Sunday, September 17, 2006

Competence and Performace

The following is a lengthy response to one of Noah's posts. I started it early in the day, and he responded in part to an earlier relevant post of mine. I have marked which parts of this were written before reading his final version.

I have spent entirely too long writing this and will need to take a break to get some schoolwork done. This post is likely to change over the next couple of days as I rethink my manner of expression, particular wordings, etc. Apologies in advance for what are no doubt copious amounts of typographical and grammar mistakes.

[Written before reading Noah's latest]

Noah's most recent post is an attempt to demonstrate that my view of what Linguistics is is too narrow. I think this is an interesting topic, so I'm prepared to give a fairly detailed response.

First, I think I would have to agree with people working primarily on the "performance" side of the question that there's something obnoxious about those of us in "competence" insisting that we are "Linguistics proper." And it's undeniable that some of the people who insist on a competence (knowledge of language) vs. performance (production and perception of language - speaking, hearing, etc.) dichtonomy do so in order to establish themselves as superior researchers. But I don't mean it that way, and I think it's perfectly clear that Chomsky didn't either. And in fact, I think the offense taken comes from some fairly trivial organizational problems at universities.

Here at Indiana University, for example, we have separate departments of Linguistics and Speech and Hearing Science. This seems to me a better way of organizing things. People who are interested in motor productions of sounds, for example, clearly belong in Speech and Hearing. People who are interested in formal Syntax obviously would be out of place in Speech and Hearing. People like Noah who are interested in both (Noah's line of research is fascinating - essentially dealing with questions of how sound waves get mapped onto internal linguistic representations - specifically in what psychological processes we can hypothesize people use to discriminate between linguisticially relevant sounds) have both departments at their disposal and can either do joint PhDs as students or have dual appointments as professors. The debate, seen from this perspective, is really one over who has the right to use the word "Linguistics" in their department name.

A couple of years ago, there seemed to be a kind of fad among certain of the "sound people" to say "Language Science" rather than "Linguistics." I had the impression that this was meant to differentiate what they do from what Syntax people do. The intent is obvious: Syntacticians aren't really doing "science." And according to some definitions, they're not. In my Methods in Cognitive Science Course, I was taught that an "experiment" necessarily involves manipulations of variables. Simple observation doesn't qualify. An experiment needs controlled circumstances so that we can be sure that the dependency we're trying to establish really is one owing to the variables and outputs involved, and not to some hidden factor, or combination of factors, or artefact of the way the numbers fall, or whatever else. In that sense, Syntax is certainly less empirical than Phonetics et al. Syntax is almost totally concerned with observation; there are few, if any, real "experiments" involved. And so people on the production side of the debate are able to counter that Syntax is "less scientific" than what they do. And in an important sense, they're right.

The use of the term "Language Science" illustrates, however, that the debate is reminiscent of political correctness in its insistence on attending to semantic issues as though they were substantive. To the general public, "Linguistics" and "Language Science" are equally vague. The "folk" definition of either is "department having to do with language-related stuff." The attempt to recast "Linguistics" as "Language Science" is nothing but an attempt to choose a word that would exclude the more mentalistic subdisciplines in the same way that the word "Linguistics" can be taken to exclude the more sound-related subdisciplines. But ultimately what is being fought over is who has the right to go to the general public and say they are a better representation of "department having to do with language-related stuff." But of course, the question could be avoided by having two different departments, each dealing with a different kind of "language-related stuff."

Now, Noah's objection would be that such a split wouldn't really make sense. It is obvious that human language will have to be studied from all possible perspectives. Insofar as most people are genetically endowed with the abilities to speak and hear, and insofar as human languages prefer these modalities when they are available, Phonetics is clearly "Linguistics/Language Science." I guess it would prefer to be in the "Language Science" department, but lots of Phonetics is also about simple observation (rather than experimenting), and so it can presumably feel at home in "Linguistics" too. But why does it have to choose at all?

[Written after reading Noah's latest]

Well, there's some truth to that. But there's also an obvious response - and that's that insofar as we have divisions in science at all, we're committed to the idea that you can't simply lump everything that's related in some way to something and expect it to form a stable, usable cateogry. I used the example of Law in my original post. It is undeniable that the legal profession is language-bound. Law is a function of language in so many ways. But you don't have to know much about Linguistics - neither what sound waves make what words nor what reason we have for believing in syntactic operations x, y or z - to function as a lawyer. That something is a thing people regularly do with language isn't a convincing argument for making it part of the study of human language - though certainly lawyers have produced copious amounts of data of human language in action over the centuries, more than their fair share. Another way to put it might be the association game. As children, we all had a good time taking turns saying words, and then another related word, and then another and so on - and the fun of the game is that you end up somewhere completely different from where you started. Literally, everything in the universe is related in some number of finite steps. It isn't enough to simply say that sounds and such are typical featuers of human language. We'd need to know that there is some useful sense in which including them in our concept of "Linguistics" makes it a more useful concept than it would have been if we had confined ourselves to competence.

Now, I don't want to get pigeon-holed into being some kind of hard-core "Language is only philosophy" person. I did say in my original post that Phonetics was Linguistics too. So I should explain that the reason I feel I need to take this position for this post is because I get a bit frustrated with the techniques that the "sound people" use to argue that they are every bit as much Linguists as we are - more even.

For example - Noah quotes heavily from Labov in his original post. Here the argument is supposed to be that competence/performance distinction are ultimately meaningless:

The terms 'idealism' and 'materialism' can be seen to be most appropriate in relation to the definitions of data involved. The idealist position is that the data of linguistics consists of speakers' opinions about how they should speak: judgments of grammatically or acceptability that they make about sentences that are presented to them....

The materialist approach to the description of language is based on the objective methods of observation and experiment.

Nice, nice. In other words, people who are concerned with "competence" base their conclusions on people's opinions, people who are concerned with "performance" base their conclusions on objective facts. Labov has chosen his words carefully to ensure that the reader gets the impression that people concerned with performance are doing real work and those concerned with competence are just taking surveys. Now, this characterization is indeed damning (never mind that syntacticians use other methods besides asking people's opinions - there are also corpus studies and observations of speech - but for the most part Labov is right) if you don't buy the performance/competence divide to begin with. But if you see merit in the divide, then it isn't damning at all. People interested in competence work with "speakers' opinions about how they should speak" because it is the only data available. If there were more objective methods available, we wouldn't hesitate to use them.

What Labov is missing here (deliberately, I suspect) is that "grammaticality" is about people's opinions of sentences WHAT ELSE WOULD IT BE? Asking people's opinions about sentences as a way of data collection is justified because that is what is being studied. "Grammaticality" is nothing other than a native speaker's intuition about the output.

This is in sharp contrast to Phonetics, which has things that it can measure. Of course, if things are available for measurement, then it makes sense that they should be so catalogued. It would be no use showing a person a readout of the soundwaves of the sentence that he just spoke and asking him what he was going for when he produced this bit here, or whether he thinks maybe the bit at the end isn't really "as English" as the bit at the beginning - because that isn't what Phonetics studies. "Speech production" has a physical definition. "Grammaticality" might, but we don't know enough about it to be sure, so we investigate it with the methods we have available.

Suggesting that these two methods are the result of a philosophical difference over what the various parties see language study as is to gorssly mischaracterize the issue. It's like suggesting that there would be something fundamentally wrong with asking a subject in a perception experiment to actually give his opinion as to, say, what color he's seeing. No, no, we measure color perception...well, HOW ELSE? If the object of study is how people discriminate between different shades of similar colors (to use experiments that Nosofsky likes to do), then the subject's opinion of what the color is is surely relevant??? Even if there were somehow some ethical method of splitting open the subject's skull and reading information directly off of his neurons, we would still need a verbal report from the subject as to what he thought he was seeing in order to properly correlate neuronal configurations with slight variations in color perception! Well, the study of "grammaticality" is like that only more so. There simply is no way to study it without making heavy refernce to subject opinions - more, in fact, than in the color perception experiments.

Now, the point of this section is just to say that characterization's like Labov's are silly. There are linguistic sounds, and there is grammaticality. Linguistic sounds have a studiable physical realization, and although grammaticality might, we're a long way from the technology that would allow us to see it directly, so for all practical purposes, it's something that's not physical. In every possible useful definition of the word "subject," these things are obviously two different "subjects." There is, in other words, a good reason to talk about a difference between competence and performance. Now, this isn't to say they can't inhabit the same university department. They can. But there is nothing ridiculous or deceptive about trying to say that they study different things and so necessarily use different types of data and different approaches to the data they study.

Now Noah pulls the same trick:

While I don't agree that this constitutes an infinite regress (it seems clear to me that a lower bound on linguistically relevant and controllable production and perception variables is establishable in principle), the general point is important, and one could easily make the case that the partition between linguistically interesting competence and mere performance typically excludes linguistically relevant knowledge.

"Linguistically interesting competence and mere performance." Take out the adjectives in that sentence and the problem goes away. Again, the competence side isn't denying that the performance side has interesting things to study. The debate is really just about who gets to use the word "Linguistics" in their department name. There is nothing "mere" about performance. Performance is extremely important, and there is a whole array of fascinating subdisciplines devoted to its study.

Noah then procedes in kind - with what I take to be the general strategy for people trying to conflate competence and performance. The process goes something like "establish that some effect of sound has been observed, mention obliquely that languages are made of sounds, assert that performance is therefore relevant as though someone somewhere had ever denied that to be the case."

"...[p] is relatively hard to hear, and [g] is relatively hard to say." This is because of aeroacoustic effects in both cases. In the former case, the shape and size of the sub-closure cavity in [p] causes the noise that accompanies closure release to be very quiet, and thereby indistinct, relative to stops at other places of articulation. In the latter case, the closure for [g] is closer to the vocal folds than most other stops. Voicing requires a pressure drop across the glottis, and stopping airflow above the glottis inhibits this, particularly so when the super-glottal cavity is small, as it is with [g].

So, at least in some cases, the means by which sounds are produced and perceived directly, if not deterministically, affects the distribution of speech sounds across languages. Speech sounds - combinations of phonological features - are at the foundation of phonological theory. If you know the phonology of, say, Dutch, you know, among other things, which sounds are part of the language and (a subset of) which sounds are not. Which is another way of saying that you know which phonological features are functional in combination with which other features. Which means that you know which rules and constraints operate when. Which had better be part of competence, or the term risks losing all meaning, at least with regard to phonology. If this is part of competence, then at least some of phonetics is linguistics proper.

This is a curious thing to say in light of the fact that in my original post I thought I had made it clear that Phonetics was part of Linguistics.

People like me, though we accept that Phonetics is also Linguistics, would insist that Phonetics is more concerned with the interface than the subject proper.

So I accept that Phonetics is part of Linguistics (and, indeed, for precisely the reasons given here). I see no reason to conclude from anything above that Phonetics is not "more concerned with the interface than the subject proper." Phonetics deals with the implementation. And features of the implementation are indeed what Noah describes here. That phonemes are representation-independent I take to be clear. Applying phonetic knowledge is useful for producing Dutch sounds and recognizing them when you hear them, to continue the analogy, but Dutch phonemes can as easily be written in Roman letters. Mute Dutch people are capable of communicating in Dutch without any knowledge of the sound content of these forms. And that's because sound is ultimately an interface. Now, it's the most important of the possible interfaces - it's the default interface to be sure. And it's important to such an extent that I say (again) I "accept that Phonetics is also Linguistics." But I remain convinced that sound has little or no effect on studies of meaning and grammar. That makes it a separate subject. If we want to include Phonetics in Linguistics, then it will be - at least for the time being - a subdiscipline (just as Syntax is a subdiscipline). The reason we are able to maintain a distinction is because the data that the two subjects use are different, and their methods necessarily also. More importantly, the object of study is different.

It's as if we were to say that Syntacticians use their mouths when they ask people their opinions of sentences, therefore they need to know some basic things about Phonetics in order to do their work. Poppycock. Syntacticians speak languages in the same sense that the girl at the coffee shop does. She doesn't need to know anything about Phonetics to serve coffee, and we don't need to know anything about Phonetics to study Syntax. Likewise, Phoneticians don't need to know anything about Syntax in order to study Phonetics. With Phonologists and Phoneticians the relationship is obviously much closer, and there may indeed be plenty of overlap.

I take it, however, to be obvious that there are interfaces between performance and competence. If there weren't, communication couldn't happen. And obviously if we study these things closely enough the lines will blur around the edges. After all, it isn't as though anyone hypothesized that sound waves hit the ear, a black magic box translated this into "competence-related stuff," the brain did its competence thing, and then another similar inscrutible box translated from output knowledge into tongue waving and air pushing. The boxes aren't black, and they are a (very interesting in my opinion) definite object of study. Look, people read. In order to read, photons have to hit their eyes. The retina has to do something with these photons, organize the information in some way as to make it understandable. Now, people become highly sensitive to a wide range of variation in writing. We are able to tell who did the writing, or what font it's in. Especially amazing is that we can play artful games (such as one often sees in advertisements) with letters - changing them to things so wholly different from what we're used to seeing that computers have an extremely difficult time discerning what's been written. We can have animated letters dance and do attention-grabbing stunts of various kinds. And yet, at some level, they're just letters. This entire range of things gets translated into linguistically relevant information. Which is, ultimately, that this is an 'a' followed by an 's' followed by another 's,' which is the 'ass' in 'assume,' etc.

And despite all this, I have never heard it seriously suggested that Optometry is part of Linguistics.

Of course, once we know a decent bit about Optometry and the Cognitive Science associated with image perception, then no doubt there will be some studiable connections. As I said, the interface isn't a giant chasm. One thing touches another touches another touches another, and there's an unbroken causal chain from external form to internal perception. I don't think anyone (least of all Chomsky) has ever sought to deny that. If anyone did, I'm sure nobody took them very seriously. But the point remains that Language happens in multiple modalities. It migrated from being sound only, to being sounds and motions, to being sounds and motions and images, to living in computers, and so on. It's hard to imagine that anything fundamental to Capital-L Language is really contained in sound waves. No doubt there are some examples of things that are, but these are not, I think, the general rule.

Anyone who speaks Japanese will know exactly what I am talking about. Because that language more clearly divorces sound and writing, there are plenty of examples of subtle differences in meaning that have no accoustic realization. My favorite example has to do with the word "nomu," which is nominally "to drink." But of course we can differentiate senses in which one drinks. One can "take in and retain," one can "sip," etc. And there is a host of at least three (maybe 5 - depending on how you define "general use") general use characters for this word, and they have different shades of meaning. They are all, however, read exactly the same way. In other words, there exists a group of related concepts in Japanese that are differentiable in expression only through writing. They have no real sound interface. And there are plenty of such examples in that language. In Natsume Soseki's excellent book I am a Cat - the actual word used for "I" is obscure in the extreme. "Wagahai wa neko de aru" is the Japanese title - where "Wagahai" means "I" and is used herein place of the more common "Ore," "Boku," or "Watashi." It is also "spelled" differently - but the interesting thing is that few Japanese speakers know that (combination of) character(s). So you frequently hear people say "Watashi wa neko de aru." And in fact, I read the book originally with no idea how to pronounce the character - I knew from context what it meant. The two characters that make up "wagahai" are related to concepts of "self" in other ways, but you never see them in modern life, and in any even I didn't know how to pronounce them at first and just read them "wata(ku)shi." It isn't the same expression of "self" that the normal words realize - something you pick up quite quickly in reading the book. I guess my understanding of the book is nowhere near as complete as a native speaker's would be (it is quite difficult to read in Japanese, and I never finished it) - but the point is simply to illustrate a complete concept I have that had, for a long time, no phonetic analogue.

Now, this isn't meant to be a knockout case. It is merely an illustration that however many examples of supposed phonetic effects on meaning and grammar there are, we kind find equally many in terms of writing and gesture. The point is that Phonetics is about Captial-L Language only because it studies the default and prefered language interface. It does not deal directly with issues of grammaticality and meaning that people like me take to be the core of Linguistics.

And again, I should stress, the debate over who gets to call their field "Linguistics" is peripheral, semantic - ultimately unconsequential. The point that I am trying to illustrate is simply that there is an important difference between these subfields. Saying that they all study the same thing is true only in the sense that the word "language" comes up a lot in both subjects. Now, I will say that I do believe as more work is done and more is learned about how the interface functions, more connections will be discovered. It may be that at some future date the distinction will vanish (though I personally very much doubt it). But this is true of many disciplines - and in any case there is reason to maintain the distinction now. Until phonetic effects on syntax become the rule and not the exception, it is quite natural to study the two as separate domains.

Now - as to the relationship between the core and the interface. Noah is willing to take my computer analogies for language seriously - so let me speak to those a bit. In his more recent post

On reading this, it occurred to me that we need to be very clear about what we think we're doing in creating theories of competence. It seems to me that these theories are about the function of grammar. That is, when we build, say, generative theories of linguistic universals, we're building descriptions and explanations of something akin to programmed functions over data arrays. The kind of data array used as input, and the kind needed as output, seem to me to have quite a bit of influence on the internal structure of the function.

If we're talking about discovering the functions and data arrays crucial to the operation of some system, in this case linguistic, we are manifestly not talking about which programming language these functions and data arrays are implemented in. The 'same' function can be implented in Scheme, C, C++, or Java.

I agree that we are talking about grammar as a functional system. What I don't buy here is that the kinds of data array used as input and the kind needed as output is all that fundamental to the processing going on. This is really the whole point about Friedman's homework yesterday. Now, at some level a program deals with blocks of bits. And if we're writing a program that reads in integers, for example, in an array, does something to them (let's say it multiplies each by the next number in the array, 1 if no number follows) and then spits them back out in another array, then do we need to know anything about how integers are implemented?

God, I hope not!

Integers, it must be understood, are themselves an abstraction, and how they are implemented in different machines is far from uniform. To deal with these differences, we naturally need some kind of an interface. This is why there are problems with running things across platforms. Java's big boast is that it runs on every platform, for example. That's a (highly) debatable point, but let's just take Noah at his word for now and say that it shouldn't matter what programming language a program is written in, we just want the algorithm to run. And it shouldn't matter what platform we're working on, we just want the algorithm to run.

Well, right - and that's the whole point. The algorithm can be specified quite well without needing to know anything at all about the platform. The algorithm is, in fact, completely invariant. And that's the point of having programming languages in the first place - so that I can write programs that don't have to know anything about the machine! When I write a program that reads in an array and does all the multiplication I talked about earlier, I want to be able to program on the level of arrays, array indices, and multiplication of integers. I don't want to have to program on the level of bits and machine words and such. Ultimately, of course, everything takes place on that level. But this isn't really what programming is about. If programming were about playing with bits, no one would have ever designed a programming language at all. They would have just used the machine language and set the bits directly.

This seems to me a good analogy for what's going on with Phonetics and Syntax. Understanding happens on the level of Syntax and Semantics. That's where the mental part of communication comes in. Now, we have to translate Syntax and Semantics into Phonetics (or writing or whatever) in order for the physical act of communication to take place, yes. And this is like the bits and the hardware. But the actual specification of meaning doesn't take place on the level of bits and hardware. My one Scheme program should run equally well on a machine that represents integers with one 16-bit machine word just as easily as a machine that represents integers in 2 32-bit words, or whatever. The program works on the level of the algorith. I tell it to read in the first item in the array (where here an "array" is a machine-independent abstraction), read in the second, "multiply" it (which can be implemented in various different ways with invariant semantics), and give me an answer, which I then store as the first item in an output array. "Language" takes place on the level of arrays and integers, but "calculation" takes place on the level of hardware, which might be different for different systems, etc. A programming language does the same thing that a human language does: it abstracts away from the various hardwares of the two things communicating (the human and the computer) and allows for concept-level communication to take place. That there is an interface is not under debate, and neither is the idea that the interface is crucially important. But it's ultimately the compiler or the interpreter that translates the program into machine-code instructions. Machine-code instructions are performance, and the program is competence.

Of course, this is an analogy, and it remains to be seen how far it can be stretched. There may indeed turn out to be some important differences between human and computer language (as such, I mean, obviously humans and computers are not the same kinds of entity at all, so there are concrete differences in the kinds of ideas they need to communicate -- indeed, it's stretching the term for most people to say computers have ideas and "communicate" at all).

I think the main point to get out of all of this is that it really is a definitional matter. Clearly (at least to me) at this stage competence and performance are different things. There may be some day in the future where our knowledge of language is finely enough detailed and the causal relationships between what we now call the "interface" and the "abstraction" are better understood to the point where it's worth doing fulltime study of how they affect each other. But for now they remain different in kind. I can use the words "competence" and "performance" consistently - the concepts are stable and reflect truth about reality.

I further think that the only reason people have these debates at all is because one side or the other feels slighted. And I realize I contribute to this problem by saying things like "Linguistics Proper" to talk about competence-related fields.

I know this post is already ridiculously long, but let me say something about why I do that briefly. What I take myself to study is Language - which doesn't necessarily have to be "human" language. At present, human language is the only known form - but I'm willing to imagine that there are other possible forms. Just like with the various programs for Friedman's homework (see previous post) - the point is that you try to figure out what's common in general to Language. Once you know that, you can write various instantiations of it. In my particular case, I'm trying to understand abstractions that will allow me to program a machine to translate between human languages. Obviously this involves instantiating linguistic knowledge in something that is not human. And so it is crucial to my field that there be something called "Language" that is not necessarily exclusively found in humans. If such a thing is not possible, then there is really nothing for me to do (except perhaps create artificial humans).

So I definitely recognize that this is a prejudice. This is an assumption that I have to make. It may not be borne out in practice, but my whole point is that at present there is no case for denying it. I am still, as it were, free to try because no one has shown me Language crucially depends on eardurms and neurons or vocal tracts or whatever. There is at least a component of it that (I assume) does not depend on these things, and what I study requires me to figure out what that is.

As I said, I think the choice of terms is regrettable. Because "Linguistics" is the default name, asserting that Phonetics is not relevant to "Linguistics" does indeed carry with it the implication that Phonetics is somehow less interesting. But that is not my intention. Phonetics is an interesting and highly useful field of study. It is not the same field I study. The aspects of language that it studies are different in kind from the aspects of langauge that I study. I wish that we had better terms for this so that people didn't get as defensive about it as they obviously do (on both sides - see example above about "Language Science").


Post a Comment

<< Home