Escolar Documentos
Profissional Documentos
Cultura Documentos
1. Introduction
In this paper we examine certain problems for minimalist Case theory that are
brought to light by Multiple Case Checking in raising constructions, as
previously studied by Massam (1985), McCreight (1988), and Ura (1996). In
general, Case theory has been developed to exclude the option of a chain
receiving anything other than exactly one Case. However, certain constructions arguably demonstrate that this is possible. In particular, Multiple
Case Checking (MCC) seems to force us to posit inflectional feature
complexes that contain seeming contradictions, such as [[+nominative]
[+accusative]]. In this paper we examine these constructions and demonstrate
that presyntactic insertion of phonological features must be abandoned. We
also show that both Case assignment and Case checking are required, and that
languages differ as to which features of a chain are realized phonologically.
The problem of MCC arises only in cases where a single chain receives
more than one Case. If two chains are involved, there is no MCC. For this
reason, we include in our empirical coverage only instances of movement
from a Case-marked position to another Case-marked position.1 We put aside
* We gratefully acknowledge support from SSHRCC (410-97-0493). We would like to thank
Mr. Harry Manamana for his work as Niuean language consultant. We would also like to thank
Jonathan Bobaljik, Jun Mo Cho, Samuel Epstein, Susi Wurmbrand, two anonymous reviewers,
and members of the University of Toronto Syntax Project for useful comments. Any errors are
our own.
1
All MCC data, including the movement cases studied in this paper, are open to an analysis of
prolepsis (Higgins 1981; Wager 1983; Massam 1985), in which two separate chains are posited
with an obligatory coreference relationship holding between them. The problem is, no
satisfactory theory of prolepsis, as distinct from raising, has been developed. There is a danger
that prolepsis analyses might be proposed simply to avoid problematic raising structures, such as
those involving multiple Case. Since at least for the Niuean data presented here, all evidence to
date points to a single thematic chain, we assume that raising is involved in these constructions
(see Massam 1985 and Seiter 1980 for further arguments against prolepsis in Niuean). We thank
Jonathan Bobaljik for discussion on this point. For all the data in this short paper, we rely on the
literature as cited for the basic analyses. We do not attempt here a reanalysis in minimalist terms.
Blackwell Publishers Ltd, 1999. Published by Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK and
350 Main Street, Malden, MA 02148, USA
66
Structural-Structural
Inherent-Structural
MCC Types
Case Chains
Case Chains
Allow MCC
A. Hungarian, Niuean
D. Icelandic
Allow MCC
B. Norwegian
E. Not examined
with syncretism
Disallow MCC
C. English
F. Not examined
Examples of the first four kinds are given below. The data is
alphabetically keyed to correspond with the typology established in the
chart (AD). Example (2) provides an example of MCC from Hungarian. The
wh-word kiket originates in subject position of the embedded clause, where it
receives nominative Case. En route to its final A-bar position, it checks ECM
(accusative) Case with the intermediate verb. Notice that, morphologically, it
realizes this second Case and not nominative.
(2)
Example (3) shows MCC in Niuean. Here the DP na tama originates in the
subject position of the embedded clause, where it receives absolutive Case, as
shown in (3a). In (3b) we see that the DP has raised to a higher position where
2
These include relative constructions where the head bears some relation to two different
Case-marked positions, and coordinate constructions, which can also be analyzed as involving
more than one chain. We also disregard Case-stacking phenomena (e.g., Korean) because it is not
clear that these really do involve multiple Case (Schutze 1997).
67
Note that these cases are different from the copy raising cases discussed by
Ura (1996) because no pronominal copy appears in the embedded position
in fact, a copy is generally prohibited (see Seiter 1980).
The B-type examples in (4) show syncretically resolved MCC.
(4)
68
C: English
a. Mary seems ti to like apples.
b. *Maryi seems (that) ti likes apples.
(i)
Pat seems as if shes in a bad mood today.
(ii) ?Pat seems as if shes absent today.
(iii) Pat seems to be absent today.
Blackwell Publishers Ltd, 1999
69
associated with phonology. Then, the single DP checks two heads one
with absolutive, and the other with middle Case. The Early Insertion
hypothesis is schematized in (7), where each number refers to a formal Case
feature and the cross-mark indicates checking with deletion. In (7a), we see
that the Computational Component (CC) selects two abstract-Case formal
features (FFs) for a single head, each associated with its corresponding
phonology. In the syntax, each of the FFs is checked in turn, as in (7b) and
(7c). We might assume that the FFs are erased by checking as shown in (7),
or that they survive to PF. Either way, remaining at PF will be the phonology
of both morphemes. This result is problematic because it violates the
principle of Full Interpretation (Chomsky 1995), because there is
phonological material which is receiving no interpretation at PF.
2.1 Problems posed by MCCs
(7)
c. Upstairs checking
d. at PF
70
turn (as indicated by the check marks in (8b) and (8c)). Again, assuming the
features do not delete, we are left with PF receiving two sets of instructions
(as in (8d)), and we need an algorithm that allows PF to choose which set to
ignore. Under our view that PF is interpretive, this is not feasible.
(8)
Late insertion 1 (= no deletion (with named cases); cf. Halle & Marantz
1993)
a. Upon selection
b. Downstairs checking
c. Upstairs checking
d. At PF
On the other hand, as shown in the schema in (9), it might be assumed that
the features are deleted when checked and do not survive to PF, or that they
are not present at all. In this case, PF will receive no instructions at all as to
how to spell out the Case on the DP.
(9)
c. Upstairs checking
d. at PF
71
72
73
except that it shows that the same structural Case can be checked twice. Finally,
the Latin example in (15) shows that in raising from accusative to nominative
the last checked Case wins, regardless of markedness.
We now turn to an account of these facts. In general, we assume the
Minimalist Program, as in Chomsky 1995, except where specified. We
propose that among the languages we are considering, Case is unique and that
at all levels of grammar, a DP may only bear one Case feature. Given this
restriction, we propose a theory of Case that is a combination of Case
checking and Case assignment.5 DPs enter the numeration with simply a
[Case] feature. Relevant functional projections, however, are entered with
[Case] and with a named feature as well, such as [acc] or [nom]. When a DP
checks Case against a functional head, automatically [Case] is checked, and
[nom] or [acc] is copied or assigned as a subscript on the Case feature of the
DP by the functional projection. Both checking and assignment must be
satisfied for convergence. As we will see, PF either implements the head of
the A-chain or the whole A-chain. A schematization of this proposal is given
in (16) where (c) and (d) show the representations fed to PF.
(16)
a. Infl pre-checking
INFL
j
CASEnom
c. Infl post-checking
INFL
j
CASEnom
b. DP pre-checking
DP
j
!
CASE
d. DP post-checking
DP
j
CASEnom
Given this, let us now consider English. If a DP were to raise from a Casemarked position, possibly attracted by a Case feature on a higher Infl, the
assigned Case would move, along with the DP. This Case would block further
assignments because the morphosyntactic Case slot is unique.6 This is not to
say that such an instance of Attract would not be possible, but the derivation
would not converge because the Case feature on the higher Infl would not be
properly checked. Thus, in English, MCCs are ruled out.
Note that Chomsky (1995) also rules out raising from a Case-marked
position to another Case-marked position by his claim that [Case] on a noun is
deleted or erased once checked. This would mean that in a sentence such as (5),
the Case feature on the matrix Infl would remain unchecked, and the derivation
5
Many other researchers on Case have been led to the conclusion that Case consists of more
than one component, often including, for example, morphological and abstract Case. Our
proposal belongs to this family of ideas.
6
Because this is a hypothetical situation, it is difficult to determine whether a copy of the Case
subscript remains on the trace. In principle, for the issues at hand, the exact nature of the trace is
irrelevant.
74
would crash. In our view, the [Case] feature itself does not become inert, but
because the assignment subscript slot is filled, it effectively does, because the
DP cannot in fact check further Cases.
Now let us look at situations where MCC does occur. Clearly, in these
instances, a Case assigned in a lower position does not block further Case
assignment. How can we permit this situation? We propose that in such
constructions the specific Case of the DP is interpretable only by virtue of being
in a checking configuration with that functional head. This assumes that the
interpretability of Case features requires that they be dominated by an
appropriate functional head. In other words, Case is interpreted
compositionally. Effectively, this means that the Case subscript is left behind
when the DP moves out of one Case-checking configuration into a higher one,
as in (17b). Thus it is eligible to receive a different case specification. This is
the sort of derivation we propose for Niuean. The structures in (17) demonstrate
this with a nominative/accusative system for simplicity.
(17) a. Checking downstairs [nom]
IP
Vmax
VP
[nom] configuration
IP
DPi
Vmax
Infl
CASE
[nom] CASE t
i
[nom]
VP
Vmax
DPi
V+v
CASE
[acc] CASE
[acc]
[acc] configuration
VP
IP
ti
Vmax
Infl
CASE
[nom] CASE t
i
[nom]
VP
75
The fact that the embedded Case was assigned is recoverable to the
grammar by its visibility on the trace left behind.7 Because it is on an empty
category, it will not be pronounced, but it can be attested in other ways for
example, by the agreement on the lower verb in the Hungarian example in
(2).8 We assume that wh-chains have access to the head of their A-chain at
PF, in order to account for the fact that wh-words bear the Case assigned to
the head of their A-chain.
Now lets consider syncretic cases, like the Norwegian examples in (4).
Here, MCCs are permitted just in case there is no morphological conflict
between the assigned Cases. This fact tells us that in this language, like
Niuean, it is syntactically possible for a chain to have more than one Case.
However, for Norwegian, we propose that all Cases on the chain are
simultaneously accessible at PF, as in (18a). In particular, this means that at
PF, the chosen vocabulary item must be consistent with each position in the
chain, where consistent with means that the Case value of the item must
not be distinct from any Case value in the chain. Crucially, we assume that
the lack of a Case value on the foot of the chain is not considered distinct,
thus allowing English examples such as (5a). In Niuean, on the other hand, as
in (18b), only the head of the chain is spelled out. Given this, it is at the level
of PF that we need to rule out sentences such as (4b,c). Recall that PF accepts
only unambiguous instructions. In Norwegian, instructions such as firstsingular [+nominative and +accusative] will simply not be able to undergo
VI, because there is no matching vocabulary item that is consistent with every
position in the chain, as defined above. Instructions such as Name
[+nominative and +accusative], however, will be interpretable at VI;
consistency is met, because the relevant vocabulary item is underspecified for
Case.
(18) a. Norwegian-type language
[IP. . .DPi [Vmax V+v. . .[[IP. . .ti. . .[Vmax. . .[ti. . .V]]]]]
j
j
j
Case
Case
Case
[acc]
[nom]
PF instruction: Spell-out chain (VI: [+nom, +acc])
b. Niuean-type language
[IP. . .DPi] [Vmax V+v. . .[[IP. . .ti. . .[Vmax. . .[ti. . .V]]]]]
j
j
j
Case
Case
Case
[middle]
[abs]
PF instruction: Spell-out head of chain (VI: [+middle])
7
We adopt the minimalist notion of trace, which holds that traces are in fact unpronounced
copies, but we use the term trace to avoid confusion with pronounced pronominal copies.
8
We assume that nominative Case and agreement are crucially intertwined.
76
At the level of PF, English is like Norwegian in that it spells out the chain,
but in English conflicting Case information in a single chain can never arise, as
discussed above. The range of language types discussed is represented in (19).
(19) Case and chain typology
Take Case
Spell-out Chain
English
Norwegian
Niuean
+
+
The unmarked setting for each parameter in (19) is indicated by [+]. This
typology is consistent with views of learnability, based on the familiar type of
positive-evidence arguments. With respect to the Take Case parameter, English
exhibits the unmarked case, whereas both Norwegian and Niuean instantiate
the marked case. Learners of the latter languages would be presented with
positive evidence for the marked value of this parameter, because these
languages allow chains to grammatically bear more than one Case.
Now consider the Spell-out Chain parameter. Norwegian and English
exhibit the unmarked situation for this parameter, where the entire chain
feeds to PF. Niuean exhibits the marked situation, where only the head of the
chain feeds to PF. It is only in Niuean that the learners would receive positive
evidence as to which Spell-out parameter is required, because they hear
sentences where the case of the NP does not match the case assigned to its position. Note that once the structure of these sentences is determined, the
child is automatically led to a [] value for the Spell-out Chain parameter.
Consider now ?-type languages. In fact, such languages would not differ
empirically from English, but we predict that such a type would never be
attested: Because the two language types are empirically nondistinct, the
child will always opt for the unmarked Spell-out parameter value and spell
out the whole chain.9
So, it appears that syncretic MCC constructions are constrained, not only
in the syntax, but also at VI, because PF can choose to spell out either the
whole chain or the head of the chain. In the Norwegian case, the Interpretive
Vocabulary Insertion Hypothesis is obviously respected in that instructions to
VI must be unambiguously interpreted. In the Niuean case, where a choice
does appear to be made at PF as to which Case to pronounce, there is no
violation of the Interpretive Vocabulary Insertion Hypothesis, because this
choice is a parameter set during acquisition and does not vary from derivation
to derivation.
Sam Epstein (personal communication) has pointed out that this constitutes an example of a
language which is provided by UG, but which is not learnable, thus supporting Chomskys claim
(1995:18) that all grammars provided by UG need not be learnable.
Blackwell Publishers Ltd, 1999
77
c. Accusative
CASEacc
b. Quirky subject
CASEnom
j
dative
d. Quirky object
CASEacc
j
dative
The actual [Case] node of the quirky argument can itself be used to check
a higher structural Case feature. In (6b) we assume the verb assigns its
underlying object a marked quirky Case. Then, when it raises to subject
position, its root [Case] node will enter into a checking relation with Infl, the
result being like (20b). Because inherent and abstract Cases are different
types, it is possible for both to appear in a single Case slot. At VI, basic
Paninian principles, as outlined by Halle & Marantz (1993) will ensure that
the_ more highly specified Case (i.e., the quirky Case) is realized. This means
that a DP with a quirky Case and a structural Case will be phonologically
realized with the quirky Case. The fact that verbs fail to agree with quirky
Case arguments can be accounted for if agreement is a reflex of the operation
of vocabulary insertion on nominative Case.
The possibility for nominative and dative to both be assigned to a single
DP supports our view that structural Cases are not true optional features (in
the sense of Chomsky 1995) but rather are assigned values for features, rather
like subscripts on a Case feature.
The shape of the hierarchy thereby determines what feature combinations
are licensed in the grammar. It rules out two structural Cases, except in
languages where Case assignment is configurational, and it allows for quirky
plus structural combinations. Thus, feature hierarchies such as those proposed
by Ritter (1997), Harley (1994), and others, are relevant to syntax as well as
morphology.
Notice finally, that covert Case checking is rendered impossible under our
analysis because Case would not yet be assigned at PF. But note that covert
Case checking is virtually impossible in all models in which morphological
Blackwell Publishers Ltd, 1999
78
References
ANDREWS, A. D. 1990. The VP-complement analysis in Modern Icelandic. In
Modern Icelandic syntax, ed. J. Maling & A. Zaenen, Syntax and Semantics 24,
165185. San Diego, CA: Academic Press.
CHOMSKY, N. 1981. Lectures on government and binding. Dordrecht: Foris.
CHOMSKY, N. 1986. Knowledge of language. New York: Praeger.
CHOMSKY, N. 1995. The minimalist program. Cambridge, Mass.: MIT Press.
HALLE, M. & A. MARANTZ. 1993. Distributed morphology and the pieces of
inflection. In The View from Building 20, ed. K. Hale & S. J. Keyser, 111176.
Cambridge, Mass.: MIT Press.
HARLEY, H. 1994. Hug a tree: Deriving the morphosyntactic feature hierarchy. In
Papers on phonology and morphology, ed. A. Carnie & H. Harley. MITWPL 21: 189
230.
HARLEY, H. 1995. Subjects, events and licensing. Ph.D. dissertation, MIT,
Cambridge, Mass.
HIGGINS, F. R. 1981. Proleptic objects and verbs of perception in Zacapoaxtla
Nahuat. In Texas Linguistic Forum 18, ed. F. Karttunen, 69277. Austin, TX:
Department of Linguistics, University of Texas at Austin.
JOHNSON, K. 1991. Object positions. Natural Language & Linguistic Theory 9; 577
636.
KISS, K. 1985. Parasitic gaps and case. Paper presented at MIT, Cambridge, Mass.
LEGATE, J. A. & C. SMALLWOOD. In press. The Case Filter meets the Minimalist
Program: Evidence for strong case features. In The minimalist parameter, ed. H.
Goodluck, G. Alexandrova & O. Arnaudova. Amsterdam: John Benjamins.
10
It has been claimed that English accusative Case checking involves covert movement.
However, other possible analyses have been proposed. (for example, Johnson 1991 and Runner
1995).
79
MARANTZ, A. 1991. Case and licensing. In Proceedings of the Eighth Eastern States
Conference on Linguistics, ed. G. F. Westphal, B. Ao & H.-R. Chae, 234253.
MASSAM, D. 1985. Case Theory and the Projection Principle. Ph.D. dissertation,
MIT, Cambridge, Mass.
MASSAM, D. 1996. Clause structure and case in Niuean. In Toronto Working Papers
in Linguistics 14, ed. P. Koskinen, 83102. University of Toronto.
MCCREIGHT, K. Y. 1988. Multiple case assignments. Ph.D. dissertation, MIT,
Cambridge, Mass.
RITTER, E. 1997. Agreement in the Arabic Prefix Conjugation: Evidence for a nonlinear approach to person, number and gender features. In Proceedings of the 1997
Conference of the CLA, ed. L. Blair, C. Burns & L. Rowsell, 191202. University of
Calgary.
RUNNER, J. 1995. Noun phrase licensing and interpretation. Ph.D. dissertation,
University of Massachusetts, Amherst.
SALTARELLI, M. 1976. Theoretical implications in the development of accusatiuus
cum infinitiuo constructions. In Current Studies in Romance Linguistics, ed. M.
Lujan & F. Hensey [Texas symposium on Romance Linguistics, 4]. Washington,
D.C.: Georgetown University Press.
TZE, C. 1997. INFL in Child and adult language: Case, agreement, and
SCHU
licensing. Ph.D. dissertation, MIT, Cambridge, Mass.
SEITER, W. J. 1980. Studies in Niuean syntax. New York: Garland.
TARALDSEN, K. T. 1981. Case-conflict in Norwegian topicalization. In Proceedings
of the eleventh annual meeting of the North Eastern Linguistic Society, ed. V. Burke
& J. Pustejovsky, 377398. Graduate Linguistics Student Association, University of
Massachusetts, Amherst.
URA, H. 1996. Multiple feature checking and grammatical function splitting. Ph.D.
dissertation, MIT, Cambridge, Mass.
WAGER, J. S. 1983. Complementation in Moroccan Arabic. Ph.D. dissertation, MIT,
Cambridge, Mass.
Susana Bejar
University of Toronto
Department of Linguistics
130 St. George St.
Toronto, ON M5S 3H1
Canada
sbejar@chass.utoronto.ca
Diane Massam
University of Toronto
Department of Linguistics
130 St. George St.
Toronto, ON M5S 3H1
Canada
dmassam@chass.utoronto.ca