Blog Rating

Selected Books by Edmund Blair Bolles

  • Galileo's Commandment: 2500 Years of Great Science Writing
  • The Ice Finders: How a Poet, a Professor, and a Politician Discovered the Ice Age
  • Einstein Defiant: Genius vs Genius in the Quantum Revolution

« Blogging as a Research Tool | Main | A Brain for Resolving Hierarchies »


Justin Colley

What you point to is just the competence/performance distinction. I can't remember phone numbers very well, but that is just a limitation of processing power. It doesn't change the fact that my numerical capacity has a discrete infinity.

You point out, correctly, that some cultures do not count past 2. Nevertheless, all humans have the capacity to do so, unless there is some evidence of which I am unaware. Some people from primitive cultures have initial difficulty interpreting two-dimensional images as three-dimensional, but there is no doubt that our visual systems are the same as ours.

Similarly, the fact that the more nested a sentence becomes, the more difficult it is to understand, entails nothing about the language faculty. It is a processing constraint, which is probably a global constraint. Think, for instance, of my difficulty remembering phone numbers, or even of patting my head and rubbing my stomach at the same time!

The very long sentence you give ('Joan knew that Peter knew that...' etc) is difficult to understand, but not impossible. Change the word order, however, and the sentence is unintelligible. The only known computational system that can make intelligible the semantic information in that sentence--or in most, if not all, of the sentences we're using--is one that has the property of recursion.

Karthik Durvasula

Justin you get what the generativists have been pointing out for ages. Very well put!

Two more comments:
1) About "No editor or schoolmarm would accept it under any other circumstances, and any theory of language that accepts it as a possible sentence does not care about how the humans produce or follow sentences."

This is an unfortunate statement. What schoolmarms or editors think is right should have little place in detailing linguistic theories. We have seen time and again (as LanguageLog points it out) how mistaken these people are. And as far as editors and "likely say that the reader gets lost in the details" is concerned, editors are far less concerned with the faculty of language anyway, and far more concerned with the maximal comprehension of language under the constraints that the environment and other production processes impose on the human system. This is hardly what a generativist calls the "language faculty".

2) About "the maximum number of nested dependencies was three (though this was very rare) and that in spoken language, multiple nested dependencies are practically absent. This suggests that '[f]ull-blown recursion creating multiple clausal center-embeddings is not a central design feature of language in use.'"

This is equally unfortunate. The authors have at the very basis of their argument, a frequentist argument. If there is one argument that has been thoroughly argued against in generative syntax, it is the argument from frequency. To still use it as a basis for our understanding of the language faculty is no better than hanging on to behaviourism in it purest form. lol!

In extending Justin's discussion of competence vs. performance. Let's say the limit on sentence depth in observed language/corpora is 3 or 4. No matter what ur pet theory of this fact is, it will inevitably boil down to a performance factor - either working memory limitation as in the classic view or attentional factors as in your view (although, honestly I don't see what it means to say attentional span independent of working memory). That is, there is an independent (non liguistic) explanation for this fact. Therefore, the simplest model of the language faculty that doesn't duplicate explanations makes no reference to sentence depth. This is the essence of the argument. Proper scientific methodology regarding avoiding duplication of explanation, force us to admit that there is no known limit on the depth of a sentence in the language faculty. The limits in the performance can already be explained thru other mechanisms. And this is the crux of the competence/performance divide.

The argument at some level is identical to why there are an infinite number of integers. No matter what text of mathematics or numbers, you will see only a finite list of numbers. But, we don't want to say the system of maths in general or that in our heads somehow details just the finite list. The real reason for this is that we can account for the finiteness in the lists to the fact that any physical material that records numbers is likely to be finite in character. That is, there is an independent explanation (performance factor) that accounts for the finiteness of the observation. Our theory of maths in general or in our heads is much cleaner/simpler if we don't account for the finiteness, and indeed go after the boundlessness/infinitude.

I do want to say no one is arguing against numbers. You can see a positive discussion of probability and frequency in the foundational document of generative linguistics (chomsky's dissertation). The issue here is on what structures/representations are numbers or probabilities or frequency calculated over. (I have conflated the three terms cos this is a very general audience)
BLOGGER: It’s great to see some comments from the generative group, although their metaphysics always drives me a bit mad. Rejecting performance-based evidence on the grounds that competence is what matters, and frequency-based evidence on the grounds of other “arguments,” and user-based evidence on what is acceptable, ensures that a priori reasoning won’t be challenged. No data need apply.

Here’s a good example of metaphysical thinking, “the fact that the more nested a sentence becomes, the more difficult it is to understand, entails nothing about the language faculty.” Passing over the interesting verb “entails” I’m left wondering what test we can do determine the nature of the language faculty.

Here’s another example, “The only known computational system that can make intelligible the semantic information in that sentence--or in most, if not all, of the sentences we're using--is one that has the property of recursion.” But the fact is the sentence is unintelligible as it stands. I suppose I could diagram it out, but I doubt that I could ever grasp these relationships whole the way I can instantly understand what’s going on all around when I read, “Peter told Jane that Jack is a liar.” But it turns out we don’t need recursion to understand this sentence.

So perhaps we don’t need recursion to be understood, but we do need it to produce the unintelligible. Is that what you’re arguing?

Justin Colley

I don't know if I could call myself a generativist, I've only just started studying linguistics. I have found their arguments to be more convincing thus far, so I guess you could say I'm a generativist-in-waiting!

A quick couple of points. The point about processing power (working memory limitations, as Karthik puts it) is empirical, not metaphysical. If these limitations apply to all mental processes--processing of numbers, pictures, music-- then it is a general limitation that tells you nothing about the language faculty specifically. If there is a limitation that only applies to the production and understanding of sentences, then it does tell you something interesting about the language faculty. As far as I am aware, the evidence indicates a general limitation. You are vastly more knowledgeable in this area, though: I would be very interested if there is evidence to the contrary.

Again, the sentence is difficult to understand, not impossible. This difficulty is--pending evidence to the contrary-- the result of limitations in working memory. Make the verbs uniform and this becomes clear. So: "Joan said that Peter said that Elizabeth said that James said that Helen said that Caulfield said that Harvey said that... x" It's pretty clear what's going on in this sentence; the difficulty is in remembering all the names! When the verbs change, as in your example, even more information is available to the listener. It is difficult to keep all this information in the head at the same time, but again, this is very probably a general constraint. So if you did diagram it out, like you suggest, I would have as much difficulty processing it as when it is expressed in English.

Note also, that to make it comprehensible in a diagram, you would be forced to recreate to recreate the 'nestedness' of the original sentence. You couldn't, for example, paint a big panorama of the scene, and convey the same information. So again, you haven't shown that the nested sentence is unintelligible at all, merely difficult to process.

(I'm afraid I don't know enough about linguistics to see the force of your example 'Peter told Jane that Jack is a liar.' From what I can tell, it's a bit like saying 'I can count to 10 and stop, therefore a discrete infinity isn't the crucial property of the human numerical faculty.' I'm sure you can see the fallacy in that. I might be missing the point here, though. Perhaps Karthik can chime in on this.)
BLOGGER: i see we use "understand" in different ways. I'm using it as a kind of gestalt experience in which the sentence is grasped as a unit. So, I'm not much persuaded by talk of attention or memory limits since these put limits on understanding. You're using it differently. Something more to argue about.

Karthik Durvasula

1) "Rejecting performance-based evidence on the grounds that competence is what matters, and frequency-based evidence on the grounds of other “arguments,” and user-based evidence on what is acceptable, ensures that a priori reasoning won’t be challenged. No data need apply."

The issue at hand is not whether performance based evidence is applicable to theory construction or not. In the end we only have performance based indicators. The issue at hand is, how do we visualize it so as to maximize the simplicity of our theory; and thereby maximize our understanding of what's happening. Also important is how much weight do we put into any performance factor. The the unsettling answer as Chomsky has repeated too many times is that no one performance indicator is the holy grail. As recent experiments in psycholinguistics show, some performance measures give conflicting results; in which case, which one do we believe? And weirdly, the answer is it is not "data first"; it is "theory first" - we believe those performance indicators that lead to theoretical explanation/elegance. If you don't like this answer and I am sure many psychologist don't, then one needs to provide another pathway to choosing between conflicting performance data.

Btw, no one said "competence is all that matters". Competence is all that matters to study the "systematicity" underlying language; because that is by definition what competence is. It is not to say, performance factors are not important or nonexistent. Again this is something that was extremely clearly laid out in Chomsky's dissertation and in Chomsky (1965)

2) Furthermore, you can give a few sentences off-hand and claim a certain (semi-formal) representation of those structures. However, any theory awaits confirmation through repeated support. The kinds of structures frequentists and even you seem to be proposing are things people discarded very early on in the generative syntax literature because they have no explanatory power; nor can they actually account for the phenomena that have been raised as things that need to be accounted for. This is where I stress the foundational documents in the field have raised these issues about 60 or so years back, and have given it a thorough discussion. My personal favorite is Chomsky 1965 - Aspects to a theory of Syntax. Chomsky very nicely raises and answers the questions that you are raising. And if you read it carefully, you see that there is almost no option but to do what is being done in generative syntax.

Again you are welcome to dispute the theoretical claims of generativists (as opposed to the methodology), but you can't dispute it until you have an account for the data that supplants theirs. And the kind of data they deal with is far more sophisticated than just a couple sentences (again, I mean no offense, just clarity); because ultimately, the important goal is to understand the system that producing the complexity of sentences observed and acceptable in natural languages.

To argue against the frequentivist and rather superficial approach, one needs to just go back to probably one of the oldest famous sentences in generative syntax:

(1) "colorless green ideas fly furiously"

This sentences and any collocation of its subparts is going to have a zero frequency. The sentence doesn't even make sense. However, nearly every undergraduate I have taught in my intro. to linguistics classes over the years has said there is something "English-like" about it. The question at hand is what knowledge has he/she depended on to come to the conclusion, for it is surely not frequency. Contrary to this no one thinks "green ideas colorlessly furiously fly" is very "English-like" (Note: Justin pointed to a similar example). Both are zero frequency, so why do we have a difference in judgements? Clearly, there is a system beyond what can be observed in the performance; and it is this system that generative linguists are after.

3) "metaphysics always drives me a bit mad" I am afraid I can't help you with this one. What both Justin as I have pointed out is that if one were to go about constructing a theory of numbers, one would do exactly what is being done in generative syntax (in fact, historically, that is what was done). What you call "metaphysical thinking" has been at the bottom of rationalist/deductive theories of science since the beginning (I refer you to Tarki 1946. Introduction to Logic and to the Methodology of Deductive Sciences; though it is more about mathematical theory than anything else).
At the core is the assumption for such a view (clearly "metaphysical" in your sense), that the object of inquiry has a "system"; and the system needs to be factored away from various external influences. This is the Gallilean method. For example, when Galileo studied objects falling to the earth; the "performance factors" (air resistance…) would never have permitted him to directly infer that the system causing it had the same influence on all objects irrespective of mass/weight - this is because the various external influences would always have shown different falling rates - he clearly wasn't privy to vacuum chambers back then. It is the "metaphysical" thinking of perceiving a system behind the madness that led Gallileo to present a theory that was not a reflection of reality as we observe it; but a description of it in the abstract. As far as I am concerned, if one doesn't share this assumption that there is a distinct system behind natural phenomena that can interact with other systems (be it gravity or language), and we don't try to describe it in the abstract, the whole venture of science becomes incoherent. The history of science is replete with countless examples of exactly the same process, which I am sure you as someone very familiar with the topic can immediately vouch for, on some introspection.

4) @Justin: "'I can count to 10 and stop, therefore a discrete infinity isn't the crucial property of the human numerical faculty.'" Yup, I think you raised the logical fallacy in the argument precisely.

Karthik Durvasula

To add to this: there is also a debate about the validity of generative syntax findings/methodology with many psychologists who don't seem to understand it.

Almeida and Sprouse (2011) run standard psycholiguistic experiments on ALL the sentences covered in Adger (2003) introductory syntax book, and show that they get the same qualitative results as linguists have gotten thru their standard elicitation methods. Therefore, the data on which the theories are constructed are equally beyond reproach.

Finally, I see a part of your argument in the subtext as "I don't understand it, therefore it couldn't be right". And if this is indeed the basis for rejecting the theory, then there is something wrong. You either properly understand it and then reject it, or not understand it and have nothing to say. To not understand it and reject it would be tantamount to pseudoscience. There have to be better arguments than that if one is to reject a whole tradition of scientific inquiry.

Verify your Comment

Previewing your Comment

This is only a preview. Your comment has not yet been posted.

Your comment could not be posted. Error type:
Your comment has been posted. Post another comment

The letters and numbers you entered did not match the image. Please try again.

As a final step before posting your comment, enter the letters and numbers you see in the image below. This prevents automated programs from posting comments.

Having trouble reading this image? View an alternate.


Post a comment

Your Information

(Name and email address are required. Email address will not be displayed with the comment.)

Bookmark and Share

Your email address:

Powered by FeedBlitz

Visitor Data

Blog powered by Typepad