Bob Port’s comment of 3/28/2006 asked, “What terms would phonological descriptions employ? To be very concrete, what will replace +/- Voice if this is not an apriori universal?”

I presented a talk at MCWOP 10 (Mid-Continental Workshop on Phonology), held at Northwestern University in October 2004, where I proposed the beginnings of a model that unites the two phonologies of signed and spoken languages into one system. Since then, I have made adjustments and refinements to it. A more recent version of the model will be published in the proceedings of the Semiotic Society of America conference of September 2006.

In order for anything phonological to be universal, it cannot relate specifically to phonetics because that excludes sign languages. Sign languages do not have +/-Voice features. This eliminates universal phonetic features from phonological systems.

If one were to claim that spoken and signed languages consist of separate sets of features in their phonologies, then we end up with two separate systems, yet there is no genetic basis for this, as the languages we end up using are not determined genetically.

It is the capacity for language that appears to have a genetic basis. That capacity must be able to handle both signed and spoken languages identically because the genetic capacities are identical.

Differences among languages must relate, at some core level, to the same genetically-endowed capacity. This indicates that there are language-specific mappings from the core phonological component (or however it is conceptualized) to the actual articulations.

The answer that I propose for the question “what will replace +/- Voice if this is not an apriori universal?” is something like “F5″ for voiced and “F6″ for voiceless. But another language might use “F24″ for voiced and “F57″ for voiceless, depending on where they fit into the respective linguistic systems.

Some particular sign language might use “F5″ for closed fingers, “F6″ for extended fingers, “F24″ for a place of articulation on the hand, and “F57″ for a lateral circle movement.

This is only a small part of the model, but it answers the question, conceptually anyway, on how phonology can be universal without referring to phonetics.

Bob Port asks what terms phonological descriptions would employ if there were no universal terms. I’ll try to clarify, but to summarize what I’m going to say: formal =/= universal.

What I have in mind is that phonological features are constructed by the language learner based on what s/he hears. There is something universal in the way humans process sounds, but I doubt whether there’s a one-to-one correspondence between what goes on in the early stages of human audio processing, and the learned (under my hypothesis) phonological features.

To take the specific example of Voice, while it’s likely that English speakers have a phonological feature we might label Voice, and Korean speakers also have a feature that we might label Voice, and the two are somewhat similar, I doubt that they’re the same in terms of some objective acoustic space.

But the fact–if I am correct–that VoiceEnglish and VoiceKorean are not the same, does not prevent their being used in phonological rules (or constraints, or what have you) in the two languages. (Of course, the fact that these features are similar but not identical has repercussions for second language learners.)

As for where VoiceEnglish and VoiceKorean might come from, I suspect the learner constructs them by something like the clustering learning methods in modern AI. (There is the possibility that what is constructed is not phonological features, but phonemes. This would mean a return to at least part of what the American Structuralists believed. But I won’t argue for that here, and I only believe it on odd-numbered days.)

To return to the algebra analogy, we need formal tokens, but they don’t have to be a priori or universal. Only the rule formalism for manipulating the tokens need be universal, and phonology can (should, IMHO) be about the universals of how those rules work (whether they’re ordered, how exceptions are treated, whether there are strata of rules, etc.).

I will add that there’s a lot of work being done in Optimality Theory that presupposes that the constraints are universal. I don’t believe that for a moment, so I won’t argue for that.

The problem we pose is that if there is no formal phonetics, there can be no formal phonology. What terms would phonological descriptions employ? To be very concrete, what will replace +/- Voice if this is not an apriori universal? How can, say, +/- Voice in English be compared with the rather similar (but not identical) +/- Voice in Korean if there is no universal unit of +/- Voice?

My view is that the only place formal systems actually exist is in artifacts of human intelligence (although many aspects of our world can be approximated using formal models). Arithmetic and algebra have been formalized using paper and pencil. We can use computers (human artifacts) to implement formal systems too. But all these systems begin with a set of formal tokens (either graphic tokens like letters or numbers, or computer bits). Formal phonology assumes (apparently) that human linguistic skill IS inherently formal. That assumption is apparently congenial to many linguists but there is little actual evidence for it and noone even bothers to make a systematic defense of it — certainly Chomsky and Halle did not. They simply asserted it.

Throughout the original article, there seems to be a fundamental misunderstanding (there, how’s that for being obnoxious, to use Bob Port’s term!) between substantive universals and formal universals. In particular, the argument seems to be that if the proposed substantive universals (distinctive = phonological features) are wrong, non-existent, or non-innate, then there can’t be any formal universals (formal system of rules or constraints, formal system for constraining rule interactions, or etc.).

This is just plain wrong. Arithmetic works regardless of whether integers are innate, or provided in advance. And closer to language, spelling changes (like i -> y / __ e) can be formalized regardless of whether alphabets are innate (and obviously they aren’t).4a7d3d609129a9296bf7ac0608c2097

Perhaps the best analogy is to algebra. The techniques of algebra work fine regardless of the numbering system you use–base 2, base 10, or even some irrational base. Phonology at least could be (and I suspect it is) like algebra, rather than like arithmetic: it provides the formal universals for how you manipulate symbols. What the symbol inventory is, and where it comes from, is a different question.

In sum, a good deal of the work in theoretical phonology would stand without change if phonological features (or some other phonological atoms) turn out not to be universal/ innate.

Now if you are doing arithmetic or writing a computer program, you can assume that formal operations (like addition of very large numbers or the repetition of a program loop) can take place flawlessly as many times as you please. But such formal operations demand flawless reading' andwriting’ of the appropriate symbol tokens. In Chomsky-Halle phonology, those symbol tokens happen to be phonetic objects, the segmental features of their universal phonetic alphabet. So these features play a role that is formally identical to the role of bits in a computer: they can be assumed to be read and written without error. The purpose of our paper is to point out that the there is absolutely NO EVIDENCE WHATEVER (I’m following Halle’s advice to always make the strongest claim possible so that you expose yourself to the maximum range of potential criticism) that such a discrete (that is, discriminable virtually without error) and universal inventory exists. In fact there is an enormous range of data that disprove such a claim.

Can it be that noone made this observation before? Well, it has been made by others at various points (see refs in our paper). Certainly Ladefoged (RIP) has pretty much made this observation at various points, but, probably out of politeness, avoided being as forceful (that is, as obnoxious) as we are being about it.

So are we saying there is no discreteness in language? Certainly not. The phonological systems of language are full of instances of discreteness. These cases demand a theoretical account. How do language learners come to interpret the continuosly variable sounds in their language as discrete and contrastive?

The issue we challenge is whether the discreteness is provided IN ADVANCE. The traditional theory simply asserts the discreteness of phonetics which then IMPLIES the discreteness of phonology. The discipline of phonology does not have to ACCOUNT for discreteness, because formal discreteness is provided by phonetics (they hope). Our claim is that where discreteness is found, we need now to find an explanation. And the traditional methods of formal phonology, presuming formality, cannot help us with this problem at all. Only empirical data and a new set of theoretical approaches (such as dynamical systems theory, for example) will be able to make sense of these phenomena.

Adam reads my earlier comment to mean “that phonology must deal with discrete phonetic units.” I do see phonology as dealing with discrete units, I don’t think those units need to be phonetic parameters themselves. My assumption about phonological units instead is that they are abstract and discrete categories which map to and from regions of continuous phonetic dimensions. (Meanwhile, I can’t fairly say whether this is what all or most or any other phonologists think).

This mapping is probably equivalent to what Adam Leary, in his above comment, calls the “operations on these symbols to produce an output that matches what you observe in the real world”. Whether the need for these operations is a good or a bad thing, I believe, is a separate question from where those symbols come from.

I also think it’s also possible to have a model of formal phonology in which the discrete units/symbols/categories aren’t universal - instead, they’re symbols that learners assign to clusters of observations in continuous phonetic space.

In fact, Port & Leary make a similar statement (p. 956): “What is universal about phonology is not any fixed list of sound types, but rather the strong tendency of human language learners to discover or create sound categories out of what they hear.” How these categories emerge is a fascinating question, and it’s one that lots of phonologists do work on. I guess some of us then make the extra step of conflating those emergent categories with symbols to be manipulated in phonological operations.

It is even more interesting to see how our thinking about phonology is shaped by the need for ‘discrete units’. Bob’s point is crucial-most phonologies assume a discrete universal inventory; even OT does.

I would like to address a few points that Bob Kennedy raises in his comments. You write, “Let’s define phonology instead as the study of inventories and distributions of discrete phonemic symbols within and across languages. (aside, I have not seen any argument against the concept of the phoneme.)”.

My reading of your comment is that phonology must deal with discrete phonetic units. Our argument points out that phonology as a separate, abstract formal system requires discrete units to work! It would also require operations on these symbols to produce an output that matches what you observe in the real world.

There are two important problems with this type of phonology. First, unfortunately, many linguists have the tendency to view such systems as ‘what is actually happening’ rather than a model. Second, starting with a discrete symbolic model of phonology requires rules to produce the phonetic output as well as some sort of clock to place these abstractions back into real time.

We are saying turn this model on its head. You don’t need discrete phonetic units to produce symbolic like behavior. It can and is done with a dynamic system that can handle timing issues and operations are a state of the attractor space. Such a system means that phonetics and phonology as separate levels of analysis is unnecessary as the separation is a matter of time scale within the same state space.

I think most importantly what we discuss is not a straw man at all. Rather, the fundamental assumptions of phonology, in whatever permutation that you may use, rest on the notion of a discrete universal set of phonetic parameters. Further, this assumption leads us to a restricted range of phonological theories. We also do make suggestions as to how we view a theory of phonology in our article as well as in other works of ours individually.

Unfortunately the two critical remarks thus far seem to miss our point. Marc quotes the last sentence of our paper but then seems to misunderstand it. And Bob Kennedy says that we (Port and Leary) characterize modern phonology as “the PURSUIT of a discrete universal inventory”. Actually, we claim that formal phonology must ASSUME the apriori existence of a discrete universal inventory before it can even get started. Noone seems to be pursuing a discrete universal inventory except a few phoneticians hoping (in vain, we would say) to make good on Chomsky and Halle’s assumption of the formality of phonology. Why is the assumption of an apriori list unavoidable? Because formal systems cannot, in principle, account for their own formality. They ASSUME formality - and that means assuming a set of apriori symbol tokens - which in this case is cashed out as a set of apriori phonetic symbols. And the utter lack of any evidence supporting such an apriori token list poses an unbearable empirical burden on formal phonology, in our view.

Adam and I are happy to respond to any questions.