Department of Philosophy, University of Montana
Missoula, Montana 59812
appel@selway.umt.edu

In a series of papers Björn Lindblom argues that phonemic structure is emergent, in the sense that it arises in a self-organizing manner. Central to this description is the claim that phonemic structure is not prespecified as a "mental blueprint". More generally, on this view, phonemic structure is not the direct consequence of the causal forces that produce it; rather, it emerges as a by-product of the interaction among sound-pattern primitives, performance constraints, and the demand for increased vocabulary. The sound-pattern primitives, called "gestures", are specified as sequences of articulatory closures and openings, corresponding to stop-consonant and vowel configurations, respectively. Thus, the gesture inputs are indistinguishable - in articulatory, acoustic, and perceptual terms - from CV syllables. This gives rise to the following objection: Phonemic structure arises from primitives which are indistinguishable from CV syllables. The presence of CV syllables seems to indicate the presence of phonemic structure. Thus, phonemic structure seems to be primitive and prespecified, contrary to the requirements of a self-organizing system.

Lindblom, anticipating this objection, denies that the gestural primitives of his model are in fact CV syllables. He acknowledges that they "resemble" CV syllables, but emphasizes that they are not "analyzed" as such; instead, the gesture is defined as a "holistic transition" and as a "Gestalt trajectory":

Although these gestures resembled stop-vowel syllables, their specification did not presuppose an analysis in terms of segments. Rather a possible gesture was defined as a holistic transition running between an arbitrary point in the universal phonetic space of 'possible closures' and a similarly arbitrary point in the universal space of 'possible vowels'. (Lindblom et al, p. 181)

My aim in this essay is to analyze and evaluate Lindblom's above response. I will first try to show that his reply, as it stands, is inadequate. I will then present two different ways his reply can be elaborated and strengthened, each of which leads to a different conception of what precisely the self-organization of phonemic structure consists in. I will conclude by suggesting that neither conception is wholly satisfactory.

The problem with Lindblom's reply is it suggests that what the emergence of phonemic structure consists in, is simply a change in perspective on pre-existent structure, rather than a change in the structure itself. It suggests that the process he describes as emergent is not itself structure-generating, but simply structure-recognizing. The inputs to Lindblom's model are CV syllables which are not analyzed as such and the outputs are CV syllables which are analyzed as such. Hence, it appears that the only thing that changes is the analysis, not the structure itself.

Compare the above with Lindblom's favorite non-linguistic example of a self-organizing system: termite nest building. The output of the latter process is an intricate structure of pillars and arches. But the inputs are just randomly distributed piles of (glutinous) sand. Thus the emergence of termite nests consists in a radical change in the organization and structure of the sand, not a change in the analysis of the sand.

Nevertheless, I think there are two stronger defenses implicit in Lindblom's reply, which I will try to make explicit below. I will call the first reply "Recombination" and the second "Asymmetric Dependence".

The Recombination View denies that phonemic structure is present from the outset, but not by denying that the primitives are physically indistinguishable from CV syllables. Rather, the Recombination View denies that physical structure is the relevant level of structure at all. According to this view, the essential feature of phonemic structure is not that it is composed of a series of physically independent units (closures and openings), but rather that the units are functionally independent. It is only once a single closure or opening by itself signifies a difference in meaning - that is, only once minimal pairs are present - that phonemic structure can be said to have arisen. By contrast, the initial input contains pairs of physically independent closures and openings, but they do not function independently.

Note that the Recombination View does not depend on how the original inputs are analyzed. In particular, it does not depend on refraining from interpreting the inputs as a sequence of two segments. Rather, the point is, it makes no difference whether we do or not, because however we analyze them, the initial inputs function as holistic units. On the other hand, once minimal pairs are present, the sound patterns are analyzed as two independent segments because they now function that way. Thus although a sequence corresponding to [ba] would not be analyzed as a CV syllable in the initial input and it would be analyzed as such in the output system, the change is not merely a change in the analysis. Rather, the change in the analysis itself reflects a change in the input - it now exhibits functional structure.

So it appears that phonemic structure is not prespecified as the initial objection charges. But is this account of how phonemic structure arises appropriately described as a self-organizing process? The local causal mechanism is recombining closures and openings. Whether this is a self-organizing process depends on whether it is best described as a case of indirect causation.

It seems safe to assume, as Lindblom does, that the process is not the result of an explicit and conscious attempt on the part of our ancestors to create phonemic structure. But this does not settle the question of how directly caused the process is. Lindblom is clear about the pressures that produce minimal pairs - demand for increased vocabulary and demand for sounds with low articulatory costs relative to perceptual benefit. The latter constraint limits the available pool of closures and openings and hence the number of patterns which do not share either a closure or an opening. When the former demand overloads the resources of the latter, reusing closures and openings is the only way to satisfy demand. So there need be no antecedent plan to generate phonemic structure; the external constraints are sufficient.

But there remains a question of whether phonemic structure arises as a direct or indirect consequence of these external pressures. It would seem to be as a direct consequence because these pressures lead directly (if blindly) to reusing closures and openings, and there is little, if any, explanatory distance between the process of recombining closures and openings and the process of generating phonemic structure. The former process just is the generation of phonemic structure.

By contrast, consider again, the termite nest building case. The local causal process in this case is termites depositing sand in response to the presence of pheromones. In so doing, they end up creating a structure of pillars and arches. But the process of depositing sand in response to pheromone presence and the process of generating an intricate pillar and arch structure are not ipso facto the same. One could easily imagine the sand structure being produced without a process of responding to pheromones, but it is more difficult to imagine phonemic structure being produced without a process of recombining primitives.

Thus, it seems that the process which generates phonemic structure is too direct to count as emergent. However, I am less concerned to press this conclusion, than to emphasize that in order to maintain that phonemic structure is a self-organizing process (on the Recombination View) one must deny this conclusion.

But there is a second way to interpret Lindblom's claim that the gestural primitives are not analyzed as exhibiting phonemic structure, and hence that phonemic structure is not prespecified. On this second account what's important is not in the first instance whether discrete segments form minimal pairs or not, but whether discrete segments are present at all. Consider that the CV syllable patterns are, in one sense, composed of discrete units and in another sense, not. They are defined as consisting of two distinct components - closure and opening. They can be distinguished in articulatory and acoustic terms.

But, from another point of view, they are holistic. One hears a syllable as a single sound contour; one doesn't hear it as two physically independent segments. More specifically, one cannot separate the syllable into two perceptually independent units insofar as the stop-consonant onset cannot be pronounced without some (i.e., even tiny) bit of the following vowel. By contrast, vowels are not similarly dependent on onsets. Vowels can be pronounced independently; a syllable can consist of a vowel only, but not of a consonant only. So one can say there is an asymmetric dependence between consonant and vowels (or between closures and openings) in the formation of CV syllables. Consonants are dependent on vowels in a way that vowels are not dependent on consonants..

This asymmetric dependence is present whether we are considering syllables in a system with or without minimal pairs. It's a physical, not functional, dependence. Yet in a system with minimal pairs - that is, in a system with phonemic structure, the consonants take on an independent functional existence as well: [pa] and [ba] is every bit as much a minimal pair as [pa] and [pi].

Once phonemic structure is present there is a mismatch between the autonomy of the segments at the physical and functional levels (at least for consonants). Prior to the emergence of phonemic structure, one has sound patterns that are holistic both physically and functionally; subsequently one has segments, but they are functional segments. According to the Asymmetric Dependence View, then, what the emergence of phonemic structure consists in is the creation of discrete structure from physically and perceptually holistic patterns.

The difference between this view and the Recombination View is that on the latter view, having holistic syllables which differ in only either a closing or an opening, but not both, just is phonemic structure, whereas on the present view, it is not phonemic structure until it is analyzed as such. On the Asymmetric Dependence view, it is only when holistic syllables are compared or juxtaposed, that segments emerge for the first time. What makes them emergent on this view is that they arise by abstraction from contrasting syllables; they are implied or inferred from the physical patterns.

According to the present view, Lindblom's claim that the initial inputs are not sequences of segments because they are not analyzed as such, means that you only get sequences of segments at the functional level; neither initially nor ultimately, do you have physical segments. And you only get sequences of segments at the functional level when you have holistic patterns that are only partially distinct. These partial distinctions are then analyzed as discrete segments.

Notice, though, that on this view, phonemic structure emerges only after the model's outputs are given a functional analysis. The outputs themselves do not yet constitute phonemic structure. Even though certain pairs of them differ only in the initial or latter part, they are nevertheless holistic syllables until discrete structure is imputed to them. And discrete structure is imputed to them only by an indirect process of comparing holistic syllables. Thus it does not seem problematic on the present view to say that phonemic structure arises by process of indirect causation. But then what the emergence of phonemic structure consists in is largely analyzing physically holistic syllables as discrete units but, significantly, the holistic syllables are already functioning as minimal pairs, insofar as pairs of them may differ in only an opening or only a closing.

On the Recombination View, phonemic structure arises as the consequence of recombining primitives. But it arises as a perhaps too-direct consequence to count as emergent. On the Asymmetric Dependence View, phonemic structure emerges by an indirect process of comparing of holistic syllables and abstracting discrete units from them. However, these holistic syllables are the outputs of Lindblom's model and already function as minimal pairs. Thus, we seem to be left with the choice between a view (Recombination) which takes phonemic structure to emerge from pre-existent discrete, physical structure, and a view (Asymmetric Dependence) which takes it to emerge from pre-existent continuous, functional structure.

 Conference site: http://www.infres.enst.fr/confs/evolang/