There have been a couple approaches to the compositional semantics of de re attitude reports -- specifically, attitude reports where individual-denoting terms correspond to the speaker's way of identifying an individual rather than the ascribee's. One of these (cf. Cresswell and von Stechow 1982, Stechow 1982) thinks of (some) attitude verbs as tripartite functions taking individuals x and y and a property P and returning the proposition that x thinks de re of y that she is P (this sort of analysis obviously underdetermines what the correct theory of thinking de re of y that she is P might be).
In these sorts of accounts, the attitude relation combines directly with the term that's to be construed de re. This implies some sort of covert scope-shifting operation like LF movement which takes a term from inside the attitude verb's complement and makes it a sister of the attitude verb, cf. (1) ('~>' relates surface and LF representations, t gives the trace of the moved proper name, and \t is interpreted as lambda abstraction over t):
(1) Ralph thinks [he should arrest Ortcutt] ~> Ralph [thinks Ortcutt] [\t he should arrest t]It's known that this sort of movement violates well-motivated syntactic constraints on scope shifting -- in particular, it's not bounded by the embedded tensed clause -- but, so far as we know, it's always been thought to be at least *semantically* plausible. We argue that this is a mistake.
The argument is parallel to some which have been taken to show that LF movement can't be necessary for transparently-construed nominal restrictors (cf. Fodor 1970, Bäuerle 1983, and Percus 2000 -- though see Keshet 2008 for another take). Namely, variable binding shows that the de re term must be interpreted inside of the complement clause:
(2) Ralph thinks no female student likes her mother.Say Ralph has several pairs of pictures in front of him, each pairing a female student (on the left) with her mother (on the right), though Ralph doesn't know that -- he just thinks oh here's some pictures of girls paired with arbitrary women. Ralph also thinks something like the following: no left member of the pair likes the right member of the pair. In this scenario, (2) isn't hard to interpret as true.
Yet the LF-movement account can't readily generate this reading. Here's why: Since, for any left member of a pair x, x's mother (NB: I'm quantifying into the quotation marks) doesn't correspond to Ralph's way of identifying the right member of that pair, either her or her mother must be construed de re'. On the LF-movement theory of de re ascription, this entails representations like (3) or (4):
(3) Ralph [thinks [her mother]] [\t no female student likes t]The problem with (3) and (4) is that the pronoun is no longer c-commanded by the quantificational phrase no female student. Given the standard sorts of rules for interpretation that we assume (cf. Heim and Kratzer 1998), this means the pronoun cannot be bound and must be interpreted referentially. In other words, for constructions like (2) the LF movement account predicts--contrary to fact--that the only way to construe the pronoun her or pronominal DP her mother 'de re is to give up the bound interpretation (where mothers vary with students).
(4) Ralph [thinks her] [\t no female student likes [t mother]]
One possible way to resist this argument is to deny that the quantificational phrase no female student must stay low. But, importantly, scoping it out of the property-denoting clausal complement (so that it can take scope over the moved de re DP) misplaces its quantificational force: the reading we're interested in guarantees that all of Ralph's doxastic alternatives w'; are each such that none of the girls on the left (i.e. the actual female students) likes her pair-mate on the right in w'. For this to obtain, no female student must be interpreted inside the clausal complement, which entails that it stays low.
One might also try maintaining that mother is construed de re but that the genitive pronoun remains in the scope of the quantifier--i.e. Ralph thinks no female student likes her right pair-mate, where the right-pair-mate-of function is contextually equivalent to the mother-of function. It's not totally clear how this would go, but even if it was made to work, it'd be insufficiently general: the point is replicated with examples which replace her mother with the anaphor herself. Absent some lexical decomposition of the anaphor, the de re nominal solution can't apply here.
Drawing on Percus & Sauerland (2003) and Anand (2006), we explore an alternative in situ theory of de re ascription. This account makes use of Percus & Sauerland's *concept generators*, silent lexical items whose semantics generates de re ascriptions without resorting to covert movement of the de re term. The LFs we end up positing look (very roughly) as follows, with G a variable over concept generators:
(5) Ralph thinks [\G no female student likes [[G her] mother] ]
Here, the pronoun stays in the c-command domain of no female student, which allows it to get bound. This sort of theory predicts, correctly, that sentences like (2) can have readings where (a) a quantifier scopes inside of the complement clause but nevertheless (b) binds a de re' pronoun/pronominal DP. Put slightly differently: sentences like (2) yield the first properly semantic argument that de re ascription is better handled with compositional mechanisms which allow the de re term to remain in situ (e.g. concept generators) than with scope-shifting.