How to organize a phonology?

Question

When I look at phonologies on Wikipedia, such as Xhosa or Chinese, it typically goes through the base I would say "features" of the sound system: what are the consonants they use, what are the vowels, if there are tones, clicks, stress, that sort of stuff. It seems at some level it is about the atoms of sound, like the consonants and vowels, each of which are for all intents and purposes are one base sound. Even with aspirated consonants, it's still overall one general sound.

But then some phonologies go beyond this atomic sort of level, and add on diphthongs, or consonant clusters like tɕ (or sometimes with the arc, t͡ɕ). These things appear as a single unit, but I would call it a second layer of composition, the second level of building blocks on top of the base sound features (consonants/vowels/stress/tone/etc.). Another second-level I would say is long vowels. In the "Xhosa vowel phonemes" chart on Wikipedia, it lists both long and short vowels. But if I were putting together a phonology, it seems it would be more "pure" to just list the base vowels which get used, at level 1. Then long vowels go into the level 2 group, along with diphthongs and consonant clusters.

Then some might go a little further depending on the language, to syllables, or things like in English, longer consonant strings that appear in words, like str or -md or things like that. I would say these are getting into level 3 of composition or something like that.

My question is, what should go into a good phonology? How should it be organized? What should it include, and what should it leave out? To narrow it down, if I just collected the base consonants (or consonant + aspiration/velarization/etc., i.e. any primitive sound whole), and left out diphthongs and consonant clusters (what I am calling level 2 stuff), would that be a useful phonology? Or must it include the diphthongs and other "higher level" things/features of the sounds produced too?

I am playing around with a conscript for writing pronunciations for a fantasy project, and have begun collecting the consonants and vowels off Wikipedia to put into "phonological charts", but I'm realizing I don't have a clear definition/sense of what that actually is in the end. Some include diphthongs, some don't. I would personally move the diphthongs and level 2 stuff into a higher level thing, maybe not called part of the phonology. But I'm not sure. What goes into a phonology in the end?

If we are going with diphthongs in the phonology, and consonant clusters, why not just go and list out every possible sound combination (of say 1 syllable) which can be used in a language? I have never seen that done, but if you take picking sound combos to the extreme, you would end up listing out every possible thing you could say in the language. Not sure why diphthongs, for example, are chosen, but not other common sound sequences. That sort of stuff.

Answer 1

This is very much an opinion question, but I have one. The first thing you need to do is decide whether you mean "Wikipedia phonology" or "professional phonology". I have nothing to say about wiki-phonologies – to the extent that they are any good, they are based on professional phonologies. In the latter realm, you have to first decide what you mean by "phonology", and second, you have to know what actually is in the phonology of the language.

In terms of the logical flow of presentation, the most important thing to do is relate the letters on the page to the sounds in the air that are the physical manifestation of a phonology. Thereafter, it really depends on what you think a "phonology" is.

Here are the two main views of what a phonology is. One is the "realization system" view – the morpho-syntax presents an abstract arrangement of roots and affixes composed of basic phonological atoms (usually feature structures, sometimes segmental letters). The phonology then performs conventional computations and yields a "phonetic output" which the phonetic component and other downstream mental systems can convert into a physical output. The second view is the "data-classification" view, which attempts to erect various data structures on top of the physical output, in aid of arriving at a structure that we could call "meaning".

One practical difference between the two views is that the realization system does not necessarily mandate encoding all observational regularities about the data into the grammar, therefore so-called "phonotactics" (statistical regularities of distribution) are not guaranteed a place in the grammar, whereas in the data-classification approach, it is a fundamental desideratum of the approach that one observes all of the statistically-significant distributional patterns (that is a goal, not a trivially-accomplishable thing that one always does).

There are very many examples of good professional phonologies out there, as well as not-so-good ones. Two weaknesses plague phonologies. One is "incompleteness", indeed illustrated in the Xhosa article (seriously? There is a huge very well documented work on Xhosa tone by Cassimjee that is not even mentioned). The wiki article suffers from the typical wiki problem of extremely superficial treatment, whereas a professional phonology will generally give you a more exhaustive treatment. Unfortunately, professional phonologies can also be incomplete. Dalgish's dissertation on Olutsootso is quite good but for some reason he just omits tone. The second plague is "over-theorizing", where so much emphasis is put on a particular piece of theory that the work is really about theory, and not the description of the language.

Classical Arabic and Vietnamese are at two ends of a certain phonology scale, related to what a phonology "does". Classical Arabic has a very rich, indeed elegant system of rules there abstract roots are assembled, and the result is changed into something pronounced by changing vowels, deleting or mutating glides, reducing vowel sequences, and it is well-described in a fairly theory-neutral albeit early 70's generative model by Mike Brame. Vietnamese doesn't change the pronunciation of roots as a function of adding this prefix vs. that suffix, you just pronounce the bits in the order that they are strung together.

Since you are doing a conlang, you have to first ask, in what sense does the language have a "phonology". Conlangs seem to be pretty much devoid of phonology in the Classical Arabic sense – there aren't intricate systems of segmental (or, horrors, prosodic) alternation, and certainly nothing requiring the learner to know an abstract underlying form. Since a conlang is deliberately constructed, you can design an engine that generates all possible "syllables", and all possible "sequences of syllables", and then make sure that the lexicon is actually populated with those outputs. Shux, I've done it as an exercise. If your system has 5 vowels and 10 consonants, and syllables can be unrestricted C(C)V(C), and words can be up to 3 syllables long, you have a rather huge set of possible words. If words have no internal morphological structure, you just need a big dictionary. If they do have structure, then you potentially face a problem, if t- is a prefix – what happens if you prefix /t-/ to the root /pra/? Delete /t/? Insert a vowel? Delete /p/, or /r/? Prohibit adding /t/ to a root that starts with a consonant cluster? So it is possible that you will have avoided the problems by making ever morpheme be /CV/. whatever the case is, what goes into the description is determined by the facts of the language.