Analyzing 'genitive/accusative + V-ing phrase (gerund-participle phrase)' as different constructions

(1) I regretted [his leaving the firm].

(2) I regretted [him leaving the firm].

(3) I regretted [leaving the firm].

(4) He didn’t bother [giving me a copy].

Regarding the above sentences The Cambridge Grammar of the English Language (Page 1190) has this to say:

If (1) and (2) are analysed as quite different constructions, with only (2) a clause, then which of the constructions would (3) belong to?

This problem would be particularly difficult to resolve with those gerund-participials where it is not possible to include an NP before the verb, as in (4). We avoid these problems by treating the optionality of the initial NP as simply a matter of the optionality of subjects in non-finite clauses.

Here, CGEL is basically arguing that the bracketed construction in (1) is no less a non-finite clause (with his as its subject) than that in (2) is (with him as its subject).

So, CGEL is basing this argument on the presumption that the bracketed portion in (2) is a non-finite clause. But I wonder why that has to be the case.

PROBLEM of CGEL’s APPROACH

CGEL’s approach cannot explain the potential semantic difference between (1) and (2), as explained in Merriam-Webster Dictionary of English Usage (as quoted in this Language Log):

The accusative pronoun is used when it is meant to be emphasized.

Because CGEL’s approach analyzes (1) and (2) as the same construction only with some difference in register (formal vs. informal), I think it fails to accommodate the semantic difference shown above.

SUGGESTED APPROACH

What if we considered the verb ‘regret’ as taking two complements in (2), one being him and the other being leaving the firm, where the former is construed as the semantic–but not syntactic–subject of the latter?

In this approach, him in (2) would be a raised object of the verb ‘regret’, whereas the verb ‘regret’ in (1) would be analyzed as taking only one complement, a non-finite clause shown in the bracketed portion.

Then, (1) and (2) would be "analysed as quite different constructions".

This way, there would be no "problem" analyzing (3) or (4).

More importantly, the suggested analysis treats (1) and (2) as different constructions, thereby possibly accommodating the semantic difference quoted in the Language Log (shown above).

QUESTION

I’d like to know what others think of this suggested approach vis-à-vis CGEL’s, and if any existing grammar employs something like the suggested approach.