In accordance with standard GB (Government and Binding) and standard HPSG (Head-driven Phrase Structure Grammar) assumptions, Polish raising constructions exhibit 'case transmission', as in (1)-(2) below, where the adjective introduced by the lower verb agrees with the case of the subject of the higher verb.

 

(1) Maria-NOM zaczęła być miła-NOM.

    Maria     started be  nice

    `Maria started to be nice.'

(2) Pięć-ACC dziewcząt-GEN zaczęło być miłych-GEN / miłe-ACC.

    Five     girls         started be  nice         nice

    `Five girls started to be nice.'

 

According to the standard assumptions of either theory, control constructions should not allow for a similar case transmission.  This prediction is only partially fulfilled in Polish: case transmission is not observed in object control constructions (cf. (3)), where the adjective occurs in the instrumental case, as it always does when the subject is PRO (cf. (4)), but it is still observed in subject control constructions, contrary to expectations (cf. (5)-(6)).

 

(3) Janek kazał   Tomkowi-DAT być miłym-INS / *miłemu-DAT.

    Janek ordered Tomek       be  nice         nice

    `Janek ordered Tomek to be nice.'

(4) [PRO Być miłym-INS] to   [PRO być głupim-INS].

         be  nice       PRED      be  stupid

    `To be nice is to be stupid.'

 

(5) Janek-NOM chce  być miły-NOM.

    Janek     wants be  nice

    `Janek wants to be nice.'

 

(6) Pięć-ACC dziewcząt-GEN chce  być miłe-ACC / miłych-GEN.

    five     girls         wants be  nice       nice

    `Five girls want to be nice.'

 

Such examples are discussed by Franks 1995, but no worked out solution is proposed, and the tentative solutions suffer from a number of drawbacks, as Franks himself notes.

 

The aim of this paper is to propose an HPSG account of such data, which modifies the standard HPSG assumptions concerning the control/raising dichotomy.  In particular, building on Hudson's 1998 observations about Icelandic and Ancient Greek, we propose to decouple two properties which are assumed to jointly differentiate between raising and control:

 

1. the raised element, unlike the controlling element, is assigned no semantic role by the raising verb (this is an HPSG analogue of GB's Theta-criterion);

 

2. the raised argument is structure-shared with its base position (an HPSG analogue of GB's movement or chain formation), while the controller is only co-indexed with the controlled element.

 

We argue that only the property 1. truly distinguishes between raising and control and that, in Polish, subject control, but not object control, does involve structure sharing.  We show how this explains the facts (1)-(6) above and we further justify this analysis citing (after Hudson) similar Icelandic and Ancient Greek data.  We formalise the account by proposing a single simple non-configurational language-specific Control Principle.

 

Finally, we suggest how to modify this principle in order to account for the fact that, in case of 'long distance' subject control and raising, also the 'non trasmitting', i.e., instrumental (as in (3)-(4)) option is possible, as in the following examples:

 

(7) a. Jan-NOM wydaje się szczęśliwy-NOM / ?*szczęśliwym-INS.

       Jan     seems      happy              happy

       `John seems happy.'

    b. Jan chce  wydawać się szczęśliwy / ?szczęśliwym.

           wants seem

    c. Jan chce spróbować wydawać się ?szczęśliwy / szczęśliwym.

                try

    d. Jan bał się nawet chcieć spróbować wydawać się ??szczęśliwy /

szczęśliwym.

           feared  even  want   try

       `John was afraid to even want to try to seem happy.'

 

 

REFERENCES:

 

Steven Franks.  (1995).  "Parameters of Slavic Morphosyntax."  Oxford University Press.

 

Richard Hudson.  (1998).  "Functional Control with and without   Structure-Sharing."  In: Anna Siewierska and Jae Jung Song (eds.), "Case, Typology and Grammar", pp.151-169.  John Benjamins.