The following remarks by Poincaré were quoted in earlier versions of my Lingua paper but not included in the final version.

***
… The physicist who has just given up one of his hypotheses should, on the contrary, rejoice, for he found an unexpected opportunity of discovery. His hypothesis, I imagine, had not been lightly adopted. It took into account all the known factors which seem capable of intervention in the phenomenon. If it is not verified, it is because there is something unexpected and extraordinary about it, because we are on the point of finding something unknown and new. Has the hypothesis thus rejected been sterile? Far from it. It may be even said that it has rendered more service than a true hypothesis. Not only has it been the occasion of a decisive experiment, but if this experiment had been made by chance, without the hypothesis, no conclusion could have been seen; and only one fact the more would have been catalogued, without deducing from it the remotest consequence. (Poincaré 1952: chap. 9, 150-151)
***

There is only one way to make sure of the validity of a chain of logical reasoning. This is to put it in the form in which it is most easily testable: we break it up into many small steps, each easy to check by anybody who has learnt the mathematical or logical technique of transforming sentences. If after this anybody still raises doubts then we can only beg him to point out an error in the steps of the proof, or to think the matter over again. In the case of the empirical sciences, the situation is much the same. Any empirical scientific statement can be presented (by describing experimental arrangements, etc.) in such a way that anyone who has learnt the relevant technique can test it. If, as a result, he rejects the statement, then it will not satisfy us if he tells us all about his feelings of doubt or about his feelings of conviction as to his perceptions. What he must do is to formulate an assertion which contradicts our own, and give us his instructions for testing it. If he fails to do this we can only ask him to take another and perhaps a more careful look at our experiment, and think again. (Popper 1959: 99, the emphasis by HH)

[The following is taken from Popper 1959 The Logic of Scientific Discovery (Routledge). (30. Theory and Experiment)]

　　　All these considerations are important for the epistemological theory of experiment. The theoretician puts certain definite questions to the experimenter, and the latter, by his experiments, tries to elicit a decisive answer to these questions, and to no others. All other questions he tries hard to exclude. (Here the relative independence of sub-systems of a theory may be important.) Thus he makes his test with respect to this one question '... as sensitive as possible, but as insensitive as possible with respect to all other associated questions. ... Part of this work consists in screening off all possible sources of error.'¹ But it is a mistake to suppose that the experimenter proceeds in this way 'in order to lighten the task of the theoretician',² or perhaps in order to furnish the theoretician with a basis for inductive generalizations. On the contrary, the theoretician must long before have done his work, or at least what is the most important part of his work: he must have formulated his question as sharply as possible. Thus it is he who shows the experimenter the way. But even the experimenter is not in the main engaged in making exact observations; his work, too, is largely of a theoretical kind. Theory dominates the experimental work from its initial planning up to the finishing touches in the laboratory.^*3
　　　This is well illustrated by cases in which the theoretician succeeded in predicting an observable effect which was later experimentally produced; perhaps the most beautiful instance is de Broglie's prediction of the wave-character of matter, first confirmed experimentally by Davisson and Germer.^*4 It is illustrated perhaps even better by cases in which experiments had a conspicuous influence upon the progress of theory. What compels the theorist to search for a better theory, in these cases, is almost always the experimental falsification of a theory, so far accepted and corroborated: it is, again, the outcome of tests guided by theory. Famous examples are the Michelson-Morley experiment which led to the theory of relativity, and the falsification, by Lummer and Pringsheim, of the radiation formula of Rayleigh and Jeans, and of that of Wien, which led to the quantum theory. Accidental discoveries occur too, of course, but they are comparatively rare. ... (Popper 1959: 107-108)

　¹ H. Weyl, Philosophie der Mathematik und Naturwissenschaft, 1927, p. 113; English Edition: Philosophy of Mathematics and Natural Science, Princeton, 1949, p. 116.
　² Weyl, ibid.
　^*3 I now feel that I should have emphasized in this place a view which can be found elsewhere in the book (for example in the fourth and the last paragraphs of section 19). I mean the view that observations, and even more so observation statements and statements of experimental results, are always interpretations of the facts observed; that they are interpretations in the light of theories. This is one of the main reasons why it is always deceptively easy to find verifications of a theory, and why we have to adopt a highly critical attitude towards our theories if we do not wish to argue in circles: the attitude of trying to refute them.
　^*4 The story is briefly and excellently told by Max Born in Albert Einstein, Philosopher-Scientist, edited by P A. Schilpp, 1949, p. 174. There are better illustrations, such as Adam's and Leverrier's discovery of Neptune, or that of Hertzean waves.

The following remarks are given at the end of Einstein's Forward to Dialogue Concerning the Two Chief World Systems -- Ptolemaic & Copernican by Galileo Galilei, translated by Stillman Drake, 1967, University of California Press.

"It has often been maintained that Galileo became the father of modern science by replacing the speculative, deductive method with the empirical and experimental method. I believe, however, that this interpretation would not stand close scrutiny. There is no empirical method without speculative concepts and systems; and there is no speculative thinking whose concepts do not reveal, on closer investigation, the empirical materiel from which they stem. To put into sharp contrast the empirical and the deductive attitude is misleading, and was entirely foreign to Galileo. Actually it was not until the nineteenth century that logical (mathematical) systems whose structures were completely independent of any empirical content had been cleanly extracted. Moreover, the experimental methods at Galileo's disposal were so imperfect that only the boldest speculation could possibly bridge the gaps between the empirical data. (For example, there existed no means to measure times shorter than a second.) The antithesis Empiricism vs. Rationalism does not appear as a controversial point in Galileo's work. Galileo opposes the deductive methods of Aristotle and his adherents only when he considers their premises arbitrary or untenable, and he does not rebuke his opponents for the mere fact of using deductive methods. In the first dialogue, he emphasizes in several passages that according to Aristotle, too, even the most plausible deduction must be put aside if it is incompatible with empirical findings. His endeavors are not so much directed at "factual knowledge" as at "comprehension." But to comprehend is essentially to draw conclusions from an already accepted logical system." (pp. xvii-xix)

What is given below is not about generative grammar; see the original passage at this page. It is the 5th Q and A there. (I have changed the [ ] parts.)

---

Q. In the current mainstream practice of [generative grammar], is falsifiability ignored entirely?

A. No. It would in a sense be easier if it were ignored altogether. Mainstream [generative grammar] pays lip service to it; that is, it always says, "Oh, yes, of course, [grammatical] theory should be confronted by evidence. And, of course, if the evidence is contrary, if it seems to refute the theory, oh, yes, we certainly pay attention to that, and we must adjust the theory or even consider possibly a new theory." So, they do not in any way disagree with the implications of falsifiability. They preach it but they do not practice it. In other words, when confronted with contrary evidence of some beloved theory, they adjust the theory, or they minimize the evidence. Sometimes they even ignore the evidence. They do not look very hard at contrary evidence, preferring to confirm rather than to look for refuting evidence.

---

The above remarks seem to apply to much of the practice in generative grammar. The last sentence, i.e., "They do not look very hard at contrary evidence, preferring to confirm rather than to look for refuting evidence," reminds us of Popper's remark cited in footnote 82 of my "Falsifiability" paper (2003) in Lingua (see the end of section 4 the paper).

What is given below is not about generative grammar; see the original passage at this page. It is the 5th Q and A there. (I have changed the [ ] parts.)

The original link, provided in April 2003, is no longer functional. I have replaced it with the current link, but it does not provide the entire article. I am therefore copying the content of the page below.]

***
05/01/98
Challenge

Magazine: Challenge, MAY-JUNE 1998
Section: THE STATE OF MODERN ECONOMICS
THE PROBLEMS WITH FORMALISM
Interview with Mark Blaug
---------------------------
Q. I think it is fair to say that you are the leading proponent of the idea that a weakness of economics is that much of its theory is not falsifiable. If you cannot falsify something, then how can you demonstrate that it might be true? What is your definition of falsifiability?

A. In confronting a theory, it should be possible to think of evidence that, if found, would falsify the theory or would lead you to abandon the theory. You would realize that it was wrong. It could be falsified by some evidence. If, for example, someone says, "I believe such and such, and there is no evidence that would ever make me abandon this theory," then whatever this belief is, it is not science because scientific beliefs, theories, hypotheses, or whatever you want to call them, should be falsifiable at least in principle. Now all the problems begin.

Q. Did Karl Popper, the philosopher, originate this criticism of economics?

A. No, he said this about other areas. He said very little about economics, and what he did say was adulatory rather than critical. No, he never criticized economics. Popper was a philosopher of science. Most of the applications of his ideas were in physics, and he was particularly interested in relativity theory, quantum mechanics, and all that. Insofar as he was interested in social science, it was the sort of grand social science. He wrote The Open Society, which had three great enemies-Plato, Hegel, and Marx. But he was not particularly interested in economics and said very little about it, and some of the things that he said about it generate more questions than answers.

Q. Then who is the first person to raise falsifiability as an issue in economics?

A. Terrence Hutchison, who at an amazingly early age wrote The Fundamental Postulates of Economic Theory. In this book, he criticized the assumption of perfect knowledge in most of economic theory and introduced Popper. Hutchison, who is still alive and in his early nineties, has remained ever since a great disciple of Popper and the application of Popper in economics. I have followed in his footsteps. He was sort of a mentor of mine.

Q. Did you study with him?

A. He was one of my Ph.D. supervisors at Columbia University. He visited Columbia. I had two supervisors: George Stigler and Terrence Hutchison. They made a wonderful contrast. But actually there was quite a lot of Popper in Stigler. So, in some ways, I owe a lot to both of them.

Q. In the current mainstream practice of economics, is falsifiability ignored entirely?

A. No. It would in a sense be easier if it were ignored altogether. Mainstream economics pays lip service to it; that is, it always says, "Oh, yes, of course, economic theory should be confronted by evidence. And, of course, if the evidence is contrary, if it seems to refute the theory, oh, yes, we certainly pay attention to that, and we must adjust the theory or even consider possibly a new theory." So, they do not in any way disagree with the implications of falsifiability. They preach it but they do not practice it. In other words, when confronted with contrary evidence of some beloved theory, they adjust the theory, or they minimize the evidence. Sometimes they even ignore the evidence. They do not look very hard at contrary evidence, preferring to confirm rather than to look for refuting evidence.

Q. Isn't that the point--that economic theory is often so malleable that it can be applied to any series of outcomes?

A. Yes. But it is false to believe that this is something unique to economics. It is pervasive in all of social science and, indeed, it is actually a much more telling point in sociology and political science. Because they do not produce such hard conclusions, their theories are even more difficult to refute than economic theories. But even in the natural sciences it is rare to find a crucial experiment that conclusively confirms or conclusively refutes. If people in a discipline take empirical evidence seriously, it piles up and eventually causes the theory to be overturned. But it is not the kind of thing where you wake up one morning and suddenly say, "By God, there is refuting evidence and now I am going to throw away the theory."

So, there is a very slow buildup, even in natural sciences. And sometimes the empirical evidence is more in the nature of big events. To give you a trivial example, the inflation and the stagflation of the 1970s did more to persuade economists that there was something wrong with Keynesian economics--that you needed supply-side policies and all that--than all the empirical evidence on the econometric studies against Keynesian economics. Sometimes you have to be hit over the head with a hammer before you give up a beloved theory.

Q. What examples can you name in which evidence has been overwhelming and yet economics has not abandoned a theory?

A. Rational expectations and the new classical macroeconomics. The implication that no government policy can possibly influence the real output, income, and employment of an economy has been refuted again and again. And the contrary evidence is indeed acknowledged to a considerable extent by those who are the leading spokespersons for this classical macroeconomics. Yet new classical macroeconomics is still taught in all the textbooks, and there are still many macroeconomists who go on fervently believing that new classical macroeconomics is based on a firm foundation and that people indeed hold expectations rationally. There is even the evidence in the stock markets. Stock markets are one of the best places to test the idea because, of all markets, they are the one that most motivates people to be well informed. Yet the stock market is littered with anomalies, with market bubbles, which are impossible to explain if all traders in the stock market hold expectations rationally.

Q. In your article in this issue of Challenge you make much of formal modeling and the evolution from the work of Kenneth Arrow and Gerard Debreu. Why do you believe that such formal mathematical modeling has taken such a firm hold? Why has the profession gone in that direction?

A. I will give you the standard answer, although I think it is a very difficult issue. This is a deep question about the history of ideas. How did we get like this? If I could answer that question--if I thought I could confidently answer it--I would certainly rush to print with the answer. I am not sure that I can confidently answer it. But a standard approach among people who have thought about this is that, sometime after World War II, economics began to model itself after hard science. It wanted to be the one social science that looked exactly like physics. This led to mathematization, mathematical modeling, formal modeling, and the resulting worship of technique and formal elegance.

But the odd thing is this explanation does not really wash because if you know anything about physics--and I am an amateur physicist--physics is not at all like that. Physics takes evidence very seriously but many physical theories are rather muddled and confused and inconclusive. They are by no means very elegant. The subject we economists really have been aping is mathematics. We have turned economics into a kind of social mathematics that employs words such as "price," "market," "commodity." It looks like economics, but when you read an article that uses such words, all the relationships are mathematical relationships; all the inferences are mathematically drawn; and no thought is given to whether these mathematical variables, concepts, functional relationships bear any resemblance to real-world observation. Deirdre McCloskey [see Challenge, January-February 1997], whose writings I do not otherwise like, has said quite rightly that economists look to the math department, not the physics department. That is absolutely true.

Q. Let us try to take it one step further, though. Why would these mathematical models, which after all are not very complex by a mathematician's standard, catch on? Does the modeling serve as some kind of fraternal initiation that bonds economists?

A. The math is not all that difficult, although it does create an entry barrier. Most people can learn the mathematics, but you do have to study it. It is just as if we wrote economics in French. It would be an effective entry barrier because some people find it rather difficult to learn foreign languages. You do not have to be terribly intelligent, but you do have to be patient, and spend a lot of time at it. So, it is an effective entry barrier.

The people who have been initiated now have a vested interest in taking the barrier seriously and paying attention to it and giving it high prestige. Otherwise, it would not serve as an entry barrier. So, after a while, they justify the entry barrier because they possess this elegant particular virtue or technique. After it is created, it justifies itself. I shall add one other thing to this: the enormous output of Ph.D.s in economics in the United States--almost a thousand every year! Many of them do not go into academic life, but all the estimates suggest that about 45 to 50 percent do seek employment in academic life. They have been taught by their teachers a particular kind of economics, and, of course, they disseminate the same economics in their teaching and also by publishing in journals.

Journal discussions that are expressed mathematically are much easier to evaluate inasmuch as you can see whether a person has really written down a consistent mathematical model. That is much easier to judge unambiguously than a piece of economic wisdom expressed in prose, which is very difficult. Two people can easily disagree on whether the prose work is valuable or not. And these journal articles, which number in the thousands--there are three hundred or more journals in the English language in economics, publishing two, three, or four issues a year--have to be refereed. Most of them are refereed by volunteers who have to pass judgment quickly (once you get on the list you read dozens of articles, sometimes in one week). It is all much easier once you share a community of standards, mathematical standards essentially. It still does not explain how it all got started. And I really do not understand it.

Q. Has the computer reinforced this tendency even more? After all, regressions are so easy to do now.

A. Insofar as it has created, curiously enough, a kind of econometrics that is no longer very interested in empirical evidence either but in econometrics theory. Econometrics has become almost as much a spectacle as economic theory, and it is a theory of a statistical kind. Yes, economists run regressions and all that. But that is not what the econometric journals are full of. They are full of fancy new statistical analyses, a much more sophisticated way of analyzing time series that are almost ends in themselves.

Q. You make some rather categorical statements that economists pay no attention to the real-world implications or applicability of their models. Are there any clear exceptions?

A. Of course. If you wish to express criticism that will have an impact on ten or twenty thousand practitioners throughout the world, you have to exaggerate slightly to make any point at all. If you modify too much, nothing is left of it.

Q. What was it about the Arrow-Debreu thesis that you single out in your article in this issue of Challenge that so influenced the profession?

A. It was extraordinarily sophisticated mathematically, and even though it was published forty-four years ago, it is an elegant article using game theory, which was then very new. It used game theory in a way that nobody had expected to prove the existence of general equilibrium. It was elegant, it was rigorous, and it appeared to solve a problem. Does general equilibrium actually exist? To be a little more technical, can multimarket equilibrium actually exist in an economy? And it seemed to establish this. After all, Walras had argued this eighty years earlier but had never been able to prove it in any satisfactory way. So it appeared a wonderful example of elegant quasi-mathematical proof, and it seemed to elevate economics almost immediately to a subject rather like mathematics, certainly applied mathematics.

Q. You argue that such models require highly simplified assumptions in order to work. And you point out that, if these assumptions are relaxed, the conclusions do not hold up. But do economists try to relax these assumptions and test their conclusions?

A. Yes. There are many economists who look at relaxing these assumptions. "Let us see if we relax the assumptions-will the answer still hold?" But this can become rather alarming because, again, all we really do is work a little harder at the mathematical modeling to get it right. I think this way of thinking about establishing economic relationships is misleading. I am not against modeling, nor am I even against mathematics. It can be a very useful handmaiden, but not as an end in itself. Not as the way we judge whether, for example, an article is worth reading or not, an argument is worth listening to or not. What I am trying to do is to alter the intellectual priority that economists assign to different kinds of economics. That is, they assign enormous prestige to any kind of economic theory that is mathematically expressed, but almost none to historical argument or a case study. This is a clever way of marshaling empirical evidence to prove a particular economic theory. That is what is wrong.

Q. What examples come to mind of real-world issues where this kind of thinking in economics has led to errors in public policy or errors in judgment?

A. We have not been very good at thinking about the transition problem in Eastern Europe because we have not been thinking about how market economies actually work and what is required to make markets function. So our advice to East European governments has been very wooden because you have to understand how markets have to be created and how property rights have to be established. We spend little time studying the institutional structures in which markets are imbedded and without which they cannot work. And so, we have given either no advice or bad or misleading advice to East European governments. The transition problem of Eastern Europe has made me and other economists extremely aware of how lopsided our approach is to markets and the market mechanism and to capitalism. Creating markets is not just Gerard Debreu's general equilibrium theory. That is not very helpful. Indeed, it is probably even misleading in thinking about these problems.

Q. Can similar statements be made about the Asian financial crisis at the moment or, let us say, the inequality of income in America?

A. The problem with the Asian tigers is partly the same: We no longer give much attention to developing economies. It was a big subject in economics in the 1950s and 1960s but has gone into an almost total decline. One of the causes for this decline is that it is not a good area to work in if you want to produce fancy, technically puzzling pieces of mathematical modeling. So many young economists find that it is just not an exciting area in which to work. You cannot get promoted publishing articles on development economics. They will admit that these are important real-world problems, but that is not the way to get ahead. It is not the way to build your career in an area that prizes technical elegance.

Q. Is your solution, then, more empiricism?

A. More empiricism, more history, more getting your nose dirty in data, surveying people, asking opinion, monitoring behavior--yes, all of that.

Q. In the United States there seems to be some movement toward more empirical experimentation anyway. Have you noticed this?

A. I think a move toward experimental economics is very hopeful. The amazing thing is the enormous resistance that it has encountered.

Q. What example can you name of that?

A. Talk to any of the leading experimental economists. They will tell you about the enormous resistance that their work has run into. It is very hard to push experimental economics because you are running against this grain of formal modeling. It has made forward progress in recent years, but I am amazed how difficult it has been to interest economists in it and how little of it is part of mainstream economics as taught in graduate school.

Q. Have we learned anything fundamental in, say, the past quarter-century of economics?

A. Sure. Particularly in macroeconomics, how to deal with inflation, how to deal with unemployment. I think we've learned an enormous amount, particularly about steering the economy in a macro sense. We certainly learned a lot from the deregulation and privatization movement. Although there is a lot more to learn, when I contrast what we now know about these things with the days when I was a graduate student in the 1950s, I think we have made forward strides. But to some extent we have also gone backward.

Q. In what specific sense have we gone backward?

A. In the sense that economics as a completely formalistic discipline has turned students off from pursuing it. I do not know about America, but in Europe economics unfortunately is attracting fewer students, whereas business studies and business management is attracting more students. It is not just that economics has become technical; it is that economics prizes technicalities above everything else and that is why I call it formalism. Formalism is the tendency to worship the form rather than the content of the argument. That is the kind of subject it has become. We care only about the form in which an economic theory or hypothesis is presented, and we care almost nothing about the actual content of the hypothesis.

Q. What are the major issues on which we have not made progress?

A. Markets and how they actually function; that is, how they adjust to match demand and supply. We in economics know a hell of a lot about equilibrium, but we really don't know how markets actually get to equilibrium.

Q. In your view, what can save economics?

A. I am very pessimistic about whether we can actually pull out of this. I think we have created a locomotive. This is the sociology of the economics profession. We have created a monster that is very difficult to stop.

Q. Could real-world empirical facts or a severe economic cataclysm change it?

A. That would certainly change it, but I do not see that around the corner. Perhaps I am too pessimistic, and it is very depressing to stay there. There does not seem to me to be any way out.

~~~~~~~~
Interview with Mark Blaug

MARK BLAUG is professor emeritus, University of London, and visiting professor of economics, University of Exeter.

Section 5, shortly after (1):
I take the major goal of generative grammar to be a comprehension of the connection between the sense experiences as reflections of the language faculty, "by the use of a minimum of primary concepts and relations." What is meant by the language faculty here is (ultimately) UG.

Re. the last sentence above, what I had in mind is something like the following.

The language faculty cannot be equated simply to UG, since it is not obvious, and it is perhaps incorrect to say, that our linguistic intuistions are reflections of UG in any direct way, even if we are, or could be, concerned only with the intuitions that clearly and strictly reflect the language faculty. However, certain aspects of our linguistic intuitions are reflections of the grammar we have (internalized). The grammar we have, i.e., the grammar of a particular language, is in turn a 'reflection' of UG, by hypothesis. It is in this sense that I made the statement quoted immediately above.

Now, how is the grammar of a particular language a 'reflection' of UG? My reasoning perhaps was that the grammar of a particular language is a function of UG (the initial state) plus a particular parameter setting in some form or another, on the basis of primary linguistic data. But if the linguistic variations are restricted to (the functional domain of) the Lexicon―the idea that seems widely adopted in the generative tradition now―what constitutes the generative procedure must be invariant across languages, with the possible exception of the linear precedence relation of some sort―the so-called head parameter (i.e., the head-initial vs. head-final parameter). The difference between Japanese and English, for example, would, under this view, be reduced to the presence and the absence of certain items (including features as well as categories) in their Lexicons, leaving aside the difference that is arguably due to the parameter just alluded to).

It is argued in a series of works by N. Fukui, including Fukui 1986 and Fukui & Sakai 2003 (in Lingua), that Japanese does not have active functional categories in its Lexicon, the thesis I adopt and have been pursuing myself. Under this view, then, the generative procedure in Japanese is no different from that in English with respect to what grammatical operations are in principle available and how they apply. Certain operations simply do not take place due to the absence of what would trigger their applications.

(1)　c.　Repeatability (or reproducibility) is attained to the extent that an outcome of an experiment that confirms the prediction is repeated (or reproduced) (over a series of instances of the same experiement --added here, HH).

(1)　b.　An outcome β of an experiment confirms a prediction γ iff β is in accordance with γ; β disconfirms γ otherwise.

We should certainly address what is meant by an 'experiment' here. The failure to articulate the basic structure an experiment in generative grammar (at least in the domain where we deal with 'meanings') seems to be a serious shortcoming of the paper when its stated goal is to illustrate how generative grammar is, or can be, an empirical science. A more satisfactory work would therefore contain (i) explicit characterizations, as it is deemed appropriate and relevant, of concepts such as observations, generalization, theory/hypothesis/analysis, predictions, falsifiability, and experimental designs and (ii) their concrete illustration on the basis of specific empirical materials.

(1)　a.　A proposal α is falsifiable iff α makes a prediction that can be confirmed or disconfirmed.

What need to be added here, perhaps, is "beyond the empirical generalizations that have led to the proposal itself, " i.e., "a prediction (beyond the empirical generalizations that have led to the proposal itself)." That is to say, the proposal in question, insofar as it has an empirical content to it, must have been advanced on the basis of a (set of) generalization(s). The claim that the generalization(s) in question is/are valid, by itself, does not make a prediction, however. We make a prediction only if (i) the generalization(s) in question (call this/them G1) is/are stated in terms of theoretical concepts, and (ii) some other generalization(s) (call this/them G2) is/are stated (at least in part) in terms of the same concepts, so that clear correlations between G1 and G2 are deduced from the theoretical statements under discussion.

Since our sense experiences, such as introspective judgments about a given sentence in a given language, most likely reflect more than the language faculty proper (such as the 'pragmatic' knowledge about the usefulness of the given utterance, the knowledge about the real world, etc. --added here, HH), such a task necessarily involves hypotheses about the nature of the relevant sense experiences, in particular, hypotheses as to which aspects of our sense experiences under discussion are reflections of our grammar (proper --added here, HH), and in what theoretical terms they are to be expressed. At a particular stage of theory construction, a given factor can be reasonably considered grammatical in nature only if it can be expressed in terms of concepts postulated within the grammatical theory being developed.² Every concept and relation postulated in the theory, in turn, must be tightly related to the native speaker's linguistic intuitions---often quite indirectly---as reflections of his/her grammar. It is in fact the tight connection between (i) theoretical concepts and relations on the one hand and (ii) the speaker's linguistic intuitions on the other that makes it possible to put forth definite predictions about the latter that are formulated in terms of the former, thereby making the proposed theory/hypothesis falsifiable."

²  Chomsky (1955/1975: 61) thus states that "a field of investigation cannot be clearly delimited in advance of the theory dealing with the subject matter; in fact, it is one of the functions of a theory to give such a precise delimitation.

What I meant to say, but did not say, was that such tight connection is conspicuously missing in much of the works in the literature, making it difficult to argue, on the basis of the general practice in the field, that generative grammar is being practiced as an empirical science.

If "the aim of science is, on the one hand, a comprehension, as complete as possible, of the connection between the sense experiences in their totality, and, on the other hand, the accomplishment of this aim by the use of a minimum of primary concepts and relations,"¹ as Einstein (1936: 293) puts it, and if generative grammar is that part of science whose aim consists of a comprehension of the connection between the sense experiences as reflections of the language faculty, it follows that one of our tasks is to identify what the relevant sense experiences are; cf. Chomsky 1955/1975: 37.

¹  The emphases are as in the original.

Some comments and related quotations will be posted as "daughters" to this posting.

Suppose that we want to demonstrate (i), rather than simply assuming it.

(i) The language faculty exists as an autonomous system, i.e., as a system whose properties, and hence its workings, are not affected at all by pragmatics and other considerations outside the language faculty proper.

We should perhaps be cautious as to what linguistic phenomena we want to deal with since it is most likely the case that not every linguistic phenomenon will lead us to a successful demonstation of (i).

Also relevant in this regard is the following passage from Popper 1959.

***
There is only one way to make sure of the validity of a chain of logical reasoning. This is to put it in the form in which it is most easily testable: we break it up into many small steps, each easy to check by anybody who has learnt the mathematical or logical technique of transforming sentences. If after this anybody still raises doubts then we can only beg him to point out an error in the steps of the proof, or to think the matter over again. In the case of the empirical sciences, the situation is much the same. Any empirical scientific statement can be presented (by describing experimental arrangements, etc.) in such a way that anyone who has learnt the relevant technique can test it.
(Popper 1959: 99, the emphasis by HH)
***

We cannot emphasize more the significance of what is intended above. When we make an empirical claim/hypothesis about the nature of the language faculty, we should present it, in principle, with a design of (a) experiment(s) by following which we/one can test whether it makes (a) correct prediction(s).

It is equally important to stress that it would not be enough just to agree with and endorse the above statement. What counts is whether one puts it into practice. If you agree with Popper's remark above, you might want to ask whether or not the most recent work that you read (fairly) carefully indeed does that. If the answer is yes, you should be able to conduct, or at least have a fairly clear idea about how to conduct, the relevant experiment(s) to test the predictions made in that work.

(My Lingua paper does not quite provide explicitly how the relevant experiment can be done. I suspect that it should not be very difficult to 'design' the relevant experiments, on the basis of the discussion there. But when we try to actually do it, we might find it more difficult than we expected.)

  A grammar determined by a linguistic theory (given data) constitutes a hypothesis concerning the speaker-hearer's knowledge of his language and is to be confirmed in terms of empirical evidence drawn, ultimately, from investigation of the linguistic intuitions of the language-user (which might, in principle, be analyzed in terms of operational tests; cf. [sections]12-15, below). The general theory, now regarded as an explanatory theory, is likewise to be understood as a psychological theory that attempts to characterize the innate human 'language faculty," and that can be tested in terms of its consequences in particular languages. (Chomsky 1955/1975: 37)

Chomsky, N., 1955/1975. The Logical Structure of Linguistic Theory. New York: Plenum Press.

  The aim of science is, on the one hand, a comprehension, as complete as possible, of the connection between the sense experiences in their totality, and, on the other hand, the accomplishment of this aim by the use of a minimum of primary concepts and relations. (Seeking, as far as possible, logical unity in the world picture, i.e., paucity in logical elements.) ... Thus the story goes on until we have arrived at a system of the greatest conceivable unity, and of the greatest poverty of concepts of the logical foundations, which is still compatible with the observations made by our senses. ... It is a matter of faith that nature—as she is perceptible to our five senses—takes the character of such well-formulated puzzle. The successes reaped up to now by science do, it is true, give a certain encouragement for this faith. (Einstein 1936:293-295)

The table of content of Language Faculty Science as of February 2014, is included in the handout. Appendix III, however, will most likely not be included as such.

I gave a talk at USC about a month ago.

Its abstract was:

Title: Language Faculty Science as an Exact Science

"It is generally agreed that it is not possible outside physics and its closely related fields to deduce definite predictions and expect them to be borne out experimentally. I argue in my on-going work that it is indeed possible. What I propose is a consequence of (i) taking as the object of inquiry the language faculty (I-language in the terms of Chomsky 1986) rather than language as an external or externalized object (E-language in the terms of Chomsky 1986), and (ii) adopting the methodological naturalist approach to the study of the language faculty―which maintains that we should approach our subject matter just as researchers in a natural science approach their subject matters (Chomsky 1986, 1993, and 1995, among other places).

In this talk, I provide a conceptual and methodological basis for the claim that language faculty science as an exact science is possible. I address issues such as what can count as facts in language faculty science, how we deduce definite and categorical predictions, and how we can expect our predictions to be supported experimentally. Empirical illustration of the viability of the proposed methodology, on the basis of actual experimental results, will not be covered in this talk―there simply is not enough time for that―although I might try to give a very rough picture of the general experimental design and actual experimental results if there is time for that."

And its handout is available here

As noted in the abstract, the handout does not cover Experimental results addressed in the book. Nor does it address how Experiments are conducted. See the table of contents of the book provided in the handout. When the book is out, it will be accompanied by a website where the details of all the Experiments discussed in the book are provided.

The title of the book that I am working on now is Language Faculty Science: how it becomes an exact science―a proposal and illustration.

The draft of Chapter 1 starts as follows, disregarding the font formatting and the footnotes.

***
1.1. The Goal
It is an indubitable fact that, once they have reached a certain maturational stage, the members of the human species, barring any serious impairment, are able to produce and comprehend sentences of the language that they are exposed to. One of the most fundamental working hypotheses adopted in the present work is that there exists the language faculty underlying this ability of ours. The existence of the language faculty has been assumed by many and it has been the point of departure of Chomsky's research program although there is also a contrary view accepted by many, as indicated in N. J. Enfield's recent (2010) review article in Science "Without Social Contexts?." Chomsky has maintained over the years that we should approach the language faculty just as natural scientists approach their subject matters. It has, however, remained unclear how hypotheses about the language faculty can be put to rigorous empirical test. The problem manifests itself most acutely once we consider how the hypothetico-deductive method―the most commonly acknowledged hypothesis-testing "method" in a mature science such as physics―can be applied to language faculty science.

The present work articulates how predictions about the language faculty can be deduced from our hypotheses, how such predictions can be tested against experimental results, what should count as the relevant data in language faculty science, whether and how such data can be of a categorical nature, what kind of experimental design would maximize the significance of the experimental results, how we can make various aspects of experimental devices maximally effective, how rigorous a match we could expect between the prediction and the experimental results, along with many other related issues. It attempts to pursue and defend the thesis that it is possible to investigate the language faculty by applying the hypothetico-deductive method, i.e., by rigorously comparing the predictions deduced by our hypotheses with experimental results and observations. Insofar as we can carry this out successfully, with compelling empirical demonstration, that will constitute support for the existence of the language faculty.

1.2. Reproducibility and measurability

In physics, what is predicted and compared with experimental results (or observations) is something that is measurable (ultimately in terms of temporal and spatial values). The measurability of the relevant "data" is what makes it possible to compare a prediction with an experimental result and also to determine how much reproducibility there is to the experimental results and observations. Given that reproducibility and measurability are two prerequisites for effectively adopting the hypothetico-deductive method, it follows that predictions in language faculty science must be about something reproducible and measurable as long as we adopt the hypothetico-deductive method.

One may wonder whether it is reasonable to apply the hypothetico-deductive method to research concerned with the language faculty. After all, it is commonly understood that the "predictions" in fields outside the extremely limited domains of inquiry including physics are about differences and tendencies and that it is not possible to deduce point-value predictions in such fields. One may thus object that physics is not the right field for us to turn to as a model of our research program. One may also point out that the hypothetico-deductive method is not the only method adopted even in physical sciences. My response, briefly put, is: if it is possible to get to know something by following the hypothetico-deductive method, why would one want to adopt a less rigorous method?
***

It continues as:
(The formatting and the footnotes are not provided below.)

***
1.3. Evidence in language faculty science

There are various types of evidence that we can in principle bring to bear on the validity of our hypotheses. Some are "experimental" while others are not. Whatever type of evidence one wishes to consider, it has to be articulated how the predicted "value" can be deduced from a set of hypotheses, how a particular experimental result or an observation can be understood as a reflection of properties of the language faculty, and finally, how we can rigorously compare the prediction and the experimental result or the observation. Without minimally satisfactory answers to such questions, it remains unclear what significance can be assigned to the experimental result or the observation.

Given the assumption that the language faculty underlies our ability to relate a sequence of sounds/signs to a "meaning," it makes sense to ask informants, including ourselves, about possible correspondences between sounds/signs and "meanings." However, in light of the fact that the informant judgments, especially when "meanings" are involved, have been known to be extremely slippery, one should naturally wonder how we can justify the use of informants' introspective judgments as crucial evidence, let alone the use of the researcher's own judgments. The present work proposes a specific way to make informant judgments qualify as evidence in language faculty science, as something measurable and reproducible.

1.4. Three heuristics

The proposal is embedded in a larger methodological proposal about how we can deduce definite predictions and test them experimentally in accordance with the research heuristics in (1).

(1) a. Secure testability.
b. Maximize testability.
c. Maximize the significance of the experimental result.

I take it for granted that, regardless of the object of inquiry, one should like to adopt the research heuristics in (1) if it is at all possible to do so. The methodology proposed in the present work is a consequence of having the language faculty as the object of inquiry and adopting (1). The present work thus shares its goal with Chomsky's research program and proposes a means to achieve the goal by pursuing rigorous testability.

1.5. Deducing a categorical prediction
1.5.1. The model of the Computational System

In order to deduce a categorical prediction about informant judgments, it is necessary for (part of) the language faculty to have a categorical nature. Chomsky's (1993) model of the Computational System indeed has such a categorical nature; it is a structure-building system such that (i) its input is a set of items taken from the mental lexicon, (ii) its only structure-building operation takes two objects and forms one, and (iii) it yields two output representations, called an LF and a PF representation. The LF representation is a hierarchical organization of abstract objects and is the formal basis of the "meaning," and the PF representation underlies the "sound/sign." Given a set of items taken from the mental lexicon, the Computational System either generates or fails to generate an LF-PF pair of representations. It is in this sense that Chomsky's (1993) model of the Computational System is categorical and can serve as a basis for deducing a definite prediction.

1.5.2. The model of judgment-making

Assuming the Computational System, which has this categorical nature, however, is not sufficient for deducing a categorical prediction about the informant judgments. It must be articulated how the Computational System is involved in the act of judgment-making; otherwise, it remains unclear how the informant judgment could be revealing about the properties of the Computational System. Such a model of judgment-making has been put forth in Ueyama 2010. Ueyama (2010) proposes that the informant, upon hearing a presented sentence α, selects a set of lexical items N, in part on the basis of her/his past linguistic experience, and once the N is selected and enters the Computational System as its input, the system generates or does not generate an LF-PF pair, as noted above. We thus have a further basis for deducing a definite prediction about the informant judgment.
***

1.6. The Task of the informant
1.6.1. Problems with dealing with 'simple' acceptability

Viewed this way, asking the informant about the acceptability of a given sentence amounts to asking, though obviously not in these terms, whether or not s/he can come up with a set of items N from the mental lexicon that would yield an LF-PF pair whose PF representation is non-distinct from the presented sentence α. For the ease of exposition, I shall henceforth refer to an LF-PF pair whose PF representation is non-distinct from the presented sentence α simply as the LF-PF.
The purpose of asking the informant to report his/her judgment on the acceptability of α is to find out whether there is an LF-PF pair corresponding to α. In order for the informant judgment to be significant, it must be a reflection of properties of the Computational System. And that means that we must make sure that (i) the "Not acceptable" judgment is due to some grammatical condition not being satisfied in the course of the "attempted" derivation―resulting in the failure to yield the LF-PF―and (ii) the "Acceptable" judgment is due to the existence of an N that would result in the LF-PF. It is possible, however, that the informant judges α to be unacceptable because s/he fails to come up with an N because of parsing difficulty. Likewise, it is also possible that the informant judges α to be acceptable because α is 'intelligible' (i.e., understandable) in some way even if there is no N that would result in the LF-PF. And, there does not seem to be a principled means to exclude such possibilities. If the reported "Unacceptable" judgment is due to parsing difficulty, the unacceptability in question would not be revealing about properties of the Computational System. The same holds if the reported "Acceptable" judgment is due to the 'intelligibility' of α. We are thus led to conclude, on the basis of these conceptual considerations, that it is not clear, in principle, what significance or how much significance can be assigned to the informant judgment on simple (un)acceptability of the presented sentence.

1.6.2. Solutions: Invoking a dependency interpretation

One way to maximize testability (see (1b)) is to consider hypotheses whose consequences are in principle testable in any language. Given the diverse differences among languages with respect to lexical items, it seems reasonable, for the purpose of maximizing testability, to turn our attention to some structural (i.e., hierarchical) property at LF, abstracting away from idiosyncratic lexical properties as much as possible. As noted above, the only structure-building operation assumed in the model of Computational System adopted here is one that combines two objects to form one, called Merge. Suppose we hypothesize that a certain intuition pertaining to two linguistic expressions a and b arises in the mind of the informant only if what corresponds to a and what corresponds to b at LF are in a structural relation directly definable in terms of the application of Merge such as: something stands in a structural relation R with something else if and only if the former Merges with something that contains the latter. To facilitate the exposition, let us refer to the intuition in question as γ(a, b).

We can try to avoid the problems noted in the preceding section by checking the informant judgment on the acceptability of α where the interpretation of α includes γ(a, b). What the informant is now being asked, though surely not in these terms, is whether s/he can come up with a set of lexical items N that would serve as input to the Computational System so as to yield an LF-PF pair such that (i) the structural condition for γ(a, b) would be satisfied in the LF representation, and (ii) the PF representation would be non-distinct from α. For the ease of exposition, such an LF-PF pair shall henceforth be referred to simply as the intended LF-PF.

We can greatly reduce the possibility that the informant's "Unacceptable" judgment is due to the parsing difficulty, by having the informant judge the same surface string as the presented sentence α, but crucially not involving γ(a, b). Let us call such a surface string as α'. The informant's "Unacceptable" judgment on α under γ(a, b) should not be due to parsing difficulty if the same informant has accepted α', identical to α. We can also have the same informant judge a surface string α'' that differs from α as follows: α'' is comparable to α in terms of its structural complexity; but there is hypothesized to be an LF representation corresponding to α'' (but not corresponding to α), in which the structural condition for γ(a, b) is satisfied. To the extent that the informant judges α'' to be acceptable under γ(a, b), we can reasonably conclude that the unacceptability of α under γ(a, b) reported by the same informant is due to the failure of the hypothesized condition for γ(a, b) to be satisfied in the LF representation corresponding to α.

Let us further note that another way to substantially reduce the possibility that the informant's "Acceptable" judgment is based on the 'intelligibility' of the presented sentence α is to have the same informant judge a sentence α''' that minimally differs from α with regard to whether the structural condition for γ(a, b) is satisfied; the condition is hypothesized to be satisfied in an LF representation corresponding to α, but not α'''. Suppose that the informant has judged α under γ(a, b) to be acceptable. Insofar as the same informant judges α''' to be unacceptable under γ(a, b), it seems reasonable to assume, on the basis of the informant's rejection of α''', that the informant understands what is meant by γ(a, b), and that her/his "Acceptable" judgment on α under γ(a, b) is indeed based on the LF representation corresponding to α satisfying the structural condition for γ(a, b), rather than being due to the mere 'intelligibility' of α. The relevant points will be elaborated and will be illustrated with a number of concrete examples in subsequent chapters, where we will discuss not only the structural condition for γ(a, b) but also the lexical condition(s).

In sum, by invoking γ(a, b), we can considerably reduce the problems noted in the preceding section concerning the significance of the experimental result. The general design of our on-line experiments is in accordance with the above considerations. In a later chapter, we will address in some depth how the specific aspects of the experiment are 'constructed' and how the experimental results are interpreted in accordance with the heuristics in (1).

1.7. The pf-LF correspondences

For the purpose of deducing categorical predictions about the informant judgment, we also need language-particular hypotheses regarding the correspondences between (i) the schematic linear sequence of linguistic expressions in the language under discussion and (ii) the schematic LF (hence, hierarchical) representation(s). Such hypotheses shall be called pf-LF correspondences and will be discussed in some depth in later chapters, in relation to the so-called scrambling construction in Japanese.

1.8. Two more issues to clarify
1.8.1. Reproducibility and intermediate judgments

There are two more interrelated issues I would like to clarify in this introductory exposition; one is how to interpret "intermediate judgments," (i.e., "Not completely unacceptable but not fully acceptable"), and the other is what should count as reproducibility of our experimental results. As in any scientific research program, identifying and working with reproducible phenomena is a minimal requirement for progress. If the informant judgments are not reproducible, they do not (yet) constitute a piece of data to account for in language faculty science. Let us thus first turn to how we should characterize a reproducible phenomenon in language faculty science.

1.8.2. Schemata

First of all, it is not the informant judgment on particular sentences that would count as evidence. If we deduce a prediction that sentences conforming to a certain schema do not give rise to a certain interpretation, such an interpretation should be unavailable in any sentence conforming to that schema, no matter what lexical or pragmatic adjustments and alterations might be made on what is left unspecified in the schema. Consider the schemata in (2a) and (2b).

(2) a. NP1 Verb [ … NP2 …]
b. [ … NP2 …] Verb NP1

The only thing specified in (2) is the positions of the two NPs, NP1 and NP2. To maximize the generality of each schema, it must be understood, for a sentence that instantiates each schema, that (i) the Verb can be inflected, (ii) any materials can in principle appear in the "…" parts, (iii) verbal or sentential modifiers can appear, and (iv) it can be embedded in a larger sentence, etc. Sentences such as John hates his book (with John=NP1 and his=NP2) and every boy was looking for a book that he had bought in New York (with every boy=NP1 and he=NP2) are among the examples conforming to (2a).

Secondly, the informant judgments on a single schema alone do not constitute a "fact" in language faculty science. I maintain that we have a "fact" in language faculty science only if we have obtained informant judgments on a set of schemata in accordance with our predictions. In light of the preceding considerations, a "fact" in language faculty science must minimally involve three schemata. Sentences conforming to one of the three types of schemata are predicted to be unacceptable. Such a schema shall be called a *Schema (which can be read as "star schema") and sentences conforming to it *Examples (which can be read as "star examples"). For any *Example, there should be no N (a set of lexical items) that would result in an LF representation in which the condition for γ(a, b) would be satisfied. It should thus be predicted that any *Example should be judged completely unacceptable under γ(a, b).

Corresponding to the *Schema, there are two types of okSchemata. One type is identical in form to the *Schema, but without the specification of γ(a, b); the other type corresponds to an LF representation in which the condition for γ(a, b) is satisfied. Corresponding to okExamples conforming to the latter type of okSchema, there must be a set of lexical items N that would yield the intended LF-PF.

Provided in (3) is an instance of a set of three schemata.

(3) a. okSchema
NP1 Verb [ … NP2 …], with γ(NP1, NP2)
b. *Schema
[ … NP2 …] Verb NP1, with γ(NP1, NP2)
c. okSchema
[ … NP2 …] Verb NP1, without γ(NP1, NP2)

In order to obtain the predicted schematic asymmetry as indicated in (3), we must be able to deduce from our hypotheses that the structural condition for the intuition γ(NP1, NP2) is not satisfied in any of the LF representations corresponding to (3b) but it is satisfied in at least one LF representation corresponding to (3a). For this deduction, we must have a hypothesis about the condition on γ(NP1, NP2). In accordance with the "Maximize testability" heuristic in (1b), we should consider γ(NP1, NP2) that is based on the structural relation that is most readily discernable in any language, i.e., one that can be defined most directly in terms of the only structure-building operation assumed in the model of the Computational System adopted here. We also need hypotheses regarding the correspondences between (i) the schematic linear sequence of linguistic expressions in the language under discussion and (ii) the schematic LF (hence, hierarchical) representation(s), i.e., pf-LF correspondences, as pointed out earlier.

If the dependency interpretation holding between every player and his (i.e., γ(every player, his)) is an instance of γ(a, b) under discussion, the English examples in (4a, b, c) can instantiate the schemata in (3a, b, c), respectively.

(4) a. okExample
Every player praised his coach. (with γ(every player, his))
b. *Example
His coach praised every player. (with γ(every player, his))
c. okExample
His coach praised every player.

1.8.3. The fundamental asymmetry
The existence of an N corresponding to the presented sentence α that would result in the intended LF-PF does not necessarily mean that the informant will judge α to be acceptable. As noted above, the parsing difficulty might result in the informant's failure to come up with such an N; it is also possible that the informant reports an "intermediate acceptability" due to the (extreme) unnaturalness of the interpretation of α, quite independently from the availability of γ(a, b). Hence, the prediction in question, which we shall call an okSchema-based prediction, is that okExamples conforming to it are not necessarily completely unacceptable under γ(a, b), while a *Schema-based prediction is that any *Example should be judged completely unacceptable under γ(a, b). The recognition of this asymmetry between the two types of predictions is one of the keys to language faculty science as an exact science.

1.8.4. Intermediate judgments
We surely would like the informant judgments on the okExamples much 'better than' "just barely acceptable" or "slightly more acceptable than" the completely unacceptable *Examples. But the fundamental asymmetry that must be recognized is that *Examples are predicted to be completely unacceptable under γ(a, b) while okExamples are not. From this recognition follows how the "intermediate acceptability" is to be interpreted; "intermediate acceptability" reported by the informant must mean that there is an N that would result in the intended LF-PF. Its less-than-perfect status must be due to extra-grammatical factors for the following reason: If the condition for γ(a, b) were not satisfied in any LF representation corresponding to it, the presented sentence α should be judged to be completely unacceptable according to the *Schema-based prediction, provided that the informant clearly understands what is meant by γ(a, b) under discussion.

1.8.5. Reproducibility
Reproducibility in language faculty science should ultimately be across-language reproducibility because language faculty science is concerned with the properties that are invariant among the members of the human species. In this sense, it must be understood clearly that a given experiment in language faculty science, which necessarily deals with (a) specific language(s), should be dealing with a universal hypothesis. Hence the reproducible result in one language should in principle be replicated in any other language. Across-language reproducibility, however, can be meaningfully pursued only if we have attained within-language reproducibility, i.e., across-informant reproducibility within a language. Across-informant reproducibility in turn can be meaningfully pursued only if we have attained within-informant reproducibility, which consists of across-example reproducibility and across-occasion reproducibility. In the absence of within-informant reproducibility, with reproducibility as characterized above in terms of confirmed predicted schematic asymmetries, it would be unclear what significance we could assign to an experimental result, no matter what statistical method might be employed to analyze the data and how massive the relevant data might be.

1.9. Summary

How we can deduce a definite and categorical prediction about the informant judgment is crucially tied to the recognition of the fundamental asymmetry between the two types of predictions. For the reasons noted above, it is the informant judgments on (examples conforming to) a *Schema that we can expect to be categorical. The measurability and the reproducibility that we can aspire to obtain in language faculty science are thus about complete unacceptability and the lack thereof. It is such conceptual articulation as given above that leads to the conclusion that the study of the language faculty can become an exact science. When it comes to how we can actually obtain categorical judgments from informants, more issues need to be addressed, including the effectiveness of various experimental devices, such as the instructions given to the informants, as will be discussed in later chapters.

The present work puts forth the following theses: The informant judgments should count as a "fact" in language faculty science only if they constitute a confirmed predicted schematic asymmetry. A confirmed predicted schematic asymmetry obtains if and only if a *Schema-based prediction has survived a rigorous attempt of disconfirmation and it is accompanied by corroboration of the corresponding okSchema-based predictions. A *Schema-based prediction is that informants judge any sentence conforming to a *Schema to be completely unacceptable under γ(a, b) (a specified interpretive relation holding between two expressions a and b). An okSchema-based prediction is that informants judge sentences conforming to an okSchema to be acceptable under γ(a, b), to varying degrees. There is a fundamental asymmetry between a *Schema-based prediction and an okSchema-based prediction. The former can be disconfirmed but cannot be confirmed while the latter can be confirmed but cannot be disconfirmed. Testability is brought about most crucially by a*Schema-based prediction. In accordance with the "Secure testability" and the "Maximize testability" heuristics in (1a, b), we should therefore pursue hypotheses that would give rise to *Schema-based predictions. Furthermore, when modifying hypotheses, we should try to retain the existing *Schema-based prediction(s) or, better yet, obtain new *Schema-based predictions, again in line with the heuristic in (1a. b). In the subsequent chapters, I will elaborate on each of these points, and will illustrate them by making reference to results of a number of on-line experiments.

Unless we start accumulating results in language faculty science based on research that rigorously pursues testability, the research program initiated by Chomsky in the mid-1950s will most likely remain to be regarded as a metaphysical speculation, at least by those outside the field. The present work is an attempt to articulate how it is possible to pursue language faculty science as an exact science. It provides a conceptual basis for how that is possible in principle and empirical illustration of how that has actually been done.

1.10. Outline of the book

What follows is the outline of the rest of the book. Chapter 2 deals with how we deduce testable predictions in language faculty science. The main methodological claim of the present work is that it is possible to pursue language faculty science by adopting the three heuristics in (1), repeated here.

(1) a. Secure testability.
b. Maximize testability.
c. Maximize the significance of the experimental result.

It follows from the heuristics in (1a, b) that we should try to deduce definite and testable predictions from our hypotheses. This leads us to adopt Chomsky's (1993) model of the Computational System of the language faculty because of its categorical nature. We will also be led to adopt Ueyama's (2010) model of judgment-making, which makes it possible to relate explicitly the informant judgment to hypothesized properties of the Computational System. It will be argued that, given the model of the Computational System and the model of judgment-making we adopt, we can deduce categorical predictions about the complete unacceptability of sentences of a certain schematic form and the acceptability (to varying degrees) of sentences that are minimally different from the former, under a specified interpretation. Among the issues to be addressed is how predictions are deduced in language faculty science. Language faculty science is concerned with the universal properties of the language faculty but individual experiments necessarily deal with particular languages; it thus follows that predictions in language faculty science are deduced at least by two types of hypotheses, universal and language-particular. In order to make a testable prediction about properties of the Computational System, we need a hypothesis that relates hypothesized properties of the Computational System to what we can observe. Statements that relate theoretical concepts to informant intuitions are called bridging statements and, as noted earlier, they play a crucial role in deducing testable predictions in language faculty science.

Chapter 3 deals with how we can maximize testability. The chapter proposes that the evidence for or against hypotheses about properties of the language faculty must be (based on) a confirmed predicted schematic asymmetry, which obtains if and only if experimental results are reproduced in harmony with the predicted schematic asymmetry, according to which sentences conforming to one type of Schema (*Schema) are always judged to be completely unacceptable under a specified interpretation while those conforming to the other types of Schema (okSchema) are not necessarily judged to be completely unacceptable.

We can try to maximize testability by focusing on hypotheses whose applicability is, in principle, universal. The c-command relation is the most basic structural relation directly definable in terms of the application of Merge (the only structure-building operation assumed in the Computational System), and hence we expect its effects to be "detectable" in any language. We can therefore try to maximize testability by focusing on intuitions that are (hypothesized to be) necessarily based on the c-command relation and pursue the relevant c-command-based hypotheses and their consequences, in line with the spirit of Reinhart's (1983: chapter 7) conjecture.

Among the topics to be deal with in this chapter are:

Confirmed predicted schematic asymmetries
The fundamental asymmetry between two types of predictions
*Schema-based predictions
LF c-command-based intuitions

Chapter 4 addresses how to maximize the significance of the experimental result. We would like the experimental result to be maximally significant with regard to the validity of the hypotheses that have given rise to the prediction(s) in question. If the result is in harmony with the prediction (i.e., in harmony with the predicted schematic asymmetry), we want to maximize the likelihood that it provides support for the hypotheses in question. If the result is not in harmony with the prediction, on the other hand, we want to make sure as much as possible that it indeed means that at least one of the hypotheses in question is at fault, in line with the "Maximize our chances of learning from errors" heuristic, which is a specific instance of the "Maximize the significance of the experimental result" heuristic.

There are basically two ways in which we can maximize the significance of the experimental result. One is by invoking a particular dependency interpretation whose availability is (hypothesized to be) necessarily based on a c-command relation at LF. This would significantly reduce the possibility that the informant's "Unacceptable" judgment is due to a parsing problem. In this chapter, I will articulate why it is necessary to invoke a certain dependency interpretation in our experiment in order to maximize the significance of the experimental result, at least at the initial stage of language faculty science. The other way to maximize the significance of the experimental result is by trying to maximize the effectiveness of the various experimental devices as well as the hypotheses that yield the relevant predictions.

The discussion in this chapter includes: (i) from what hypotheses and assumptions we deduce definite predictions, (ii) how the experimental results get affected not only by the choice of the relevant hypotheses and assumptions but also by the instructions given to the informants and the quality of the informants. It is for the purpose of designing experiments, interpreting the experimental results, and maximizing the significance of the experimental results that a synthesis is called for of all the considerations addressed in this work. It is in the context of such a synthesis that we can see most clearly the crucial role played by the heuristics in (1) and truly appreciate the abstract nature of our experiments despite the surface appearance that our experiments deal with specific sentences of a specific language.

As noted, although a given experiment necessarily deals with (a) particular language(s), what is really at stake is the validity of universal hypotheses. In order for the experimental result to be significant in language faculty science, it is therefore imperative that we have established as clearly and rigorously as possible that the language-particular hypotheses we have invoked in deducing the prediction in question have contributed to an independent confirmed predicted schematic asymmetry in "preliminary experiments." In order to obtain a rigorous and categorical experimental result in accordance with the prediction (i.e., the predicted schematic asymmetry), informant-classification or informant-calibration is necessary by means of "preliminary experiments." It cannot be emphasized more that such informant-calibration and informant-classification are for the purpose of maximizing the effectiveness of the experimental result, which in turn is for the purpose of maximizing the significance of the experimental device. The main purpose of this is to leave as little room as possible for making "excuses" for failing to obtain an experimental result in accordance with our prediction. Those researchers who do not pursue rigorous testability and pursue "compatibility-based research" seem to think that informant-calibration and informant-classification are for the purpose of obtaining the experimental results that we want. But the real purpose of informant-calibration and informant-grouping is to maximize the significance of the experimental result, and ultimately to maximize testability.

The topics that will be covered in this chapter include:

Invoking a dependency interpretation γ(a, b)
Universal and language-particular hypotheses
Bridging statements
Assumptions about the effects of lexical choices not included above
Maximizing the effectiveness of the experimental devices
The lexical choice
Instructions to the informants
Informant resourcefulness

Chapter 5 deals with various aspects of experiments in language faculty science. The sentences (the *Examples and the corresponding okExamples) used in our experiment are constructed on the basis of a predicted schematic asymmetry, which consists of a *Schema and the corresponding okSchemata, constructed on the basis of the relevant universal and language-particular hypotheses and the bridging statement(s). This chapter addresses what formal and non-formal considerations enter the construction of the *Examples and the okExamples in the experiment. The chapter also illustrates the general experimental design in accordance with the proposed characterization of language faculty science.

The chapter addresses the following topics:

Predicted Schematic Asymmetries
Constructing *Schemata and okSchemata
Constructing *Examples and okExamples
Experimental Design

Chapter 6 illustrates the proposed methodology in relation to the so-called scrambling construction in Japanese, one of the most extensively discussed "topics" in Japanese syntax in the generative tradition. Insofar as one considers that the derivation of the so-called scrambling construction (simply OSV) is more complex than that of its 'non-scrambled' counterpart (simply SOV), it is necessary to identify the effective experimental devices and hypotheses dealing with SOV before we begin rigorous research on OSV. And this in turn makes it necessary that we be able to determine whether a particular surface order of expressions (or somewhat more precisely, that of the major constituents of the sentence) is "basic/un-marked" or "non-basic/marked." The availability of the bound variable construal has been the most extensively used empirical means for this purpose over the years.

I will first articulate what hypotheses must be adopted in order to make the relevant predictions regarding the availability of the bound variable construal. It will then be shown that it is possible to obtain robust judgments from informants in line with the predictions. On the scale of 0-100, where "0" corresponds to "complete unacceptability" and "100" to "full acceptability," we have been able to obtain in a number of experiments the average score of around 5 on the *Schema, contrasting sharply with the corresponding okSchemata. The significance of such results should be appreciated especially in light of the fact that there seem to be no experimental results in English that are even remotely comparable to our experimental results dealing with the same type of "phenomena." It will be pointed out that a key to making progress in language faculty science is to identify, utilize, and refine the effective experimental devices as we proceed from a simple to more and more involved experiments. Some illustration of these points will be provided in relation to long-distance OSV, resumption in OSV, and local disjointness effects in Japanese.

It will be noted that Japanese seems to be a language that is well suited for language faculty science as conceived here because (i) unlike languages like English, there is an overt (i.e., morphological) means in Japanese to identify an expression β that must be dependent on another expression α at LF as long as β and α are not co-arguments of the same predicate, and (ii) the relevance of LF for the "interpretations" can be seen more transparently in Japanese than in a language like English.

The chapter addresses the following specific topics:
The LF c-command condition on bound variable construal
SOV in Japanese
OSV in Japanese
Preliminary experiments
So vs. a-NPs as the bindee
Split antecedence
The effectiveness of the various experimental devices
Hypotheses
Instructions to the informants
The choice of the binders
The choice of the bindee
Informants
Resumption in OSV in Japanese
Long-distance OSV in Japanese
Local disjointness and OSV in Japanese

In language faculty science, the researchers are consciously probing into the properties of their own mental organ (the language faculty), a feature not shared by a physical science. In Chapter 7, I will briefly review the proposed methodology in relation to various issues in philosophy of science, including those discussed by Lakatos, Popper, Duhem, Peirce, and Feynman, and address what may be special about language faculty science. I will consider how Feynman's (1965: 142) three necessary conditions for progress in science apply to language faculty science―the ability to experiment, honesty in reporting results, and the intelligence to interpret the results―in light of the discussion in the preceding chapters. In connection to this, I will also discuss the relation between the three heuristics in (1) and the hypothetico-deductive method, which will help us understand the relation between language faculty science and 'language science'.

The topics the chapter deals with include:

Lakatos' scientific research program
Popper's falsificationism
Duhem's under-determination thesis
The Hypothetico-deductive method and language faculty science

If it is indeed possible to pursue language faculty science as an exact science, it is worth considering its implications on the relation between language faculty science and 'language science', and more significantly, its much broader implications for the possibility of an exact science outside the extremely limited domains of inquiry. Chapter 8 briefly addresses how the proposed methodology, especially its emphasis on *Schema-based predictions, might prove to be useful in other research areas that deal with the human mind.

End of the draft of chapter 1.
As noted earlier, the footnotes are not provided and much of the formatting is lost here.

It will be pointed out that a key to making progress in language faculty science is to identify, utilize, and refine the effective experimental devices as we proceed from a simple to more and more involved experiments.

And the experimental devices include the assumptions and hypotheses that are used in designing our experiments. It goes without saying that we cannot expect to learn much about anything from the result of an experiment that is designed crucially based on assumptions and hypotheses that have been shown not to be valid. The book I am currently working on addresses how we CAN obtain confirmed predicted schematic asymmetries. But my JJL paper (Hoji 2010 "Hypothesis testing in generative grammar: Evaluation of predicted schematic asymmetries") and Hoji 2006 "Assessing Competing Analyses: Two Hypotheses about 'Scrambling' in Japanese," both of which are available here, discuss what experimental devices we should not use if we want to make progress in language faculty science (unless we can control the noise so as to be able to obtain a confirmed predicted schematic asymmetry based on such assumptions/hypotheses).

See the remarks in section 1.9 in [42415] in the Methodology board.

The paper that I submitted to a journal last August -- I have not received the reviews yet -- ends with the following sentence, slightly adapted here, to avoid reference to other parts of the paper.

"Most importantly, crucial reference to confirmed schematic asymmetries as "basic units of facts" against which we evaluate our hypotheses about the Computational System of the language faculty makes us hopeful that we might be able to make generative grammar an empirical science, or to put it more accurately, to make language faculty science possible, where [the hypothetico-deductive method can be rigorously applied.^FN"

FN: One may take this statement as contentious if one thinks that generative grammar has already established itself as a science of the language faculty. As suggested in the preceding pages, my own assessment is rather different from such a view as long as we mean by "science" a field in which a hypothetico-deductive method is applied rigorously and hypotheses in question are subjected to careful and robust empirical tests. I do not, by any means, claim, however, that the method suggested here is the only way to study the language faculty. There may be other approaches to the language faculty and there may, and perhaps should, be other types of evidence beside informant intuitions that can be used for or against hypotheses about the language faculty. One might, for example, hypothesize the core property of the language faculty that is different from the model of the Computational System adopted here; one might also have a model of judgment-making that is distinct from what we adopt. The relevant hypotheses and the alternative models in question, however, must be articulated with respect to how informant judgment or other types of evidence, if relevant, is/are related to the hypothesized properties of the language faculty so that we can make testable predictions and aspire to make progress in our endeavor to understand the properties of the language faculty in a research program of language faculty science as an exact science.

Under this thread, I will post excerpts from the draft of the book I hope I will finish soon.

Let us record in (38) the two conclusions we have reached.

(38) a. The crucial contrast is between β=0 and β=/=0.
b. The failure to obtain the 'necessary' judgments has a qualitatively different significance depending upon whether the 'necessary' judgment is β=0 or β=/=0, as indicated in (37).

These are extremely important consequences, especially in light of what is often claimed in the literature in regard to the significance and the relevance of a contrastive judgment and how to handle judgmental variations (including unexpected judgments). It is often claimed that the empirical evidence for or against a given proposal is a contrastive judgment and that as long as there is a clear enough contrast found in a given paradigm we are justified to take the paradigm as valid evidence for or against a given proposal. It is also often maintained that judgmental variations are inevitable and we should not expect to obtain totally clear and consistent judgments among our informants.
(Copied from one of the chapters of the book draft.)

Since the relevant discussion makes use of some technical terms, I cannot reproduce the entire discussion in a short posting like this. While the main point here is as in (38), the above remarks are followed immediately by (i).

(i) What is being proposed here share some aspects of such views.

But I will proceed to emphasize the differences between 'such views' and my own.

β=0: the informant's judgment that the sentence is totally unacceptable under the intended interpretation
β=/=0: the informant's judgment that the sentence is not totally unacceptable under the intended interpretation
'necessary' judgments: the informants' judgment that should obtain in order for an alleged generalization to be valid or in order for the predictions not to be disconfirmed or to be confirmed (<==This is a rough characterization of what is intended. A more 'precise' (i.e., close to what is stated in the book draft: chapter 5) characterization would require the introduction of some technical terms.)

The book addresses intellectual honesty, albeit somewhat indirectly.
In my view, one of the main reasons why generative grammar has not yet become an empirical science and seems to remain to be what Popper (1982: 32) calls "metaphysical research program," i.e., a "research programme ... which [is] not yet testable" is the fact that the field tolerates, if not encourages, intellectual dishonesty, not by accident but by design.

It is interesting to observe in this context that Boeckx 2006: chap. 3 tries to characterize the Minimalist Program in the terms of Lakatos' 'scientific research programmes'. Given the conspicuous absence of the concern for making the research program progressive (theoretically or empirically; see Lakatos 1970/1978), hence for making it a scientific research program in the terms of Lakatos 1970, 1978; cf. the [30221] posting "Lakatos 1978, applied to generative research" here, it seems to me to be more appropriate to regard Boeckx's conception of the current state of the Minimalist Program as being analogous to Popper's (1982) "metaphysical research programme" than to Lakatos' "scientific research programme." Adapting Popper's (1982: 161-162) phrasing, we might call the current state of the Minimalist Program as conceived of by Boeckx (2006: chap. 3) speculative anticipations of testable theories of the language faculty.

Boeckx, Cedric. 2006. Linguistic Minimalism: Origins, Concepts, Methods, and Aims, Oxford University Press, New York.
Popper, Karl. 1982. Quantum Theory and the Schism in Physics: From the Postscript to the Logic of Scientific Discovery, edited by W. W. Bartley, III., Routledge, London and New York.
Lakatos, Imre. 1970. "Falsification and Methodology of Scientific Research Programmes," in I. Lakatos and A. Musgrave (eds.), Criticism and the Growth of Knowledge, Cambridge University Press. pp. 91-195. (Reproduced in Lakatos 1978.)
Lakatos, Imre. 1973. A Radio Lecture: Science and Pseudoscience. (The audio is available at http://www.lse.ac.uk/collections/lakatos//Default.htm, and a slightly revised version is published in Lakatos 1978, as "Introduction.")
Lakatos, Imre. 1978. The Methodology of Scientific Research Programmes: Philosophical Papers Volume 1, edited by John Worrall and Gregory Currie, Cambridge University Press.

http://neurotheory.columbia.edu/~ken/cargo_cult.html
"Cargo Cult Science"
Richard Feynman
From a Caltech commencement address given in 1974
Also in Surely You're Joking, Mr. Feynman!

(You can actually get to its content at various places. Just Google "Cargo Cult Science." To understand the first paragraph below, you may need to either look at the preceding paragraph of the document available at the site noted above or check "Cargo Cult" in Wikipedia (or something like that) unless you already know what it means. But even without understanding it completely, you will get the basic point. The basic point seems quite relevant when we consider much of the practice in generative grammar.)

***
Now it behooves me, of course, to tell you what they're missing. But it would be just about as difficult to explain to the South Sea islanders how they have to arrange things so that they get some wealth in their system. It is not something simple like telling them how to improve the shapes of the earphones. But there is one feature I notice that is generally missing in cargo cult science. That is the idea that we all hope you have learned in studying science in school--we never say explicitly what this is, but just hope that you catch on by all the examples of scientific investigation. It is interesting, therefore, to bring it out now and speak of it explicitly. It's a kind of scientific integrity, a principle of scientific thought that corresponds to a kind of utter honesty--a kind of leaning over backwards. For example, if you're doing an experiment, you should report everything that you think might make it invalid--not only what you think is right about it: other causes that could possibly explain your results; and things you thought of that you've eliminated by some other experiment, and how they worked--to make sure the other fellow can tell they have been eliminated.

Details that could throw doubt on your interpretation must be given, if you know them. You must do the best you can--if you know anything at all wrong, or possibly wrong--to explain it. If you make a theory, for example, and advertise it, or put it out, then you must also put down all the facts that disagree with it, as well as those that agree with it. There is also a more subtle problem. When you have put a lot of ideas together to make an elaborate theory, you want to make sure, when explaining what it fits, that those things it fits are not just the things that gave you the idea for the theory; but that the finished theory makes something else come out right, in addition.

In summary, the idea is to give all of the information to help others to judge the value of your contribution; not just the information that leads to judgement in one particular direction or another.
***