Beruflich Dokumente
Kultur Dokumente
LINGUISTIK AKTUELL
This series provides a platform for studies in the syntax,
semantics, and pragmatics of the Germanic languages
and their historical developments.
The focus of the series is represented by its German title
Linguistik Aktuell (Linguistics Today).
Texts in the series are in English.
Series Editor
Werner Abraham
Germanistisch Instituut
Rijksuniversiteit Groningen
Oude Kijk in ’t Jatstraat 26
9712 EK Groningen
The Netherlands
E-mail: Abraham@let.rug.nl
Volume 27
° zi†ka
Rudolf Ru
° IC
RUDOLF RUZ KA
University of Leipzig
This study has grown out of a series of papers starting with “Remarks on
Control” (Linguistic Inquiry, 1983), and reflects a gradual process of clarify-
ing my insights into the phenomena of Control and what is related to it. I have
presented my views in lectures and courses at the Universities, academic
institutions or Congresses in Brighton (Sussex), Leeds, Stockholm, Lund,
Gothenburg, Prague, Kiew, Zagreb, Hamburg, Frankfurt, Berlin, Munich,
Leipzig, and Jena.
I wish to acknowledge the helpful comments and criticisms of a number
of colleagues and friends, including Adriana Belletti, Manfred Bierwisch,
Guglielmo Cinque, Milka Ivić, John Lyons, Jarmila Panevová, Barbara
Partee, Luigi Rizzi, Inger Rosengren, Petr Sgall, Peter Suchsland, and Ljubov
Vladimirovna Zlatoustova. I am particularly indebted to Werner Abraham for
his valuable suggestions and comments on both content and exposition.
Naturally, the final decisions have all been mine.
Finally, I want to express my gratitude to María del Mar Rodríguez
Fernández and Ralf Růzicka for much technical help.
Contents
Chapter 1. Introduction 1
Chapter 2. Preliminaries
2.1. Reflections on Control and its domains 3
2.1.1. Control and economy conditions 3
2.1.2. On the array of Control – Its central domain – 5
2.1.3. Other major domains of Control 6
2.1.3.1. Controlled complement clauses with filled C(P) 6
2.1.3.2. Controlled clauses as adjuncts 9
2.1.3.2.1. Infinitival S-structures 9
2.1.3.2.2. Control in free participial and nominal adjuncts 10
2.2. Approaches to Control 12
2.2.1. Emergence of Control 12
2.2.2. The configurational approach 13
2.2.2.1. Manzini’s theory 13
2.2.2.2. Kayne’s similar approach 16
2.2.2.3. On the arbitrariness of PROarb 19
2.2.2.4. Configurational and binding approaches extended 22
2.2.3. Questions posed to Control 23
2.2.4. Clausal status of the controlled phrase or the propositional
vs. the attributive account of Control 23
2.2.5. Θ-roles and Control 24
2.2.6. Specification of thematic properties and Control domains 25
2.3. Organization of the study and languages considered 27
Notes 187
References 197
Subject index 205
Chapter 1
Introduction
Preliminaries
The empirical problems usually assembled under the notion of Control cannot
be properly understood and accommodated in a chosen theoretical framework
unless the linguistic devices are identified and surveyed from which the
grammatical phenomena are isolated that have been given that half-metaphori-
cal name. Theorizing on Control, we find ourselves committed to facts and
principles of economy in the design of language, which empirically legitimize
economy of derivation and explanation. Indeed, they manifest themselves
consistently in the grammatical areas to be considered, demonstrating more
conspicuously than elsewhere that they “… play a significant role in account-
ing for the phenomena of language and its use.” (Chomsky 1992)
Why is Control a particular instantiation of economy? Speaking in terms
of the external interface levels, that is, the articulatory-perceptual level PF and
the level that constitutes the instructions for the conceptual-intentional system,
LF (following Chomsky & Lasnik (1993) and Chomsky (1992)), we are
concerned in principle with economy conditions displayed in pairings of PF
and LF representations. Language(s) have to cope with a particular task of
verbalization, which consists in rendering the intended conceptual relationship
between two events, where event is understood in the abstract sense, as, e.g. in
Davidson (1987), Higginbotham (1985), or in Bierwisch (1989), the latter
interpreting event as the instantiation of a proposition. The key situation
invoking Control as a phenomenon of “… understood reference for an argu-
ment position … if there is no expression in that position …” (Higginbotham
4 Chapter 2
1992: 79) arises when one and the same entity is involved in both interrelated
events figuring as an NP-argument in both event descriptions. Setting aside
conjunction or embedding as a finite clausal complement or adjunct, we are
concerned with a different syntactic option, namely, infinitival complements or
adjuncts, and participial adjuncts. This option is not taken on the chance of the
reference of the “empty” element being reconstituted, but on condition that it
readily can.
The choice, e.g., of an infinitival S-complement creates the problem of
reconstituting its empty subject. But we rather have to proceed in the opposite
direction by claiming that the choice of the nonfinite complement (or adjunct)
is determined by the presumed identifiability of the empty element. Syntactic
organization is guided by economy principles in that phonetic absence of the
recurring element in subject position is forced by the nonfinite verb. Note that
if economy can dispense with assignment of phonetic shape, it cannot exclude
the resulting empty category from the “overt” syntax, since it is pertinent to
the LF-component, which channels instructions for cognitive “reality”.
Which factors should naturally be expected to impose conditions or
restrictions on the theoretical reconstruction of the elements left empty. In
other words, how are the particular interface conditions realized in the rel-
evant pairing of PF- and LF-representations possibly projected into concep-
tual-pragmatic mapping. The outward mechanism of control hinges upon the
syntactic status of the nonfinite embedded clause. If it is an infinitival
S-complement (and sentential argument) of the matrix verb, control is affected
by the relationship set between the two events, more exactly the event descrip-
tions. A second consequence that affects control directly: The controllee is an
argument in the predicate-argument structure of the complement clause and
carries a thematic characterization of its θ-grid. The complement clause in
turn is a (sentential) argument of the matrix predicate and carries a θ-role
assigned by the matrix verb. Since the matrix (control) verb determines
thematic and, possibly, semantic properties of the potential controller NP as
well as of the clausal complement whose empty subject NP argument (at S- or
LF-structure) is the controllee, the identification of both NP’s is, presumably,
sensitive to compatibility restrictions. For example, if the controlled NP is the
subject of an active infinitival S-complement of promise, it can take only the
subject of the matrix active clause as its controller. It is cross-clausal quasise-
lectional restrictions of this basic type which, developed into self-consistent
devices, legitimize the economy of dispensing with the assignment of pho-
Preliminaries 5
ing effects on control are, for example, verbs like promise, ask, persuade, try,
endeavour, signal, teach, threaten, help (and their cross-linguistic counter-
parts), which are distributed among lexical classes. It will become predictable
from the character of the respective control conditions that the controlled
element (PRO) is bound to be an argument: “Why is PRO always an argu-
ment?” (Brody 1993: 2)
On the opposite side of the classificatory spectrum, we find verbs which
do not seem to exercise any influence or impose selectional restrictions on
thematic properties of the controlled subject of their infinitival S-complement.
These are verbs like, e.g. hope, wish, be afraid, expect, hate, like, choose,
want.
hates
kiss her )
(1) He is afraid (PRO to be kissed
likes
It is this indifference which is responsible for, or at least involved in, what has
been called “non-obligatory” control (Williams 1980: 208).
(2) They expected (him to cheat her)
(3) They hated (him to do that)
The infinitival verb is often interchangeable with a gerund complement,
which, possibly, is another consequence of specific properties of these verbs.
See (4) and (5).
(4) She was afraid of (PRO asking for help)
(5) He dislikes (PROi being spoken to ti like that)
If one can roughly speak of this group of verbs as attitudinal verbs, the
indifference indicated above seems plausible.
(12) shows that in the former only John is available as a masculine gender
controller of PRO, which cannot bind herself. In (12) John and Anne are
available as not quite equally felicitous controllers of PRO. In (13), tell allows
arbitrary, that is, context-related PRO and, thus, oneself. PRO, if controllable
by John and binding himself, results in a bizarre interpretation.
*herself
(11) John asked how to behave himself
himself
(12) John asked Anne how to behave herself
herself
??
(13) John told Mary how to wash himself
oneself
(18) Henryi gave Mary many of his books (PROi to reduce his library)
Cases that cannot be disambiguated without resorting to discourse and situa-
tions are quite in order as the Russian example (19) shows.
(19)R On prines kniguj (Oj (PRO pocitat’ tj))
he brought a book to read
It is open whether he brought a book to read it himself or to give it to
somebody to read. The choice is clear in the Russian example (20).
(20)R On prines mnei (svoi stichi) (Oj (PROi procitat’ tj))
[dat]
he brought me his poems to read
I will conclude the illustration by two straightforward instances of this type of
control, in Russian and English, and a subtle one in Czech. See (21) – (23).
(21)R Jai narocno ne zakavycival eti
I on purpose not put into quotes these
stroki (ctoby (PROi zaintrigovat’ citatelja))
lines in order to arouse the curiosity of the reader
‘I purposely did not put these lines into quotes in order to
arouse the curiosity of the reader.’
(22) ((PROi To tell you the truth) Ii have never really thought of
them that way.)
(ex. from Stump (1984))
(23)C Zenui (pro) povezou do Budejovic (Oi (PRO spálit ti))
[acc] [3 ps pl perf.]
the woman (they) will take to Budejovice to burn (her)
(ex. from Svoboda (1962))
The pro-subject of povezou, which is [+plural], is not necessarily identical
with PRO. ‘They will take the woman to B. for her to be burnt there.’ pro
refers to a discourse-related domain of people no single person of whom is
identified.
The concept and term control circumscribe phenomena that have figured in
traditional grammar under different notions like, e.g. subject and object infini-
Preliminaries 13
tive, which betray the fact that (the interpretation of) the infinitival VP must
have recourse to the higher subject or the object depending on the lexical class
of its embedding verb (e.g. promise or persuade). If “the theory of control is
concerned with the choice of antecedents of PRO” (Chomsky 1982: 7), PRO
in turn is understood as a term coined for the theory of control. PRO is the
symbol for a particular empty category which, at the relevant point in the
derivation is present in the subject position of nonfinite VP’s at S-structure
and/or LF, if no lexical subject can take this place (“obligatory” control).
Being invariably an argument (Brody 1993: 2), PRO needs identification with
an antecedent or binding by a quantifier. If neither an antecedent nor a
quantifier is available, some other interpretation must be looked for, a case
which has been treated under the general heading of arbitrary (arb) control.
The theory of control is concerned, then, with the nonredundant predictability
of identifying or, more generally, understanding what is the content and
reference of each instance of representing PRO. Nonredundancy is important
for laying claim to a theory of its own. Chomsky/Lasnik (1993: 73) suggest
that “… there is considerable evidence for the existence of a distinct Control
module in the theory of grammar.” Crucially, nonredundancy requires isolat-
ing issues of control from problems of identifying or binding empty categories
of types different to PRO.
On the general assumption of autonomous syntax, predictability in con-
trol has been thought to be achievable by an essentially configurational theory.
The question has been whether solutions to the problems raised by the empty
category PRO can be attained by subsuming them in theories already existing
for other empty categories, or by taking or adapting concepts from them while
upholding the claim to a distinct theory of control. I would like to sketch now
the generalized main approaches to control with a view to appreciating how
they probe into what are the substantive conditions of Control.
S′j Inflj VP
(Agr)
PRO
Manzini’s ingeniously constructed syntactic device accounts for the funda-
mental difference between control in object sentences and control in subject
sentences with no more effort than drawing on the i-within-i condition and its
effect in a newly introduced domain-governing category. Two generalized
binding conditions, (28) and (37) (Manzini’s (20), (30)) take up notions of
binding theory in Chomsky (1981). Under this approach, one must be pre-
pared to judge, e.g. (29), (32) and (33)–(36) to be equally wellformed on
syntactic grounds.
Once the cost in either empirical adequacy or syntactic wellformedness is
tolerated, Manzini’s configurational approach is important since she has laid
the emphasis on the gulf in control conditions existing between object and
subject clauses. The question is, whether she has put it in its true perspective,
that is, whether substantive factors regulating Control must still be conjec-
16 Chapter 2
“Knowledge of Language”, Kayne (1991: 675 ff.) emphasizes that “the PRO
theorem follows from the strict parallelism between Principles A and B of the
binding theory. To the extent that strict parallelism fails to hold over some
range of environments, the PRO theorem will fail to hold for that range. More
specifically, it will fail to hold for any subject PRO governed by a lexical
category found within the category of which PRO is the subject, since in such
a case the governing category for PRO qua anaphor will not be identical to the
governing category for PRO qua pronoun.” In other words, if the PRO
theorem is violated only by violating the conjunction of Principles A and B,
distinct (local) domains for A and B immunize against such violations. If PRO
in particular is lexically governed by the infinitival verb, the IP of which it is
the subject is not its governing category insofar as PRO is anaphor, since
within this local domain it could not satisfy the role of an anaphor, that is, it
could not satisfy the binding theory with some indexing. Thus, the term
governing category would be senseless with respect to the anaphorical proper-
ties of PRO. The relevant governing category for an expression α has been
identified with “… the least CFC (complete functional complex) containing a
governor of α in which α could satisfy the binding theory with some indexing
…” (Chomsky 1986b: 171) For PRO qua pronoun, however, the IP of which it
is subject is a legitimate governing category, and if PRO is lexically bound by
the infinitive, there is no violation of the PRO theorem, the IP being the
smallest category that contains a subject position and the governor of the
pronoun PRO.
Kayne’s original construal seems to run the risk of partly voiding the
PRO theorem or what remains of it by associating PRO qua anaphor with a
governing category within which it could satisfy (contain a suitably accessible
binder) as well as violate the PRO-theorem, whereas PRO qua pronoun is
given a governing category in which it necessarily satisfies it. This asymmetry
is reinforced in Kayne’s modified version of the PRO theorem by virtually
excluding PRO qua pronoun from it, which might turn out to be the price of
this approach: “… the PRO theorem continues to hold for all PROs other than
those that are in subject position and governed by an internal governor.”
(Kayne 1991: 679) If such PROs are not governed, they must be made to be
governed: “All controlled PROs are governed at some level of representa-
tion.” (Kayne 1991: 679, numbered 92) In French, then, for cases like Kayne’s
(91) Jean veut aller au cinema (Jean wants to go to the movies), the infinitive
must move at LF, by analogy to S-structure movement in Italian, since at
18 Chapter 2
S-structure it does not govern PRO, which means, following Kayne, that PRO
has an antecedent only at LF. Kayne gives as “… reason for the existence of
(92) … that it is via government that PRO qua anaphor receives a governing
category … that an antecedent for PRO must be within PRO’s governing
category (that is, that an ungoverned PRO would be able to be associated with
any antecedent at all).” (Kayne 1991: 679)
These assumptions, while allowing Kayne to accomodate control to the
binding theory, seem to beg the question whether control and the “reversed”
PRO theorem can still be taken to follow from the conceptual rationale and
range of binding. Kayne’s claim, however, about locality of the controller
(antecedent), that is, about the local domain of control, is important, since it
establishes what might be called an Adjacency principle of control: “In assign-
ing to PRO qua anaphor the next IP up as governing category, the binding
theory adopted here excludes the possibility that the antecedent of PRO …
could be taken to be a subject NP two IPs up.” (Kayne 1991: 677)
I have dwelt on Kayne’s approach because it may be considered to be the
most intricate attempt to incorporate important aspects of control into binding
theory and to derive variations in the existence of control structures (e.g.
between Italian and French) from independent parametric differences of in-
finitive moving.6 But there are crucial aspects of control about which Kayne’s
approach says nothing, in particular the eligibility of the controller, or as
mentioned above, the fundamental question of subject and object Control. It is
those aspects though not elucidated under this name, which will receive the
main attention in this study, again with respect to the problematic claim on a
distinct theory of control.
Under this approach, we share Chomsky/Lasnik’s (1993: 72) suggestion
that “… Control is different enough from anaphor binding that a separate
mechanism for antecedent assignment is, in fact, justified.” Their arguments
seem conclusive: The optionality concerning choice of binder does not regu-
larly exist for choice of Controller:
herselfi
(41) John j told Mary i about himself
j
antecedentless PRO exists. The former kind of PRO in (45) is contrasted with
the latter in (44). Both instances of PRO, the antecedentless arbitrary and the
controlled one, are not independent of the distinct adjectivals, easy and useful,
respectively. That is why (44) and (45) differ in their respective control
relations, which are the basis of Brody’s argument. The principled common
dependence of control behaviour on lexical properties seems to effect the
control status of antecedent and antecendentless PRO. The disjointness effect
in (44) and the lack of it in (45) are indeed by-products of control conditions.
As to the basic motive for Brody’s argument, there can be no doubt that
“antecedentless PRO with arbitrary interpretation in fact exists”. The real
problems are whether and how “arbitrariness” could be qualified or specified,
and more importantly, in which cases an antecedent must be assumed and
represented although it has no phonetic shape.
By way of a second preliminary illustration of differences between a
configurational, in particular Case-theoretic approach and the approach pur-
sued in this study, which does not exclude configurational restrictions, let us
take a look at two examples juxtaposed by Chomsky/Lasnik (1993: 71).
(46) *It is likely (PRO to solve the problem)
(47) It is important [PRO to solve the problem]
(Chomsky/Lasnik’s examples (255), (256))
“Since PRO in (255) is not in a configuration of Case assignment (a lexical NP
is impossible here), that example might be expected to be grammatical,
presumably with an “arbitrary” interpretation for PRO as in (256) … And
(257) (our (48): RR) might be expected to be grammatical with an arbitrary
interpretation, or possibly with PRO controlled by John, given the general
lack (or at least amelioration) of condition A effects in clauses with expletive
subjects, as illustrated in (258) (our (49): RR).
(48) *John believes [it to be likely (PRO to solve the problem)]
(49) Johni believes (it to be likely [that pictures of himselfi will be on
display])
If the crucial factor determining the distribution of PRO is government and PRO
must be ungoverned (Chomsky 1981) “… this will entail that PRO will not be
Case-marked. But the requirement is now broader, since there is government
without Case-marking. This is what we find in … (255), (257) … The
distribution of PRO is thus correctly described.” (Chomsky/Lasnik 1993: 72)
Preliminaries 21
(1990: 122) Following Bouchard (1984), she assumes that there are two types
of PRO, anaphoric PRO (locally controlled PRO) and pronominal PRO (“long
distance controlled PRO” and “arbitrary PRO”). On this assumption, “…
antecedent-PRO relations fall under the binding theory if ‘governing cat-
egory’ is replaced by ‘binding theory’”. (Iwakura 1990: 123)
Iwakura argues in support of her approach by analyzing a series of
control verbs. She adapts Chomsky’s (1980: 33) control rule to suit her
analysis, modifying the i-within i condition in the attempt to maintain the
configurational essence of Control. Her analyses observe important accompa-
nying restrictions on Control, but I will claim do not concentrate on the
substantive conditions of Control.
2.2.4. Clausal status of the controlled phrase or the propositional vs. the
attributive account of Control
crucial for preferring the propositional account. I would like to provide some
more arguments in favour of PRO, capitalizing first on evidence that accep-
tance of PRO is consequent on the assumption of traces for Raising. Thus,
consider (53), which illustrates an interesting interaction between Control and
Raising, when a Raising construction is embedded under a control verb (cf.
Jacobson 1992: 188). See (53)–(53′)
(53) John tries to seem to be nice.
(53′) Johni tries ((PROi to seem (ti to be nice))).
Assuming that Raising is involved in (53), the Raised subject must end up as
PRO-subject of the clausal complement of try. PRO is the head of the A-chain
(PRO, t) and should receive a θ-specification originating from try (see below,
6.1.1.). Although much is unclear, Control and Raising match with each other
through the interdependence of the trace and PRO. Second, the appearance of
PRO seems imperative if ambiguous long-distance and local binding of re-
flexives is to be represented in the most natural way. Consider Russian (54).
(54) Komendanti prikazal ad-jutantuj ((PROj
commandant ordered (his)aide-de-camp
sojedinit’ sebjai,j s nacal’nikom štaba))
to connect him(self) with the chief of staff
°
(cf. Ruzicka 1973: 444 – 481)
Coindexing of the local binding variant requires PRO to be available. Other-
wise, this kind of subject-oriented anaphora would not find its local anteced-
ent represented. Most recently, E. Torrego (1996: 119) presents “… several
types of evidence, drawn from the morphology, syntax and semantics of the
floating unit, that confirm that control clauses have a PRO subject.”
A fundamental result that has emerged in recent work is that control in at least
a major class of relevant construction types is not independent of thematic
relations meshing with semantic roles. Originating with Jackendoff (1972,
1974) and Gruber (1965), this basic approach was supplemented with, or
diverted to, a search for configurational solutions (Chomsky 1980). But
Chomsky followed up his ideas with “… the natural suggestion … that choice
of controller is determined by θ-roles or other semantic properties of the verb,
Preliminaries 25
°
Taking up the thread of previous work (Ruzicka 1983a,b, 1986, 1987), I
will develop a strongly modified and extended framework, with the particular
intention to relate general principles to cross-linguistic variation, considering
interaction with independent factors as well as idiosyncratic developments.
Control is assumed to be mainly a lexically driven syntactic process. And if
“… θ-marking … amounts to establishing a correspondence between syntac-
tic and conceptual arguments of a verb …” (Jackendoff 1987: 409), “… what
grammatical relations (alternatively, syntactic positions, RR) a θ-role links to
depends on the verb. However, for a particular verb, linking is largely predict-
able.” (Carrier-Duncan 1985: 6), (Ostler 1979). These ideas suggest chains as
the target of control (Chomsky (1986b: 135). Roughly speaking, a mapping
relation established between pertinent lexically semantic properties and func-
tional chains can be the target of control conditions. CHAINS consisting of an
expletive-argument pair at D-structure Chomsky (1986b: 135), cannot be a
target of control conditions which are operative at LF-structure. I will assume
that θ-marking includes or consists in transferring θ-specifications from the θ-
grids of lexical entries to functional chains. The operation of control principles
must include LF, if only because of the fact that PRO can be the variable
bound by a quantifier. Clearly, if θ-grids are “… devoid of semantic content
… just an indexing device” (Jackendoff (1987: 379): Jackendoff refers to
Higginbotham (1985: 555)), they can hardly be suitable for sustaining control
relations. On the other hand, the traditional global characterizations of the-
matic roles merging in (names of) semantic roles, such as Agent, Goal, Theme
etc., have not proven to be appropriate for a theoretical implementation of the
relational concept of control. I deviate here from former work (1983a: 311 ff.).
°
In Ruzicka (1987), I suggest that the relevant conditions, that is, the thematic
Identity condition and the thematic Distinctness conditions should be replaced
by thematic feature specifications. I will take up that suggestion and modify
and elaborate it in the much broader framework of this comparative study.
Thematic feature specifications in the spirit of Gazdar et al. (1985: 21), i.e.,
ordered pairs of the form 〈feature, feature value〉, will be the conceptual
material out of which the generalizing constraints are constructed. As we said
above, the lexical subclasses established on the grounds of properties deter-
mining control behaviour come to be definable by pertinent respective con-
straints associable with each of them. The format of the constraints is such that
they can be applied to the pairs of DP’s potentially constituting the control
relation, alternatively, to the corresponding chains. The conjunction of values
Preliminaries 27
of the relevant θ-specifications in the matrix and the controlled clause respec-
tively, provides the appropriate format.
The crucial relation between the θ-specifications of the potential control-
ler and PRO is particularly effective iff the complement clause of which PRO
is the subject is itself an internal argument of the matrix “control” verb, θ-
governed and/or L-marked by it. Subject complement clauses of which PRO is
the subject, which are external arguments and selected rather by VP, are less
able to sustain the relation formulated in the constraints. Subject complement
clauses are indeed a case in point: “A PRO in a subject sentence (co)refers
freely.” (Manzini 1983: 424)).
I will argue, then, that the substantive Control conditions can be ex-
pressed by pairings of specific feature values which, while basically grounded
on conceptual thematic or semantic content, may function as relations of
abstract F(eature) values. Crucially, distinct parameterized language particu-
lar properties may react differently to the general constraints of Control.
The former specification is projected from (the θ-grid of) (po)prosit’, the latter
from zestokij, characterizing its intentional meaning, analogous to Polish (61)
and Russian (62).
(61) pro poprosiÑem Janai [PROi być szczerym ]
[1 ps sg pret] [acc/gen] [instr masc]
I asked John to be sincere
(Example from St. Franks)
(62) Maša poprosila Vanjui [PROi spat’ golym]
[acc] [instr masc]
Masha asked Vanja to sleep naked
Example (56) obeys (A) with α = –. The licit controller (ja) carries intentional
action (α = –, thus ~α = +). The chain consisting of the head PRO and its trace
is assigned 〈intact, –〉 (α = –). I repeat (56).
(63) Jai poprosila ego [PROi ne byt’ iskljucennoj ti iz
[sg fem pret] [sg fem pass]
I asked him not to be expelled from
školy]
school
Clearly, constraints like (A) can be taken here as matching conditions requir-
ing that designated pairings of values of a given feature specification are
observed. With the variable α = {+, –} the shape of constraints analogous to
(A) extends to, and exhausts, the relevant pairs of matching values.
(B) a. 〈F,α〉con & 〈F,α〉pro
b. 〈F,-α〉con & 〈F,α〉pro
The Control feature F, interpreted as 〈intact, (α)〉, is available in all the
hypothesized constraints. It has proven workable in a family of distinct lexical
classes of Control verbs, where in every case both predicates, the Control verb
and the complement verb discharge the respective contents of 〈F,α〉. Distinct
Control behaviour of the lexical classes of Control verbs which are embraced
by (B) is captured by the different values of the feature specifications as they
are distributed among the potential Controller and PRO. I will claim that in
this way intrinsic unified Control principles can be kept up which establish the
interface between lexically semantic restrictions and their syntactic obser-
vance.
The theory outlined 33
contained in the theta-grid of the complement verb for the marked application
of constraint (B,a) and (B,b) to license wellformed control. As we shall see
later (D) extends its force to the positively set parameters (G) and (H).
I will discuss now and motivate its particular application to (C,b″). If
PRO is moved and heads an A-chain, it cannot transfer a theta-specification
〈intact, +〉pro, which is assumed to be linked to D-structure subject position. If
the moved DP, that is PRO, does carry 〈intact, +〉, it must come from some-
where else, non-locally. Now consider (64).
(64) Hei asked him ((PROi to be believed (ti to be his follower)))
Assume that the chain headed by PRO is theta-marked by (be) follower of …
with the coda of the chain being in the external argument position of the
sentential argument theta-marked by believe. The theta-grid of believe con-
tains the theta specification which is linked to the “subject” of the epistemic
predicate, say Experiencer. In (64), the recipient of the putative specification
is implicit. The matrix object him is very likely to corefer with, or to be
included in the referent of, the implicit nominal argument of believe, depend-
ing on discourse-related factors. Note that if this is the intended coreference
relation, it would embody the identification which in active embeddings of
believe emerges as the unmarked control relation licensed by (C,b′).
(65) He asked himi (PROi to believe that …)
In (64) it manifests itself as an at least conceptually suggested coreference that
is backed by the incidental work of the control constraint (C,b′): 〈intact, –〉con
& 〈intact, +〉pro.
I have tacitly assumed that the implicit believer-argument of believe
receives 〈intact, +〉, which in conjunction with 〈intact, –〉 of him would satisfy
— in the active configuration (65) — the constraint (C,b′). If we are not
prepared to acknowledge the need for providing PRO in (65) and, conse-
quently, the corresponding implicit argument in (64) with the specification
〈intact, +〉 (〈F,+〉), we might lose what can be considered a prerequisite for a
rational interpretation of ask (ask somebody to resolve to believe). I will leave
this case of (co)reference solution for the moment. Control in (64) is instanti-
ated in its marked version. (C,b″) licenses he as controller. What is problem-
atic is the thematic specification of PRO that is relevant to control. A-moved
PRO carries along the assignment it receives from be follower, presumably
〈intact, +〉. Consider (66).
36 Chapter 3
Further, were it not for the conditon (D), (67) could meet the marked con-
straint (C,b″), with he licensed as controller. Marking the subject argument of
resemble or sneeze as 〈F,+〉 is taken to be a decision exclusively at the
pragmatic level, which would lead to the acceptability of (67), since the
pragmatic marking in turn is referred to by the grammatical um constraint
(B,b). Thus, (D) distinguishes, e.g. (64) from (67). In (64), 〈intact, +〉, as
required by (D) and (E) is linked, possibly, to the argumental passive mor-
pheme of the controlled clause as proposed by Jaeggli (1986) and Baker et al.
(1989). (67), since there is no position for 〈F,+〉 to be projected, is ruled out
grammatically, if, possibly, acceptable at a conceptual-pragmatic level. Not
much of a serious problem seems to be left. Now consider (68).
(68) ?*He asked him (PRO to receive the whole sum)
PRO, if assigned GOAL, by (D) does not receive 〈intact, –〉, since the theta-
grid of receive lacks 〈intact, +〉. Thus, (68) is excluded by (C,b″), since it
cannot obey (D). Nor can it meet the um version (C,b′) of course. (68) does not
warrant a well-formed interpretation, or it is semantically undefined.
Corollary (D) applies to unaccusative verbs as well as to passive verbs in
the controlled clause, which is not surprising if, following Belletti (1988: 6)
“… verbs with passive morphology can be considered unaccusative in the
same theoretical sense as lexical unaccusative verbs — namely, passive verbs
do not have the capacity to assign structural accusative.”
Assume that PRO is the head of the relevant A-chain that carries 〈intact,
–〉. Then, again, condition (D), will prevent the marked application of (B,b)
from licensing control, as the theta-grid of the unaccustive verb does not
contain 〈intact, +〉. Recall that the thematic feature specification pertinent to
control should be closely associated with the theta-grid of (classes of) verbs.
Agent, e.g., could be assumed to entail 〈intact, +〉. (69) is excluded by the
um(j,j) and, by virtue of (D), by the m(i,i) application of (B,b).
(69) *He asked himj (PROi,j to arrive home)
i
(71) ?We asked him (PRO to rest assured that we will do all we can).
Compare further (72) and (73) with (74).
(72) *We asked them (PRO to come of age)
(73) *We asked him (PRO to come of a good family)
(74) We asked him (PRO to be patient)
Clearly, if Agent entails 〈intact, +〉, the latter does not entail the former, as in
(74). The control principles cannot be reduced to θ-roles (cf. above 3.2.1.).
An hierarchical restriction is imposed on marked applications of (B,b).
(F) 〈intact, +〉con is required to be carried by a DP in (D)-structure
subject position.
The hierarchical condition could be expressed as m-commanding of PRO by
the controller.
(75) *The coachi was asked ti by Peterj (PROj to be included in the select
team)
The transparent linking of 〈intact, +〉con to the basic subject position must be
maintained to be targeted on by the “weaker” marked applications of (B,b). As
we shall see later (5.1.), a DP in adjunct-(argument) position that carries 〈F,+〉
may be accessible to the um application of constraint (B,a), which is associ-
ated with other lexical classes. Clearly, markedness distinctions related to the
projection of thematic properties into syntax will prove to be sensitive to
parameterized variation.
addressee of the request and the argument associated with the person(s)
expected to comply with the request. The former is linked to the direct object-
DP or, in the passive, to the A-chain headed by the subject-DP binding its
trace in object position. The latter is projected as the PRO-subject in nonpas-
sive infinitival complements. The clear intended coreference of the two DP-
arguments functions as control relation licensed by the um application of (B,b)
if the first argument is rightly characterized as 〈intact, –〉 and the second (PRO)
as 〈intact, +〉. The second pair consists of the argument referring to the
person(s) who do(es) the asking, and of the argument presumed to represent
the benefactive of the fulfilled request. The latter is very likely to be coreferen-
tial with the former if embodied as PRO in a passive infinitival complement
clause. This coreference, again predictable from the lexical structure of ask-
verbs, is employed as the control relation to be licensed in turn by the marked
application of (B,b) together with corollaries (D) and (E): The matrix subject
is assigned 〈intact, +〉, PRO receives 〈intact, –〉. With respect to the unmarked
constraint (B.b) in passive complements, the lexically induced coreference
pairing the addressee-DP and the Agent-DP, which backs the um application
of (B,b), has no syntactic counterpart through which it could be channeled into
the (unmarked) control relation: The relevant DP involved in the pairing
cannot be PRO in passive complements.
The principled distinction between unmarked and marked applications of
the constraint (B,b) clearly correlates with lexical and syntactic facts including
considerations of economy: The um version can be met by syntactically less
complex structures, that is, active and passive matrix clauses and nonpassive
complement clauses. It is substantiated by the prominent lexically induced
coreference condition between arguments. The marked version is met by
syntactically less transparent control relations, mainly occuring in passive
controlled clauses. They are substantiated by the less conspicuous lexically
induced coreference pairing of arguments. Recall that the marked application
of (B,b) needs to be specified by corollaries (D) and (E). The correlation with
independent thematically induced coreference provides strong factual and
conceptual support for the chosen constraint type analysis and its markedness
specification. We shall see that this approach maintains its strength in other
domains of control.
40 Chapter 3
Spanish in contrast with German and Russian. The latter shun this ambiguity.
Subject control that violates the m version of (B,b), as in one reading of (86)a
and in (87), as opposed to subject control obeying it, as in (64), is completely
ruled out. As to Spanish, numerous informants do not agree with Chomsky’s
(1988) judgement that “… sentence (61b) (= (88)) in Spanish has a single
interpretation: María must be understood as the subject of hablar. Here the
subject of hablar in the mental representation is PRO, and it is a semantic
property of the verb pedir that the PRO subject of its complement must be
bound by the subject of pedir, rather as in the English constructions ‘María
asked permission of Juan to speak to the boys’, where María must be the
subject of speak. Note that the situation in English is different. Here, Juan
would normally be taken as the subject of speak that is, the antecedent of the
PRO subject of speak, in (61b). The verbs pedir and ask are thus slightly
different in their syntactic-semantic properties.
(88) María le pidió a Juan [hablar con los muchachos]
María him asked to Juan [to speak to the boys]
‘María asked Juan to speak to the boys’.” (Chomsky 1988: 127,
128)).
Ambiguity of (88), as in (86a), seems to be sufficiently supported by judge-
ments, which can be systematically traced to the complex working device
(B,b).
This somewhat bewildering picture raises the issue of selecting the appropri-
ate theoretical tools that could capture the greatly diverging cross-linguistic
facts. The conflicting effects of the violation of m (B,b) strongly suggest that
to the extent that they significantly differ from each other they might provide
criteria for deciding whether and at which point to cross the boundaries of
grammar proper. Importantly, if violation of m (B,b) marks a crossroads at
which different “solutions” can be chosen, one seems to feel justified in
crossing from grammar to pragmatic considerations. A fair justification could
arise if the specific setting of a syntactic parameter prevents a given language
from developing a particular pragmatic pattern. Assume that in Spanish a
pattern of conventionalized politeness has developed that allows the control
The theory outlined 45
The analysis proposed for the control behaviour of ask, R. prosit’ receives
strong support from the situation which obtains in “pro-drop”-languages in
which finite complements can adjust to control relations. The empty pronomi-
nal category pro takes the place of PRO. Indeed what structural and functional
properties pro can display in the role of the controllee presents enough
evidence to prove that the constraints and principles advanced above are able
to capture the empirical facts and explain them. Serbo-Croatian is a null-
subject-language suitable for demonstration.
(a) If the um constraint (B,b) is satified, pro must appear, not the overt
pronoun ona.
(108)S-C Petar je molio Marijui da proi (*onai) ostane
Petar Aux asked Marija comp stay
na veceri
at dinner
‘Petar asked Maryi to stay for dinner.’
(The examples (108) – (110) are from Zec (1987) slightly modified.)
(b) If m constraint (B,b) is violated, under the construal (89) and in the
strong version demonstrated by the Spanish example (88), pro must
appear, not the pronoun. Clearly, in finite complement control, this
peculiar option, described above has a better chance to be realized.
(109)S-C Petari je molio Mariju da proi (*oni) ostane
Petar Aux asked Marija comp (he) stay
na veceri
at dinner
It is exactly in the respective cases (a) and (b) that on(a) must be dispensed
with. Independently of control, pro can of course occur iff agreement (or φ–)
features of the finite verb transferred to pro identify it, which implies that they
are different from those inherent to either the potential subject or object
controller in the immediate matrix clause.
52 Chapter 3
(iii) (112, 113), which indeed are of particular interest, whereas (i), (111) is
not directly pertinent to the contrast between them and their respective analy-
ses. The (non)occurrence of +WH COMP is not the only difference.
In (112), ask used as query or request information does not belong to the
lexical control class subject to constraint (B,b). There is no inherent cross-
clausal thematic relationship between two arguments that could be captured
by such a constraint. PRO is considerably less constrained in choosing an
available and suitable antecedent, though of course not freely identifiable.
Essentially, control may be logophoric and split up nonlocally, or be arbitrary.
The fact that PRO is ungoverned if Spec(C) is filled, is incidental to the
control situation as described under my approach. Indexing PRO in (112) by
both Johni and Billi is induced by the reciprocal.
Why is (113), (iii) ruled out? First, since in English the null object
parameter is fixed negatively, in accordance with Bach’s Generalization no
object controller is available for the “object” control verb ask. Second, Bill is
the only possible and local controller, and it is licensed by marked application
of (B,b) in virtue of construing ask (permission) as obeying m (B,b), which has
been discussed in detail and can be understood as setting the pragmatic
parameter (H) (4.1.) positively. Note that judgements may be subtle, as is
natural when boundaries of grammar are crossed. John, in the casual embed-
ding John said, is irrelevant to control: Being no argument of the control verb
ask, it is not targeted by constraint (B,b). Third, since Bill remains as the only
possible controller of PRO, each other is not bound by a plural antecedent.
Replacing Bill in (113), for example, by Bill and Steven would make the
sentence acceptable with the above proviso.
Chapter 4
The control effects of lexical properties of ask-type verbs are easily distinguish-
able from those of persuade-type verbs. Members of the ask-subset are, e.g.,
beg, request, implore, conjure, beseech, entreat; Russian: uprosit’/uprašivat’,
umolit’/umoljat’; German: ersuchen (request), beschwören (implore). For
example, (114) obeys the m constraint exactly as, e.g. (87), or (89).
(114) The boyi begged [[PROi (to be allowed) to come with us]]
persuade-type control verbs, though constituting an opposite subset of direct
object verbs in regard to control conditions, are subject to the generality of
constraint (B,b) as well as ask-verbs. But the subsets mainly differ in exhibit-
ing complementary control effects in the passive infinitival clause.
(115) Peterj persuaded Billi (PROi,*j to be examined by a doctor)
(116) Peterj persuaded Billi (PROi,*j to be replaced at half-time)
(117) Peterj asked Billi (PROj,*i to be replaced at half-time)
(118) Peterj asked Billi (PROj,*i to be examined by a specialist)
At least three problems are posed by the control behaviour of persuade-type
verbs:
(I): First, (115) is perfectly acceptable though violating the um constraint
(B,b). (II): the second problem is posed by the ungrammaticality of the lexical
and syntactic counterparts of (115), (116) in German and Russian. (III): Last
but not least, persuade complement control does not enjoy wellformedness in
accord with the m version of contraint (B,b): “Subject” control is widely
56 Chapter 4
carries unchangeable 〈intact, –〉. Recall that violations of (B,b) are defined on
the feature value of PRO.
We feel justified in hypothesizing that a particular parameter is involved
in the persuade subclass. Its values concern the interface between grammar
and pragmatics. This leads to the second problem mentioned above. German,
Russian and Czech, for example, fix this presumed parameter differently, that
is, negatively. They grammatically resist the assignment of 〈intact, +〉 to PRO,
which appears to be a natural option as well considering that the only theta-
theoretical and functional interpretation of passive structures on which there is
general consensus amounts to suppressing or defocusing the agent. This
seems to do the reverse of what the presumed secondary assignment of 〈intact,
+〉 to PRO does. Note that PRO is structurally prominent as the argument in
the highest A-position of this clause. The negative parameter setting keeps the
original thematic specification of this position as induced by passive immune
to conceptual-pragmatic change.
*Petr
ugovoril
(119)R Tanjui (PROi byt′
ubedi
[acc]
Peter persuaded Tanja to be
osmotrennoj vracom.
[part pass instr fem sing] [instr case]
examined by a doctor
(ex. from Comrie (1981)).
*Peter
überredete
(120)G Tanjai (PROi von einem Artzt
überzeugte
Peter persuaded Tanja by a doctor
untersucht zu werden
examined to be
*Petr
presvedcil
(121)C otcei (PROi být osetren
umluvil
[acc]
Peter persuaded father to be examined
v nemocnici)
in a hospital
58 Chapter 4
Petr
umluvil
(122)C otcei (PROi nechat
presvedcil
Peter persuaded father to have (somebody)
se ošetrit v nemocnici)
himself examine in a hospital
‘Peter persuaded his father to have himself examined in a
hospital’
Peter
überzeugte
(123)G seinen Vateri (PROi sichi
überredete
Peter persuaded his father himself
von einem Facharzt untersuchen zu lassen
by a specialist examine to have
‘Peter persuaded his father to have himself examined by a
specialist’.
Petr
ugovoril
(124)R otca (PRO podvergnut′
ubedil
[acc]
Peter persuaded father to submit
sebja osmotru vracom)
[dat] [instr]
himself to an examination by a doctor
‘Peter persuaded his father to undergo medical examination.’
Thus, German, Russian and Czech rely on a radical syntactic and lexical
solution to the conceptually inherent aspect of their respective counterparts of
persuade. In all examples (122) – (124) PRO is assigned 〈intact, +〉 contained
in the θ-grid of Czech nechat, German lassen and Russian podvergnut’ sebja,
respectively. They are licensed by um (B,b).
The parameter associated with the lexical subclass of persuade-type
control verbs, which interfaces grammar and pragmatics, can be formulated as
The persuade subclass and cognate ones 59
tic persuade-type subclass of control verbs, and for its general resistance to
being licensed by the marked constraint (B,b).
Persuade and its ilk lend themselves by far less easily to a construal in
which the matrix subject can be seen as benefactive of the action described by
the passive infinitiv. Thus, referential pairing of the interested “parts”, that is,
identification of the matrix subject and moved PRO, is not normally amenable
to a marked control relation. That is why (116) does not exhibit subject control
as, for example, (117). Persuade-type events, as opposed to ask-events,
characteristically contain arguing and reasoning, which rather neutralize as-
pects of the action directly related to its benefit for the person(s) doing the
arguing. Thus, the derived PRO-DP that is assigned, say, theme or goal in the
passive complement is not readily construed as referentially identical with the
subject-DP.
Still, if the chances for subject control in passive controlled clauses, in
accord with m (B,b), are slim, persuade-contexts do not fully exclude it and
may sporadically exhibit repercussions of the thematically driven coreference.
Sensitivity to peculiar conceptual construal of thematic roles and their interac-
tion as well as resulting fragility of control judgments have to be taken into
consideration. Take for illustration an example from Rochette discussed by
Vanden Wyngaerd (1987: 92).
(125)F Guillaume a persuadé Béatrice [PPde [NP[S’ PRO l’ aider à
faire ses devoirs]]]
“In this example, control is ambigous, (25) (=(125), RR) being translated as
follows:
a. Guillaume persuaded Béatrice to help him to do his homework.
b. Guillaume persuaded Béatrice to be allowed to help her do her
homework.
… we are dealing with a case of nonobligatory control … (the complemen-
tizer) de appears irrespective of the control properties of the matrix verb. Thus
the exact nature of the structural contrast remains to be determined.” (Vanden
Wyngaerd 1987: 92). As far as a structural contrast is concerned, it can be
captured by the two versions of constraint (B,b): (125) satifies its um version
in the a. interpretation. In the b. interpretation, if correct, we could wonder at
the interpolation of be allowed, which, on this approach, would seem to
induce “object” control as in (115), (116). A reading of (125), as paraphrased
The persuade subclass and cognate ones 61
unmarked (B,b), requiring 〈F,+〉PRO, that is, sets the parameter negatively.
Ambiguity is excluded, then.
(H) Languages (do not) allow violations of marked (B,b) to be
construed as matching it in the subclass ask, subject to the (B,b)
constraint. Positive value is linked to modality, in particular
deontic switch.
If they do not allow the construal under the conventionalized “pressure” as
suggested above (3.3.), judgements are determined by the um application of
(B,b) alone.
(H) can be assumed to interact with the null object parameter as it is
described by Rizzi (1986). Languages which set this parameter negatively,
that is, in which there is no syntactically “active” pro capable of control are
more susceptible to setting (H) positively, as described above. Let us shortly
compare (G) with (H).
(G) is a substantive interface parameter: Grammatical linking of thematic
specifications gives way to conceptual-pragmatic “reconstruction” (English)
or resists it (German, Russian). The parameter is associated with a subclass
subject to constraint (B,b), covering persuade-type verbs. If it is set positively,
repair is effected by conditions on, and recovering of, thematic specifications.
Acceptability is attained by an allocation of feature specifications which is
referred to and licensed by, the um constraint that was violated.
The parameter (H) presumed to be amenable to the ask-type subclass of
control verbs that is subject to (B,b) is of a different kind. It arises from
violations of marked (B,b) and can be assumed to be purely pragmatic. Set
positively, its work must end up in a thematic configuration that satisfies the
very m constraint (B,b) it has violated. Restoration to interpretive wellformed-
ness, that is, acceptability, consists in the conceptual arrangement of the
thematic conditions on which matching the m constraint depends: PRO must
be made thematically compatible with the subject of the matrix clause. Conse-
quently and more precisely, it must be warranted to be identifiable with the
higher subject. The clearest conceptual interpretation, namely be allowed,
which effects identification and in some examples occurs overtly, is highly
susceptible to conventional compression omitting it and conceptually restor-
ing it. Positive setting of the pragmatic parameter associated with ask-type
verbs (e.g. English, Spanish) consists in the conceptual interpolation of some
equivalent of be allowed under highly conventional favorable auspices. Take
The persuade subclass and cognate ones 63
Russian, German and English (cf. Cole 1987: 603)), play a minor role in
control (cf. 3.3.). Thus, the strongly marked contrast between English and
German arising in the control behaviour of ask and bitten insofar as it hinges
on the null object disappears in the case of persuade and its German counter-
parts. The Russian counterparts, however, allow null objects, which are
syntactically active and license control.
(129)R On ze ugovoril proi ((PROi otdat’ Rodju
[acc sg]
He (emphasis) (had) persuaded pro to send Rodja
v korpus a Laru v gimnaziju))
[acc sg]
to the military college and Lara to the gymnasium
mat′ k
(131)R proi ugovoril pro PRO j,k ostat’ sja
j
persuaded to stay
doma))
at home
‘ He persuaded him / her / them (mother) to stay at home’.
djadjui
(132)R pro ugovorili ((PRO j,i k obedu
pro j
[+plural][pret pl] [acc sg]
one uncle
persuaded for dinner
they him / her / them
ostat’sja))
to stay
Note that in all these examples the pro-object is not available in English and
German, This fact, owing to lexical properties of (counterparts) of persuade,
is independent of the null object parameter and their distinct values of the
interface parameter (G), in contrast with (H).
We take convince, urge, enjoin, incite, and some other verbs and their
cross-linguistic counterparts to be in the persuade-subclass and follow this
verb and their control behaviour most closely: They are subject to constraint
(B,b), show a more or less recalcitrant attitude to its marked application and,
crucially, their cross-linguistic differences in control are parameterized at the
interface level with regard to violations of the unmarked (B,b).
Dividing the class (B,b) up into further subclasses, separating them in
particular from the persuade-type verbs above is motivated (1) by semantic
selection, which allows some verbs to take a DP marked [–human] as their
subject, and by dative case assignment to the controller argument DP (sub-
class under 4.4.4.).
Some recent ideas of Larson (1991: 131-135) about control properties of teach
and ask suggest again that a purely configurational approach to control may
fall wide of the mark. Larson observes that “The verb teach is similar to ask in
allowing interrogative as well as nominal and infinitival complements …
Furthermore, again as with ask, an intuitive interrogative semantics appears to
assert itself with infinitives and certain nominals.” (1991: 133). These com-
mon configurational and distributional properties lead Larson to “… suggest
that at least some examples with teach like (61b) (Mary never taught John to
sing) might be analysed parallel to ask …” (p. 134). As we have shown above
(4.4.1.), control properties of ask differ from those of teach with respect to the
unmarked and marked application of constraint (B,b). Larson continues:
“However, there is an important additional fact about teach that bears on
control. Unlike ask, teach is a true dative-shifting verb …” (p. 134). Larson
supposes this fact to indicate that, in addition to control conditions shared with
ask, teach has control properties at least similar to those of the dative-shifting
verb promise. He puts it even more strongly: “Since teach has not only an
interrogative pattern like ask but also a dative pattern like promise, it follows
The persuade subclass and cognate ones 73
that we cannot rest with saying that teach shares the former’s control behav-
iour. Clearly, we must also say something about why it doesn’t seem to show
the control behaviour of promise. Since teach exhibits both of the basic
complementation patterns, why doesn’t it exhibit both control paradigms as
well” (p. 134). Thus, Larson expects that teach combines the control behav-
iour of ask and promise, but it does not come up to his expectations with either
verb. The basic reason is that those expectations are based on the wrong
assumption that the configurational and distributational properties stated by
Larson are essentially related to, or determine, control properties.18 While
teach in fact shares at least the standard unmarked control relation with ask as
with many others, in accord with um constraint (B,b) (cf. John taught Felix to
sing), it has nothing in common with the control pattern of promise. Larson
has some difficulty in accounting for these facts adverse to his approach. Not
unexpectedly, he reverts to what is relevant for control in trying to explain the
difference: “… teach does in fact realize both its ask-type complementation
and its promise-type complementation with infinitives … the latter is repre-
sented by infinitives on their (injunctive) reading. This then suggests why the
expected subject control behaviour of teach is missing. If the promise-type
infinitives with teach always have an injunctive reading … then their under-
stood subjects are always in a sense antecedently fixed. As a matter of
semantics, they must always be construed as referring analogously to generic
you or one in examples like You should always look both ways before crossing
the street … The expected infinitival complementation pattern with teach is
thus present, but promise-type subject control is excluded on independent
grounds.” (Larson 1991: 135)). Thus, promise-type constructions do not
exhibit promise-type subject control. Whatever syntactic similarity joins teach
with promise, it has little effect upon control. Larson has got side-tracked here
following consequently his bent for dealing with control syntactically.19 What
Larson calls the “injunctive” reading of teach has some indirect effect on
control in that it selectively permits a nonhuman subject, which of course is
indifferent to 〈intact,α〉 and immune to m constraint (B,b). The range of
candidates of controller narrows down to one. See (148) – (149).
(148)R Zizn’ naucila egoi ((PROi cenit’ ljudej
life (has) taught him to judge people
ne po vnešnosti
not by (their) exteriors.
74 Chapter 4
The latter group of control verbs subtly differs from teach in that they more
readily adopt the control case analyzed as an instance of cross-level param-
etrized variation on verbs like persuade and considered to be a violation of the
um version of constraint (B,b), that is, they instantiate the positive value of
parameter (G) which presupposes satisfying (D).
The collapse
prompted
(153) him i (PRO i to be examined by a doctor).
compelled
The unability of the subclass of control verbs under discussion to accept
licensing by the marked constraint (B,b) is substantiated by the possible
selection of DP-subjects with a [–human]-head noun. Such DP’s of course can
not receive a thematic role as benefactive, the characterization considered to
be a relevant precondition of the m application (B,b) to be successful (3.2.3.).
Selecting a nonhuman DP-subject disqualifies control structures in the
given subclass of verbs from taking part in correlations between “nonreflex-
ive” and “reflexive” verbs. This situation emerges, for example in Russian,
Czech and other Slavic languages. Cliticized “weak” pronominal reflexive
elements regulate grammatical functions and corresponding linkage of the-
matic roles. Compare (154) with (155).
(154)R Ochotnik naucil synai (PROi streljat’ iz
[pret masc sing] [acc]
The hunter taught (his) son to shoot (from)
malen’kogo ruzja pocti bez promacha)
[genitive] [genitive]
a little gun nearly without missing
(155)R Syni naucil-sja (PROi streljat’ iz malen’kogo
[pret masc sing-refl]
son learnt to shoot (from) a little
ruz’ja)
gun …
(156)C Naucila se (PROi bruslit’)
[fem sing pret] [refl clitic]
she taught reflexiv to skate
‘(she) learned to skate’
76 Chapter 4
At PF, the clitic takes second position in Czech, but it is always affixed to the
verb in Russian.
(157)C Nebo se naucila správne mluvit
[refl.] [pret fem sing]
for pro (she) learnt correctly to speak
“Subject”-control obtains after A-movement of the object-DP, which is the
controller in (155)-(157), in accord with um (B,b). Since se (sja) appears to
trigger the movement to nonthematic subject position, it can naturally be held
responsible for, or at least involved in, the mechanism that prevents Case
assignment. Since a DP receiving theme and describing what is taught to
somebody, is Case-marked [+dative] in Russian and Czech, I take the corre-
sponding clausal argument to occupy the same structural position in these
languages. Note that the way the theta roles of the internal arguments of teach
are linked to the direct and oblique object positions in English is the very
opposite of Russian and Czech. The theta role assigned to the DP that is
“dative shifted” in English, is borne by the direct object in the two languages.
That is the reason why I adopt a ternary verbal structure here. (cf. Emonds
1993: 226, fn. 15). X’-licensing, that is, V’-licensing of the dative DP, or
Larson’s (1988, 1990) “shell analysis” are alternatives. The choice of either of
them does not bear on the issue discussed.
I turn to another subclass of control verbs within the domain of constraint (B,b):
verbs that in languages considered here inherently assign dative to the NP
which is theta-marked goal, benefactive or addressee. The control verb allow,
promise and their respective cross-linguistic counterparts will be singled out in
(sect. 4.5. and chapt. 5.).
(158)P Krystyna radzila kolezankomi (PROi obejrzeć
[dat pl]
Krystyna advised (her) collegues to see
nowy film)
the new film
The persuade subclass and cognate ones 77
Judgments are delicate in languages cognate with languages that set the
parameter negatively like Russian. Sensitivity to lexical environment and the
cross-linguistic parameterized distribution give additional support to a prag-
matic assessment of this type of control. Crucially, we have evidence that
control here is maintained inspite of adverse parametric setting, e.g., in
Russian, when the clausal complement is not passive, although the conceptual
construal is clearly analogous to passive. This is the case in some reflexive
structures that potentially receive a passive interpretation (cf. Růzicka 1992).
Not being in a dethematizised position, their PRO subject is receptive to
〈intact,+〉, satifying um constraint (B,b) at D-structure.
On
rekomendoval
(163)R Borisui (PROi
sovetoval
[dative]
he
recommended
Boris
advised
posmotret’sja)
[reflexive]
to be examined (by a doctor)
Typical representatives of this subclass of dative object control verbs within
the grasp of constraint (B,b) are order and forbid, command, tell, instruct,
direct, charge commission. Their cross-linguistic counterparts mostly have an
oblique dative object.
(164)C (pro) rozkázal mui (PROi mlcet)
[3 ps sing pret] [dative]
he told him to be silent
(165)R Major Berezkin prosnulsja, …, pozval avtomatcika
[acc]
major Berezkin woke up, …, called submachine-gunner
Gluškova i velel prinesti sebe pomyt’sja
[3 ps sing pret] [refl dat]
Gluškov and ordered to bring himself to wash (himself)
‘Major Berezkin … ordered him (Gluškov) to bring him
(major Berezkin) things to wash with’
The persuade subclass and cognate ones 79
[dative] [acc]
The principal ordered (us) to nominate
sebjaj clenom novoj komissii)
[instr]
him(self) a member of the new comission
Sebja is long-distance bound by the nonlocal antecedent nacal’nik (Růzicka
1973).
The persuade subclass and cognate ones 81
What, for example, (168) illustrates again is the fact that though anchored
within a firm framework of constraints control can give way even to particular
pragmatical situations choosing the alternative controller. Thus, by meeting
the marked constraint (B,b), (168) does not take the path predicted for the
subclass, namely, violate the um constraint, a violation that is restored to
acceptability by some kind of reconstruction if the corresponding parameter
(G) is set positively. Still, all possible cases are in the grasp of the two versions
of constraint (B,b) and the associated parameters.
Allow-type verbs belong in the subclass that selects human and nonhuman
subjects. Finer grained selectional features may help resolve the familiar but
not clear-cut deontic versus epistemic ambiguity of allow/permit verbs.
Larson (1991: 136ff.), following Mittwoch (1976), proposes to distinguish
between the two interpretations by identifying a postulated implicit dative
argument either with the Exceptional Case Marking subject (deontic), or by
construing it as a generic or arbitrary pronoun (equivalent to one) (epistemic):
“… allow/permit [α to VP] (to β)
a. Deontic: β=α
b. Epistemic:β=proarb” (Larson 1991: 137)
This analysis is of course tailored to the assumed English ECM-structure of
the sentential argument: “… ECM-structures with allow are actually oblique
dative structures with the ECM complement corresponding to the theme
object and with an implicit oblique dative argument whose identity determines
the understanding of the complement as deontic versus epistemic.” (Larson
1991: 137) Control is not involved at all: “We see that despite superficial
similarity, the pair in (74) (our (173), RR) involves control only in the first
member; the second is an ECM structure.
(173) a. John promised Mary to leave.
b. John allowed Mary to leave.”
(Larson 1991: 137)
The ECM-analysis implies an overt complement subject that “absorbs” the
overt appearance of a matrix oblique argument. Only one of them can “materi-
82 Chapter 4
alize”: Either the oblique argument in the shape of the ECM subject and
identified with it by a. (deontic), or the independent ECM subject, with the
oblique DP reduced to the obligatory proarb (epistemic). Compared to the
Control analysis, the empty category is exchanged, the identification of PRO
is reversed: The identification of PRO with the (overt) oblique object, effected
by control, is replaced by the identification of the oblique part (to ß) with the
ECM subject. The reversed “control” is made optional. It holds for, and “…
determines the understanding of the complement as deontic …” (Larson 1991:
137) Larson seems to face the contingency that a given sentence may receive a
twofold understanding, that is, its ambiguity cannot be resolved. More impor-
tantly: Does Larson’s analysis, or some elements of it, generalize crosslinguis-
tically, in particular apply to languages in which there is no ECM-structure, a
question which may bear on the correctness of the ECM analysis of comple-
ments of allow in English?
First, in non-ECM-languages, the identity condition a. for deontic read-
ings would have to be reinterpreted if possible. As indicated above, Larson’s
deontic reading condition β=α emerges from the control relation and its
reversed reflexion in the shape of an ECM complement in which the ECM
subject takes the “controlling” part identifying the presumed “implicit” ob-
lique argument pro. The ECM-infinitival-subject must be lexically filled and
since no other position is available that could be lexically filled, the oblique
argument must be “silenced” as pro. Second, how can the epistemic reading
condition β=proarb with proarb in the fixed shape of the oblique dative argu-
ment in English ECM-structures, be related to its presumed counterpart in
control structures, for example, in Russian? Epistemic proarb, of course, is not
excluded from corefering with the ECM subject, but is referentially indepen-
dent of it in principle. Under the corresponding control analysis, the dative
oblique argument, lexically overt or covert (pro), implies identity with the
PRO subject of the clausal complement by definition, and that in both deontic
and epistemic variants. Consider (174), a case of possibly ambiguous reading.
(174) Congress has passed a law allowing [juveniles accused of
violent crimes to be tried as adults.]
The presumed proarb does not corefer with the ECM-subject, but may be
construed as corefering with the DP of an implicit by-DP phrase. The nearest
counterpart of (174) in Russian would be (175).
The persuade subclass and cognate ones 83
and the lexically overt ECM-subject can be revealed only in the oblique
position of the matrix clause. In other words, the obligatory lexical appearance
of the ECM-subject does the work done by control, e.g., in Russian. It follows
that under the ECM approach there is no means of constructing or computing
a syntactic direct counterpart of the twofold “empty” configuration of pro
identifying PRO as in (175)′. If the controlling element in the matrix clause
itself is empty as pro, its counterpart in English could turn up only out of the
ECM-subject position. In (174), a position that takes an argument correspond-
ing to the argument in the D-structure subject position is available as the
implicit by-phrase which satisfies the wanted nonlexical status. Thus, the
passive shape of the (ECM-) complement is the only syntactic device for
coping with the pro to PRO relation as effected in (175)′. The control relation
in turn has to pay for its privilege by ruling out just the passive complement
clauses considered to be its counterpart to the English hypothesized ECM-
structures as in (174). Before coming to conclusions that again relate to
parameterization of a type already suggested, let us consider another example.
“As is well known, promise is one of a small number of verbs in English that
select an object and an infinitive and show subject control.” (Larson 1991:
103)). Thus, subject control of promise, (g. versprechen, r. (po)obešcat’, c.
slíbit, p. obiec(yw)ać etc.) seems to be taken for granted. Admittedly, these
verbs have raised problems which linger on in current discussion, but they
cannot be plainly considered verbs of subject control, as has been shown in
much previous work (e.g. Růzicka 1983, Farkas 1988, Wegener 1989,
Köpcke/Panther 1993). Besides notorious examples like (181), there are many
others which, on equal terms, would justify calling promise a verb of object
control. I will provide some cross-linguistic facts first.
(181) Johni was promised ti (PROj to be allowed tj (PROk to leave))
(i = j = k)
For discussion see Růzicka (1983a: 310 ff.).
(182)R bez nego (Stalina) obešcano
[part pret pass neuter sg]
without him (Stalin) (it) promised
bylo [NP nyne zivušcemu pokoleniju]i (PROi
[pret neuter] (to) now [part pres dat sg] [dat sg]
was living generation
okazat’sja v Kommunizme)
to find themselves in communism
‘… in the absence of Stalin (without his having had a hand in
the matter) the now living generation was (had been) prom-
ised to find themselves in communism.’
88 Chapter 5
unmarked and the negative one as the marked version, in both (B,b) and (B,a).
Thus, (B,a)′ is the unmarked, (B,a)″ the marked version of (B,a).
(B,a)′ 〈intact, +〉con^ 〈intact, +〉pro
(B,a)″ 〈intact, −〉con ^ 〈intact, −〉pro
(B,a)′ accounts for standard examples like (190) and the particular case
of (189).
(190) Peteri promised John (PROi to leave).
(B,a)″ accounts for “unexpected” cases like (181), (183) – (186). (182) will be
discussed in the context of broader evidence (below, 5.4.). Marked application
needs a lexical controller: The head of the controlling A-chain must be
“pronounced”.
Crucially, the (un)marked versions of (B,a) and (B,b) respectively, are
licensing instruments that generalize about varying empirical conditions of
control which are associated with respective (sub)classes of control verbs. In
the domain of (B,b), for ask-type verbs, a substantial justification for pursuing
the markedness approach, namely the lucid availability of a second layer of
semantic characterizations pairing, that is, identifying the “interested parts”
(see above, 3.2.3.), subtly divides these verbs from persuade-type verbs, (cf.
above, p. 59ff). Violations of the unmarked version of (B,b) again lead to
different cross-linguistic fates within this event type (see above, 4.1.). In the
domain of (B,a), for promise-type verbs or events, there is no analogous
subdivision either with respect to marked or to unmarked control. In the
former case, controller and PRO are identified through involvements in prom-
ise-type events that are complementary to the unmarked twofold agentive
specification and are captured by 〈intact, –〉con/pro.
The general symmetry of the proposed solution to promise-type control
with the account given for ask/persuade-type control extends to the restriction
that PRO can receive 〈intact, –〉 only from a predicate whose theta-grid
contains 〈intact, +〉, that is, corollary (D) applies to this subclass as well. (D),
thus, unifies two broad domains of complement control defined by the the-
matic constraints (B,a) and (B,b), respectively, singling out their common
interthematic restriction as a generalization about marked control behaviour.
Concerning promise, the clear intuitive basis for (D) is the fact that the
property of the event described in the controlled clause may offer no hold for
the promise to be fullfilled intentionally. If (B,a) and its corollary (D) are not
92 Chapter 5
English — but the argument carrying the θ-role Goal or Recipient is left in
oblique (dative) object position — though excluded from controlling by PRO
violating m (B,a) —, have no subject rivalling for control. The counterpart to
the agent phrase is not hindered from controlling in accord with um (B,a) (see
ex. (189), (196) – (198)). In English, when the hierarchical superiority of the
(derived) subject disappears as in the nominals illustrated in footnote (23), the
DP in the by-phrase can maintain control.
The descriptive generalization (I) capturing these differences of control
between English on the one hand, and, e.g., Russian, Czech, German, on the
other, is a specification on the unmarked application of (B,a).
(I) The first controller-related conjunct in the um (B,a) can be
satisfied by a demoted agent phrase (Case-marked as instrumental
or PP), iff there is no raising to subject position, that is, the
subject position is not filled by the head of an nonsingleton
A-chain.
This is the case, for example, in Slavic languages and German (see ex. (189,
(196), (197)), in which the argument-DP corresponding to the DP that is raised
to subject position in English, is assigned morphological dative Case in the
resulting impersonal passive. No syntactically dominant rival for control
appears. Still, control by the demoted agent argument marked Instrumental
Case in Slavonic languages or shaped as a PP as in German may be judged to
be marginal.
Note that (I) is not complementary to (F) (3.2.2.), which syntactically
specifies the marked application of (B,b) to a pertinent subclass of control
verbs. (F) cannot be cancelled under structural consequences of the distinctly
parameterized morphosyntactical treatment of Case, which, e.g., results in the
different phenomena of passive under discussion. Consider first (200).
(200) *Frankyj was asked by mei (PROi,j to be made his partner)
(200), in the i-indexed interpretation, is ruled out by (F) specifying m (B,b).
The unmarked application of (B,b) fails in the offending PRO-conjunct (in-
dexed j).
From a syntactic point of view, we might expect that control verbs falling
under constraint (B,b) which inherently assign dative Case retained in the
impersonal passive (verbs under 4.4.4.), by analogy to (I), override (F) satisfy-
ing the first conjunct of m (B,b) by the demoted agent-DP. This prediction is
96 Chapter 5
maximal subject DP cannot contain a licit controller (ex. 203). In other words,
a potential controller within the by-phrase cannot control beyond the maximal
DP containing it, that is, into the predicating infinitival clause. Control in
(204) is licit, since (204) sharply differs from (203) in that the controlled
infinitival clause is part of the maximal DP, with promise heading all the rest.
Assuming by analogy to the hierarchy in clausal structures like (199) that the
DP in the agent by-phrase cannot control beyond its maximal DP by virtue of
the “superiority” of the “subject”-phrase and in particular the Specifier posi-
tion available in the latter, we should predict that a controller marked 〈intact,
+〉 is licensed if within the Specifier position of the subject phrase. Thus we
have the contrast between (205) (=(203)) and (206), (207).
(205) *The promise by Bill was (PROi to leave)
(206) Bill’si promise was (PROi to leave)
(207) Ouri promise was (PROi to work immediately)
It follows that a subject-DP containing an article or other Determiner as in
(202) (repeated here as (208)) but lacking an agent phrase can provide an
implicit controller.
(208) The promise to Bill was (PRO to leave).
(206), (207) satisfy um constraint (B,a), (208) requires the implicit controller
to receive 〈intact, +〉 conceptually.
We should expect now that the marked version of constraint (B,a) applies
as well and, by analogy to (181), licenses experimental sentences like (209), at
least marginally. This expectation seems to be fullfilled.
(209) ?The promise to Billi was (PROi to be allowed to leave).
The account given for the facts illustrated by the contrast between (203) and
(204) and between (202) and (203) fits in with the suggested condition of
immediate proximity (see above, 2.2.2.2.).
The drastic shift of control would seem to follow from substituting the NP dem
Gefängnis (prison) for dem Urlaub (his holidays) in the [PPX aus NP]. The
choice dem Urlaub implies the construal of standard subject control, the
choice (dem Gefängnis) inclines one to prefer object control. (213) is a
diagnostic example. The standard control relation with der Anwalt controlling
may withstand the bizarre interpretation connected with it in accord with um
(B,a) and from prison. But subject control may be readily overcome by a
simple construal which links PRO aus dem Gefängnis zu kommen (interpreted
as be discharged from prison) to detainee and, thus, may be construed to
satisfy m constraint (B,a), with PRO being conceptually marked 〈intact, –〉PRO.
straints and lexically effected construal. It would appear, under this approach,
that if a certain amount of interpretive work can be done that achieves some
conformity to the unmarked or marked constraint (B,a), considering their
respective semantic and/or conceptual underpinnings, acceptability can be
lent to deviating structures independently of further grammatical limitations.
But I have not discussed so far that other syntactic or thematic conditions may
prevent sound conceptual-pragmatic construals from being implemented, in
particular that failing Corollary (D) may not be overcome by any pragmatic
construal. Let us focus for a moment on the interpretation of (214) and its
German version (215), where syntactic and pragmatic factors come into
conflict with each other.
(214) Gladys promised Hubert to experience victory again.
(215)G Gladys versprach Hubert, noch einmal einen Sieg zu erleben
Comrie (1985: 63) comments on (214): “The most coherent interpretation of
(59) (= (214), RR) pragmatically, given that a promise is to the benefit of the
addressee, would be that Gladys is promising Hubert that he, Hubert, will
experience victory in the future, but English syntax excludes this interpreta-
tion … Abraham (1983: 221) cites German equivalents of (214) … as fully
grammatical and acceptable in the pragmatically coherent interpretation.”
Does English provide the alternative, but “incoherent” interpretation, or reject
(214) altogether? Or, put differently, why does the possible “coherent” inter-
pretation have the effect of acceptability in German, but not in English? What
is it in English syntax that resists the coherent pragmatical “escape” —
assuming that Comrie’s judgment of (214) is correct? In German, acceptabil-
ity of (215) is not independent of lexical features arrayed as the marked
constraint (B,a). But the construal by which it can be achieved requires a
considerable amount of interpretive work, since corollary (D) is not met.
Gladys must be assumed to have a hand in Hubert’s experiencing victory.
Since PRO is neutralized, which is expressed by violating corollary (D), the
conceptual effort to attain coherence does not seem to override the unmarked
control pattern (B,a), though not complying with it. This is the English value
of the syntactic-pragmatic interface parameter. The contrast, which may rank
as a parameterized variation, in general should be ascribable to the more
unyielding syntactic skeleton of a strongly marked configurational language
like English compared with German.
The case of promise 101
Such borderline cases of control demonstrate that rather subtle and even
empirically controversial instances of control crystallize around the basic
constraints and their respective m and um interpretations. Even in cases
licensed by m (B,a) like (218), Comrie prefers the “syntactic” reading.
(218) Penelope promised her son to be allowed to leave.
“… for me the interpretation ‘Penelope promised her son that she (Penelope)
would be allowed to leave’ is available, and it is in fact preferable to the
interpretation ‘Penelope promised her son that he (her son) would be allowed
to leave’ … Of course, on the preferred interpretation it is necessary to assume
that Penelope has some influence over whether or not she is given permission
to leave.” (Comrie 1985: 53). Note that in the less preferable (for Comrie)
interpretation it is necessary to assume that Penelope has influence over
whether or not her son is given permission to leave. In the preferable (for
Comrie) control relation, um (B,a) is violated. Acceptability or interpretability
may be restored in the way described above (ex. 212). On the interpretation,
clearly preferable in German, Russian and Czech, in which her son is the
controller, m (B,a) is satisfied.
This discussion strongly suggests that confining oneself to a treatment of
control phenomena that exclusively relies on structural notions of grammar
proper simplifies, even distorts the picture. It seems to be typical of control
phenomena that some questions raised in this section may be unanswerable if
not meaningless on a sound grammatical basis. Grammar, of course, is respon-
sible for fuzzy edges of interpretive capacity in the control design, which
possibly arise because grammar provides and makes available wellformed
bare structural skeletons that satisfy enough principles to leave the linguistic
rest to varying judgments of convergence. I draw two conclusions from these
considerations. First, we need a device that mediates between the fixed, the
“hard” syntactic control structures and the “soft” areas of conceptual-prag-
matic manipulation. Constraints (B,b) and (B,a), if specified by (D) - (I),
constitute these intermediary principles. Second, restriction to configurational
syntactic aspects must fail, since they do not reach and get down to some
substantive or critical issues. We have to face situations which make it very
difficult to intuitively distinguish between interpretability, acceptability and
grammaticality of control expressions.
I would like to speculate that control phenomena have conspired to create
an experimental plot trying out and straining conceptual-pragmatic interpret-
The case of promise 103
ability beyond what can be restrained grammatically. But any such facts,
being observed and grasped in nongrammatical terms, in turn are referred to,
and finally elucidated by, appropriate grammatical principles, or principles
that can be related to grammatical ones under a unifying approach. For
example, if, in a given discourse situation, the choice of a controller in fact
depends on the intersection of the interlocutors’ respective actual knowledge
sets and in particular, on “… sets of (their) matters of current concern”
(Yokoyama 1986: 141 ff.), such factors can be related to general constraints in
the shape of theta-theoretical specifications. A case in point is (219) (example
from Farkas).
(219) The mother promised the children to stay up.
The mother is a licit controller by um (B,a). The discourse situation is quite
usual, of course, which suggests that the children figures as licit controller.
The pragmatical construal by virtue of which the children qualifies as control-
ler amounts to nothing else than interpreting the complement clause as some-
thing like to be allowed to stay up, which can be referred to by the m constraint
(B,a) (cf. ex. (184), (218)). This is exactly the reason why the syntactic and
lexical counterpart in German precludes object control, but of course not
subject control.
(220) *Die Mutter versprach den Kinderni (PROi aufzubleiben)
In German, as already noted (3.3.), a lexical nonpassive equivalent for be
permitted/allowed to is available. The construal of object control is “superflu-
ous”, in German, whose only equivalent to (219), interpreted with object
control, is (221).
(221) Die Mutter versprach den Kinderni (PROi aufbleiben
to stay up
zu dürfen, …
to be permitted
Here is an exact parallel to the preclusion of word for word German counter-
parts of (91), repeated here.
(91) Hei asked PROi to leave
*Eri bat PROi zu gehen
(91) can only be rendered by (222).
104 Chapter 5
Larson (1991: 125 ff.) takes great pains to defend his configurational theory of
control and in particular his version of Rosenbaums’s (1970) “Minimal Dis-
tance Principle”24 in the face of “… two well-known lines of criticism, one
involving the specific verb promise and one involving other superficially
similar ‘double object’ verbs like ask, teach, and allow … The structural
relations among the complements of promise in (35a) (my (223): RR) do not
appear different from those in the nonpassivized example (35b) (my (224):
RR); hence, a shift in control is unexpected:”
(223) John promised Mary to be allowed to leave.
(224) John promised Mary to leave.
Under our approach, (223) is clearly licensed by m constraint (B,a). Larson
suggests “that the interpretation of the infinitival in (43a) (= (223): RR) is
determined not by control, as in (43b) (= (224): RR), but rather through
entailments, as in (43c) (= (225): RR).
(225) John promised Mary permission to leave.
“… On the suggested view, what governs object construal with promise is
simply the standard set of entailments that this verb carries by virtue of its
dative status. In general, then, construal of the understood subject of an
infinitive has no specific principles here. Construal by control follows from
the (independently motivated) shape of D-structure. And construal by entail-
ment follows from the general entailments of the verb.” (Larson 1991: 126,
128, fn. 17) The “… general ‘division of labour’ suggested here between
control and entailments seems plausible and seems moreover to answer to the
general intuition that examples of ‘shifting control’ differ in status from their
controlled counterparts — that although they can be made acceptable to a
The case of promise 105
degree, they are never as natural as the latter …” (op. cit. 129) Apart from the
vague qualifications of “naturalness” and “status”, Larson’s endeavour to
draw a neat line between independent D-structure shape and construal by
lexical entailments fails to capture their intrinsic relatedness and their interac-
tion and, contrary to his expectation, gives away the chance of drawing a line
of demarcation that marks their interface. If examples like (223), (225) are
held to be determined not by control, but rather through entailments specific to
the lexical element promise, then, on perfectly equal terms, one can claim that
(224), being determined by control, is determined through entailments of the
lexical item promise. In (223) and (224), lexical properties are differently
activated by, and related to, general principles such as (B,a), as a consequence
of the distinct syntactic and functional structures of the embedded predicate.
Larson has to admit that “One important question that I must leave unresolved
here, but which must be addressed ultimately if this suggestion is to amount to
an actual solution, is when precisely an infinitival complement is subject to
control versus construal by entailments.” (op. cit. 129) This question is
misleading. Its empirical rationale and problems can be dealt with in a distinct
and unified system, as should be clear from our approach.
Comrie (1985: 59) observes that “… indeed the main difference between a
promise and a threat is whether the speaker believes the outcome to be to the
advantage or to the disadvantage of the addressee.” S/he who threatens does
not commit her/himself to fulfil the “promise” even if the addressee does not
do or does what the threat tried to press or force her/him to do and not to do,
respectively. Still, threaten and its cross-linguistic counterparts mirror the
control behaviour of promise. In English, idiosyncratically, an addressee-NP
does not go with a controlled infinitival clause: “English threaten is a control
verb only in the absence of a main clause object …” (Comrie 1985: 52).
Counterparts to threaten in other languages should be expected to be licensed
by observing the same constraint as promise, that is (B,a). (226), e.g., an
example from Fauconnier (1976), discussed also by Vanden Wyngaerd (1987:
82), meets m constraint (B,a).
106 Chapter 5
complement, provided that the results of the pragmatic construal in turn can be
referred to by the respective constraints which license them. Note that in either
case, the quasi theta-specification of Benefactive supports identification in
accord with the m constraint, and that on the “negative” interpretation: What is
threatened is not to the benefit but to the disadvantage of the addressee (cf.
5.6.)25
(248) and (249) illustrate typical standard cases in Russian.
(248)R Vitalijaj proi ottuda iskljucili za
[gen-acc] [acc sg]
Vitali (they) from there excluded for
nedisciplinirovannost’ i proi prigrozili (PROi
[3 ps pl pret]
undisciplined behavior and threatened
otpravit’ proj s marševoj rotoj na front)
[acc]
to send (him) with a company on the march to the front
proj (j ≠ i) is a null pronoun Case-marked accusative, which is identified with
the matrix object Vitali. proi is a discourse-related null subject pro, marked
nominative and [+plural, 3ps] in Russian. It refers to a domain of persons no
individual of whom is identified.
(249)R Oteci grozilsja (PROi menjaj vygnat’ i
Father threatened me to turn out and
lišit’ proj nasledstva
[gen sg]
deprive (me) of (my) inheritance
The slightly non-standard form grozit’sja with the attached reflexive affix
closes the position for an object-NP.
Chapter 6
No Choice of Controller
If only one NP-argument is available one has no choice but to take or reject it
as the controller. The question whether a single argument is a suitable control-
ler, has to be answered in any theory of control.
Dowty proposed an entailment pattern for try that is analogous to (10) (see fn.
15).
“…
(250) ∀x ∀P [δ(P) (x*) ⇒ α(x*)]
∀x ∀P [δ(P) (x*) ⇒ β ([P(x*)] (x*)]
In the case of δ = try’, for example, an instance of the entailment α (x*) would
be ‘x acts intentionally’, and an instance of ß ([P(x*)]) (x*) would be ‘x intends
that x’s action will bring about P (x*)’” (Dowty 1985: 299). (“δ is an intransi-
tive Equi verb, P a variable over VP-meanings …, x an individual variable, α
a property of NP-meanings, and ß a relation between NP-meanings and
propositions.”)
Though stated in terms of a M(ontague)-analysis), Dowty’s entailments
can be related to control conditions in the style of this approach.
I will argue that verbs of the try-type (e. g. manage, dare; R umudrit’sja
(manage, contrive), osmelit’sja (dare) assign the theta-specification 〈intact,+〉
to their respective external arguments and to the s-structure subject position
(PRO) of their infinitival complement. Thus, they “generate” their own con-
straints on control26 .
114 Chapter 6
(253) He
managed
to reach the semi - finals
failed
What is common to manage and try, to fail and hesitate? Let us tentatively call
the property metaactive. It will be represented as the lexical feature (value)
〈metaactive,α〉. Associated with a lexical class of control verbs, it will trigger
the cross-clausal double assignment of 〈intact,+〉. Its intuitive understanding
can be paraphrased as the relation of an agent designate to bringing about, or
causing to happen, the event described by the embedded verb. For illustration,
I will give a, of course, incomplete list of the verbs in this control class in
English, Russian, Czech, German.
(a) try, attempt, seek, endeavour, aim; (b) dare, venture; (c) manage, fail; (d)
hesitate, hurry; (e) be able to; (f) arrange, plan; (g) be determined, be
prepared to; (h) agree, refuse
Russian: (a) (po)pytat’sja, probovat’, starat’sja, stremit’sja; (b) osmelit’sja,
otvazit’sja; (c) udat’sja, (+dative-DP), ne uda(va)t’sja (fail); (d) lenit’sja (be
too lazy to), spešit’; (e) umet’; (f) rešit’; (g) zatrudnjat’sja/zatrudnit’sja (find
it difficult to); (h) soglašat’sja; (i) byt’ gotov (be ready)
Czech: (a) pokusit se, hledet (seek), snazit se (endeavour); (b) odvázit se
(dare); (c) dokázat, nedokázat (manage/fail); (d) váhat (hesitate), pospíšit si
(hurry); (e) umet (know (how to VP)), být s to; (f) chystat se.
German: (a) versuchen, sich bemühen (endeavour); (b) wagen, sich erkühnen;
(c) (es ) (nicht) fertig bringen, (es) (nicht) schaffen (manage/(fail)); (d)
zögern, sich beeilen; (e) fähig sein; (f) planen, vorhaben (g) entschlossen sein,
bereit sein.
This approach, which presupposes a propositional view of infinitives, helps
cope with a spurious puzzle that has arisen with this view and has been used as
116 Chapter 6
Clearly, (264), (265), with active and passive controlled clauses, are perfectly
acceptable, whereas, to say the least, we have to do a certain amount of work
to impose an interpretation on the passive variants of (266), (267), which has
an effect on their acceptability. The conflict between 〈intact,–〉 and 〈intact,+〉
coexisting in PRO reflects this situation. Its settlement depends on pragmatic
settings of discourse, situation and knowledge. There appears some appar-
ently paradoxical evidence in favour of this approach on inverting the strat-
egy: Passive controlled clauses may prove to be more readily acceptable than
corresponding active ones. Compare (268) and (269).
staralsja
(268)R Nikto ne pytalsja (PRO zapodozrit’ ego
stremilsja
Nobody (not)
tried
to suspect him
endeavoured
v eresi)
of heresy28
No Choice of Controller 121
staralsja
(269)R On pytalsja (PROi ne být’ zapodozrennym t i
stremilsja
[part pret pass masc sing]
tried
he endeavoured not to be suspected
v eresi)
of heresy
(269) is less bizarre than (268). In (268), zapodozrit’ does not seem to assign
〈intact,+〉 to PRO. If the theta role experiencer is assigned to PRO it will clash
with 〈intact,+〉 projected from the matrix verb. The minor or absent oddness of
(269) can be attributed to the natural behaviour of trying to avoid suspicion. In
other words, the specification 〈intact,–〉 assigned to the chain (PROi, ti) from
the theta-grid of zapodozrit’ in (269) is easier to reconcile with 〈intact,+〉 than
the internal specification of PRO is in (268). (cf footnote 26).
When 〈intact,+〉 is not available in the controlled clause, the clash between the
feature value 〈intact,+〉 assigned to PRO by the matrix verb try or its ilk and the
internal theta role it carries, cannot be handled in grammatical terms alone.
The coexistence, compatibility and conceptual-pragmatic balance of the two
characterizations modulate acceptability, more or less ready acceptance or
rejection of the pertinent sentences. The problem is again whether the event or
state described in the controlled clause can be construed as conceptually
compatible with, or accessible to, trying it. Lasnik and Fiengo (1974: 553)
mark (270) as ungrammatical.
(270) Billi tried (PROi to resemble Harry)
(271)R Chotja jai i pytajus’ (PROi byt’ pochozim
although I try to be like
na Christa …)
Christ …
‘… to resemble Christ’
122 Chapter 6
Koster (1984: 431, 433) argues that both are ruled out by the common labor of
binding theory (A) and Control theory. Accounting for (279) and (280)
distinctly, I assume that (279) is a violation of the theta-criterion, following
Vanden Wyngaerd (1987: 76): “… because try does not subcategorize for an
NP object, i. e. because the argument John receives no theta-role.” In other
words, a wrong lexical property (of try) is projected. Similarly, Iwakura
(1985: 31) argues against Koster’s analysis. As regards (280), its ungrammati-
cality revolves around two parameters.
The first can be roughly characterized as lexical sensitivity to the matrix
impersonal passive. I will sketch the problem first. “It is not possible at the
moment to give necessary and sufficient conditions for such structures. As
Williams (1980) points out, not all verbs with for-complements permit this
construction. But if these constructions are possible at all, the complement is
usually a for-complement.” (Koster 1984: 434)).
Williams (1980: 215) illustrates the unclear situation: “Decide is another
verb with an antecedentless passive: (53) It was decided to have dinner at 6. …
But decide does not allow lexical NP: (54) *John decided (for) Bill to get the
prize.” (Williams’ enumeration). (281) is an example in which the passive is
not antecedentless.
(281) It was decided by John to behave himself.
(281) is judged grammatical by Manzini (1983: 427) and Vanden Wyngaerd
(1987: 121), but a very similar example (282) is taken to be not acceptable by
Chomsky (1980: 35).
(282) *It is preferred to arrive a few minutes late.
Possibly, the ergative verb is responsible for the judgement. (283), however, is
considered grammatical by Williams (1980: 215) and Iwakura (1985: 47).
(283) It was preferred to have dinner at six.
German (284) is perfectly acceptable.
(284)G Es wurde vorgezogen, schon 7 Uhr zu frühstücken
it was preferred already at 7 to have breakfast
The range of well-formed impersonal passive control structures of type (281),
(283) does not seem to correlate with the occurrence of a complementizer
(for). The complementizer effects are epiphenomenal. Cross-linguistically at
No Choice of Controller 125
essayé de
secourir
(285)F * Il a été Paul.
visiter
it has been tried to help
Paul.
to visit
The German counterparts are fully grammatical.
poprobovano
pomogat’ Pavlu
(287)R *Bylo
posetit’ Pavla
[part pret pass neuter]
(it) was tried to help / visit Paul
Thus, control could be well-formed, as in the active, wherever the structural
type is grammatical, which depends on a parametrized variation whose factors
are not completely lucid. (For details, cf. Růzicka 1986a, 1988, 1992). The
parameter of impersonal passive constructions, if set positively, is still sensi-
tive to lexical properties (and projected syntactical effects, e. g. unergative vs.
ergative).
The second parameter that can be set up with respect to impersonal
passive constructions as well as to some associated control regularities con-
cerns conditions of well-formed occurrence of (the respective counterparts) of
by-phrases. Following Jaeggli (1986: 600) and Roeper (1987: 276 ff.), I
assume that “… an NP in a by-phrase is assigned the ‘pure’ external theta-role
126 Chapter 6
of the passive predicate.” (Jaeggli) This NP, overt or implicit, receives 〈intact,
+〉 from the theta-grid of pertinent predicates like try in the same way as the
external argument in subject position at D-structure in the active. The hypoth-
esized (cf. Dowty 1991: 605 ff; Burzio 1986); Růzicka 1992) parameter
consists in observing this parallelism or not. Clearly, the second parameter is
dependent on the first. But, as we will see immediately, they cannot be united.
Importantly, the parameters are effective in particular lexical classes, e.g. for
try, versuchen; dance, tanzen.
English sets the first parameter and, consequently, the second negatively,
differing, e.g. from German. If the parameter is fixed positively, (K) applies to
the passive as to active matrix clauses, 〈intact,+〉 being assigned to the NP of
the counterparts of by-NP and to PRO.
(288)G Es ist [PP von [vielen]i] versucht worden
it is (has) by many tried been
([dieses Problem]j (PROi tj zu lösen))
this problem to solve
This treatment extends to functional verb passive and verbal noun.
(289)R Byla sdelana popytka [NPe]i (PROi
was made (an) attempt (by somebody)
ugovorit’ ego)
to persuade him
Under Safir’s (1987: 562, 584) approach, the implicit (empty) instrumental-NP
is not mapped onto a syntactic position. The notion of “lexically satisfying a
theta-role” (Rizzi 1986) might be invoked here. I leave this problem unresolved.
The status of the second parameter is slightly complicated by cross-
linguistic variations in the (im)possibility of overt appearance of (the counter-
part to) a by-phrase with the NP carrying 〈intact,+〉. This is evidenced by
modern Czech. Judgments on the grammaticality of the occurrence of the
agent instrumental-NP vary.
(290)C … aby bylo *úcastniky provozu)
[pret neuter] [instrum pl]
in order that (it) was/be by the participants of the traffic
dbáno predpisu°
[part pret pass neuter] [genit pl]
observed (the) regulations
No Choice of Controller 127
We should hesitate to allow the assignment of a feature value, that is, 〈intact,+〉
to an implicit NP that can never appear overtly.
Instances of control that involve Raising, like (291) – (292), provide evidence
and further support for this approach.
(291)R Borisi pytalsja (PROi pokazat’sja (ti molodym
Boris tried to seem (to be) (a) young
[sportsmenom])INSTR
(sportsman)
(292)R Jai riskuju (PROi pokazat’sja vam (ti suevernym
I risk to seem (to) youDAT superstitious
i nedalekim)INSTR
and limited
(293) John tries to seem to be nice
(example from P. Jacobson 1990)
R. riskovat’ (riskuju) is among the verbal class which is subject to (K).31
Following Levin/Rappaport (1986: 646), I assume that “The XP assigns a
theta-role to the NP of which it is predicated, so this NP may be called the
external argument.” XP in (291), (292) is AP suevernym, NP molodym
sportsmenom).
The NP affected by Raising ends up as PRO in a non-theta-position, but
receives a theta-characterization through the chain (PROi ti). PRO now is
assigned 〈intact,+〉 by cross-clausal projection rule (K), e.g. in the derivation
of (291), (292). The two theta-specifications, that is, 〈intact,+〉 assigned across
clause boundaries and the internal specification of the chain (PRO, t) pro-
jected from the theta-grid of the predicative A(P) in (292) – (293), have to
come to terms with each other. The arising conflict is rather one-sided with the
theta-specification assigned by the AP- or NP-predicate to its external argu-
ment, the coda of the chain (PROi, ti), being quite a feeble characterization.
Thus, 〈intact,+〉, which PRO has received by virtue of (K), prevails over it to
its suppression, and the interpretation of, e.g., (291), (293), clearly, takes the
attempt (pytalsja, tries) to be directed towards giving the impression
128 Chapter 6
(pokazat’sja, to seem, to appear to) that Boris is a young sportsman, and John
is nice. This, indeed, is (part of) the well-formed meaning of (291) – (293).
The cross-clausal assignment of 〈intact,+〉 to PRO in virtue of (K) bears the
empirical burden of accounting for this meaning, capturing the special case of
two theta-specifications competing with each other.
Chapter 7
In sentences like (294) – (296) the crucial issue is how the controller is
“concealed” and how it is identified in the visible matrix clause structure. I
Reflexive clitic impersonal clauses in Control structures 131
have assumed so far that PRO is controlled by the pro-subject and its accom-
plice se whose interpretation is straightforward: It has either an “anonymous”,
that is, indefinite specific, or a generic reading. The latter is illustrated in the
Slovak example (298).
(298)SL Najviac sa dnes (pro) umiera na
[reflexive] [3 ps sg]
mostly today (one) dies of
srdcové chorobyACC
heart diseases
The assumption that in (294), (295) pro is the controller implies that it is an
argument. How does it relate to the reflexive clitic se or sa, siøe (Polish)? The
clitic is a lexical nominal element and I follow Cinque (1988: 536) in assum-
ing that it “… is part of a CHAIN with [NP,IP] …” The concept of CHAIN is
theoretically suitable to account for what the clitic does. Certain properties of
the zero subject-DP are a function of the reflexive clitic.
In the examples (296), (298), se, sa “conspires” with pro to furnish the
(subject) position occupied by the latter with a designated content, that is, to
reduce it to “anonymous” or generic meaning. This is nearly equivalent to
saying that the reflexive clitic enables pro (and the rest of the clause) to satisfy
F(ull) I(nterpretation) (see Chomsky 1986: 98 ff.): In an otherwise unchanged
context without the reflexive clitic, one would either have to look for some
quite different non-autonomous, e.g., discourse-anaphorical interpretation of
the zero subject, or consider the sentence defective. Compare (299) with (300)
(299)C (pro) Zadalo se, ze S
[neuter sing] refl
‘one demanded that S’
Agr, share relevant features and index with both. Besides [+person] (personal
Agr), there is no other (number) feature specification required in the CHAIN.
The reflexive structures of this type are incompatible with reciprocal anaphors
(see Růzicka 1992: 152). But we don’t have to specify pro as [–plural], since in
Slavic languages [3 ps sing (neuter)] is the default Phi-feature agreement
specification if the s-structure subject position is occupied by a non-lexical
(pro)noun lacking an inherent number feature [+pl] or by a (pro)noun carrying
[–nominative].32
The nonargumental clitic does what Cinque (1988: 530) — with respect
to Italian si — calls “… serve(s) as a syntactic means to supplement personal
Agr with the features able to ‘identify’ … the content of pro as an unspecified
(generic) person pronominal, an interpretation that would not be available to
the ordinary person inflection paradigm.” Note that the feature [+human] is
not a result of s-selection. The typical range of [+human]-external argument
interpretation rather is the result of an interplay of principles identical with, or
complementary to, those that result in passive. In other words, the particular
type of structures “inserts” compatible verbs that in turn select [+human]
subjects. Or, we might say, the reflexive formative licenses pro carrying
[+human].
I will try to elucidate how the impersonal structures under discussion
interrelate with the passive. The question arises how the reflexive clitic comes
to function in this way. I would like to suggest that what ends up as the two
structures just mentioned are consistent arrays resulting from the (co-)opera-
tion of identical and complementary principles, the latter being associated
with distinct sets of lexical predicates (verbs), transitive vs. nonergative. In the
reflexive passive, for example in Czech, the subject A-position is dethema-
tizid by the clitic se, which, being argumental in these structures, binds or
“withholds” (Cinque 1988: 535) the external theta-role, and becomes the
target of object movement. With intransitives, in particular unergative verbs,
the subject position is held open again by constituting a CHAIN with the clitic,
but in this case no object is available to move in. So if the reflexive clitic need
not dethematize the subject position and can be taken to be nonargumental, it
seems to be obvious from the viewpoint of linguistic economy that what the
reflexive clitic provides for by being part of a CHAIN with [NP, IP], namely
the empty subject position be utilized as an argumental pro for Full Interpreta-
tion: pro is an argument to be interpreted so to speak according to its phonetic
content, namely as “anonymous”, indefinite specific or generic. In both
Reflexive clitic impersonal clauses in Control structures 133
related structures, the theta role Agent is affected, with its respective
DP-recipients being demoted and suppressed (in the passive) or deprived of
referential readings (in the impersonal reflexive structures). As we shall see,
the reflexive clitic se (or sa, siøe) on which these properties hinge must be
marked [+person] in the reflexive passive as well.
(301)C Kongresi se peclive pripravil ti
[nom masc sing] [+person] [masc sing]
congress self carefully prepared
‘The Congress
has been
carefully prepared.
was
structural Case position of the direct object is closed, it seems natural that
another argument position, the subject position, must open in compensation.
Controlled clauses are incompatible with the reflexive passive in Slavic
languages, a fact which conforms with Cinque’s (1988: 561) observation that
“Infinitival control structures are generally taken to be incompatible with
passive si …” Cinque continues “… passive si appears to be compatible with
control … where … pure middle interpretation is possible …” But, for example,
in Russian, even control cases in which an embedding noun “prompts” pure
middle interpretation seem barred.
(302)R *Eti okna imejut svojstvo (PRO legko myt’sja)
these windows have the property easily to wash-refl
(303) illustrates that reflexive non-middle passive cannot be realized in con-
trolled clauses.
Capturing the difference between the two passive constructions and their
respective CHAINS34 hinges on the elements which are linked to the clitics.
In Slavic, “transitive” reflexive passives have the as yet unfilled subject
position e linked to, and dethematized by, the argumental reflexive clitic, an
operation falling under the notion of functional CHAIN. In -en-passives, the
by-phrase, its counterparts, or the corresponding implicit (IMP) is linked to the
clitic -en. In these considerations, I assume that the Russian or other Slavic
“analytic” passive is similar to the -en-passive in the relevant aspects. In the
-en-passive (and its counterparts), the morpheme is an argumental affix tied to
the Infl node which unequivocally “identifies” the argument and the theta-
role, as the subject-NP explicitly does in the corresponding active. In contrast,
the occurrence of counterparts to the by-phrase is strongly restricted and in
some languages even excluded in the reflexive passive. Besides, object move-
ment, common to both types of passive, is unique to passive structure only in
the analytic type (cf. Růzicka 1992: 138).
Crucially, the passivization process is not unequivocally determined by
the reflexive element, that is, the surface subject of verbs to which the
reflexive clitic is attached may retain external argument position, and it is
privileged to contain a moved d-structure object of ergatives (cf. Růzicka
1992). Properties of the s-structure subject and other contingent lexical prop-
erties take essential part in determing a passive or non-passive interpretation
of reflexive structures. In general, passive interpretation in reflexive structures
is impeded by the presence of [+human] in the PRO-matrix. Control verbs,
however, predominantly are accompanied by [+human] for PRO in their
complement. As a consequence, pertinent Control constraints, in particular
(K), often have no unequivocal theta-specification, namely 〈intact,–〉 to apply
to in potential reflexive passive structures, in contrast with the analytic pas-
sive. Cf. (307).
(307)R *Otec popytalsja (PROi vozvrašcat’sja
[+refl.]
‘(Their) father tried to be given back
detjam)
[dative]
to (his) children.’
This seems to be the intuitive empirical underpinning of the preclusion of
controlled reflexive passive, beyond its possible Case-theoretical account.
Chapter 8
There are two basically distinct situations that circumscribe the range of
control and constitute two distinct subsets (a) and (b):
138 Chapter 8
complement. In other words, only one argument or at least only one theta role
is available, basically (the chain headed by) PRO, which either is resumed in
the matrix clause in overt shape or is left in the impliciteness or “arbitrariness”
of PRO. Note that the evaluation associated with the matrix DP is inseparable
from, or identical with, the evaluation of the intentional action of the comple-
ment clause and valid only with respect to it. Expressions like Peter is clever
are understood as relating to his cleverness at something. Stowell (1991: 112)
clearly recognizes that “… MP (mental property, RR) adjectives cannot
attribute a property to an action without simultaneously attributing this prop-
erty to the agent of that action … the human argument in question must be
understood as the agent of the action-denoting argument.” I would like to put
forward the suggestion that the matrix PP, e.g.. in it is stupid of John does not
occupy a theta-position, and that its DP is raised from the embedded subject
position leaving the chain (Johni, ti). By of-insertion, which adjectives permit
(Chomsky 1986b: 89), the Case filter is met. The PP s tvoej storony is a clearly
idiomatic variant. I will not pursue this radical proposal of a unified theta-role
here, returning to a standard treatment, which can be transformed into the
proposed one.
Quite independently of whatever thematic specification PRO receives as
an argument of the complement predicate (VP or AP), its intended referent is
accessible to the matrix characterization only if s/he is understood as “a
decision taking” individual, a fact which might be expressed by PRO being
provided with 〈intact,+〉, via projection from the matrix predicate somehow
inherently marked with it. But this proposal cannot be fully maintained.
Consider (322), in which PRO receives 〈intact,+〉 originally from the theta
grid of otsizivat’sja.
(322)R No edva lirazumnee bylo (PROi otsizivat’sja
[pret neuter]
but hardly more sensible(it) was to take shelter
odnimi bez oruzija sredi etoj pustyni)
[gen sg]
alone without arms in the middle of this desert
The dative plural of the semipredicative odnim, which is coindexed with PRO,
refers to the discourse-related set of people to the exclusion of others. It must
not be exactly this set of people whose potential decision is judged compara-
tively. But now consider (323), which must be ruled out since no decision
“Self-control” with evaluating and attitudinal predicates 145
taking individual or set of people are available. Only PRO could provide one
saving (323) if it were supplied with 〈intact,+〉 on the strength of feature
transfer or discharge.
(323)R Bylo glupo (PROi byt’ priglašennym ti na etu
(it) was stupid to be invited to this
konferenciju)
conference
Compare: It was stupid to invite him to this conference. Thus, transclausal
projection of 〈intact,+〉 to PRO from matrix predicates of type (b) cannot be
admitted generally. This restriction will be stated as (L), which is similar to
(K), a specification on the application of the latter with respect to predicates of
type (b)
(L) The presence of 〈intact,+〉 somewhere else in the complement clause
prevents another instance of it from being assigned to PRO, (across
clausal boundaries)
As a third case, consider (324), which is perfectly acceptable.
glupo
(324)R Bylo by nerazumno (PRO predavat’ sja
[modal particle]
stupid
(it) would be insensible to indulge
melancholii
[dat sg]
in melancholy
This short chapter has been included in the investigation of Control since
independent solutions to the grain problem as it emerges in Control, and to
anaphoric contexts in VP-deletion, are perfectly consistent with the theory put
forward in this study, providing instances of its formal indexing conse-
quences.
The problem pointed out by Higginbotham (1985) and called “grain” problem
“… reasserts itself in an even more pernicious form when simple control
constructions, such as those in (45) and (46), are brought into the picture.
(45) John and Mary persuaded each other to leave.
(46) John and Mary promised each other to leave,
… But the rub is that (45) and (46) are not ambiguous in the same way as
(43). …
(43) John and Mary told each other that they should leave.
… this sentence is at least three ways ambiguous between what we termed the
‘I’, ‘you’, and ‘we’ readings … (45) has only a ‘you’ reading, (46) only an ‘I’
reading …” (Heim, Lasnik, May (1991: 79)) The authors continue that “It
would seem then that unless we build the theory of control (redundantly) (my
emphasis, RR)) into the semantics of these sentences, we are left without a
satisfactory account of their interpretation.” Any description of control, of
course, must account for the distinct control properties inherent to persuade
and promise, respectively. Since the reciprocal each other occurs in object
152 Chapter 9
position but in fact conceals a subject-object relation, it must be taken apart for
“subject”- and “object”-control to be theoretically reconstructed. But the
complex reciprocal cannot be simply taken asunder, as each “instance” of
other is determined by, and must be related to, its distinct “distributing”
partner each. Heim, Lasnik and May (1991: 80) present a basic representation
of reciprocal sentences which “… provides three loci of anaphoric binding:
relative to the distribution index contributed by each, relative to the index of e
other, the reciprocator, and relative to the range index of the antecedent
subject NP.”
Resolving the complex reciprocal into the subject-part each and the
object-part e other, with the empty element coindexed with each, enables the
authors to construe the respective LF-representations (50), (51) (their number-
ing) for (45), (46).
(50) (John and Mary1 each2) (persuaded (e2 other)3 (PRO3 to leave))
(51) (John and Mary1 each2) (promised (e2 other)3 (PRO2 to leave))
PRO is coindexed with the subject-part each for promise, and, for persuade,
with the object-part (e other) of the reciprocal each other. Either isolated
element is to be considered an argument and recipient of a theta-specification,
that corresponds to their respective subject and object positions.
Thus, both of them are accessible to the generalized lexically-driven
constraints (B,b) and (B,a), respectively. These constraints and the above
indexing and binding device at LF-representations combine to give a full
account for the control behaviour of the two classes of verbs in reciprocal
constructions. An analogous analysis would be practicable, e. g. for the
German, Czech, Russian counterparts of each other control constructions:
g. ein-ander, c. jeden druhý, -ého, and r. drug druga. The Slavic languages
exhibit the object function of the second part by morphological Case36 .
I would like to stress that the success of the intriguing indexing system of
Heim, Lasnik, May (1991) on its application to differences of scope interpre-
tation and to control properties interacting with reciprocal constructions,
substantially relies on “… inherent semantic properties of the controlling verb
and properties of control PRO … narrow scope is excluded when the control-
ling NP must be under the scope of a distributor, as a function of the lexical
meaning of the predicate of which it is an argument … the possibility of broad
or narrow scope is a function of the semantic status of the controller of PRO.”
(Heim, Lasnik, May 1991: 99). Thus, the main claim of this study that, in the
Aspects of control in the “grain problem” and in VP-deletion 153
What has received the slighty metaphoric name control is a linguistic syn-
drome rather than a module or subsystem of grammar that could be cast in a
unified form. The complex of problems posed by the phenomena standardly
subsumed under control nevertheless can be disentangled and organised into
an array of constraints that determine control behavior and set aside factors
which are epiphenomenal to, if necessarily accompanying, control.
The natural prerequisite of attaining descriptive generalizations about
control conditions is to specify the structural descriptions that must or can
display control. The specification proposed divides the realm of control into
relevant domains and selects for analysis those that are related and subject to
principled constraints which can extend a certain homogeneity to the whole
field. Crucially, the interplay between lexical semantics, thematic structure,
syntax and explicated pragmatics hinges on the respective constraints and
their format.
In the major syntactic control domains, nonfinite adjunct clauses differ
fundamentally from complement clauses. In the former domain, control prop-
erties of participial adjuncts differ from those of purpose or rationale infiniti-
val clauses. Control in nonfinite adjuncts has not been dealt with here — for
modern Standard Russian, a detailed description of control in participial
°
adjuncts is Ruzicka (1982) with further literature — but the theoretical spirit in
which it could be approached, as well as the methodological line to be taken
have been suggested (2.1.3.).
In the area of infinitival nonadjunct clauses, control in subject clauses
differs considerably from control in object clauses. “Object position is pro-
156 Chapter 10
jected only from thematic structure …” (Rothstein 1995: 500). Subject posi-
tion is projected syntactically and need not be thematically related to the
predicate. Object clauses but not subject clauses exhibit strongly marked
lexical determination of control, a fact which has to do with the assumption
that the theta role of the subject — if it has one — is determined composition-
ally by the VP (Chomsky 1981: 105). Object complement clauses are internal
sentential arguments of the control verb which theta-marks them. Their PRO-
argument, which is the head of an A-chain, needs identification. It has been
shown in this study that the identification of PRO ultimately hinges on
specific properties and functions of the internal clause argument, the con-
trolled clause. That is why there is a fundamental difference between control
in object sentences and control in subject sentences, which, among other
things, has contributed to the use of notions like arbitrary and logophoric
control.
The distinct types of control verbs, by thematically differentiating be-
tween their respective clausal complements, indicate how the content of the
latter is to be incorporated in the sentence. In the literature, distinctions
between thematic properties of clausal complements that arise from lexical
characteristics of control verbs and in turn induce control conditions have not
been given careful consideration. The clausal complements of, e.g. ask, prom-
ise, or try and their respective crosslinguistic counterparts have been treated
with a certain disregard of such particular pairing of thematic specifications as
could possibly attend or determine control conditions. Distinctions and influ-
ences that have this effect are taken to be a precondition of the main general
thesis of this study:
Thematical properties of clausal complements that are determined by their
respective matrix control verbs are projected to, and distributed among, argu-
ment positions of the clausal complement and come out clearly to the effect
that they impose distinct restrictions on control relations.
Thus, conditions of coreference for the complement PRO are created by
lexically thematic properties of the matrix control verb. PRO is required to
have thematic specifications meeting those of a licit controller. How it must
meet them depends on, and varies with, lexical classes of control verbs, which
in turn are established by the respective types of referential interdependence
they command.
Summary and conclusions 157
verbs as against object control verbs. Ask. beg, for example, are verbs of
“object” control if control is licensed on the unmarked application of (II), but
they are verbs of “subject” control on the marked application of it, (sec-
tion 3.2.4.2). Promise-type verbs behave in the opposite way with regard to (I)
(section 5). I have explained why this is so. Importantly, if we rely on a
markedness concept as specified here for complement control, various forms
of more intricate and subtle control behavior can be included in, and captured
by, the basic system of constraints. Specifications on the unmarked and
marked application of (B, a) and (B, b) provide us with criteria substantiating
a subclassification of the main control classes associated with the respective
constraints.
I will present now a survey of the lexical control classes associated with the
respective constraints and specifications on their application. More or less
exemplary lists of members of the respective classes are added. The introduc-
tion of a systematic cross-linguistic perspective has justified the attempt to
formulate interface (conceptual-pragmatic) and syntactic parameters, which
suggests that pragmatics could be parameterized. The combined framework of
the respective constraints, corollaries specifying their application, and param-
eters, namely (B) – (L), is taken to account for the complement control mecha-
nism. (Order of succession as in the main body of the study)
parameters (G) and (H). Marked control under (II) is substantiated by the
activation of the “rest” of available thematical specifications, which display
complicity in control by inducing a less “conspicuous” referential pairing of
arguments (section 3.2.3.).
(1)G Siei hatte gebeten (PROi im Kloster
she had asked in a convent
ti untergebracht zu werden)
accomodated to be
‘She had asked to enter a convent.’
A natural restriction put on the “weaker” marked control in this subclass is
expressed by (F): The more transparent linking of the relevant thematic
feature value 〈F, +〉CON to the underived subject position must be maintained
for it to be targeted by the marked application of (II) (3.2.2.).
Violations of marked (II) (*…〈F, +〉PRO ) give rise to a parameterization in
pragmatics which consists in rescinding them by means of the allowance
construal — which is considered the positive setting of the parameter — and
ends up in a thematic two-place configuration that satifies the very marked
constraint (II) that has been violated. The presumed parameter
((H), section 4.1.) is contingent upon the strength of the effect the null object
parameter has on the availability of (marked) subject and (unmarked) object
control. Positive setting of that parameter, as in Russian, may exclude marked
(subject) control — via the allowance construal as in the English example
(91) — since the null object satisfies unmarked (II).
(2)R Ja ne prosil proi (PROi menja spasat’)
I not asked me to save
‘I did not ask (anybody) to save me.’
English, for example, which has no syntactically overt pro excludes it from
functioning as an unmarked controller in accordance with Bach’s Generaliza-
tion. If in Russian or German, for example, control by the pro-object is
mandatory — given that unmarked (II) is met —, English has to rely entirely
on “derived” marked (subject) control, a situation which induces the allow-
ance pragmatic-conventional construal (see example 91). Spanish evidences a
strong instantiation of the conventionally pragmatic parameter (H) which is
not linked with the null object parameter (section 3.3.). The presence of a
lexical object argument does not exclude the interpretation licensed by the
160 Chapter 10
guages. ″b. Both Italian and English have an understood null dative that is not
restricted to generic contexts, can be pragmatically interpreted (can be dis-
course-related, RR), and can function as a controller, but not as a binder.”
(Rizzi 1986: 551)
In German and Slavic languages (morphological) dative is the counter-
part of the direct object controller in English, a fact which simply means that
for (II)c the internal dative argument carrying the indirect object theta role
gains broader access to control. For the passive matrix sentences, it follows
that derived subject control in English corresponds to unchanged dative
control in the languages mentioned above. The result is impersonal sentences,
since no object is moved.
(10)R Emui bylo poruceno (PROi napisat’
[dative] [partic. pass. neuter]
(to) him (it) was directed to write
neskol’ko bumag v St. Peterburg
[genitve pl.]
some papers to St.-Peterburg.
‘He was charged to write some papers to St.-Peterburg.’
The dative controller can occur as a pro-DP in the passive as well. I would like
to emphasize that dative control is not affected by theories taking dative to be
configurational, not inherent.
Whereas the ask-type and persuade-type subclasses of (II) are subtly set
apart by their respective reactions to, and remedies against, violations of
unmarked and marked application of (II), which imply cross-linguistic varia-
tion, there is no analogous division in the domain of (I) for promise-type
verbs. The marked application of (I) is straightforwardly complementary to
the unmarked one, reversing the value of α.
Example (181–184) illustrate marked control under (I) with passive
matrix clauses, (185–188) do the same with active ones. The particularly
elucidating example (182) combines constraining grammar and conceptual
construal by reinterpreting the complement verb in a way that matches marked
(I). Passive is impersonal since the internal dative object is not moved to
subject position. If no nonpleonastic subject-DP competes, as in German and
Slavic languages, unmarked control by the demoted Agent-DP is facilitated
(example 189, 196–198). In English the derived subject argument may be
responsible for ruling out unmarked control under (I) by the demoted agent in
the by-PP (cf. English glosses of (196) and (198)). The descriptive generaliza-
tion (I) (in 5.2.) captures the cross-linguistic variation, that is, Case distinc-
tions resulting in different control behavior.
If marked (I) (B,a) is not met because (D) is not satisfied — recall that (D)
is a condition or specification on the marked application of (I) and (II) as well
as on the operation of (G), (H), acceptability can only be salvaged by a
conceptual construal which amounts to the restitution of a structure that in turn
meets (D) and can be referred to by the marked constraint (I): (187, 188) are
handled by analogy to (193, 194) respectively. Another striking example is
(219) as compared with (218). The former is clearly ambiguous, with the
second reading which takes the children as controller being forced on the
sentence by its pragmatic naturalness. The “natural” interpretation is not
indifferent to grammar. To become acceptable, it must find its way to a
licensing constraint, which is the marked application of (I). The conceptual
construal can be made visible by the insertion of be allowed taken as a
metalanguage “bridge” from pragmatics to grammar. (218) has be allowed in
its structure and apparently meets marked (I) including (D). It should prove
unproblematic. Still, there are judgements that prefer “subject” control. This
situation may throw some light on the conceptual status and firmness of the
grammatical notion of Control. Control may prove susceptible to a complex
array of factors in which the syntactic ones appear more reliable but can lead
astray. The available control structures may offer options of Full Interpreta-
Summary and conclusions 165
cally, vorschlagen is a verb of control that belongs under control class (B, b)
(= II). The application of the pertinent constraint, however, needs refinement.
The oblique (dative) argument, the Addressee argument of the event of
proposing, is (one of) the controller(s) whenever the corresponding DP is
lexically overt. The strength of the probability that the proposing person will
include herself into the participants of the proposed event or action is fully
discourse-related depending on the pragmatic situation. Even third persons
may be viewed as participant(s) of the proposed event, but still the addressee’s
consent or agreement about what is proposed is requested. Other persons may
come into question again depending on the discourse situation.
(24)G Ichi schlage vor (PROi,X in der Bibliothek zu arbeiten,
I propose, suggest in the library to work
während du das Zimmer aufräumst)
while you the room are tidying up
In German, the occurence of reflexive versus logophoric anaphors in the
complement clause may clear or disambiguate the possibilities of coreference
(Kamp/Roßdeutscher 1994). The issue of varying pronouns and their possible
help in interpreting coreference is beyond the scope of this study.
The Case of PRO has been under discussion (see in particular Chomsky/
Lasnik 1993). The results remain inconclusive: “The state of PRO with re-
spect to Case marking and visibility is problematic … the problem arises in
particular in Chomsky’s (1981, 1986b) theory with respect to what constitutes
a visible head of a chain … Either PRO is directly stipulated to be visible, or it
is stipulated to receive some suitably abstract type of Case, so that visibility
follows on standard Case-marking grounds.” (Manzini 1992: 144). Pesetsky
(1993: 558) leaves open the question “… whether quirky zero Case is the
same as null Case for PRO recently discussed by Chomsky and Lasnik.”
Avoiding a theory-internal stipulation, a solution based on empirical
evidence might receive support, for example, from Icelandic and Slavic
languages in which “… PRO may be not only Case-marked, but in fact
marked with one of the standard Cases rather then a special ‘null Case’”.
(Baltin 1995: 242) Baltin points out Sigurdsson’s (1991) work on Icelandic.
In Russian, e.g., the subject of infinitivals is marked standard Dative Case. If
PRO receives a special sort of Case, which is taken to be peculiar to subjects
of infinitives, one theory-internal advantage of this account is — considering
that disjunctive statements are not desirable — that “… a disjunctive require-
ment that A-chains must be headed by Case or PRO may be dispensed with for
one that says that A-chains must be headed by Case.” (Safir 1996: 314) In a
recent publication on Control, Petter (1998: 244) endorses Chomsky/Lasnik’s
(1993) assumption that “… if PRO is syntactically represented it must also be
assigned structural Case.”, but she specifies Case assignment quite differently:
“… we may conclude that PRO is assigned structural nominative Case … in
both non-finite and verbless clauses a(n abstract) Tense head assigns struc-
tural nominative Case to the subject in its specifier position.” (1998: 245). The
conclusion is mainly supported by Icelandic data which “… show that predi-
cate agreement only occurs when the predicate is related to a subject with
structural nominative Case … this type of agreement also occurs when the
subject is PRO.” (245) I perfectly agree with Pepper about drawing on the
nominative as a likely candidate for PRO. Still, language particular and
structural requirements seem to modify and restrict possible generalizations.
Thus, as mentioned above, in Russian, infinitivals strongly suggest structural
Dative Case for PRO, whereas for PRO in participle adjuncts and verbless
constructions which are closely related to them structural nominative is the
°
appropriate Case. (cf. Ruzicka 1982)
Case-marking of PRO, which never is an expletive, does not affect the
174 Chapter 10
central aspects of control discussed in this study. Nor does the categorization
of PRO. I would like to regard PRO as a pronoun like pro, which seems to
come nearest to the truth considering the available theoretical and in particular
conceptual tools. PRO can be an instance of coreference anaphora and it can
be a case of bound variable (Heim, Lasnik, May 1991: 99). “It can enter into
anaphoric relations without binding.” (Reinhart/Grodzinsky 1993: 73) PRO is
replaced by pro, which takes over control in noninfinitival complements. The
need for control principles and constraints, their interaction with syntax and
pragmatics, does not depend on the final categorization of PRO.
Chapter 11
sive suffix (cf. Baker/Johnson/Roberts (1989)), has the effect of causing 〈F,+〉
to prevail over 〈F,–〉 on PRO, which results in a being satisfied. In other
words, PRO receives 〈F,–〉 in object position and takes it to subject position by
passive movement. Here the contrary value 〈F,+〉, for which try is responsible,
is joined to it. PRO, on which two competing values meet, benefits from the
internal presence of 〈F,+〉, e.g. in passive structures, to establish dominance of
〈F,+〉, which is a natural general characteristic of try-event structures. The
relationship can be taken as a necessary condition: 〈F,–〉 is available on PRO
only if 〈F,+〉 is present in the same predicate structure (see (D) in 3.2.2.). I
assume that grammaticality (or acceptability) judgments are less straightfor-
ward (see, e.g. (267)) if a is satisfied via “strengthening” 〈F,+〉 on PRO. This
will become clearer on discussing (270). PRO in (270) receives 〈F,+〉 on the
strength of intrinsic lexical properties of try implemented by (K) via an
appropriate checking configuration, but — in accordance with the above
condition — does not receive 〈F,–〉. Whatever feature is discharged on PRO
by the predicate resemble, the conceptual distance between it and the properly
interpreted feature value 〈F,+〉 is relevant to judgments on (270), which have
ranged from “anything goes” (Vanden Wyngaerd1987: 75) to “ungrammati-
cal” (Lasnik/Fiengo 1974: 553). Still, we are not faced with a grammatical
decision. Positing the cooccurrence of the two features on PRO is all grammar
can do. Readiness to construe a conceptual-pragmatic interpretation and con-
textual elements (see ex. (277), (278)) decide on (degrees of) acceptability.
Just as try and the like (6.1.1.) are not exactly verbs of subject control (cf.
(284)–(289), so the subclass of ask, beg, etc. (cf. 10.3.1.1.) does not classify
object control verbs either. Criteria for the latter are coherent with those of the
former in the logic of control. Whatever judgement is made on (270) entails
the same judgement on (2).
(2) ?*John asked Bill to PRO resemble Harry.
Nevertheless, the situation changes a lot with verbs of this subclass. Most
importantly, they are tagged in the lexicon with the (meta)feature ~α. Recall
that the feature values are distributed to PRO and to arguments in the immedi-
ately superordinate part of the sentence. Arguments of two distinct argument
structures can enter a principled control relationship since the argument
structure of the complement predicate constitutes a clausal argument of the
matrix predicate. Metafeature ~α, with α ranging over values of a properly
interpreted (intentional action or originator (Manzini/Roussou 1997)) feature
F, projects lexical properties of the ask-class on basic conditions of
(un)marked control.
(2) is gibberish or made acceptable by pragmatic construal on equal terms
with (270): constraint ~α would require that the controller NP have the reverse
feature value to that on PRO, a requirement which no more can be satisfied
than constraint a in (270). Neither John nor Bill in (2), can take a feature value
reverse to that of PRO, since they have 〈F, +〉 and 〈F,–〉 respectively, and PRO
is marked neither 〈F,+〉 nor 〈F,–〉. Recall that the latter is available depending
on the former. Thus, PRO cannot have 〈F,–〉 and be controlled by virtue of 〈F,
+〉 on [NP John]. PRO in derived subject position after passive movement
carries feature value 〈F,–〉 with ~α of ask selecting the DP-subject in situ as
proper controller.
(3) The injured forwardi asked the coach to PROi be substituted at half-
time.
Recall the hierarchical restriction (F) on ~α (ex. (75), (82); 3.2.2)
(4) *The coachi was asked ti by Peterj to PROj be included in the select
team.
180 Chapter 11
11.4.4. Markedness
The approach adopted in the main body of this study is not wedded to the view
that numerations to be mapped to control structures which satisfy output
conditions at the LF-interface must contain PRO. What I think is needed is an
apparatus in which the entities are available on whose properties and interac-
tion control hinges conceptually and empirically and without which an
account of control cannot be a natural explanation. In the spirit of the
minimalist program, positing as well as dispensing with an entity like PRO
should not be the outcome of invoking construction-specific assumptions. If
182 Chapter 11
To investigate and specify the nature of the compatibility constraints has been
the main concern of this study. In terms of selectional constraints, the matrix
Turning to the Minimalist Program 183
1. Stump (1985) has documented and interpreted grammatical (including semantical) re-
strictions on the variability of the logical relations between free adjuncts or absolutes and
their respective matrix clauses.
2. “An obvious problem with (20) (our (28), RR) as a sufficient descriptive condition is that,
while (20) predicts that both subject and object control are possible in all of (12)–(15)
(our (29)–(32)), only object control is possible in (12) and (15), and only subject control
is possible in (13) and (14).”
3. (20) (our (28)) is based on a series of formalizations. The c-domain is defined as follows:
″(1) γ is the c-domain of α iff γ is the minimal maximal category dominating α.
(2) α c-commands β iff the minimal maximal category dominating α dominates β.
(3) α governs β iff
a. α is a lexical category, and
b. α and β c-command each other.”
Manzini defines a domain-governing-category for an element as in (i):
″(i) (her (31), RR)
γ is a domain-governing category for α iff
a. γ is a governing category for the c-domain of α, and
b. γ contains a subject accessible to α.
Further,” … PROs like NP-traces are pure anaphors …
″(ii) (her (32), RR) An anaphor without a governing category is bound in its domain-
governing category.” (p. 424)
The examples with PRO in object sentences as in (our) (29)–(32) and (33)–(36), involve
configuration (iii) (her (33)):
(iii) S
NP Infl VP
V S′
PRO
188 Notes
“In (33) the c-domain of PRO is S′, since S′ is the minimal maximal category dominating
PRO. Further, the governing category for the c-domain of PRO is S; for S is the minimal
category containing S′, a governor for S′ (V), and a subject accessible to S′ (NP or Agr).
Finally, S, the governing category for the c-domain of PRO, contains a subject accessible
to PRO (NP or Agr again); for NP and Agr c-command PRO and coindexing of NP or Agr
with PRO does not violate the i-within-i condition. Hence, by (31) (our (i), RR), S is the
domain-governing category of PRO; and by (32) (our (ii)), PRO is correctly predicated to
be bound in S.” (Manzini, l.c. pp. 421, 424, 425)
4. Such cases are illustrated by (i)–(iv) (Manzini’s (26)–(29):
(i) [PRO to behave oneself in public] would help Bill
(ii) [PRO to behave himself in public] would help Bill
(iii) Mary knows that [PRO to behave herself in public] would help Bill
(iv) [PRO to behave himself in public] would help Bill’s development
“The PRO can have arbitrary reference as in (i); or it can corefer into S, as in (ii), into a
phrase superordinate to S, as in (iii), or into a phrase subordinate to S, as in (iv).”
(Manzini, l.c.: 424)
5. “In (34) (our (38)), as in (33) (= (iii) in fn. 3), the c-domain of PRO is S′ and the
governing category for the c-domain of PRO, S′, is S, where the governor and accessible
subject for S′ is Agr. But in (34), S, the governing category for the c-domain of PRO, does
not contain a subject accessible to PRO; for S′ and Agr c-command PRO, but
cosuperscripting of S′ or Agr and PRO would violate the i-within-i condition. Hence, by
(31) (= (i) of fn. 3) PRO does not have a domain-governing category; (32) (= (ii) in fn. 3)
then does not apply, and PRO is correctly predicted to (co)refer freely.” (l.c.: 425)
6. The account of the Italian-French contrast with respect to control in the presence of se/si
(Gianni non sa se andare el cinema, ‘Gianni NEG knows if to go to the movies’ vs French
*Marie ne sait pas si aller au cinema) hinges on Kayne’s assumption “… that in the
infinitive-clitic languages like Italian, the infinitive will in the general case move into a
position that is hierarchically closer to PRO — left adjoined to the l′ — than the position
it moves into in the clitic-infinitive languages. I would like to propose now, that in so
doing the infinitive in (87) = (… se … [IP PRO … [l′ Vinf + [l’ … (Cl+) l …) blocks off
government of PRO by C0 and thereby eliminates the potential PRO theorem violation
induced by that C0.” (Kayne, 1991: 674)). Italian se, thus, does not govern PRO, because
Vinf is a closer governor, with the notion of Minimality slightly modified. The PRO
Theorem is not violated.
7. (c) includes Farkas’s (1) and (4) of “… the main questions any theory of control has to
address … (1) What are the principles which determine the set of possible controllers? (4)
What determines the choice of the controller in case the set of possible controllers is of
cardinality greater then 1?” (Farkas, 1988: 27, 28))
8. Here is a collection of pertinent statements. Control theory “… crucially involves θ-grids
… Control is a syntactic process, consisting of the coindexing of PRO with the designated
θ-role; therefore, the designated θ -role must be visible in the syntax, when the control
module applies.” (Rizzi, 1986: 552). “The system of thematic relations … forms an
important basis for a significant class of control phenomena.” (Nishigauchi, 1984: 217).
Bresnan (1982: 404) assumes “… thematic constraints on anaphoric control …”. A
particular treatment of control in the spirit of thematic relations has been advanced by
Jackendoff (1987: 371 ff.): “All of this is by way of justifying the introduction of
Notes 189
thematic relations as part of the account of control … The fundamental point, from which
all else proceeds, is that thematic relations are part of a level of semantic/conceptual
structure, not part of syntax.” Interestingly, Williams (1992: 316) releases PRO from
being the target of identification in adjunct control: “The adjunct control cases stand apart
as cases that do not involve PRO, but rather direct theta role assignment.” This radical
step is very revealing. The reason seems to be that in adjunct control thematic properties
of lexical items that determine or influence control cannot be systematized and be
marshalled into generalized superordinate contraints. They rather must operate on their
own considering characteristics of individual events, deictically or logophorically fixed
understood reference. Recall the discussion of examples (15)–(16), or consider
Higginbotham’s (1992: 105) example (80) (our (i)), which he adds to Bach’s example
(ii):
(i) Is that a book [PRO to read to each other]?
(ii) Here’s a book to read to each other.
Higginbotham comments on both examples: “We have seen that ‘pragmatic control’ of
PRO is permitted as in Bach’s example (ii). But is it PRO itself that is pragmatically
controlled, or rather an understood benefactive, which in turn serves as antecedent of
PRO? The thrust of my discussion suggests the latter, but, it is not easy to test the
question. Even an example such as (80) (our (i), RR) could, it seems, contain a benefac-
tive … containing within the predicate nominal an understood ‘for NP’, with NP serving
as antecedent of PRO.”
Still, the basic problem of control remains the same. The antecedent of PRO is established
by discourse-related construal. The thematic specification which is assigned to PRO by
read is of course compatible with benefactive, it even predicts the latter for the under-
stood or overt NP in the PP of the matrix clause. In this situation, we might say that
control is reversed with PRO “controlling” the antecedent which it creates and which
necessarily matches it thematically. A similar “reversal” of control will come up even in
complement control (section 8).
Naturally, not the whole community of investigators of control are in unison with a θ–
role-driven approach to it. Emonds (1985: 106) tries to accomodate control to principles
holding in other modules: “It can thus be concluded that all obligatory control can be
reduced to subcategorization and the principles which govern it.” Koster (1984: 431)
relies on control theory and on binding theory in accounting for his (31) *John was tried
[e to go]: “This is, however, the point where the independent theory of control comes into
play … it is this theory that explains (31) …” With respect to the same sentence, he draws
on binding: “It is here that the independent binding theory comes into play.”
9. The above interpretation of “Intentional action” implies the belief that the identified
referents of PRO and its licit controller have control over the action described in the
complement clause, a characteristic feature figuring in Farkas’ (1988) analysis.
Davidson (1989: 46) stresses that “It is a mistake to suppose there is a class of intentional
actions: if we took this tack, we should be compelled to say that one and the same action
was both intentional and not intentional … Hamlet intentionally kills the man behind the
arras, but he does not intentionally kill Polonius.”(p. 46).
10. This assumption finds support in Døabrowska’s discussion of “Dative and nominative
experiencers” (1994: 1029)
11. Larson (1991: fn. 19) states that “… ask is normally (my emphasis, RR) object-con-
trolled. Passive in its complement appears to license subject construal …” Since subject
190 Notes
control in such cases does not fit in with Larson’s straightforward configurational
account of control (in which c-commanding and subject control are harmonised by
assuming a D-structure that places the inner object (of Double object construction) in a
non-c-commanding position relative to the infinitve, and by fixing Controller choice at
D-structure with the help of his Minimal Distance Principle, as in Larson’s
(1991: 108-122) analysis of the verb promise), Larson removes it from the realm of
control altogether: “Suppose then that examples like (ib) John asked Mary to be allowed
to leave, like their counterparts with promise (see below 5.5., RR), do not in fact involve
control but instead involve construal by ‘tranfer of possession’ entailments’. Then we
expect the subject to be associated with the infinitive under (an appropriately modalized
form of) the entailment:
(iii) X – asks – Y – for – Z ⇒ X gets Z″ (Larson, 1991: 133, fn. 19).
I will discuss Larson’s radical step, which he repeats for the recalcitrant behaviour of
promise, below (5.5.).
12. “A predicate phrase must agree with the object of a transitive verb phrase and the subject
of an intransitive verb phrase in number, gender, and person.” (Bach, 1979: 520)).
13. “The German sentence is ambiguous and on one interpretation, subject control, is
synonymous with the English sentence …” (Comrie, 1985: 50).
14. A peculiar but illuminating instance of control is presented by a Greek passage from the
gospel to St. John (4, 9) and its respective translations.
(i) Πωσ ~ σ ὺ ´Ιουδα~ιοσ rν παρ´ ݵου~ πε~ιν αßτε~ισ … ?
How is that you a Jew being from (of) me to drink (you) ask
‘How can you, being a Jew, ask me to drink’
(ii) Old Church Slavonic
Kako ty ijudei sy ot mene piti prosiši …?
(iii) Latin
Quo modo tu ludaeus cum sis, bibere a me proscis ?
The external argument (συ, you) is the controller. The request δüσ µοι πε~ιν give me to
drink) precedes (i). The addressee of the request shows up in the PP παρ ݵου idiomati-
cally linked to the verb αßτÝω (ask) and copied in the translations. Control in (i)–(iii)
seems analogous to the English and Spanish constructions discussed above (3.2.5.),
which retain full acceptability though violating the marked Constraint (B,b). In some
languages, as in Spanish and English, pairing of the interested parts in the ASK-type
events is construed by a conceptual operation, which in the given Greek example is
modified, even strengthened, to something like (iv):
(iv) [… ASK mei [PROi [to give youj [[ something]k [[operatork] [PROj to drink tk]]]]]]
15. We can take advantage of Dowty’s (1985: 300) entailment patterns (his numbering):
″(10) ∀x∀y∀P [δ(P) (y*) (x*) → γ(y*)]
∀x∀y∀P [δ(P) (y*) (x*) → ζ ([P(y*)])(x*)]
“P” is a variable over VP-meanings… For example , if δ=persuade, then an instance of
γ(y*) would be “y is an agent capable of forming intentions to act”, and an instance of
ζ([P(y*)]) (x*) would be “as a result of x’s action, y comes to intend to act so as to bring
P(y*) about.”
(10) can be illustrated by the control relation in (i):
(i) persuade him i [[PROi to VP …]], Russian … ugovorit’ egoi ((PROi VP)) or by (ii):
(ii) Johni was persuaded ti [[PROi to leave]]
Notes 191
〈intact,–〉 is assigned to the tail of the A-chain (Johni ti) and transferred to John. It follows
from (10) (δ=persuade), if applied to (ii), that PRO carries a value α of the theta
specification distinct from that of the internal argument John. If syntax is to satisfy (10),
y must be mapped onto the direct object-NP of the superordinate clause at D-structure.
(The referent of) this NP reproduces the relevant identity in (10) by being the target of x’s
(implicit in (ii)) actions and the agent intending to act so as to bring about P. In other
words, (ii) satisfies um (B,b).
16. Recall that if the chain headed by PRO has received 〈F,+〉 (〈intact,+〉) itself, which must
have been carried along with PRO moved from its departure site to its non-theta-position,
it is assigned 〈intact,–〉 by (D) and (E) automatically (see above, the analysis of (64)).
17. A theta-position is generally permitted to receive multiple theta-roles (cf. Chomsky,
1986b: 97), a license that need not be drawn upon in postgrammatical quasi-θ-assign-
ments.
18. In Russian, for example, the counterparts of teach have distributional properties different
from obešcat’ (promise) (no dative-shift, inherent dative case of the theme-NP) without
effect on control.
(i)R Neuzeli ja vas dolzen knjaz’, ucit’ delikatnosti
[acc pl] [dat sg]
Really I you must, prince, teach tact
‘Do I really have to teach you, prince, tact’
Nor do we have in the German counterpart of teach the analogue to how to + infinitive
construction. In German, embedded question-CP’s are not infinitival generally.
(ii)G *Johannes lehrte ihn, wie zu spielen.
John taught him how to play.
19. Russian obucat’ (teach) has a reflexive partner obucat’sja which is a distinct lexical entry
and has subject control. Its English counterpart is to learn:
(i)R Obucali egoi ((PROi plavat’))
(They) taught himi to swim
(ii)R Oni obucalsja ((PROi plavat’))
He learnt to swim
Note that the attached reflexive clitic -sja is not a quasi-trace of movement.
20. Köpcke/Panther (1991: 162) observe that the verbs empfehlen (recommend) and raten
(advise) are not only unmarked with respect to the factor Benefactive for the matrix
subject, but even exhibit a morphosyntactic marker (dative), which indicates the role of
Benefactive of the matrix object. (“Die Verben empfehlen und raten sind nicht nur
unmarkiert hinsichtlich des Faktors ‘Benefiziens’, für das Matrixsubjekt, sondern weisen
sogar eine morphosyntaktische Markierung auf (Dativ), die auf die Benefizientenrolle
des Matrixobjektes deutet.”)
21. Discourse-related conceptual construal of a domain of individuals is a paraphrase of this
domain being “understood”. A slight modification of the (interpretation of the) projection
principle might be necessary: “A consequence of the projection principle is, to put it
informally, that if some element is ‘understood’ in a particular position, then it is there in
syntactic representation, either as an overt category that is phonetically realized or as an
empty category assigned no phonetic form …” (Chomsky 1986b: 84).
192 Notes
22. The translations (i) and (ii) of (i)’ and (ii)’ respectively, assumed in Gazdar et al.
(1985: 203) do not seem to capture the whole spectrum of control of these two verbs and
cognate ones. Still, the distinct hierarchical positions of the NP argument which may
receive the theta-role goal reveals a basic intuition about the different status the object-
NP has in the respective configurations projected by the two verbs.
(i) persuade’ (leave’) (us*) (Kim*)
(i)′ Kim persuaded us to leave.
(ii) promise (us*) (leave) (Kim*)
(ii)′ Kim promised us to leave.
Control is thought to be captured by (iii):
(iii) “The first NP argument to combine with a functor in which a VP occurs is the
(semantic) controller of that VP” (Gazdar et al. 1985: 202).
23. Bresnan (1982, 405) notes: “The first attempt in transformational grammar to explain the
deviance of examples like (87) (= (i) b. RR) appears to be that of Jenkins (1972, 200 ff.),
who proposed a constraint stating that the object of by cannot be coreferential with an
implicit or expressed subject of a complement.
(i) a. John promised Mary to be on time.
b. *Mary was promised by John to be on time.
If this were true, it would itself require explanation, but the following examples show that
this by-phrase constraint does not express the correct generalization:
(ii) (=(94))
a. John had been promised by Mary that she would meet him at the station.
b. John expects a promise by Mary to remain faithful to him.
c. An attempt by the gang of four to advance themselves now would be foolhardy.
In all of these examples, the object of by is or can be understood as coreferential with the
subject of the complement.”
24. “Minimal Distance Principle (MDP): An infinitive complement of a predicate P selects as
its controller the minimal c-commanding noun phrase in the functional complex of P.”
(Larson, p. 115)
25. Modals even do more, their interference in control behaviour is not restricted to a
particular control type or class. For example, dürfen can intervene in control relations of
counterparts to persuade as in counterparts to ask and promise as above, licensing control
in cases in which it cannot be made acceptable pragmatically in English:
(i) Scipio persuaded the Senate (PRO to have a free hand)
Comrie (1985: 63) comments on his example (61) (= (i)): “… real-world knowledge of
Scipio’s relations with the Senate make the only plausible interpretation that Scipio
persuaded the Senate to allow him, Scipio, to have a free hand, but once again English
syntax disallows this interpretation.” In German, the counterpart of (i) is made grammati-
cal by “inserting” dürfen:
(ii) Scipioi überredete den Senat (PROi frei handeln zu dürfen)
‘… to be permitted to have a free hand’.
(ii) thus satisfies the m version of Constraint (B,b), but this type of modal interference
seems to create the only possibility that persuade-type verbs meet m (B,b) (cf. above 4.1.)
In English, which has to choose passive in paraphrasing dürfen, (ii) seems questionable,
cf. (iii):
(ii) ?*Scipioi persuaded the Senate (PROi to be permitted to have a free hand).
(iii) Scipio persuaded the Senate i (PROi to be examined by a commission).
Notes 193
26. What is implemented in this way compares with independent phenomena observed by
Zubizarreta (1982), it is a kind of their directional reverse: “Zubizarreta explains these
facts in terms of a projection … the rationale clause projects agency onto an argument in
the matrix clause … the subject of the matrix clause receives an agent role from the
rationale clause.
(i) John was arrested by the police to impress his mother.
(ii) Jesus died to save our souls.
(iii) The boy fell to deceive his mother.
(iv) The train derailed to save the child”.
(Roeper 1987: 298), with the examples from Zubizarreta)
27. This question has not been raised by Grewendorf/Sabel although they continue to claim
“… that long scrambling in German is licensed only by members of a special class of
verbs. In the following example with the matrix verb vorwerfen ‘accuse’, long-distance
scrambling out of the complement clause into the matrix clause is not possible:
(i) a. weil keiner diesem Manne [CP PRO den Triumph auszukosten]
since nobody this manDAT the triumphACC to-savor
vorgeworfen hat
accused has
‘since nobody has accused this man of savoring his triumph’)
(i) b. *weil [[den Triumph] i keiner diesem Manne [CP PRO ti
since the triumphACC nobody this manDAT
auszukosten] vorgeworfen hat]”
to-savor accused has
(i)a., (i)b. = Grewendorf/Sabel’s (37a), (37b)
Vorwerfen does not belong, then, in the special lexical class presumed by Grewendorf/
Sabel. Under my approach it is a control verb of the persuade-type subclass that, in
German, contains an oblique (dative) object.
28. Sentence types like (268), (130), in which PRO is bound by a quantifier, support
Nishigauchi’s (1984: 237) assumption that “… it seems clear that we need a mechanism
to assure that PRO is represented as a variable bound by a quantifier or a lambda
operator.”
Incidentally, topicalization at S-structure and IP-adjunction at LF of (negated) quantifier
phrases, with the subject pronominal (nikto) within the scope of the quantified time
adverbial, may end up as conjoined phrases at the top of the sentence at PF, blurring their
respective hierarchical and local origins.
(i)R Nikto i nikogda ne stremitsja eticeski
nobody and never (not) endeavours ethically
ocenit’ javlenija prirody.
to evaluate phenomena of nature
‘Nobody ever ethically evaluates phenomena of nature.
29. The “oddness” of (i), (ii) noted by Hornstein and Lightfoot (1987: 33) is a case in point
again.
(i) John tried to receive the gift.
(ii) John tried to know the answer.
(iii) is the Russian counterpart of (ii):
(iii) Dzon pytalsja znat’ otvet.
Prefixing u to znat’ gives the perfective aspect uznat’ (to learn, to get to know), which in
its result-oriented interpretation cancels the conceptual conflict. (iv) is readily accepted:
194 Notes
34. (“a chain is a CHAIN …” (following Chomsky, 1986: 132)). Baker et al. (1989: 225) use
a notion of “chain” slightly different from Chomsky (1986: 132 ff.).
35. Borer’s (1986, 395) analysis is similar to Epstein’s: “Though Epstein assumes that that
operator (the operator that binds the infinitival [NP, S] position, RR) is specifically a
small pro …, I will simply assume that it is an abstract operator, the precise nature of
which requires further investigation. I will further deviate from Epstein’s analysis in
assuming that the movement involved is syntactic (rather than LF) and that it adjoins the
null arbitrary operator to S. Further adapting Epstein’s proposal to the system proposed
here, I will assume that the abstract operator A-binds the anaphoric infinitival Agr and
that its index is thus transmitted to the embedded PRO … This S-structure representation
is thus given in (36) (Borer’s numbering, RR):
(36) OPERATORi (it is not easy [e]i (S′ [e]i Agri to solve this problem))”
Differing from Epstein, who treats pro as a universal quantifier, Borer associates a set
interpretation with the null operator: given a set S, it is not easy for a member/members of
S to solve this problem. (1986: 395)
36. A particular analysis, which I will not undertake here, would be necessary for Czech and
Polish variants of reciprocal expressions:
(i)C Navzájem si pomáhají.
[dat] refl clitic [3 ps pres]
(they) each other help
(ii)C Slíbili si navzájem navštívit prednášky toho druhého
[dat refl. clitic] [gen]
(they) promised mutually to visit the lectures (of) the other
‘They promised each other to visit each other’s lectures’.
The adverb navzájem “reciprocally” marks the reciprocal meaning of the clitical dative
reflexive si.
(iii)P Obiecali sobie wzajemnie (PRO pójšć do lekarza)
[dat]
(they) promised themselves reciprocally (mutually) to go to the doctor
They promised each other …’
196 Notes
References
Brody, Michael. 1993. Θ-Theory and Arguments. Linguistic Inquiry 24: 1–23.
Burzio, Luigi. 1986. Italian Syntax. Dordrecht: Reidel.
Carrier-Duncan, Jill. 1985. Linking of Thematic Roles in Derivational Word Formation.
Linguistic Inquiry 16: 1–34.
Chierchia, Gennaro. 1984. Topics in the Syntax and Semantics of Infinitives and Gerunds.
Doctoral Dissertation, University of Massachusetts, Amherst.
Chierchia, Gennaro, and Richard Turner. 1988. Semantics and Property Theory. Linguistics
and Philosophy 11: 261–302.
Chomsky, Noam. 1980. On Binding. Linguistic Inquiry 11: 1–46.
Chomksy, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris.
Chomksy, Noam. 1982. Some Concepts and Consequences of the Theory of Government
and Binding. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 1986a. Barriers. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 1986b. Knowledge of Language: Its Nature, Origin, and Use. New York:
Praeger.
Chomsky, Noam. 1989. Some Notes on Economy of Derivation and Representation. In I.
Laka and A. Mahajan, (eds.), MIT Working Papers in Linguistics 10. Functional Heads
and Clause Structure, 43–74.
Chomsky, Noam. 1993. A Minimalist Program for Linguistic Theory. In K. Hale and
S.J. Keyser, (eds.), The View from Building 20. Essays in Linguistics in Honor of
Sylvian Bromberger, MIT, Cambridge, Massachusetts.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge: M.I.T. Press.
Chomsky, Noam, and Howard Lasnik. 1993. Principles and Parameters Theory. In Joachim
Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann, (eds.),
Syntax: An International Handbook of Contemporary Research, 506–569, Berlin:
Walter de Gruyter.
Chvany, Catherine. V. 1975. On the Syntax of BE-Sentences in Russian. Cambridge, Mass.:
Slavica Publishers.
Cinque, Guglielmo. 1988. On Si Constructions and the Theory of Arb. Linguistic In-
quiry 19: 521–581.
Cole, Peter. 1987. Null Objects in Universal Grammar. Linguistic Inquiry 18: 597–612.
Cole, Peter, Gabriella Hermon and Li-May Sung. 1990. Principles and Parameters of Long-
Distance Reflexives. Linguistic Inquiry 21: 1–22.
Comrie, Bernard. 1981. Language Universals and Linguistic Typology. Oxford:
B. Blackwell.
Comrie, Bernard. 1985. Reflections on Subject and Object Control. Journal of Semantics 4:
47–65.
Conrad, Rudi. 1969. Transformationsanalyse russischer Infinitivkonstruktionen.
Linguistische Studien. Halle.
Culicover, Peter, and Wendy Wilkins. 1986. Control, PRO and the Projection Principle.
Language 62: 120–153.
Dùabrowska Ewa. 1994. Dative and Nominative Experiencers: Two Folk Theories of the
Mind. Linguistics. 32(6): 1029–1054.
Davidson, Donald. 1989. Essays on Actions and Events. Oxford: Clarendon Press.
Diesing, Molly. 1988. Bare Plural Subjects and the Stage Individual Contrast. Unpub-
lished manuscript. University of Massachusetts, Amherst.
References 199
Hale, Kennth, and Samuel Jay Keyser. 1993. On Argument Structure and the Lexical
Expression of Syntactic Relations. In Kenneth Hale ans Samuel Jay Keyser, (eds.), The
View from Building 20. Essays in Linguistics in Honour of Sylvain Bromberger, 53–
109. MIT, Cambridge Massachusetts.
Hajicová, Eva, Jarmila Panevová, and Petr Sgall. 1987. A Remark on Control. In Studies in
Honour of Robert Busa S. J., Linguistica Computazionale IV, V, Pisa, 59–69.
Heim, Irene, Howard Lasnik, and Robert May. 1991. Reciprocity and Plurality. Linguistic
Inquiry 22: 63–101.
Higginbotham, James. 1985. On Semantics. Linguistic Inquiry 16: 547–593.
Higginbotham, James. 1992. Reference and Control. In Richard K. Larson, Sabine Iatridou,
Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 79–108.
Dordrecht, Boston, London: Kluwer.
Hornstein, Norbert, and David W. Lightfoot. 1987. Predication and PRO. Language 63:
23–51.
Hust, J., and Michael Brame. 1976. Jackendoff on Interpretive Semantics. Linguistic
Analysis 2: 243-277.
Ickovic, Viktor A. 1970. Norma i ee kodifikacija. Moskva.
Ijbema, Aniek and Werner Abraham. 1998. Die syntaktische Funktion des infinitivischen
zu. Festschrift für Peter Eisenberg. Tübingen: M. Niemeyer.
Ivić, M. 1972. Problematika srpskohrvatskog infinitiva. (Problems related to the Serbo-
Croatian infinitive) Zbornik za filologiju i lingvistiku 15/2. Novi Sad.
Iwakura, Kunihiro. 1985. The Binding Theory and PRO. Linguistic Analysis 15: 29–55.
Iwakura, Kunihiro. 1990. Binding Theory and Two Types of PRO. Linguistic Analysis 20:
121–138.
Jackendoff, Ray 1972. Semantic Interpretation in Generative Grammar. Cambridge,
Mass.: MIT Press.
Jackendoff, Ray. 1974. A Deep Structure Projection Rule. Linguistic Inquiry 5: 481–506.
Jackendoff, Ray. 1987. The Status of Thematic Relations in Linguistic Theory. Linguistic
Inquiry 18: 369–411.
Jacobson, Pauline. 1990. Raising as Function Composition. Linguistics and Philosophy 13:
423–475.
Jaeggli, Osvaldo. 1982. Topics in Romance Syntax. Dordrecht: Foris.
Jaeggli, Osvaldo A. 1986. Passive. Linguistic Inquiry 17: 587–622.
Jenkins, Lyle. 1972. Modality in English Syntax. Doctoral Dissertation, MIT, Cambridge,
Mass.
Johnson, Kyle. 1985. A Case for Movement. Unpublished Ph. D. dissertation. MIT.
Joseph, Brian D. 1992. Diachronic Perspectives on Control. In Richard K. Larson, Sabine
Jatridon, Utpal Lahiri, and James Higginbotham, (eds.), Control and Grammar, 195–
234. Dordrecht, Boston, London: Kluwer.
Kamp, Hans, and Antje Roßdeutscher. 1995. Leider hatten wir vorgeschlagen, diesen Satz
zu analysieren. (ms. Berlin).
Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht: Foris.
Kayne, Richard S. 1991. Romance Clitics, Verb Movement, and PRO. Linguistic In-
quiry 22: 647–686.
References 201
Růzicka, Rudolf. 1980. Studien zum Verhältnis von Syntax und Semantik im modernen
Russischen. Berlin: Akademie-Verlag.
Růzicka, Rudolf. 1982. Kontrollprinzipien infiniter Satzformen: Infinitiv und Gerundium
(deepricastie) im Russischen. Zeitschrift für Slawistik 27: 373–411.
Růzicka, Rudolf. 1983a. Remarks on Control. Linguistic Inquiry 14: 309–324.
Růzicka, Rudolf. 1983b. Autonomie und Interaktion von Syntax und Semantik. In R.
Růzicka, W. Motsch, (eds.) Untersuchungen zur Semantik, studia grammatica XXII:
15–59. Berlin: Akademie-Verlag.
Růzicka, Rudolf. 1984. Illokutionäre Kraft und Subjektkontrolle - Bitten und Fragen.
Linguistische Arbeitsberichte 44: 7–10, Leipzig.
Růzicka, Rudolf. 1986. Control in Competing Frameworks. In J. L. Mey, (ed.), Language
and Discourse: Test and Protest, A Festschrift for Petr Sgall, 101–121, Amsterdam:
Benjamins.
Růzicka, Rudolf. 1986a. Typologie der Diathese slavischer Sprachen in parametrischen
Variationen. Welt der Slaven XXXI, 2: 225–274.
Růzicka, Rudolf. 1987. The Semantic-Pragmatic Interface and Interaction in Control Areas.
In Albrecht Neubert and R. Růzicka, (eds.), Topics on the Semantic Borderline, Lin-
guistische Studien, Reihe A, 166: 128–177, Berlin.
Růzicka, Rudolf. 1988. On the Array of Arguments in Slavic Languages. Zeitschrift f.
Phonetik, Sprachwissenschaft und Kommunikationswissenschaft 41, 2: 155–178.
Růzicka, Rudolf. 1992. Slavic and Italian Impersonal Constructions with Reflexive Clitics.
In I. Zimmermann, A. Strigin, (eds.), Fügungspotenzen, studia grammatica XXXIV:
133–161. Berlin: Akademie-Verlag.
Růzicka, Rudolf. 1994. Gegen die Aufgabe von PRO. In Anita Steube and Gerhild
Zybatow, (eds.), Zur Satzwertigkeit von Infinitiven und Small Clauses, 13–18.
Tübingen.
Safir, Ken. 1987. The Syntactic Projection of Lexical Thematic Structure. Natural Lan-
guage and Linguistic Theory 5: 561–601.
Safir, Ken. 1996. Derivation, Representation and Eesumption: The Domain of Weak
Crossover. Linguistic Inquiry 27: 313–339.
Sag, Ivan. A. and Carl. Pollard. 1991. An Integrated Theory of Complement Control.
Language 67: 63–113.
Sgall, Peter, Eva Hajicová, and Jarmila Panevová. 1986. The Meaning of the Sentence in its
Semantic and Pragmatic Aspects. Academia Praha.
Siebert-Ott, G. 1985. Bemerkungen zu den Elementen einer Theorie der Kontrolle. In
Werner Abraham, (ed.), Erklärende Syntax des Deutschen. Studien zur deutschen
Grammatik 25: 255–270, Tübingen.
Sigurdsson, Halldór A. 1991. Icelandic Case-marked PRO and the Licensing of Lexical
Arguments. Natural Language and Linguistic Theory 9: 327–363.
Stowell, Tim. 1991. The Alignment of Arguments in Adjective Phrases. In Susan
D. Rothstein, (ed.), Syntax and Semantics 25. Perspectives on Phrase Structure: Heads
and Licensing, 105–138. Academic Press, INC.
Stroik, Thomas. 1994. Kayne’s Universal Word Order and Psych-Verbs. (ms. Berlin).
Stump, Gregory. T. 1985. The Semantic Variability of Absolute Constructions. Synthese
Language Library 25, Dordrecht, Boston, Lancaster: Reidel.
204 References
D M
demoted agentives as controllers, 89, markedness, 34 f. 38 f. 59, 161, 180
93-97, 192 note 23 minimal distance principle, 22, 23, 104,
deontic switch, 62, 110, 180 192 note 24
discarding PRO, 181 f. modal verbs intervening, 109, 192 note
discourse-related interpretation, 7, 191 25
note 21 modality switch, 181
division of labor, 104
N
E null object parameter, 43-49, 52 f.
ECM or control, 81 interacting with interface control
entailment, 172, 190 note 11, 14 parameters, 58-62, 65
206 Subject index
O S
object of by and counterparts as control- scrambling and control, 114 f.
lers, 93-97, 192 note 23 selectional cross-clausal underpinnings
oblique object control, 76, 148 of control, 19, 38, 172
specification on the marked application
P of main constraints, 34
parameterization of pragmatics, 2, 5, stage-level predicates, 142
55-63, 180 subject complement clauses, 27
pragmatics, 44, 189 note 8 symmetry and distinction between main
pro as controller and controllee, 51, constraints, 91 f.
129-136
propositional versus attributive account, T
23, 116 f. turn of approach to control in terms of
psychological state predicates, 141 thematical specifications, 25
Q U
questions of control, 23 unaccusatives, 37, 64
R V
raising, 23 f.. 106 verbal noun control, 49 f., 96 f.
reciprocality, 151 f. violation of (marked) constraints, 40,
reflexive impersonal clauses, 129-136 43 f., 165
reflexive passive controlled clauses, VP-deletion, 152
133 f.
In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titles
have been published thus far, or are scheduled for publication:
1. KLAPPENBACH, Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie.
Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei
Beiträge von Helene Malige-Klappenbach. 1980.
2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion und
Beispielanalyse. 1982.
3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from
the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen,
January 1981. 1983.
4. ABRAHAM, Werner & Sjaak De MEIJ (eds): Topic, Focus and Configurationality.
Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986.
5. GREWENDORF, Günther and Wolfgang STERNEFELD (eds): Scrambling and Barri-
ers. 1990.
6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic Phrase
Structure Phenomena in Noun Phrases and Sentences. 1989.
7. ÅFARLI, Tor A.: The Syntax of Norwegian Passive Constructions. 1992.
8. FANSELOW, Gisbert (ed.): The Parametrization of Universal Grammar. 1993.
9. GELDEREN, Elly van: The Rise of Functional Categories. 1993.
10. CINQUE, Guglielmo and Guiliana GIUSTI (eds): Advances in Roumanian Linguistics.
1995.
11. LUTZ, Uli and Jürgen PAFEL (eds): On Extraction and Extraposition in German. 1995.
12. ABRAHAM, W., S. EPSTEIN, H. THRÁINSSON and C.J.W. ZWART (eds): Minimal
Ideas. Linguistic studies in the minimalist framework. 1996.
13. ALEXIADOU Artemis and T. Alan HALL (eds): Studies on Universal Grammar and
Typological Variation. 1997.
14. ANAGNOSTOPOULOU, Elena, Henk VAN RIEMSDIJK and Frans ZWARTS (eds):
Materials on Left Dislocation. 1997.
15. ROHRBACHER, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I
raising and pro-drop. 1999.
16. LIU, FENG-HSI: Scope and Specificity. 1997.
17. BEERMAN, Dorothee, David LEBLANC and Henk van RIEMSDIJK (eds): Rightward
Movement. 1997.
18. ALEXIADOU, Artemis: Adverb Placement. A case study in antisymmetric syntax.
1997.
19. JOSEFSSON, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swed-
ish. 1998.
20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Ad-
verbs, pronouns, and clause structure in Romance and Germanic. 1998.
21. KLEIN, Henny: Adverbs of Degree in Dutch and Related Languages. 1998.
22. ALEXIADOU, Artemis and Chris WILDER (eds): Possessors, Predicates and Move-
ment in the Determiner Phrase. 1998.
23. GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998.
24. REBUSCHI, Georges and Laurice TULLER (eds): The Grammar of Focus. 1999.
25. FELSER, Claudia: Verbal Complement Clauses. A minimalist study of direct perception
constructions. 1999.
26. ACKEMA, Peter: Issues in Morphosyntax. 1999.
°
27. RUZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study.
1999.
28. HERMANS, Ben and Marc van OOSTENDORP (eds.): The Derivational Residue in
Phonological Optimality Theory. 1999.
29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal
noun. 1999.
30. BEUKEMA, Frits and Marcel den DIKKEN (eds.): Clitic Phenomena in European
Languages. 2000.
31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000.
32. ALEXIADOU, Artemis, Paul LAW, André MEINUNGER and Chris WILDER (eds.):
The Syntax of Relative Clauses. 2000.
33. PUSKÁS, Genoveva: Word Order in Hungarian. The syntax of È-positions. 2000.
34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization.
2000.
35. HRÓARSDÓTTIR, Thorbjörg. Word Order Change in Icelandic. From OV to VO.
2000.
36. GERLACH, Birgit and Janet GRIJZENHOUT (eds.): Clitics in Phonology, Morphol-
ogy and Syntax. 2000.
37. LUTZ, Uli, Gereon MÜLLER and Arnim von STECHOW (eds.): Wh-Scope Marking.
2000.
38. MEINUNGER, André: Syntactic Aspects of Topic and Comment. 2000.
39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, ‘‘Self’’, and
Interpretability. 2000.
40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton
van der WOUDEN (eds.): Perspectives on Negation and Polarity Items. 2001.
41. ZELLER, Jochen : Particle Verbs and Local Domains. n.y.p.
42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and
ergativity. n.y.p.
43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001.
44. TAYLAN, Eser E. (ed.): The Verb in Turkish. n.y.p.
45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds.): Issues in Formal German(ic)
Typology. n.y.p
46. PANAGIOTIDIS, Phoevos: Pronouns, clitics and Empty Nouns. ‘Pronominality’ and
licensing in syntax. n.y.p.