In English wh raised from, or in situ in, a direct object or prepositional object, you can almost always use "who" at least as well as "whom",1 and in some cases you can only use "who":
- Who/whom did you meet?
- Who/whom were you talking to?
- To who/whom were you talking?
- I asked her who/whom to talk to.
- I asked her who/?whom she met.
- I asked her who/whom she was talking to.
- What did you give to who/whom?
- You gave it to who/*whom?!
- I already told you about what to talk about to who/whom.
- I already told you about what to talk about, to who/whom.
- I already told you about what to talk about, and to who/whom.
But there seems to be at least one case where whom is actually required: when pied-piping a to-adjunct out of a relative clause, or at least an infinitive one:
- I already told you about the man to *who/whom to talk.
Is there any principled reason for this one construction to be different from all the others?
More specifically, is there any way to account for it in mainstream generative grammar?
As I understand it, all of the examples above are usually explained by the fact that "who" is marked with (abstract) accusative/ACC or oblique (depending on your favorite flavor of Case theory), but modern English allows realizing accusative "who" as either "who" or "whom", or similar. But then why does it not allow it in this one sentence?
Or, alternatively: In a construction grammar it's easy to account for: the pied-piped infinitive relative clause construction just overrides whatever is inherited from the parent construction(s), so that it requires "whom".
But such an unmotivated idiosyncrasy should eventually die away.2 And "whom" has had a long time to do so, and all the other uses of whom are well on their way, but this construction seems to be stubbornly insisting on it without exception, even today. So, why?
My best guess—which seems like a pretty bad guess—is the very minor phonological infelicity of "to who to", which doesn't appear in any of the other examples.
1. At least by actual speakers of English—even in formal writing. Most prescriptive grammar writers claim it's incorrect, but then they also mandate hypercorrecting to "whom" even in instances that are clearly nominative, like the infamous "Whom shall I say is calling?" And of course most of them don't allow wh-raising without pied-piping in the first place, so half of these examples can't even arise for them.
2. Also, of course any CxG framework will leave room for a few nuts that just don't crack right, because languages don't have to be perfect. But that's a last resort fallback, not the default assumption.