In the recent volume on Phonological typology (Hyman & Plank (eds.) 2018), the editors complain that phonology is not given sufficient attention by morphosyntax-heavy mainstream typology, so it may perhaps be reassuring to note that “phonology is not different” in one respect: The nature of the things to be compared is often unclear.
This blogpost begins with Maddieson’s (2018) contribution, because it directly addresses a question that I have been wondering about more generally: “Is phonological typology possible without (universal) categories?” Maddieson’s answer is (basically) no, but what does he mean by “universal” and by “categories”? More needs to be said about this.
By logical necessity, comparison is not possible without (universally applicable) comparative concepts, and the main questions in linguistics are (i) whether we should use hypothesized innate categories as comparative concepts, and (ii) whether we can use the comparative concepts also for describing/analyzing languages (or, conversely, whether we can use the descriptive categories also as comparative concepts). (I think the answer to both questions is no.)
Ian Maddieson’s concern in his recent paper is primarily with the distinction between categorical and scalar phonetic/phonologocal concepts for comparing rhythmic patterns and vowel-consonant patterns in the world’s languages. He surveys a range of studies and finds that all (or almost all) studies that compare languages based on continuous variables (= scalar concepts) are ultimately based on categorical variables. He could have mentioned his own WALS chapters as well, which compare languages with respect to such quantitative (and relative continuous) variables as the size of consonant inventory.
All of this is unproblematic, I think, but Maddieson tries to relate it to the question of universality of categories, as discussed in my earlier papers:
“Much of the terminology used in the phonetic sciences and applied in phonological analysis refers to categories that are determined outside the scope of an individual language. That is, they seem to fit the bill of being “pre-determined categories” of the sort that Haspelmath (2007) declared “do not exist”. Haspelmath (2010) argues that cross-language comparison, and hence any form of linguistic typology, cannot be based on “descriptive categories” but must instead be based on “comparative concepts”. This seems like a distinction without a difference.“ (Maddieson 2018: 108)
But Maddieson is not interested in language-particular description/analysis, and what I asked in the 2007 paper was whether particular languages should be described in terms of a pre-established (innate) set of categories (my answer was: probably not – because we have no good evidence for them). Of course, comparative concepts must also be universal (because otherwise they wouldn’t be applicable to all languages), and they must be formulated in advance of the comparison. But they need not correspond to innate categories – and here it seems that Maddieson would agree (I have never heard him endorse anything similar to Chomskyan/Hallean ideas in phonology).
Thus, what Maddieson’s article is really about is the question whether comparative concepts can be fully scalar/continuous, or whether they necessarily have a categorical component. There is no disagreement with the proposal that descriptive categories and comparative concepts should be kept apart.
It would have been more interesting to hear how Maddieson would respond to a trenchant critique of his approach to segment typology (Maddieson 1984 and much subsequent work, including his 2005/2013 WALS chapters), as expressed by my Jena colleague Adrian Simpson in a widely cited (1999) paper. Simpson notes (also following Lass (1984)) that the systems compared by Maddieson are not commensurable, because they are based on contrasts and not on phonetic substance. A phoneme is typically represented by a single phonetic symbol (e.g. the English phoneme /t/), but it may correspond to rather different phonetic realities ([t], [tʰ], [ɾ], and [ʔ]), so that it is unclear why it should be represented by one rather than by all the allophones.
This issue is also highlighted by a paper by Paul Kiparsky (2018, also in the Hyman & Plank volume), who provides a beautiful (but very complex) discussion of languages with “vertical vowel systems” (such as, famously, Kabardian), among other things. From the perspective of a phonological analysis, such languages have been said to have only central vowels (/a/, /ə/), though when speakers talk, one hears front and back vowels as in other languages.
From my perspective, the conclusion is clear: One needs to compare the phonological properties of languages in the same way as their morphosyntactic and semantic properties: on the basis of special (etic) comparative concepts, rather than on the basis of language-particular (emic) analyses. General meanings (= emic meanings) of words or grammatical markers are rarely suitable for comparison, and grammatical constructions must also be reduced to fine-grained comparative variables, rather than comparing them through the same categories which are used to describe them.
Simpson (1999) and Kiparsky (2018) give somewhat different answers, but both distinguish different “levels” of phonetic/phonological analysis. Simpson (1999) distinguishes between a phonetic, a functional phonetic and a phonemic level, and he sees comparison as unproblematic only for the phonetic level. Kiparsky distinguishes an m-phonemic level (= generative underlying representation) and an l-phonemic level (= Lexical Phonology level), and he argues that universals must be stated at this latter level. While Simpson’s phonetic level could probably be straightforwardly interpreted as a “comparative level” (not suitable for description/analysis), Kiparsky distances himself explicitly from the idea that comparison and description should be separated:
“All putative phonological universals are framed in terms of theory-dependent categories, and defined on some theory-dependent level of representation, most often the phonemic level. Therefore the linguistic descriptions on which they are based cannot be theory-neutral or atheoretical. The approach of “describing each language in its own terms” is at best aspirational. With one exception, all grammars I am aware of draw heavily on existing descriptive frameworks.” (Kiparsky 2018: 97)
By “theory-dependent”, Kiparsky evidently means “dependent on a general theory of innate categories” – in other words, he defends the old Chomskyan/Hallean idea that innateness is the only basis for universality. In addition, he reasserts the standard generative view that language-particular analysis should be in terms of a general set of innate categories.
So while both Maddieson and Kiparsky seem to represent positions that are incompatible with mine, they do so in very different ways: Despite appearances, Maddieson actually has no disagreements with me (and he would probably also agree with Simpson that comparison must ultimately be phonetically based, and that this is what he approximates in his database of inventories). By contrast, Kiparsky continues to defend the idea of phonologocal categories as innate natural kinds, even though he recognizes that
“reconciling typologies with descriptive frameworks and the analyses dictated by them can involve a labyrinth of choices” (Kiparsky 2018; 54)
This echoes the recognition elsewhere in the field of Chomskyan linguistics that there are many “moving parts” of the enterprise. Whether one is encouraged to pursue this enterprise probably depends on the subjective assessment of success so far. Personally, I find explanatory successes such as vowel dispersion theory (Liljencrats & Lindblöm 1972) a good reason to pursue the functional-adaptive path, while reserving final judgement about the possibility of discovering the natural kinds of grammar.
Thankfully, Kiparsky’s formulations leave no matter for doubts on his views, and he even distances himself from Larry Hyman, one of the volume editors:
“Since there are no theory-neutral grammars, there is no theory-neutral typology. In terms of Hyman’s (2008) distinction, there are no “descriptive” universals of language. All universals are analytic, and their validity often turns on a set of critical cases where different solutions can be and have been entertained. … It follows that the search for better linguistic descriptions, more illuminating typologies, and stronger cross-linguistic generalizations and universals should go hand in hand.”
Instead of “theory-neutral universals” or “descriptive universals”, I would talk about “empirical universals based on comparative concepts” (because they are NOT based on descriptive categories, and of course there is no “theory-neutrality”), and instead of “analytic universals”, I would talk about “natural-kind universals”. Since we don’t know yet what the natural kinds of grammar are, the latter are extremely contested and progress is very slow (if there is any at all), but thanks to Kiparsky’s clear statements, there is no denying that this is a comprehensible research programme.
Thus, Hyman & Plank’s (2018) volume can be highly recommended as a source for some quite different views of how phonological typology should be practiced. Phonology does not seem to be different (cf. Bromberger & Halle 1989)…