6. Evaluating the papers collected in this volume in light of Hoji 2015
6.1. Introduction
　　It would be important and useful to evaluate the papers collected in this volume in light of the methodological proposal advanced in Hoji 2015. For each paper, we can ask whether and how it makes a definite and categorical prediction. In the terms of Hoji 2015, we can ask whether it offers a predicted schematic asymmetry, and if it does, what universal and language-particular hypotheses give rise to it. We can also ask whether the prediction is experimentally supported, i.e., whether we obtain a confirmed predicted schematic asymmetry in the terms of Hoji 2015. To put it in somewhat concrete terms, whenever we see an example sentence that is claimed or assumed to be unacceptable (with the specified interpretation), we can ask the questions in (12)-(14).

(12)The fundamental schematic asymmetry:
a.What is the *Schema that the example sentence in question instantiates?
b.What is the corresponding okSchema?

(13)The prediction-deduction:
What universal and language-particular hypotheses make the *Schema and okSchema in (12) a *Schema and an okSchema, respectively?

(14)Experimental results:
a.Does the *Schema-based prediction survive a rigorous attempt at disconfirmation? That is to say, is any sentence that we can construct instantiating the *Schema completely unacceptable (under the specified interpretation), no matter how hard we try to make it acceptable?
d.Is the okSchema-based prediction confirmed? That is to say, can we construct a sentence instantiating the okSchema that is more or less acceptable (under the specified interpretation)?

　　Trying to answer such questions would be a useful exercise for the purpose of evaluating a given paper with regard to its potential contribution to, and/or its relevance to, language faculty science. Even apart from the issues regarding language faculty science, addressing such questions will help us understand what testable predictions are made with what hypotheses, how explicitly each of those hypotheses is formulated, as well as what is assumed to be a valid generalization and whether it indeed qualifies as a confirmed (predicted) schematic asymmetry.

6.2. General remarks
　　I will now make brief, and not particularly systematic, remarks on the papers collected in this volume from the perspective of Hoji 2015. The discussion is not intended to be self-contained because it not possible to fully illustrate here the methodology for language faculty science proposed in Hoji 2015. I would like to refer the reader to Hoji 2015. The accompanying website (http://www.gges.xyz/hojiCUP/) provides information about Hoji 2015, including the designs and the results of every Experiment discussed in Hoji 2015.
　　In the terms of Hoji 2015, Hoji 1985 tried to identify as good probes as possible in discovering the universal properties of the language faculty through the investigation of Japanese, and used the probes thus identified to argue for the thesis that the Japanese phrase structure is strictly binary branching. Clearly, I was not thinking in those terms when I wrote Hoji 1985. But this now seems to me to be a reasonable interpretation of what I was trying to do in Hoji 1985.
　　It may be interesting to note that there seems to be a general tendency in the generative tradition that when one works on a language other than English, one addresses generalizations in her/his language in relation to what seem to be analogous generalizations in English. Of course, there is nothing inherently wrong with comparing two or more languages. But if the comparison or analogy is based on shaky empirical grounds, it is unclear what genuine insight we can expect to obtain about what formally underlies the intuitions of the speakers of the different languages, i.e., about the aspect of the language faculty responsible for the speaker intuitions under discussion. One might suggest that loosely "established" "generalizations" in a number of languages can lead us to an insight into general properties of language. It is unclear, however, how one can pursue rigorous testability in such research if, as discussed in Hoji 2015, rigorous testability is closely related to the deduction of definite predictions and experimental testing of the predictions.
　　If one "analyzes" a certain linguistic "phenomenon" in Japanese, for example, as being analogous to a phenomenon in English which has been characterized in terms of highly theoretical notions, one's long-term contribution depends in part upon (i) how robust the alleged generalization is and (ii) how aspects of the theoretical account of the generalization are motivated independently of the "phenomenon" at issue. The parasitic-gap analysis in Hoji 1985: Ch. 2 of what would later be called the A-Scrambling construction in Japanese is a good example of making an analogy of some loosely understood "phenomenon" in a non-English language to a phenomenon in English that is analyzed in highly theoretical terms. If the descriptive generalization in English itself is not as robust as one wishes it to be, and if aspects of the theoretical characterization of the phenomenon in question are not independently motivated on empirical grounds, the "theoretical characterization" of the loosely understood phenomenon in Japanese is bound not to survive the test of time.
　　Suppose that one's theoretical characterization of a phenomenon in Japanese were based on a solid empirical and experimental basis, in the form of confirmed predicted schematic asymmetries in the terms of Hoji 2015. That would mean that we have a prediction in the form of a predicted schematic asymmetry that is deduced from universal and language-particular hypotheses and that we have obtained experimental results precisely in accordance with our prediction. That would in turn mean that we now have an empirical basis that is almost entirely independent of a particular conception of grammar; see Hoji 2015: Chapter 3: note 33 and the discussion in the text thereabout). If the theoretical characterization of the phenomenon in English changes (over time), we would therefore be in a good position to check the empirical consequences of the theoretical change; hence we might be able to tell whether or not the change in question is sound and progressive in the terms of Lakatos 1970.
　　Rigorous testability can be pursued regarding our hypotheses only if our hypotheses are about something that has definite and categorical properties. Likewise, in order to pursue rigorous reproducibility (among informants as well as within an informant), we must deal with something that has definite and categorical properties. This is certainly true if we want to deduce definite predictions about the individual informant from our hypotheses about the language faculty.
　　In the terms of Hoji 2015, Papers 1-7 tried to identify as good probes as possible in discovering the universal properties of the language faculty through investigation of Japanese, and used the probes thus identified to argue for various hypotheses about Japanese phrase structure. If we take FD as our object of inquiry, we can try to determine what might be a good probe for investigating the properties of FD for a given informant, and for a given experimental set-up. This is the shift from (i) analyzing linguistic phenomena in terms of theoretical concepts to (ii) studying the nature of a theoretical (i.e., hypothesized) object by means of linguistic phenomena. Hoji 2015 articulates a conceptual and methodological basis for how we can do the latter and expect our predictions to be supported empirically. It also provides experimental demonstration for the viability of the proposed methodology.

6.3. Paper 1
　　From the perspective of Hoji 2015, the testability of one's research on the Binding Theory can be attained only if we can specify how co-indexation is related to the interpretation detectable by the informant. Furthermore, rigorous testability can be pursued only if we can identify, independently of binding-theoretic considerations, what expressions in the language in question have the [+anaphor] feature or the [+pronominal] feature. It is also imperative that we try to motivate the structural properties of a particular language under discussion that the binding conditions make crucial reference to, again independently of binding-theoretic considerations.

6.4. Paper 2
　　In accordance with the methodology proposed in Hoji 2015 for language faculty science, an okSchema alone, or the confirmation of an okSchema-based prediction alone, does not constitute a fact in language faculty science. An okSchema would be part of a fact in language faculty science only if it is combined with the corresponding *Schema, and only if we obtain a confirmed predicted schematic asymmetry―i.e., only if the *Schema-based prediction has survived a rigorous attempt at disconfirmation and the corresponding okSchema-based prediction has been confirmed. See Section 5 for a brief discussion of the proposal in Hoji 2015, which include various notions introduced here.
　　If we do not have a confirmed predicted schematic asymmetry involving the NOC, we do not (yet) have a fact to explain, according to Hoji 2015. The inclusion of Sections 3 and 4 in Paper 2 is due to the lack of a clear understanding of the significance of the *Schema-based prediction and, I might say, also due to the fact that I was still a linguist then, not a language faculty scientist. When a linguist demonstrates that an alleged generalization is not valid by showing that there are acceptable *Examples instantiating the *Schema that is part of the alleged generalization, s/he often encounters a reaction like the following: "Okay. You have shown that the generalization under discussion is not valid. But there remains a contrast, at least to some degree, between the *Examples and the corresponding okExamples that originally motivated the generalization. You have shown that the generalization has exceptions and you have concluded that the hypotheses that accounted for the generalization should not be accepted as they stand. But, what is your alternative account of the tendency that the generalization in question points to?" The linguist tends to feel compelled to respond. The inclusion of Sections 3 and 4 in Paper 2 seems to me to be largely due to the tendency of a linguist responding in a context like that.
　　Papers 1-6 occasionally contain discussion of English paradigms. They are often used as a basis of the Japanese paradigms. One might suggest that the inclusion of the discussion of English paradigms in Papers 1-6 were prompted by the absence of the rigorous testability-seeking research orientation and the lack of the strong internalist commitment; see the remarks above regarding the "comparative research" as practiced in Hoji 1985 (in reference to its "parasitic gap" analysis). That seems to me to be a basically accurate characterization of the research orientation pursued in Papers 1-6. But the remark at the end of Paper 2: Section 1, "…in what follows I will refer mostly to O&W [which deals with Japanese] rather than to Huang 1988, 1991, only because I cannot evaluate the relevant data in Chinese in the way I have been able to evaluate the relevant data in Japanese," indicates that I already had the internalist inclination although I was not as committed to it as I am now and I did not know at that point how to try to pursue rigorous testability in research that deals with language or the language faculty.

6.5. Papers 3 and 4
　　Paper 3 and Paper 4 deal with the sloppy-identity reading and argue for the various points including those mentioned in (2). The two papers address the sloppy-identity reading and the Mix-reading pattern in "ellipsis constructions" in relation to the points listed in (2).
　　From the perspective of Hoji 2015, I should note that there is inherent difficulty in designing an experiment dealing with the sloppy-identity reading. Even if we deal with the simple cases of the sloppy-identity reading―e.g., cases that do not involve the Mix-reading pattern―it will be significantly more difficult to design an experiment on the sloppy-identity reading than on the BVA case. The reason has to do with how our experiment, which necessarily consists of a Main-Experiment and its Sub-Experiments, is to be designed. Our Experiments test a predicted schematic asymmetry. A predicted schematic asymmetry is given rise to by a set of hypotheses. A crucial part of the predicted schematic asymmetry is how the phonetic sequence (ps) in question is "represented" in the mind of the informant―more precisely, in what we call the LF representation corresponding to the ps. The crucial part of the consequences of the hypotheses is thus what condition(s) is/are or is/are not satisfied in the LF representation in question.
　　The Experiment involving the sloppy-identity reading necessarily involves at least two sentences, and the notion of parallelism is crucially involved. We, therefore, have to have hypotheses, not only about the LF representations corresponding to each of those sentences but also about how the two LF representations are related by the informant, especially in relation to the parallelism in question. This makes it qualitatively more difficult to design an effective experiment when dealing with the sloppy identity reading than when we are dealing with a single sentence.
　　Furthermore, when dealing with a single sentence, we can conduct a single-researcher-informant experiment and check various sentences with non-sense words and get very clear judgments despite (and sometimes because of) that. That, however, seems very difficult (if not impossible) to do if we dealt with the sloppy-identity reading because, as noted above, the sloppy-identity reading is crucially related to the notion of parallelism holding between two sentences and it is difficult to determine the parallelism in question without having some pragmatic context specified. The Mix-reading pattern poses an additional problem because of the complication of the relevant judgments and also because of the lack of hypotheses that can serve as a basis for designing an effective experiment which consists of a Main-Hypothesis and Sub-Hypotheses.
　　In order to be able to use the sloppy-identity reading as a good probe into the properties of the CS, we must therefore have a minimal articulation of how we can design an experiment dealing with the sloppy-identity reading that consists of a Main-Experiment and its Sub-Experiments, and how the result of the Main-Experiment is to be interpreted on the basis of the results of its Sub-Experiments. It is hoped that the articulation provided in Hoji 2015 will serve as a good basis for our future research in this domain and help us obtain confirmed predicted schematic asymmetries dealing with the sloppy-identity reading.

6.6. Paper 5
　　The main concern of Paper 5 is the three conditions on FD, two structural and one lexical. In the terms of Hoji 2015, the probe used in Paper 5 for testing the hypotheses in question was BVA. As discussed in Hoji 2015 and also in Paper 7, the choice of LG (i.e.  and  in BVA(, )) affects the effectiveness of the probe for a given informant. What is predicted is not about the individual informant's judgments on specific Examples of instantiating a particular Schema, but it is about the correlations of the individual informant's judgments "across" the three conditions and "across" different LGs (and in some cases "across" different SGs). The significance of the correlation of judgments is addressed in Paper 7. But in the paper, it is not articulated how we can obtain correlations of judgments in the terms of confirmed predicted schematic asymmetries.
　　Hoji 2015 offers experimental demonstration of the correlation of judgments across the lexical condition and the LF c-command condition on FD. Hoji in preparation tries to address the correlation of judgments across more dimensions, including the anti-locality condition on FD, as well as different LGs and different SGs, providing further support for the claim made in Paper 5.

6.7. Paper 6
　　By focusing on the sloppy-identity reading in surface anaphora, rather than in deep anaphora, Paper 6 tries to deal with something that has definite and categorical properties, i.e., the Computational System of the language faculty. FD is hypothesized as a formal object underlying the sloppy-identity reading observed only in surface anaphora. A number of operational tests were applied to various "ellipsis constructions."
　　It seems safe to say that the main concerns of Paper 6 were with the logical issue of testability. The relevant hypotheses were tested in a single-researcher-informant experiment (with myself being the informant) and in multiple-researcher-informant experiments of a rather limited scale, in the terms of Hoji 2015. Once one tries to design an experiment to test the empirical predictions made in Paper 6 with regard to the Mix-reading pattern, one understands that it would be quite challenging to design a Main-Experiment and its Sub-Experiments dealing with it and attain a confirmed predicted schematic asymmetry in the Main-Experiment in a multiple-informant experiment. The difficulty in question is directly related to the difficulty regarding what universal and language-particular hypotheses lead to definite and testable predictions about the individual informant's judgments about the Mix-reading pattern. When I started conducting on-line experiments in 2004, as an attempt to see how we can replicate robust judgments among informants, I decided not to deal with the sloppy-identity reading because of the additional complications that such an attempt would invoke.

6.8. Paper 7
　　Paper 7 adopts Ueyama's (1998, 2003) analysis of the so-called Scrambling, i.e., OSV, in Japanese. The paper, however, discusses only a portion of the empirical consequences discussed in Ueyama' 1998, 2003. For example, it does not address 'multiple-scrambling' and 'long-distance scrambling' in any depth; but see Paper 7: footnotes 30 and 84 for brief remarks. As noted in the preceding pages, I had come to realize by the time of preparing Paper 7 that we must focus on the *Schema-based prediction to pursue rigorous testability and reproducibility. Ueyama's analysis, like other analyses in the field, however, does not give rise to a *Schema-based prediction if we limit our discussion, as we do in Paper 7, to the simplex OSV, i.e., the OSV without involving an embedded clause or without involving the multiple "Scrambling." To make a *Schema-based prediction with regard to the simplex OSV, I addressed "resumption" in Paper 7.
　　As noted in Section 4, the period between Paper 7 and Hoji 2015 was a slow process of my realizing (3), repeated here.

(3)a.If we want to pursue rigorous testability, we should be engaged in a study of the language faculty rather than language or languages.
b.In language faculty science, so-called linguistic phenomena are not the object of our investigation; rather, they are probes in our investigation of the properties of the language faculty.
c.Being concerned with the language faculty as our object of inquiry, we must be an internalist.
d.Being an internalist, we should be concerned with making and testing predictions about individuals.

After Paper 7, I started my attempt to obtain reproducible experimental results in accordance with the various empirical generalizations presented in Paper 7, initially in a multiple-researcher-informant experiment of a rather limited scale and then in a multiple-non-researcher-informant experiment. Although I was at one point working with the average among a group of informants, as in Hoji 2006a, 2006b, and 2010, the subsequent recognition of (3c, d) led me to focus on the reported judgments by individual informants, eventually leading me to the methodological proposal in Hoji 2015.
　　As suggested, Paper 7 paved the way to Hoji 2015 in terms of my conceptual understanding of what to pursue in language faculty science as an exact science and also in terms of what would serve as an empirical and experimental illustration of the methodology for language faculty science. There is a qualitative difference between Paper 7 and Hoji 2015 in terms of my understanding and articulation of various conceptual issues. The substantial difference between Paper 7 and Hoji 2015 in terms of the level of my understanding and articulation of various issues, however, makes it difficult to present an assessment of the methodological contribution of Paper 7 from the perspective of Hoji 2015. I would like to try to do so on a separate occasion.

6.9. Summary
　　During 1985-2015, the concern and the focus of my research have slowly shifted, eventually leading to Hoji 2015. During the research that led to the papers collected in this volume, I came to be increasingly concerned with methodological issues, as indicated by the titles of Papers 6 and 7. The shift can perhaps be characterized as being from compatibility-seeking to testability-seeking research.
　　One might point out that compatibility-seeking and testability-seeking are not mutually exclusive. One can test the degree of compatibility. The difference between the two I intend here has to do with whether one deduces definite predictions and aspires to obtain definite experimental results in accordance with the definite predictions. What is referred to here as compatibility-seeking research does not aspire to do so. It typically proceeds based on rather loose compatibility among various observations, a collection of which is regarded as constituting a generalization, and on a rather loose sense of compatibility between such "generalizations" and the theory under discussion (which in turn is often rather loosely formulated). It typically addresses "predictions" that have not been deduced from hypotheses in a rigorous fashion, and the formulation of their hypotheses is typically independent of whether the hypotheses lead to definite and testable predictions.
　　The difference between testability-seeking research and compatibility-seeking research can also be understood in relation to what is typically considered as supporting evidence (for hypotheses) in each type of research. The testability-seeking research tries very hard to look for ways in which its hypotheses can be shown to be invalid. What constitutes evidence in support of its hypotheses under the testability-seeking research is the definite prediction made under the hypotheses having survived a rigorous attempt at disconfirmation. In order for a given hypothesis to have the chance to receive empirical support, it must be possible for the hypothesis, in conjunction with other hypotheses, to gives rise to a definite prediction. When a hypothesis is put forth, one of the first questions to be considered is, therefore, how it can be put to rigorous empirical test, i.e., how its validity can be tested experimentally. Under this approach, the formulation of hypotheses and even the choice of the specific research topic are severely limited by the desire to seek testability and the desire to deduce definite predictions from hypotheses.
　　The compatibility-seeking research, on the other hand, does not make (rigorous) attempt at disconfirmation of the predictions made under its hypotheses. Instead, it typically seeks confirming evidence for its hypotheses. What constitutes confirming evidence depends in part upon how rigorously one carries out one's research. But it may be the confirmation of an okSchema-based prediction in the terms of Hoji 2015 or the identification of some pairs of Examples that seem to exhibit a contrast in the direction of what is suggested, though not necessarily deduced, by the hypotheses in question, and often despite the fact that the contrast does not constitute a confirmed predicted schematic asymmetry in the terms of Hoji 2015. There are a number of issues that deserve serious discussion in relation to this, but I cannot pursue the discussion further in this essay.