as well. If we were to fully specify a phonetic transcription at the absolute, unaltered surface, we’d just print a wave form and spectrogram. But in linguistics we understand that as the pure sounds are translated into phonetic categories, there has already been a perceptual filter applied, so we never make it perfectly detailed.

The most obvious example of this is in vowels. As I have said on the podcast before, vowels in real languages are never defined as points in the vowel space as the IPA chart would have you think. A vowel, perceptually, will fall into a range of values in a region of that space, determined relative to other vowels. Languages with fewer vowels will have a wider margin of error — As an Arabic speaker’s /i/-vowel may fall down to the region of English /ɪ/ or even /e/. In some cases, the distinctions may be important, but in many cases, we really don’t bother.

Consider the English /u/-vowel. In many dialects of English. This vowel is fronted. For some speakers it is so far front that it is in the same region as the /i/ vowel. You would think that we would sometimes transcribe it as central [ʉ] or even as front [y] sometimes, but the fact is that even in relatively narrow transcriptions, you almost never see that. The reason for that is that the fronting of /u/ isn’t really phonological. There aren’t really phonological rules governing it, and it doesn’t have an affect on other sounds. It seems to just happen on the phonetic level without the higher level processing of phonology. So, unless we are specifically talking about this phenomenon, we don’t worry about it so much. We just transcribe it as [u].

Linguists can also disagree on a transcription, or transcribe differently depending on their perspective. For example, if I’m going by ear, I would phonetically transcribe the vowel in Mandarin Chinese zhong as a mid back vowel, [o] or maybe [ɔ]. San Duanmu instead transcribes it as [u], and I have also seen [ʊ]. Duanmu’s justification is that it is underlyingly /u/, and I won’t get into the reasoning there, I do agree with it. I don’t know if my lack of native intuition gets in the way there, but I would probably differ as I think that the lowering is a phonological rule we’d have to take into account.

Now, vowels are notoriously fuzzy. Tonal systems are even moreso — I would not really want to give a confident transcription of a tone on any level without figuring out the tonal system first. But even consonants can have these. For instance, in English, [ʃ] is almost always rounded. If you are a native English speaker, just make a shushing sound and you’ll see it. But linguists never transcribe it with a rounding diacritic, because it is not affected by any rule and it doesn’t condition any rule. The theory is that the lip rounding enhances the distinction between [s] and [ʃ].

Finally, we can even disagree about a sequence of sounds. What is the rhyme in her. Is it [ɹ͈]? Or rhotacized schwa [ə˞ ]? Or is it a sequence of [əɹ]? It depends in part on how you think English works. I have seen many times a question something to the effect of what the difference is between a palatalized consonant and a sequence of a consonant plus [j]. And the fact is, the difference is mainly phonology — is that glide a property of the consonant or its own vowel. Now, the “glide” will often be phonetically shorter if we’re talking about palatalization, but that’s a side effect. How you transcribe it really depends a lot on the phonology.

There are a couple of takeaways I want conlangers to come away with here. First, is that, when reading linguistic descriptions of languages, look for descriptions of the sounds if you want an accurate view of the phonology. The transcriptions will only go so far. Even if the linguists are using IPA (and not all do), their use of it will be informed by their theories and by traditions in transcribing that language. Also understand that just because something is in square brackets, doesn’t mean that represents precisely what comes out of a speaker’s mouth. It’s still filtered.

Second, yes, use standard IPA values, but understand that IPA is just a tool, and an imperfect one. Don’t stress about finding exactly the right symbol with diacritics and whatnot. Transcribe what you need to transcribe, and write up a description of how things are pronounced. This basic principle follows for other parts of your grammar. Need to name your cases? Find labels that fit well enough, then give detailed descriptions of their usage. Syntax? You can try some high-level typological labels, but you need to also give example sentences and explain how they are working. In all these cases, the labels are just to give you a shorthand — the real work is in the description.

One Response to “Conlangery SHORTS #28: Fuzzy Phonetics”

  1. Okuno Zankoku

    Overall: yes, absolutely! I do want to push back in just one place, though.

    Something that conlangs can suffer from is not having a clearly defined accent (just take a look at the various approaches to Na’vi within a single tribe, and ever a single family). I therefore spend a fair bit of time trying to understand how I want a conlang to feel in my mouth, and then I need to remember and communicate that information.

    At that point, I start using Canepari’s transcription system; it’s all about precision, to a frankly ludicrous degree. If I’m making a recording, having a very narrow transcription is really the only way I can break out of my normal accent and give the illusion that the fictional speakers of a conlang have their own set of emotional reactions to different voices.

    Of course, I still use regular IPA in a very normal way when all that detail is less relevant. There are plenty of reasons for that: it’s more approachable for other conlangers, it’s more suitable for accent-neutral description, and perhaps most importantly/frustratingly it’s far easer to typeset. I would say that recordings are the best means of transmitting this information, but actually… there are times when I can’t perform an accent I’m imagining, and then I can use the transcription to describe it anyway.


Leave a Reply

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

This site uses Akismet to reduce spam. Learn how your comment data is processed.