The last two paragraphs of Ch. 6: section 6 (the last section of the chapter) are as follows:
(The formatting is lost and the footnotes are not provided here.)

***
　　Reproducibility under discussion, of course, is not confined to one particular language that we happen to have investigated. To the extent that our experiments in English deal with universal hypotheses, we should be able to test the validity of those hypotheses in other languages as well, provided that we have a minimally necessary understanding of the language-particular properties in those languages that are relevant to, and necessary for, the testing of the universal hypotheses. If we can manage to replicate our experimental results in other languages, by means of classifying informants in accordance with how the predicted schematic asymmetry under discussion has been deduced and how the Main-Experiment is carried out, we will have more confidence not only about our hypotheses that have given rise to the predicted schematic asymmetries in EPSA [31]-11 (=[31]-4) but also about the proposed methodology for language faculty science, especially with regard to its crucial use of informant classification.
　　It must also be pointed out that we must try to increase the number of "reliable" informants by overcoming the issues addressed in (77b). I.e., we do not want to continue to "leave aside" those informants who have failed to qualify as reliable informants for the purpose of testing the validity of the Main-Hypothesis/ses in our Main-Experiment. We should try to enhance the resourcefulness and the effectiveness of the informants so that a greater number of our informants qualify as reliable informants with regard to the testing of the Main-Hypothesis/ses in our Main-Experiment. Given our working hypothesis that the properties of the CS are universal, such should be possible, in principle.

The last section the Appendix to Chapter 7 (which is now the only Appendix to Chapter 7) is like this:
(The formatting is lost.)

In this Appendix, I discussed two more bridging hypotheses about Japanese and two more about English and provided further illustration of the role of bridging hypotheses as hypotheses about effective probes for investigating properties of FD and hence of the CS. The discussion addressed a wider context of research in which the Main-Experiments discussed in this book have been designed and conducted. In the course of the discussion, we also addressed the issue of theory-laden nature of language faculty science and the abstract nature of the research program. We cope with the theory-ladenness, as addressed in the previous subsection, by proceeding in our research by establishing and accumulating confirmed predicted schematic asymmetries. What exact significance we can assign to a given confirmed predicted schematic asymmetry in a given Main-Experiment might depend upon the hypotheses that give rise to the relevant predicted schematic asymmetry, including the bridging hypothesis in question, how the Sub-Experiments are designed and conduced, and what informant classifications we have used. But that is how we must proceed in language faculty science as an exact science. In all this, it is predicted schematic asymmetries and confirmed predicted schematic asymmetries that allows us to pursue rigorous empirical testability and that provides us with an empirical basis for our theorizing.

The last few paragraphs of the last section (Section 6) of Chapter 7 are currently as follows:
(The formatting is lost and the footnotes are not supplied here.)

　　The merit of employing notions of Main-Hypotheses and Sub-Hypotheses and those of the Main-Experiment and its Sub-Experiments should be appreciated in light of the theory-laden nature of language faculty science, as just reviewed. I.e., they are for the purpose of making the result of our Main-Experiment as significant as possible with regard to the validity of the Main-Hypotheses tested in the Main-Experiment, as discussed in the preceding pages.
　　We would naturally like to see (i) whether there are properties of FD other than those specified by [U1] and [U2], (ii) whether there are dependency interpretations that must be based on FD but is distinct from BVA, and (iii) whether there are other formal objects/relations that must be based on LF c-command but is distinct from FD. To the extent that we have affirmative answers to these questions and to the extent that we find out about properties of what is mentioned in (i)-(iii) by the "Guess-Deduce-Compare" method, we will have more confidence about our hypotheses including the bridging hypotheses.
　　As to (i), there seems to be another structural condition on FD, which prevents a and b of FD(a, b) from being co-arguments of a predicate. As to (ii), there seem to be two other types of anaphoric relations that are based on FD; one is the so-called sloppy-identity reading and the other is a particular instance of coreference. As to (iii), what underlies one type of the scope dependency interpretation seems to be one such object/relation. Discussion of each of (i)-(iii) involves complications of a substantive magnitude, and the space limit prevents me from even illustrating the relevant phenomena here, let alone discussing the designs and the results (that we have obtained so far) of the Experiments testing the relevant predicted schematic asymmetries. I can only state here that the results reported in this chapter forms a basis for our investigation of issues pertaining to (i)-(iii), and that the Experimental results we have obtained so far dealing with (i)-(iii) provide support for the hypotheses discussed in this chapter, and hence, indirectly, for the proposed methodology for language faculty science as an exact science.

Due to space considerations, Appendix II of Chapter 7 will not be included in the book. I will try to make the point sated in [44528] in some other place in the book.

The last few paragraphs of Ch. 7: Appendix II: Further illustration of the purpose and the effects of informant classification are as follows, at the moment.
(The formatting is lost.)

***
　　The practice in the field seems to be based on the concession that we cannot obtain definite and categorical experimental results. This seems to have led to the reliance on "statistically significant contrasts in "empirically-oriented" research. If a "statistically significant contrast" suffices as evidence, however, the contrast between the %(Y) on Schema A and that on Schema B in the initial result of EPSA [5]-1 would qualify as such. As we have observed in this Appendix, informant classification did not lower the %(Y) on Schema B and the %(I) in EPSA [5]-1 close to the predicted 0, unlike English EPSA [33]-10 and Japanese EPSA [33]-9, for example.
　　There is no guarantee that a given "contrast" we observe is a reflection of a hypothesized grammatical principle/condition/etc. Only with careful improvement of the experimental device, in the form of successively rigorous informant classification as discussed in this book, can we expect to obtain a confirmed predicted schematic asymmetry. And only when we do so, we can have some hope that the observed "contrast," now in the form of a confirmed predicted schematic asymmetry, rather than a mere (but "statistically significant") contrast, is indeed a reflection of hypothesized properties of the CS.
　　This has far-reaching implications with regard to how we can try to accumulate knowledge of our subject matter whose underlying properties we can investigate only by means of abstract theorizing and by means of experiments focusing on informant reactions or something else that we can observe but can be related only indirectly to the hypothesized properties about the subject matter. See the remarks at the end of Chapter 8.

Section 1 of Chapter 6 is as follows:
(The formatting is lost, and the footnotes are not provided here.)

***
　　In this and the next chapter, I will illustrate how we have obtained Experimental results that are very close to, or precisely in accordance with, our definite and categorical predictions in the form of predicted schematic asymmetries. The demonstration will be crucially based on the notions introduced in the preceding chapters, especially the idea of Main- and Sub-Experiments and informant classification based on the results of Sub-Experiments.
　　Section 2 reviews how the predicted schematic asymmetries tested in our Main-Experiment have been deduced. Our Main-Experiment is concerned with structural hypotheses about FD, whose hypothesized properties we try to investigate by checking the availability of a particular dependency interpretation of BVA(A, B), with specific choices for A and B. The section provides the design of the Main-Experiment, including its Schema groups and Lexical groups, along with its result without informant classification.
　　Section 3 offers initial illustration of informant classification, including its justification and effects. The idea behind informant classification in interpreting the result of our Main-Experiment, based on the result of its Sub-Experiments, is that we want to pay attention to the informants whose judgments in the Main-Experiment are significant with respect to its Main-Hypotheses. One of the Sub-Experiments for our Main-Experiment EPSA [31]-4, EPSA [31]-1, is concerned with the lexical hypotheses about FD. The other Sub-Experiment, EPSA [31]-7, is concerned with the effectiveness of the instructions to our informants, including how we express the intended dependency interpretation. In section 3.2, I will report that the informant classification based on the result of EPSA [31]-1 makes the result of our Main-Experiment EPSA [31]-4 significantly closer to our prediction. I will also report that the result of EPSA [31]-1 in turn becomes considerably closer to our definite and categorical prediction than its "original" result with informant classification based on EPSA [31]-7.
　　In section 4, I turn to the effects of informant classification based on across-occasion reproducibility in individual informants' reported judgments. We pay attention to whether an individual informant's reported judgments on the same set of Examples in the same Experiment are consistently in accordance with our predictions when the same Experiment is conducted at different times. We will observe that, by paying attention to across-occasion reproducibility in the result of our Sub-Experiments, the results of our Main-Experiment can become quite close to, or precisely in accordance with, our definite and categorical prediction.
　　The discussion up to this point is concerned with *Schema-based predictions, which are the most crucial part of the predicted schematic asymmetry. The okSchema-based prediction, however, also plays a vital role in language faculty science. Suppose the %(Y) on Schema B is 0. That does not disconfirm the *Schema-based prediction. But if the %(Y) on Schema A is also close to 0, we cannot attribute the %(Y) on Schema B being 0 to the grammatical reasons hypothesized to underlie the predicted schematic asymmetry. As it turns out, the %(Y) on Schema A in one of the two Schema groups in EPSA [31]-4 is rather high (26%) in its result without informant classification. We will observe in section 5 how informant classification affects the %(Y) on the Schema A in question. More specifically, we will demonstrate that informant classification can result in a higher %(Y) on Schema A than in the "original" result. The demonstration in this section provides support for the level of LF representation as a formal basis of meaning, as postulated in the model of the CS in Chomsky 1993.
　　I have stressed in Ch. 4: section 5 and Ch. 5: section 5.1 that informant classification is not for the purpose of obtaining the predicted Experimental results but it is for the purpose of obtaining Experimental results that are as reliable as possible. In sections 3-5, the informant classifications do lead to results that are in accordance with our predicted schematic asymmetries. The Appendix to this chapter illustrates a case where informant classification fails to give us Experimental results as "predicted." We turn to so-called Binding Condition C/D, whose effects are tested in one of our Sub-Experiments, EPSA [31]-3 (=[31]-10), for our Main-Experiment EPSA [31]-4. We will observe that the results of EPSA [31]-3 and [31]-10 do not come close to the "prediction" about the so-called Binding Condition C/D effects even with the "best" informant classification that we have used in sections 3-5 that have turned the %(Y) on Schema B and the %(I) in EPSA [31]-4 into the predicted 0.
　　Language faculty science deals with I-language. "I" of I-language is meant to stand for the "I" of "individual," "internal" and "intensional"; see Chomsky 1986: 2.3 and 2.4 and Chomsky 2000: 70, 118, for example. The internalist approach to "language" and the language faculty thus leads us to the view that our research efforts should start with an attempt to establish confirmed predicted schematic asymmetries within a single-researcher-informant experiment and to replicate the experimental results in multiple-informant experiments. A natural place for me (a native speaker of Japanese) to start is therefore to deal with Japanese and try to obtain confirmed predicted schematic asymmetries in Japanese with myself as an informant and try to replicate them in multiple-informant experiments.
　　The articulation and the presentation of the proposed methodology in chapters 2-5, however, make reference to English rather than Japanese. The initial empirical/experimental illustration of the proposed methodology in this chapter is also based on English, not Japanese. The tactical decision to make reference to English is due to readability considerations. Assuming that the readers of this book are not necessarily familiar with Japanese, heavy reference to Japanese in the conceptual articulation of the proposed methodology and its initial empirical illustration would make the average reader's task more demanding than necessary. The reliance on English has an unwanted consequence of preventing me from addressing experimental results of a single-researcher-informant experiment with myself being the sole informant. It is hoped that the discussion in the next chapter, where we deal with Japanese, will make up for that shortcoming.

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).

And that is exactly the goal of my forthcoming book Language Faculty Science.

The last two paragraphs of Chapter 8 are as follows:
(The formatting is lost here.)

***
　　In addressing social sciences, Feynman remarks as follows: ^FN14


　　　"Because of the success of science, there is, I think, a kind of pseudoscience. Social science is an example of a science which is not a science; they don't do [things] scientifically, they follow the forms―or you gather data, you do so-and-so and so forth but they don't get any laws, they haven't found out anything. They haven't got anywhere yet―maybe someday they will, but it is not very well developed, … I may be quite wrong, maybe they do know all these things, but I don't think I'm wrong. You see, I have the advantage of having found out how hard it is to get to really know something, how careful you have to be about checking the experiments, how easy it is to make mistakes and fool yourself. I know what it means to know something, and therefore I see how they get their information and I can't believe that they know it, they haven't done the work necessary, haven't done the checks necessary, haven't done the care necessary. I have a great suspicion that they don't know, that this stuff is [wrong] and they're intimidating people. I think so. I don't know the world very well but that's what I think." (Feynman 1999: 22)


What I have proposed in the preceding pages is how we can "do the work necessary, do the checks necessary, and do the care necessary" for obtaining reliable information from our experiments about properties of the language faculty.
　　What I envisage is a time when we will be able to deduce hard predictions (predicted schematic asymmetries) in various languages, will be able to evaluate by experiments the validity of our universal and language-particular hypotheses, and will be able to formulate hypotheses of a successively more general nature, without losing rigorous testability. When something like that has become the norm of the research program, an experiment dealing with one language can be understood clearly in terms of the universal hypotheses (along with language-particular hypotheses) in question so that the implications of the result of an experiment dealing with a particular language can be transparent with respect to other languages. Researchers "working with" different languages will at that point share (many of) the same puzzles and issues pertaining to universal properties of the language faculty. They will know precisely what necessary care and checks they need to do in order to design effective experiments for testing the validity of the same universal hypotheses. That will enable us to proceed in a way much more robust than what has been presented in the preceding chapters, still on the basis of confirmed predicted schematic asymmetries. The field will at that point be widely regarded as an exact science, and everyone will take that for granted. And I also suspect that, at that point, other fields of research that deal with the brain and the mind pay close attention to the research results and methodology in language faculty science because they find it useful to try to learn from the categorical nature of the experimental results in language faculty science and its methodology that has guided its research efforts.^FN15

FN14: A video clip containing these remarks can be found at: http://www.youtube.com/watch?v=IaO69CF5mbY (last accessed on 7/25/2013).

FN15: This reminds us of Chomsky's (1975: 5) remark that "it is not unreasonable to suppose that the study of this particular human achievement, the ability to speak and understand a human language, may serve as a suggestive model for inquiry into other domains of human competence and action that are not quite so amenable to direct investigation."

The Summary section of Chapter 5 is as follows.
(The formatting is lost here and the footnotes are not provided here.)

***
　　In this chapter, I have discussed how our Experiments are designed and conducted and how the Experimental results are interpreted, in accordance with the proposed methodology for language faculty science. The general design of our Experiment is that it consist of a set of three Example sentences; one *Example instantiating the *Schema and one okExample instantiating the okSchema, both with the dependency interpretation under discussion, and another okExample, which is (as) identical (as possible) to the *Example but without involving the dependency interpretation. That our Experiment consists of Examples instantiating Schemata is in accordance with the fundamental schematic asymmetry recognized in Ch. 2, in pursuit of rigorous testability in a research program that aims at discovering properties of the universal aspect of the language faculty by dealing with judgments of speakers of a particular language. We check the acceptability of Example sentences with a dependency interpretation that we hypothesize to be crucially based on an LF c-command relation, so as to maximize the testability of our hypotheses stated within the general conception of the Computational System of the language faculty suggested in Chomsky 1993. The fact that our general Experimental design allows us to check effects of variables along two distinct dimensions is in harmony with the view that the hypothesized LF relation/object FD, which is hypothesized to underlie the dependency interpretation of BVA(A, B) with particular choices of A and B, is constrained by a structural condition as well as a lexical condition.
　　In language faculty science as proposed here, we work with confirmed predicted schematic asymmetries, instead of a statistically significant contrast. The confirmed predicted schematic asymmetry is about an individual informant. We thus focus on the %(Y) on Schema B (and to a lesser degree, on that on Schema A) and %(I), instead of the "average" responses among a group of informants. In (52), I provide a review of what is meant by "%(Y) on a Schema," "%(I)," and "N(I)."

(52)　a.　The "%(Y)" on a Schema stands for the percentage of Yes answers (i.e., the reported judgment that the example in question is acceptable at least to some extent (with the intended BVA in the case of Schema A and Schema B in [31]-4, for example)) among all the answers/judgments given on the Examples instantiating the Schema in question.
　　b.　The %(I) stands for the percentage of the informants who gave a Yes answer on at least one *Example in a given Experiment.
　　c.　The "N(I)" is the number of the informants who have provided answers on the Examples being considered.

Our prediction is that the %(Y) on Schema B is 0 and those on Schema A (and on Schema C) are not 0 in our Main-Experiment as long as the informant clearly understands what is intended by our instructions (including the intended dependency interpretation) and for whom the Sub-Hypotheses in the Main-Experiment are valid. It follows that the predicted %(I) in a multiple-informant Experiment is also 0 in our Main-Experiment as long as we focus on such informants classified in its Sub-Experiments, provided that everything else about our Experiments are done properly and correctly and that our hypotheses are all valid.
　　As discussed in section 4.4, a multiple-informant experiment is a collection of single-informant experiments. Its purpose is to see if the result of a single-researcher-informant experiment is replicated, with regard to whether we obtain a confirmed predicted schematic asymmetry, rather than to see if there is a significant difference between the average responses among a group of informants on the *Examples and the okExamples. In Chapter 7, I will present experimental results―both in English and in Japanese―that show that the significance of a contrast among a group of informants that falls (far) short of a confirmed predicted schematic asymmetry can only be assessed in light of the results of Sub-Experiments. The discussion there includes a demonstration that two seemingly identical-looking Experimental results exhibiting a contrast can turn out to have radically different interpretations once we consider the results of Sub-Experiments and the informant classification it leads us to.
　　In the next two chapters, I will illustrate the proposed methodology for language faculty science, on the basis of actual experiments. The general experimental design and the specific aspects of actual Experiments still need improvement. The fact that we have been able to obtain experimental results that are quite close to our definite and categorical predictions, however, provides support for the viability of language faculty science as proposed here.

Section 1 of Chapter 5 is as follows:
(The formatting is lost here.)

　　The subject matter of language faculty science is the language faculty. The language faculty is hypothesized to underlie our ability to relate sounds and meaning. Because the language faculty is internal to the mind of an individual, language faculty scientists are internalists by definition. Taking seriously the methodological naturalist approach, we aim at accumulating knowledge about the language faculty by putting forth hypotheses so as to make predictions that are as definite as possible. More specifically, we aspire to accumulate knowledge about the language faculty by adopting the hypothetico-deductive method, dubbed by Feynman as the "Guess-Compute-Compare" method. As stated in the preceding chapters, the proposed methodology for language faculty science as an exact science is a consequence of taking the language faculty as the object of inquiry and adopting the "Guess-Compute-Compare" (or "Guess-Deduce-Compare") method for testing our hypotheses.
　　In Chapters 2-4, I have articulated how language faculty science as an exact science is possible. As discussed in Chapter 2, our desire to seek as much generality as possible with regard to an individual informant's judgments, in pursuit of discovering universal properties of the language faculty, has led us to work with schemata and recognize the fundamental schematic asymmetry as indicated in [P].

[P]　The fundamental schematic asymmetry
　a.　　The *Schema-based prediction:
　　　Every example sentence instantiating a *Schema is unacceptable with the specified interpretation pertaining to two expressions.
　b.　　The okSchema-based prediction:
　　　Some example sentences instantiating an okSchema are acceptable at least to some extent with the specified interpretation pertaining to two expressions.

A combination of a *Schema-based prediction and its corresponding okSchema-based prediction is called a predicted schematic asymmetry. If we obtain informant judgments in accordance with the predicted schematic asymmetry, we obtain a confirmed predicted schematic asymmetry, which I have suggested should be considered as basic units of facts in language faculty science.
　　While we are concerned with universal properties of the language faculty, our predictions are about judgments by an individual informant on the relation between sounds and meaning in a particular language. Our predictions―in the form of predicted schematic asymmetries―must therefore be given rise to, at least, by the following three types of hypotheses.

[H]　a.　　Universal hypotheses
　　b.　　Language-particular hypotheses
　　c.　Bridging hypotheses

Bridging hypotheses relate a particular interpretation detectable by the informant to some hypothesized concept/relation.
　　I have adopted in Chapter 3 Chomsky's (1993) model of the Computational System (CS) of the language faculty and Ueyama's (2010) model of judgment-making by informants, as general bases deducing definite and categorical predictions about individual informants' judgments. The two most crucial aspects of Chomsky's (1993) model of the CS are: (i) that the formal basis of meaning is LF and (ii) that the only structure-building operation is Merge. Specific instances of [H-a], [H-b], and [H-c] that we pursue all make crucial reference to LF c-command, and that is for the purpose of attaining maximal testability, as discussed in Ch. 3: section 7. Because of its categorical nature, Chomsky's 1993 model of the CS provides us with a basis for extracting, among the myriads of linguistic intuitions that we may observe, linguistic intuitions that we can characterize and describe in terms of our hypotheses about the CS couched within Chomsky's 1993 model of the CS, more specifically, in terms of LF c-command. Facts in language faculty science thus qualify as such only if they constitute a confirmed predicted schematic asymmetry. A confirmed predicted schematic asymmetry is based on predicted schematic asymmetry, which is deduced from our hypotheses. It is in this sense that facts and hypotheses in language faculty science are inseparable.
　　We want to focus on, and hence want our experimental result to be revealing about, the validity of (a) particular hypothesis/ses among what has given rise to the predicted schematic asymmetry.^FN1 We refer to such (a) hypothesis/ses as the Main-Hypothesis/ses. Our Main-Experiment is concerned with the Main-Hypothesis/ses among the hypotheses that give rise to the predicted schematic asymmetry. The other hypotheses, which we refer to as the Sub-Hypotheses, are tested in Sub-Experiments.
　　As discussed in Chapter 4, failure to obtain a confirmed predicted schematic asymmetry suggests that (i) one or more of our hypotheses that have given rise to the predicted schematic asymmetry is/are not valid, (ii) the "instructions" to the informant in our Experiments, including how we specify the intended dependency interpretation, are not effective, or (iii) the degree of the informant resourcefulness is not high enough, or any combination of those. In order to make the result of the Main-Experiment revealing about the validity of the Main-Hypothesis/ses, it is therefore imperative that we ensure as much as possible that the failure to obtain a confirmed predicted schematic asymmetry would not be due to (some of) the Sub-Hypotheses not being valid, the "instructions" to the informant not being effective, or the informant not being resourceful enough. Sub-Experiments are conducted for that purpose.
　　As discussed also in Chapter 4, the key to obtaining definite and categorical experimental results is the reliability of the experimental device. Unlike a physical science, we do not have a physical device the reliability of the design, construction and operation of which we can check, at least at the moment. As pointed out in Ch. 4: sections 1 and 5, our informants and our instructions are part of our experimental device. We can consider the result of our Main-Experiment revealing about the validity our Main-Hypotheses only if we focus on the informants for whom the instructions are effective and for whom the Sub-Hypotheses seem valid, judging from the results of the Sub-Experiments. Interpreting the result of the Main-Experiment without reference to those of its Sub-Experiments would be like conducting experiments without taking necessary care and without doing necessary checks; see the Feynman quote given in Ch. 8: section 2.
　　In summary, the key to obtaining definite and categorical experimental results in accordance with our predicted schematic asymmetry is the recognition that the result of our Main-Experiment must be interpreted by making reference to the results of its Sub-Experiments, as introduced in Chapter 3 and further discussed in Chapter 4. As pointed out in Chapter 4, our predictions are not about any informant. It is about those informants who are reliable for the purpose of testing the Main-Hypothesis/ses in the Main-Experiment. Crucial reference to the results of Sub-Experiments is for the purpose of making the result of the Main-Experiment as significant as possible with respect to the validity of the Main-Hypotheses tested in the Main-Experiment, and that is analogous to enhancing the reliability and the precision of the experimental device in a physical science, as discussed in Ch. 4: section 1. What has led us to recognize the concept of Main-Hypotheses and Sub-Hypotheses as well as that of Main-Experiments and Sub-Experiments is the desire to be able to focus on (a) particular hypothesis/ses among those that give rise to the predicted schematic asymmetry. It stems from our desire to assign maximal significance to our experimental result with respect to such (a) hypothesis/ses. We want our experimental result to be as significant as possible, regardless of whether it turns out to be in line with our definite and categorical prediction.
　　In Chapters 6 and 7, I will illustrate the methodological proposal for language faculty science, drawing from a number of on-line Experiments, in support of its viability. The two chapters deal with English and Japanese, addressing the same universal hypotheses. Before we turn to empirical illustration of the proposed methodology, we will go over in this chapter the general design of our Experiments and how it reflects the proposed methodology. More in particular, I will addresses how experiments are designed and conducted in language faculty science and how the experimental results are interpreted, all in accordance with the proposed methodology for language faculty science as laid out in Chapters 2-4.
　　 The general design of our experiments has been developed by Ayumi Ueyama. The first template was created in 2004 and it has since undergone three major changes so as to reflect various stages of the methodological articulation that has eventually led to the proposal laid out in the preceding 4 chapters.^FN2 The innovations reflect (i) how we tried to obtain reproducible informant judgments among ourselves (the researchers) and in multiple-non-researcher-informant experiments and (ii) how we tried to ensure the significance of the informant judgments with regard to the validity of (a) particular hypothesis/ses under discussion. The former issue obviously concerns reproducibility/repeatability and the latter testability/falsifiability. ^FN3

FN1: See the discussion in Ch. 3: section 2 on the under-determination-of-theory-by-data issue raised by Poincaré and Duhem.
FN2: The experimental results presented in Hoji 2006a and 2006b are based on an earlier version of the program.
FN3: Hoji 2003a and 2003b were attempts to articulate my understanding of the relevant issue prior to 2004.

One of the main features of Language Faculty Science is its attempt to stick to intellectual honesty as addressed by Richard Feynman in his "Cargo Cult Science" among other places, and also addressed by many other people.

I copy part of [35922] "Cargo Cult Science by Richard Feynman" here, which is under [34437] "Intellectual Honesty" under [34390] "Sneak previews of the book I hope I will soon finish." (As noted in [42418], I did not publish the book...)

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.

The thread was under a larger thread about my 2009 ms. A Foundation of Generative Grammar as an Empirical Science. In Language Faculty Science, I try my best to present my methodological proposal and its empirical illustration so as to make it as easy as possible for others to find out what might be wrong with what I claim. I am right now revising the templates for summary charts of experimental results precisely for that purpose.

Please see [44447] "The issue of testability" and [44414] "Advertisement and scientific integrity" for related remarks, under [44413] "A key to language faculty science as an exact science" at the "General Remarks" board.

Chapter 5 addresses the general experimental design that is used in actual Experiments discussed in the book.

Chapter 6 and Chapter 7 discuss English Experiments and Japanese Experiments, respectively, as an illustration of the proposed methodology of language faculty science as an exact science. The presentation and the discussion of the Experiments, especially their results, are abbreviated in the book for reasons of space. But the designs and the results of all the Experiments discussed in the book will be provided in the accompanying website.

Chapter 8 provides "summary remarks," returning to some of the general methodological issues addressed in Chapters 1-4.

Section 1 of Chapter 4 is as follows:
(Most of the formatting is lost here.)

　　Our desire to deduce definite and categorical predictions stems from our desire to find out about the language faculty by adopting the methodological naturalist approach (summarized as the "Guess-Compute-Compare" (or "Guess-Deduce-Compare" method)) to the study of the language faculty and our belief that it is indeed possible to do so. The methodological proposal and its empirical illustration in this work is an attempt to substantiate this belief.
　　Feynman (1965/94: 152-153) remarks, "Another thing I must point out is that you cannot prove a vague theory wrong. If the guess that you make is poorly expressed and rather vague, and the method that you use for figuring out the consequences is a little vague―you are not sure, and you say, 'I think everything's right because it's all due to so and so, and such and such do this and that more or less, and I can sort of explain how this works ...', then you see that this theory is good, because it cannot be proved wrong! Also if the process of computing the consequences is indefinite, then with a little skill any experimental results can be made to look like the expected consequences." As we discussed in Chapters 2 and 3, the key to deducing definite and categorical predictions about the informant judgment is the recognition of the fundamental asymmetry in [P], repeated here.

[P]　　The fundamental schematic asymmetry
　a.　　The *Schema-based prediction:
　　　Every example sentence instantiating a *Schema is unacceptable with the specified interpretation pertaining to two expressions.
　b.　　　The okSchema-based prediction:
　　　　Some example sentences instantiating an okSchema are acceptable at least to some extent with the specified interpretation pertaining to two expressions.

Without its recognition, it would not be possible to deduce definite and categorical predictions about the informant judgment and expect them to be supported experimentally.
　　Just as in the case of our desire to deduce definite and categorical predictions, our desire to obtain experimental results in accordance with our definite and categorical predictions also stems from our desire to find out about the language faculty by adopting the methodological naturalist approach. Our definite and categorical predictions are in the form of a predicted schematic asymmetry, consisting of a *Schema-based prediction and its corresponding okSchema-based prediction, as in [P], repeated above. We now turn to how we can expect to obtain a confirmed predicted schematic asymmetry, i.e., an experimental result in accordance with [P], about individual informants' judgments.^FN1
　　As noted, one of the two defining features of the present work is its commitment to the methodological naturalist approach to the language faculty.^FN2 The "Guess-Compute-Compare" method was introduced in reference to the following remarks by Feynman.

　　　　　In general, we look for a new law by the following process. First we guess it. Then we compute the consequences of the guess to see what would be implied if this law that we guessed is right. Then we compare the result of the computation to nature, with experiment or experience, compare it directly with observation, to see if it works. If it disagrees with experiment, it is wrong. In that simple statement is the key to science. It does not make any difference how beautiful your guess is. It does not make any difference how smart you are, who made the guess, or what his name is―if it disagrees with the experiment, it is wrong. That's all there is to it. (Feynman 1965/94: 150)

Feynman continues the above passage as follows:

　　　　　It is true that one has to check a little to make sure that it is wrong, because whoever did the experiment may have reported incorrectly, or there may have been some feature in the experiment that was not noticed, some dirt or something; or the man who computed the consequences, even though it may have been the one who made the guesses, could have made some mistake in the analysis. These are obvious remarks, so when I say if it disagrees with experiment it is wrong, I mean after the experiment has been checked, the calculations have been checked, and the thing has been rubbed back and forth a few times to make sure that the consequences are logical consequences from the guess, and that in fact it disagrees with a very carefully checked experiment." (Feynman 1965/94: 150-151)

　　When a physicist tries to make sure that an experiment is done very carefully, one of the first things they check is whether their experimental device, i.e., the device to measure what they are measuring, is properly designed, built, and operated. When a language faculty scientist tries to make sure that an experiment is done very carefully, what do they check? The experimental device in language faculty science, as it is being pursued here (at the present development of the research program), does not involve any hardware equipment, except computers for collecting informant judgments on-line and for some basic analysis of the results. But an experiment involves (i) example sentences, (ii) instructions to the informants including the intended interpretation in question, and (iii) the informants. The checking of how carefully the experiment has been conducted in language faculty science must therefore involve the testing of how effective our instructions are for a given informant and how clearly each informant understands the instructions as intended.
　　Recall that a predicted schematic asymmetry in language faculty science is given rise to by three types of hypotheses as in [H].

[H]　　Types of hypotheses:
　a.　　Universal hypotheses
　b.　　Language-particular hypotheses
　c.　　Bridging hypotheses

Given this, the issue raised by Poincaré and Duhem in relation to physical sciences, i.e., the under-determinacy-of-theory-by-data issue, applies to language faculty science clearly and straightforwardly, as discussed in Chapter 3: section 2. As also pointed out there, one way to cope with the issue is to focus on (a) particular hypothesis/ses among those that give rise to the predicted schematic asymmetry under discussion. We can do so effectively only by assuming the validity of the other hypotheses and by testing it in a Sub-Experiment, independently of the Main-Experiment in which the validity of the Main-Hypothesis/ses is tested.
　　As noted, an experiment in language faculty science involves (i) example sentences, (ii) instructions to the informants including the intended interpretation in question, and (iii) the informants. The example sentences are constructed in accordance with a *Schema and its corresponding okSchema, and the *Schema and the okSchema are constructed based on the hypotheses that give rise to the predicted schematic asymmetry in question, as discussed in Chapter 2: sections 3 and 4.
　　The Main-Experiment is designed under the assumption that the Sub-Hypotheses are valid. In order for the result of the Main-Experiment to be significant, it must be the case that our informants clearly understand our instructions, including the intended interpretations. If the Sub-Hypotheses are shown to be invalid in a Sub-Experiment, the result of the Main-Experiment cannot be taken as significant with regard to the validity of the Main-Hypothesis/ses. Likewise, if it is shown in a Sub-Experiment that the instructions including the intended interpretation in question are not effective for particular informants, the result of the Main-Experiment cannot be taken significant for those informants. It is in this sense that a Sub-Experiment to test the validity of the Sub-Hypotheses and the effectiveness of the instructions for individual informants can be understood as being analogous to the physicist's checking the effectiveness of their experimental device. Sub-Experiments and informant classification based on their results are, therefore, our way to maximize the reliability of our Main-Experiment in language faculty science.

FN1: Recall that it is our commitment to the internalist approach that leads us to make definite and categorical predictions about an individual informant's judgments.

FN2: The other is its commitment to the internalist approach. We are concerned with whether an individual informant's judgments are indeed in accordance with our definite and categorical predictions. We are not concerned with the "average" of a group of informants although we can address the average insofar as it is revealing about the judgments of individual informants. Illustration will be provided in Chapters 6 and 7.

The level of the conceptual articulation about the methodology for language faculty science as an empirical science is qualitatively different before and after (inclusive) Language Faculty Science.

My works including:

"Hypothesis testing in generative grammar: Evaluation of predicted schematic asymmetries" Journal of Japanese Linguistics 26.

reflect earlier stages of my understanding of various issues.

I plan to provide critical reviews of those works, including Hoji 2003 (Lingua) in due course. I am hoping that interested people can critically evaluate published works by others -- assuming that they are meant to be an attempt to discover properties of the language faculty -- once they have read Language Faculty Science and my critical discussion of my own works prior to Language Faculty Science.

The postings under this thread address some key ideas that will be included in the book Language Faculty Science. I expect the book to be published by the Cambridge University Press toward the end of this year or next year.

The first section and the first part of the last section of Chapter 3 are as follows:
(The footnotes are not included.)

***
1. Introduction
By considering the informant judgments as being about schemata (not about example sentences), we came to recognize the fundamental asymmetry between the *Schema-based prediction and the okSchema-based prediction, as indicated in [P], repeated here.

[P]　a.　The *Schema-based prediction:
　　Every example sentence instantiating a *Schema is unacceptable with the specified interpretation pertaining to two expressions.
　b.　The okSchema-based prediction:
　　Some example sentences instantiating an okSchema are acceptable at least to some extent with the specified interpretation pertaining to two expressions.

The considerations of the fundamental asymmetry between [P-a] and [P-b] have led us to recognize the relative significance of the four possible results of an experiment in a single-informant experiment, as discussed in Chapter 2. Those considerations are independent of how our predictions are given rise to.
　　　Experiments are designed to test predictions and predictions are deduced from hypotheses. The significance of the experimental result should therefore be understood in relation to what hypotheses have given rise to the prediction in question―in our case, a predicted schematic asymmetry. In this chapter, I will discuss how a predicted schematic asymmetry is deduced in language faculty science. I will first briefly address the structure of prediction-deduction in language faculty science by making reference to what types of hypotheses give rise to a predicted schematic asymmetry. I will then provide further illustration of the point on the basis of specific hypotheses that give rise to predicted schematic asymmetries.



8. Summary
　　　The two points of departure in the present work are: (i) the object of inquiry is the language faculty (not "languages" or "language" as they are commonly understood) and (ii) we want to find out about (the universal aspects of) the language faculty by the "Guess-Compute-Compare" method (of testing hypotheses). They have led us to the need to work with schemata and then to the fundamental asymmetry between the *Schema-based prediction and the okSchema-based prediction, as discussed in Chapter 2.
In this chapter, we addressed another consequence of those basic premises, in relation to how we make a predicted schematic asymmetry. A predicted schematic asymmetry is given rise to (at least) by the three types of hypotheses in [H].

[H]　a.　　Universal hypotheses
　　b.　　Language-particular hypotheses
　　c.　　Bridging hypotheses

The desire to be able to assign maximal significance to the experimental result with respect to just one of the hypotheses that have given rise to the predicted schematic asymmetry has led us to recognize―though only at a conceptual level so far―the concept of Main-Hypotheses and Sub-Hypotheses as well as that of Main-Experiments and Sub-Experiments.
　　　I have then adopted Chomsky's (1993) model of the CS. With its categorical nature, it allows us to formulate hypotheses so as to make our predicted schematic asymmetry testable. The two most crucial aspects of Chomsky's (1993) model of the CS are: (i) the formal basis of meaning is LF and (ii) the only structural-building operation is Merge. A specific instance of a Universal hypothesis ([H-a]) makes reference to an LF object/relation FD(a, b) whose necessary condition is that a c-commands b at LF. A specific instance of a Language-particular hypothesis ([H-b]) relates the linear schematic representation of a phonetic string to a c-command-based hierarchical schematic representation at LF. A specific instance of a bridging hypothesis ([H-c]) is a hypothesis concerning how a particular interpretation detectable by the informant is based on FD.
(The rest of the section is not included here.)

The first two paragraphs and the last three paragraphs of Chapter 2 are as follows:

***
Introduction

The present work attempts to pursue and defend the thesis that it is possible to investigate the language faculty by applying the hypothetico-deductive method, which Feynman puts as the "Guess-Compute-Compare" method. The passage from Feynman 1965/94 quoted in Chapter 1 is repeated here.

In general, we look for a new law by the following process. First we guess it. Then we compute the consequences of the guess to see what would be implied if this law that we guessed is right. Then we compare the result of the computation to nature, with experiment or experience, compare it directly with observation, to see if it works. If it disagrees with experiment, it is wrong. In that simple statement is the key to science. It does not make any difference how beautiful your guess is. It does not make any difference how smart you are, who made the guess, or what his name is―if it disagrees with the experiment, it is wrong. That's all there is to it. (Feynman 1965/94: 150)

More specifically, I argue that the language faculty can be studied as an exact science. By an "exact science" I mean a research program in which definite/categorical predictions are deduced from hypotheses and are tested against experimental results (or observations). Such a research program will be called language faculty science, as noted in Chapter 1. Insofar as we can carry this out successfully with compelling empirical demonstration, that will constitute support for the existence of the language faculty.
　　Given the assumption that the language faculty underlies our ability to relate sounds and meaning, it seems reasonable to ask informants about the relation between sounds and meaning and consider the informants' reported judgments as evidence for or against our hypotheses about the language faculty. One should, however, naturally wonder how we can justify the use of informants' introspective judgments as crucial evidence, in light of the observation that the informant judgment can be unstable, especially when we consider "meaning." The present work proposes how informant judgments can constitute hard evidence in language faculty science, providing conceptual articulation of the claim and its empirical demonstration. This and the subsequent two chapters provide an overview of the proposed methodology for language faculty science.


Summary

　　The internalist approach we adopt to the study of "language" leads us to investigate properties of I-language, rather than E-language, in the sense of Chomsky 1986. Since the I-language of a speaker is internal to the mind of the speaker, our research deals with an individual informant. Since we are concerned with what underlies our ability to relate sounds and meaning, we have decided to work with an individual informant's judgment on the relation between sounds and meaning as it seems to be the most direct means to discover properties of our subject matter.^{FN 6}
　　While we deal an individual informant's judgment on the relation between sounds and meaning in a particular language, we want to make the individual informant's judgment as revealing as possible about universal properties of the language faculty. Our desire to seek as much generality as possible with regard to the judgments of an individual informant has led us to work with schemata. Schemata are schematic representations of sentences in terms of the linear precedence relations among the relevant expressions, sometimes aided by the use of the notation of containment for the sake of clarity of what is intended by the schema.^{FN 7} Once we focus on Schemata, we realize that our predictions are about an individual informant's judgment on sentences instantiating a *Schema and those instantiating an okSchema. Sentences instantiating each Schema are infinite, as long as the Schema is stated with a minimal degree of generality. They can be as simple as the Schema allows it but they can also be infinitely complicated as long as they instantiate the Schema in question. This leads us to recognize the fundamental asymmetry between the *Schema-based prediction and the okSchema-based prediction. We predict that any sentence instantiating the *Schema is unacceptable. We cannot, however, predict that any sentence instantiating the okSchema is acceptable. By making the sentence in question sufficiently complicated, one can make it unacceptable. It is therefore not possible to predict that any sentence instantiating an okSchema is acceptable. Any sentence that instantiates a *Schema, on the other hand, is predicted to be unacceptable, regardless of how simple we might make the sentence. We are thus led to recognize that the *Schema-based prediction is a universal prediction while the okSchema-based prediction is an existential prediction.
　　A combination of a *Schema-based prediction and its corresponding okSchema-based prediction is called a predicted schematic asymmetry. When the *Schema-based prediction has survived a rigorous attempt of disconfirmation and the corresponding okSchema-based prediction has been confirmed, the individual informant's reported judgments are said to constitute a confirmed predicted schematic asymmetry. I suggest that confirmed predicted schematic asymmetries are basic units of facts in language faculty science. Questions remain as to what counts as "a rigorous attempt of disconfirmation" of the *Schema-based prediction, how we can aspire to replicate the confirmed predicted schematic asymmetry that we have obtained in a single-informant experiment in a multiple-informant experiment, among other questions. Before I begin to address such questions, it is necessary to discuss what hypotheses give rise to our predictions, how that should affect the way we design our experiments and how we should interpret the experimental results. We will turn to those and related conceptual issues in the next two chapters.

FN 6: As noted earlier, we leave open the possibility that other types of evidence may serve the same purpose and provide converging evidence for our hypotheses. It must be made clear, however, as also suggested in Chapter 1 (see remarks around footnote 6 therein), how we can deduce definite predictions about it (at least in part) on the basis of hypotheses about the language faculty and how such evidence can be revealing about universal aspects of the language faculty.

FN 7Though schemata are not intended to express the hypothesized structures of the sentences in question, the schematic representation allows us to address the hypothesized structural relations among the parts that are specified in the schemata. In this sense, working with schemata (not with actual example sentences) can be considered as the first step toward working with the abstract mental (structural) representations expressed in terms of universal concepts and relations that underlies an individual speaker's ability to relate sounds and meaning.
***

The first paragraph and the last two paragraphs of Chapter 1 are as follows:

***
Once they have reached a certain maturational stage, every member of the human species is able to produce and comprehend the language to which s/he is exposed, barring any serious impairment. Underlying this ability of ours to relate linguistic sounds/signs (henceforth just "sounds" to make the exposition simpler) and meaning is the language faculty: this is one of the most fundamental working hypotheses adopted in the research program initiated by Noam Chomsky over half a century ago. The aim of Chomsky's research program is to discover the properties of the language faculty, in its initial state and in its steady state. It is hypothesized that in its initial state, the language faculty, as the genetic endowment that underlies our ability to relate sounds and meaning, is uniform across the members of the species and that, in its steady state where its non-trivial "growth" has stopped, it varies in accordance with one's linguistic experience, within the limit imposed by the genetic endowment. The research program is also concerned with how the universal properties in question might be related to laws that govern the nature, beyond the language faculty per se, and how the language-particular properties are acquired. This book's main concern is how hypotheses about the language faculty can be put to rigorous empirical test. I will propose how we can deduce definite/categorical and testable predictions, and illustrate how we test our predictions and how we can obtain experimental results that are very close to our definite/categorical predictions. In short, the goal of the book is to show that it is possible to pursue a study of the language faculty as an exact science in the sense just noted. I will refer to a study of the language faculty as an exact science in this sense simply as language faculty science.


　　There is an accompanying website, where the experimental designs and experimental results discussed in subsequent chapters are made available in more detail than in the book. The website is intended to make it possible for others to examine the validity of the book's empirical claims more thoroughly than is made possible in what follows, and hence, indirectly, the viability of its methodological proposal. The website provides, among other materials, "raw data" of the experimental results in the form of Excel files so that interested people can analyze them by the statistical techniques of their choice. The book and its accompanying website are meant to show, for the first time in my view, how it is possible to investigate the language faculty as an exact science in the sense noted above. Language faculty science thus turns out to be much closer to physics than to social and behavioral sciences, and this should have far-reaching implications for research that deals with other aspects of the mind. No other work in the literature claims that we can deduce definite and testable predictions about the judgments of an individual informant on the acceptability of sentences and expect them to be supported by experimental results. Nor is there any work in the literature, as far as I am aware, that proposes how to design experiments and interpret the experimental results so as to obtain robust experimental results in accordance with our definite and categorical predictions about the judgments of the individual informant as a reflection of universal properties of the language faculty.
　　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 am going to argue that it is indeed possible. The book's slogan is: language faculty science as an exact science is possible; yes, it is. Some may say that I am a dreamer. But I am not the only one. I hope upon reading the rest of the book some of the readers will join us.
***

The two threads under this have been moved from the "Past Postings" board.

I am adding postings to those threads, as can be seen by the dates of the postings.

I must have intended to make postings about specific issues addressed in each section of my 2003 (Lingua)"Falsifiability and Repeatability in Generative Grammar."

The level of my conceptual articulation of language faculty science as an exact science has reached a certain (I think minimally satisfactory) level, and I have also accumulated enough experimental results for its concrete illustration. And that is why I have been able to complete my book manuscript Language Faculty Science" is about to be finalized, and am in the process of finalizing it for publication.

It would therefore be an interesting exercise to go over what is in the 2003 Lingua paper from the perspective of my forthcoming book Language Faculty Science, for the purpose of critically assessing the paper and also for understanding how some basic methodological points made in the 2003 paper have been articulated in Language Faculty Science, with concrete empirical illustration on the basis of experimental results.

I have moved this thread to this board from the "Past Postings" board because I plan to address some of the issues addressed in the postings therein in relation to the proposal in my forthcoming book Language Faculty Science.

Under this, I will post what was once in earlier versions of the Lingua paper, in various forms, but are not included in the final version, some/many of which are direct quotations by works by others.

A passage from Chomsky, Noam. 1979 Language and Responsibility: Based on conversations with Mitsou Ronat, Panthen Books, New York, 73 is given below.

***
I should also mention work on history and philosophy of science, which has begun to furnish a richer and more exact understanding of the manner in which ideas develop and take root in the natural sciences. This work -- for example, that of Thomas Kuhn or Imre Lakatos -- has gone well beyond the often artificial models of verification and falsification, which were prevalent for a long time and which exercised a dubious influence on the "soft sciences," as the latter did not rest on the foundations of a healthy intellectual tradition that could guide their development. It is useful, in my opinion, for people working in these fields to become familiar with ways in which the natural sciences have been able to progress; in particular, to recognize how, at critical moments of their development, they have been guided by radical idealization, a concern for depth of insight and explanatory power rather than by a concern to accommodate "all the facts" -- a notion that approaches meaninglessness -- even at times disregarding apparent counterexamples in the hope (which at times has proven justified only after many years or even centuries) that subsequent insights would explain them. These are useful lessons that have been obscured in much of the discussion about epistemology and the philosophy of science.
***
It is easy to foresee objections to the main methodological claim in the Lingua paper, on the basis of the above passage. One might, for example, maintain that the serious (or possibly fatal) shortcoming of the Lingua paper is that it adopts an artificial model of falsification.
I will try to respond to such a (hypothetical) objection at some later point.

I should also mention work on history and philosophy of science, which has begun to furnish a richer and more exact understanding of the manner in which ideas develop and take root in the natural sciences. This work -- for example, that of Thomas Kuhn or Imer Lakatos -- has gone well beyond the often artificial models of verification and falsification, which were prevalent for a long time and which exercised a dubious influence on the "soft sciences," as the latter did not rest on the foundations of a healthy intellectual tradition that could guide their development. It is useful, in my opinion, for people working in these fields to become familiar with ways in which the natural sciences have been able to progress; in particular, to recognize how, at critical moments of their development, they have been guided by radical idealization, a concern for depth of insight and explanatory power rather than by a concern to accommodate "all the facts" -- a notion that approaches meaninglessness -- even at times disregarding apparent counterexamples in the hope (which at times has proven justified only after many years or even centuries) that subsequent insights would explain them. These are useful lessons that have been obscured in much of the discussion about epistemology and the philosophy of science.

In order to evaluate the above passage, we need to understand what is meant by (1).

(1)　a.　the often artificial models of verification and falsification
　　b.　the "soft sciences"

My guess is that what is meant by (1a) is the models by philosophers in which verification and falsification of a scientific hypothesis is to be done on the basis of some concrete observational facts, or something like that (this is the thesis of "logical positivism," I understand). I assume that what is meant by (1b) are inquires beyond the 'natural sciences' (such as social sciences, perhaps including linguistics, as it is practiced in much of the field -- with generative research not being exempted here, I would hate to say).

It might not be an overstatement to say that how one understands (2) below in the context of generative grammar, and more in particular in the context of assessing one's own research, may 'define' one's orientation as a researcher.

(2)　　[A]t critical moments of their development, [the natural sciences] have been guided by radical idealization, a concern for depth of insight and explanatory power rather than by a concern to accommodate "all the facts"― a notion that approaches meaninglessness ― even at times disregarding apparent counterexamples in the hope (which at times has proven justified only after many years or even centuries) that subsequent insights would explain them.

One might, for example, take (2) to mean that we should not be too concerned with, or impressed by, empirical observations since, after all, empirical observations alone never determine the fate of a scientific theory, anyway.

Let us try to compare the approach taken by those who try to do generative grammar as an empirical science (with progress in mind) (henceforth the GGES approach, abbreviating Generative Grammar as an Empirical Science) with a non-GGES approach, by raising the following questions for each of the two approaches.

(3)　a.　Are they guided by radical idealization?
　　b.　Are they guided by a concern to accommodate "all the facts"?
　　c.　Are they guided by a concern for depth of insight and explanatory power?
　　d.　Do they at times disregard apparent counterexamples?

We should also ask whether the answers to these questions for each approach would hold only "at critical moments of their development."
　　Re. (3a), it seems that our research is always guided by radical idealization and it is not only "at critical moments of [the] development [of our theory]," and this seems to apply to both approaches although there may be some difference in regard to the 'degrees'―I would think that the GGES approach tends to do more idealization but this point would need further elaboration, on the basis of actual examples, and I must return to this on a separate occasion. Re (3b), neither approach seems to be guided by a concern to accommodate "all the facts." So, neither (3a) nor (3b) seems to distinguish between the two approaches.
　　What about (3d)? Well, both approaches do seem to disregard apparent counterexamples at times. But how they do it might not be the same, and that might be one of the crucial differences between the two approaches. My Lingua paper (Hoji 2003) perhaps can be understood as providing a very rough illustration of the difference, if we consider the discussion in the earlier sections of the paper as an illustration of how apparent counterexamples are treated by GGES researchers. It must be noted that the apparent counterexamples discussed in that paper are not simply disregarded. While it is true that they are put aside without a theoretical characterization of them, a minimal descriptive characterization of those counterexamples (to the generalizations pursued there) is indeed offered, thereby making it possible to control the unwanted factors in conducting the syntactic experiments that are intended to probe into the nature of the speakers' linguistic intuitions that are hypothesized to arise from the language faculty proper.
　　It may be worth considering here what is meant by a counterexample. There are two types of empirical observations that go against what we claim. One type is against the generalization that we put forth, and the other type goes against the predictions of our hypothesis in the context of a wider theory in which the hypothesis is formulated. We may decide to ignore both types of counterexamples. But to ignore the former type would seriously undermine the empirical basis of our hypothesis, and we must be aware of that. If we do not have a way to characterize the former type of counterexamples in some way, we will not be able to conduct experiments to test our hypothesis; see the discussion on otagai in my Lingua paper.
　　For the latter type, I would be more sympathetic to, and I can perhaps agree with, what Chomsky states. But here too it is important to recognize the possibility that the empirical evidence in question (i.e., the results of the experiments that do not confirm the prediction) might ultimately lead us to reject the approach we may be pursuing at a given time. We should at least try to formulate the hypothesis and deduce predictions in such a way that we will in principle be ready to modify or reject our hypothesis if the predictions do not get confirmed. This seems to me to be the basic attitude we should have in constructing and formulating our hypothesis and predictions, and in carefully designing our experiments to test our hypotheses.

　　Notice that the content of the crucial notions in (3c) are quite obscure and can be very subjective. Stating or endorsing something like (4) should therefore give one cause for concern, to say the least.

(4)　　 "[A] concern for depth of insight and explanatory power" should guide our research more than a concern for how our predictions are borne out.

As long as we articulate our hypothesis rigorously enough -- in the context of a well-articulated theory -- we should be able to make definite and testable predictions; cf. severe tests in Mayo 1996 (Error and the Growth of Experimental Knowledge). By carefully designing an experiment to test the predictions, we should be able to tell, fairly objectively, whether or not the predictions are borne out. We would face a very different situation if we tried to determine how insightful a given hypothesis/theory is or how much explanatory power it might have.
　　If Chomsky (1979: 70-71 ) is claiming, in his remarks on C. Peirce¹, that we must have, and hence can rely on, some intuition about "correct theories," such a position would, in my view, come close to declaring that generative grammar cannot be an empirical science (with progress in mind). I suspect that Chomsky is comparing, in his remarks above, generative grammar on the one hand and structural linguistics and some behaviorist/empiricist-oriented research activities on the other.² Insofar as we understand his remarks that way, one might conclude that comparison between the GGES and non-GGES approaches is somewhat out of context here. I would however like to think that his remarks above and the brief discussion I have provided above might in fact be quite suggestive as to what Chomsky considers (his) generative grammar to be and what GGES aspires to do. Conspicuously missing in Chomsky's remarks above (and remarks here and there in Chomsky 1995: chap. 1, as well) are concerns about how to deduce predictions from our hypothesis, and more importantly, how to test those predictions. I would be inclined to think that "depth of insight and explanatory power" can be meaningfully addressed only when our research activities are accompanied by those concerns and the practices that reflect them. The last point must be elaborated on, and I will try to do that in a separate posting.

FN1: The relevant part is "Peirce argues that to account for the growth of knowledge, one must assume that "man's mind had a natural adaptation to imagining correct theories of some kinds," some principle of "abduction" which "puts a limit on admissible hypothesis," a kind of "instinct," developed in the course of evolution."
FN2: It may be the case that what Chomsky had in mind here was not just the structuralist/behaviorist approach in general but also generative semantics specifically.

[This directly follows the passage in [8348] in Poincaré 1952.]

　　Now, under what conditions is the use of hypothesis without danger? The proposal to submit all to experiment is not sufficient. Some hypotheses are dangerous,--first and foremost those which are tacit and unconscious. And since we make them without knowing them, we cannot get rid of them. Here again, there is a service that mathematical physics may render us. By the precision which is its characteristic, we are compelled to formulate all the hypotheses that we would unhesitatingly make without its aid. Let us also notice that it is important not to multiply hypotheses indefinitely. If we construct a theory based upon multiple hypotheses, and if experiment condemns it, which of the premises must be changed? It is impossible to tell. Conversely, if the experiment succeeds, must we suppose that it has verified all these hypotheses at once? Can several unknowns be determined from a single equation? Poincaré 1952: chap. 9, 151-152)

(1) If we construct a theory based upon multiple hypotheses, and if experiment condemns it, which of the premises must be changed? It is impossible to tell. Conversely, if the experiment succeeds, must we suppose that it has verified all these hypotheses at once?

It is interesting to note that something like (2) is commonly seen in the literature in generative grammar.

(2) In this work we have argued for (A). The argument is based on various theoretical assumptions that have been made in the literature, and if successful, it provides further evidence in favor of the assumptions. Thus, we have further evidence for (B), (C), and (D).

The author(s) of a passage like (2) might therefore respond to the second question in (1) as in (3).

(3) We do not say that it verifies all these assumptions/hypotheses at once. We only say that it provides evidence in favor of the assumptions/hypotheses.

Quite relevant to the discussion here is the "deceptively easy" part of footnote *3 in [8152], repeated in (4) below, without reintroducing the font formatting (see [8152]).

(4) *3 I now feel that I should have emphasized in this pace 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.

It perhaps requires some discussion before we assess the content of each of (1)-(4), but it seems to me that how one reacts to a passage like (2), and what position one takes in regard to the issues raised above would define, at least partially, what kind of research one wants to do and what type of researcher one wants to be.