[This is directly related to the issues addressed in my JK paper; see the Passive board.]

(i) a. disconfirmation
b. confirmation
c. corroboration
d. falsification

As I understand it, Popper's use of the terms is as follows. Suppose a hypothesis H makes a testable prediction P. P gets confirmed or disconfirmed. If confirmed, the result corroborates H. If disconfirmed the result falsifies H.

Turning our attention to a negative prediction P (as discussed in my JK13 paper as well as in the WCCFL2004 handout (and the paper, to be placed here soon)) and a hypothesis H that gives rise to P, we can use (i) in the same way as above.

Now, if the factor that crucially contributes to P is removed, we no longer predict P, hence whatever was predicted to be impossible, say G, is no longer predicted to be impossible. Suppose G is indeed found possible in such cases. Since H does not predict G, this result does not corroborate H, according to the way Popper uses the terms in (i), as pointed out to me by A. Ueyama (p.c. May 2004). But I was using corroboration in such a way that this result corroborates H. I was in fact not using corroboration when P gets confirmed... I noticed this, I think for the first time, when one of the students in EALC 547 used corroborate in the 'standard Popperian' way.

As A. Ueyama (p.c. May 2004) pointed out, it is perhaps advisable to keep to the standard uses of these standard terms. If we do, how would we call the significance of obtaining G as the result of removing the factor that crucially contributes to a negative prediction P? We may use the term plausibility enhancement.

Incidentally, it is perhaps the case that plausibility enhancement is not needed in, and is not part of, the Popperian discourse, and here we are perhaps looking at a fundamental difference between (i) a natural science and (ii) a linguistic science (well, to be more precise, that part of linguistic science where 'interpretations' (i.e., the speaker intuitions about the 'meanings') of linguistic forms constitute the core part of the data to account for).

One of the most serious problems in the practice in generative grammar at the moment, it seems to me, is that linguistic intuitions that only have the significance of plausibility enhancement are often erroneously taken to be corroboration for the theory in question.