A fundamental question about 'subject-to subject raising'

Question

Raising is defined in Oxford Dictionary of English Grammar as follows:

The displacement of a noun phrase from a position within an embedded clause to a position in a higher clause.

...

With subject-to-subject raising we have a grammatical subject that carries a semantic role that is associated with a verb in a lower clause. Thus in

Henriette seems [ _ to like Paul]

the subject of the matrix clause (Henriette) is said to have been raised out of the bracketed clause (from the position indicated by ‘_’), and its semantic role, that of experiencer, is linked with the verb like, not with the verb seem (the meaning is ‘It seems that Henriette likes Paul’).

(Boldface mine.)

I specifically question the validity of the portion in bold, which can be expanded to:

Semantically, the subject Henriette is not linked with the verb seem.

But is this really true?

The meaning of seem is defined in Oxford Dictionary:

1 Give the impression of being something or having a particular quality.

[with complement] ‘Dawn seemed annoyed’

According to the definition, you can easily say that Dawn gives the impression of having a quality of being annoyed. Likewise, you can as easily say that Henriette gives the impression of having a quality of liking Paul. Therefore, the subject Henriette is semantically linked with the verb seem in Henriette seems to like Paul.

Is the emboldened portion in Oxford Dictionary of English Grammar false? If so, is subject-to-subject raising somehow a wrong concept at least when it comes to the verb seem?

I know that Cognitive Grammar treats Raising constructions as "just a special case of control constructions" but not for the reason set forth in this question. If there's a grammar or a linguistics theory that does without Raising constructions for the above-mentioned reason, please let me know.

EDIT

In response to @Tim Osborne's answer, here are some follow-up questions too long for comments:

a. How does the embedded infinitive semantically selecting the subject necessarily mean the matrix seems exercising no semantic influence over the subject? What's wrong with saying that the semantic role of the matrix subject is linked with the matrix verb as well as the embedded infinitive?

b. Sentences (1) and (2), though semantically equivalent, are not syntactically equivalent. Why do we have to even look at sentence (1) in an effort to reject that seems semantically selects Henriette, when I've shown straightforwardly that seems does semantically select Henriette in (2), which is the only sentence we're analyzing?

c. As for (6) and (10), could, or even should, their semantic anomalousness be explained not in terms of seems having no semantic control over the matrix subject but in terms of the embedded infinitives being semantically incongruous with the matrix subject as clearly shown in (5) and (9)?

d. Is there any single case in English in which a simple sentence without a superordinate verb is semantically deviant at the same time that the same sentence with an added superordinate verb is semantically good? If not, how could I, and why should I, demonstrate that adding seem as a superordinate verb can fix a semantically deviant subordinate clause?

Answer 1

The part in bold from the Oxford Dictionary of the English Language is, in my view, more right than wrong. It is more right than wrong because it can be demonstrated beyond a doubt that the embedded infinitive is semantically selecting the subject noun phrase at the same time that the matrix seems exercise no semantic influence over the subject. First, though, note that some raising verbs like seem allow the it-extraposition construction (as mentioned in the Oxford Dictionary):

(1) It seems that Henriette likes Paul.

We see in this case that there is no way to view seems as semantically selecting Henriette in such a sentence. It is, rather, more accurate to view seems as semantically selecting the proposition expressed by the embedded finite clause as a whole. With this insight in mind, rejecting the notion that seems semantically selects Henriette in the sentence

(2) Henriette seems to like Paul.

becomes more plausible.

There are additional reasons for rejecting the possibility that seems might be semantically selecting Henriette in (2). Consider the following contrast:

(3) Fido the dog barks a lot.

(4) Fido the dog seems to bark a lot.

(5) #Paul the man barks a lot.

(6) #Paul the man seems to bark a lot.

The presence of seems does nothing to change the fact that it is bark that is semantically selecting the subject noun phrase. Another, similar example:

(7) That it rained is obvious.

(8) That it rained seems (to be) obvious.

(9) #That it rained is difficult.

(10) #That it rained seems (to be) difficult.

We again see that the matrix predicate, obvious or difficult, is necessarily responsible for semantically selecting the subject, which is a finite clause in these cases. The presence of seems has no impact on this necessity.

The question is suggesting that the raising verb seems might also in fact be responsible for semantically selecting the subject nominal, contrary to received wisdom. I think it is incumbent on the one advocating for such an understanding of raising verbs like seem to demonstrate that there are cases in which the simple sentence without seems is semantically deviant at the same time that the same sentence with seems is semantically good. I do not think it is possible to do that, though.

Where I can perhaps agree with the subtext in the question concerns the role of an actual approach that assumes that the raising phenomenon really involves ‘raising’ in the literal sense, i.e. movement up the tree. Any theory of syntax that rejects transformations and movement is going to produce an analysis of raising that rejects the notion that raising is real. Thus, HPSG or LFG approaches certainly pursue non-movement accounts of raising phenomena.

---

Additional points

In response to the additional points raised in the question in reaction to my answer above, more insights and data are now produced here. These insights revolve around the distinction between control and raising. Essentially, the subtext of the question is misconstruing raising predicates as control predicates. Consider a typical control predicate:

(11) Evergrande's stock price is going down again.

(12) #Evergrande's stock price is attempting to go down again.

The contrast here is due to the fact that attempting is semantically selecting the subject. It requires its subject valent to be capable of volition. Compare this situation with that of a true raising predicate:

(13) Evergrande's stock price seems to be going down again.

A similar data set is next:

(14) The music has been turned up.

(15) #The music refuses to be turned up.

(16) The music seems to have been turned up.

What these examples show is that refuses semantically selects its subject valent, whereas seems does not. One last data set:

(17) There are two men at the door.

(18) #*There promises to be two men at the door.

(19) There seems to be two men at the door.

The raising verb seems is exercising no semantic influence over the subject valent, whereas the control verb promises is.

To summarize, the subtext in the question is suggesting that raising verbs are in fact control predicates, which they clearly are not.