Peter Szűcs University of Debrecen On the typology of control-constructions from an LFG-perspective In Lexical-Functional Grammar (LFG), “control” is a general term which covers grammatical constructions where a main clause constituent also fills a grammatical function in a (usually non-finite) embedded clause. So it covers both cases where this main clause element is an argument of the matrix verb (this is what is called “control” in the GB/MP tradition) and cases where it is not. The labels which LFG uses for these are inherited from the chomskyan tradition: “equi” and “raising” respectively. These terms solely serve to name the constructions, and do not mean a commitment to any particular analysis. An example for equi (1a) and raising (1b) are shown below: (1) a Michael agreed to quit. b Michael seemed to quit. Traditional tests to tell the two types apart are alternation with an expletive variant and the availability of idiomchunks in them. Equi fails these tests, while raising passes them. (2) a *It agreed that Michael quit. b *The cat agreed to be out of the bag. (“the secret was revealed”) (3) a It seemed that Michael quit. b The cat seemed to be out of the bag. In the standard LFG analysis of equi, the thematic argument in the matrix clause is anaphorically identified with a “pro” subject in an embedded closed clausal complement function (COMP), while in raising, a non-thematic main clause-argument is functionally identified with an subject in an embedded open complement (XCOMP). The latter may also be seen as “structure-sharing”, a single element fills two grammatical functions. This is graphically indicated by the line connecting the two subject functions in Figure 2. It is important to note that these identifications take place in the f(unctional)-structure, not in the c(onstituent)-structure (there are no movement operations in LFG, as it is a nontransformational framework). In the f-structures below, thematic arguments appear inside the angled brackets after the predicates. PRED agree <(SUBJ)(COMP)> PRED ‘seem’ (SUBJ) <(XCOMP)> SUBJ

Michaeli

SUBJ

COMP PRED quit <(SUBJ)> SUBJ

proi

Figure 1. F-structure of Michael agreed to eat a hamburger.

‘Michael’

XCOMP PRED quit <(SUBJ)> SUBJ Figure 2. F-structure of Michel seemed quit.

However, the LFG architecture allows for more variation. Falk (2001) shows that there are equi-type verbs which use functional identification with their complement clauses. An example is try, as in Matt tried to eat hamburgers. It clearly has a thematic subject like agree, but some syntactic differences indicate that its complement is a functionally controlled XCOMP. One such difference is the obligatoriness of the controller (in contrast with agree): (3) a It was agreed to eat hamburgers. b *It was tried to eat hamburgers. On the other hand, the LFG-architecture predicts that there should be no constructions where the controller is nonthematic and the identification is anaphoric. In other words, there should be no raising verbs with anaphoric identification. The reason for this is that in this case, the resulting f-structures would violate LFG’s Semantic Coherence-condition: every meaningful element (one that has a PRED attribute) must be assigned a thematic role by some predicate. In a structure with anaphorically identified raising, the main clause subject wouldn’t get a thematic role (for the main verb, it would be an athematic argument, and the embedded predicate would have its own “pro” subject). A further aspect of potential variation has to do with the finiteness of the complement clause. While control in the standard sense only involves non-finite embedded clauses, nothing in the LFG-architecture precludes finite controlled complements. In my talk, I will argue that above the standardly recognized non-finite versions of raising and equi, all types that are allowed by the LFG-architecture actually exists. Some have already been suggested by the literature and I will argue that a Hungarian construction also occupies slots in the taxonomy.

As for finite raising, the standard assumption in the GB/MP tradition is that raising cannot take place from finite complements, Zeller (2006) argues that in the Bantu language Nguni, the modal verb fanele does have a raised subject and a finite embedded clause. (5)a Abantwana ba-fanele ukuthi ba-fund-e. (Nguni) child2 SM2-ought that SM2-study-SUBJ (SM: subject marker) ‘The children must study.’ Arguments from fanele being a raising verb comes for instance from its alternation with an expletive variant and the availability of idiom-chunks as main clause subjects. Equi with finite complements have also been found in several languages, e.g. Turkish (Ince 2006). I’d like to argue that a Hungarian construction also occupies this taxonomical slot: (6) Mihályt mondtad, hogy hamburgert eszik. Michael.ACC said.2SG that hamburger.ACC eat.3SG You said of Michael that he eats a hamburger. The construction is similar to prolepsis as argued by Salzmann (to appear): (7) Ich glaube von ihm, dass er ein ganz guter Trainer ist. I believe.1SG of he.DAT that he a quite good coach is ‘I believe of him that he is a pretty good coach.’ The crucial difference is that while in Salzmann’s approach, the proleptic element (von ihm) is a PP adjunct, here the main clause controller is a direct object which is arguably a thematic argument of the main verb (no idiom chunks can appear in the construction and no expletive variant exists). This main clause element may be either anaphorically or functionally identified with the embedded subject. Evidence for this variation comes from the fact that Hungarian speakers are divided whether they accept (anaphoric identification) or reject (functional identification) plural agreement on the embedded verb, if the main clause controller is a quantified noun phrase (first observed by Gervain 2002). (The plural agreement must be a result of an embedded resumptive pronoun, as quantified NPs in Hungarian only trigger singular agreement in simple sentences). (7) Két lányt mondtál, hogy jön / % jönnek. two girls.ACC said.2SG that come.3SG come 3.PL ‘You said of two girls that they come.’ So the taxonomy predicted by LFG and attested by cross-linguistic data is this: Control type Example functional identification fanele in Nguni finite complement anaphoric identification predicted not to exist raising (athematic main clause canonical raising (e.g. functional identification argument) seem, believe) non-finite complement anaphoric identification predicted not to exist functional identification Hungarian prolepsis finite complement Hungarian prolepsis, anaphoric identification Turkish finite equi equi (thematic main clause canonical equi, verbs like functional identification argument) try non-finite complement canonical equi, verbs like anaphoric identification agree REFERENCES Falk, Y. N. (2001). Lexical-functional grammar. Stanford: CSLI Publications.  Gervain, J. (2002). Linguistic methodology and microvariation in language: The case of operator-raising in Hungarian. MA Thesis, University of Szeged.  Ince, A. (2006). Direct Complement Clauses as Object Control Structures in Turkish. In Erin Bainbridge. & Brian Agbayani (eds.), Proceedings of the thirty-fourth Western Conference On Linguistics (pp. 208-221). Department of Linguistics, California State University, Fresno.  Salzmann, M. (to appear). Prolepsis. In Martin Everaert & Henk van Riemsdijk (eds.), The Companion to Syntax, 2nd edition. Wiley-Blackwell.  Zeller, J. (2006). Raising out of finite CP in Nguni: the case of fanele. Southern African Linguistics and Applied Language Studies, 24(3): 255–275.