And most of the work in generative grammar, at least in so-called generative grammatical studies of Japanese, seems to me to have abandoned the hope just noted. Otherwise, how would it be possible to maintain hypotheses whose negative predictions are so clearly disconfirmed, and so demonstrated in published works, and to keep on using them in a crucial way?

As noted in [28637], further discussed in the document mentioned there, what is most crucial is the recognition that there is a qualitative difference in significance between judgmental variation on (i) what is predicted to be impossible for a grammatical reason and (ii) what is not so predicted. The lack of understanding of this leads to a practice in the field, not uncommon unfortunately, to take it to be significant that some (or even many) speakers find [the examples that are predicted to be impossible due to some property of the Computational System] to be highly marginal. After all, mere complexity of a particular example under a specified interpretation might be a big enough factor to negatively affect the speaker judgment. Much more significant would be the speaker judgment that the examples that are predicted to be impossible are acceptable especially when the speakers in question consistently detect a predicted contrast in paradigms that are independent from but are crucially related to the paradigm in question; see Hoji and Ishii 2004: note 4.

"What Gets Mapped to the Tripartite Structure of Quantification in Japanese," (2004) Hoji & Ishii WCCFL 23. Downloadable at this HP.

When faced with such disconfirmation of the negative prediction in a crucial experiment, one might try to save the hypothesis by introducing an additional assumption, a common practice in natural sciences. "If we assume such and such, the unacceptability of the examples under discussion (with the specified interpretation) is no longer predicted and hence the fact that some speakers accept them would no longer pose a problem to the proposal under discussion. Unless the introduction of the new assumption leads to new discovery/insight presumably by making new negative predictions, this would be what we might call, slightly adapting Lakatos' (1978) terminology, a 'degenerative' move, rather than a 'progressive' move. The situation is actually quite serious if the negative prediction that appears to have been disconfirmed is the only negative prediction that the proposal under discussion makes and can be tested at the moment.

Miyagawa and Arikawa 2005: footnote 6 seems to provide a nice illustration of (i) the lack of understanding of the significance of negative predictions and (ii) a 'degenerative' move of the sort noted above.

Miyagawa and Arikawa 2005 LOCALITY IN SYNTAX AND FLOATED NUMERAL QUANTIFIERS. http://web.mit.edu/miyagawa/www/pdfs/FNQFINAL.pdf

Further discussion would have to address issues such as the following, and it will have to take place in the "Further Discussion" board instead of this board.

(iii) how to evaluate reports of different speaker judgments, in terms of their theoretical significance, repeatability, etc.
(iv) whether or not the proposal under discussion, e.g., that in Miyagawa and Arikawa 2005, makes any negative predictions other than the one that is being discussed in their footnote 6 for which a crucial experiment can be conducted.