Core grammar, Case theory, and markedness
Core grammar (CG) is universal grammar’s substantive contribution to the grammar of a particular language. It consists of a set of general rules or rule schema (e.g., “move α”), conditions on rules (e.g., the recoverability condition for deletions), and filters (e.g., *[that [NP e]])—all of which provide a severely limited set of possible grammars from which certain properties of languages follow.1 The contents of the grammar are organized into various components, each with its specific properties. The various components interact modular fashion, in a way determined by the general organization of grammars. Following Chomsky and Lasnik (1977), we adopt the organization in (1).
maps base structures onto S-structures, the output of the transformational component. (II) maps S-structures onto phonetic representations; (III) maps them onto representations in logical form (LF), one aspect of semantic representation. (Henceforth we will refer to (I) as “the Syntax,” (II) as “the Phonology,” and (III) as “LF”). (1) constitutes an empirical hypothesis about CG. Different organizations of the components will have different empirical effects. For example, (1) predicts that deletion plays no role in determining semantic representation. The opposite prediction would hold in an alternative where deletions are part of the transformational component.
By itself, CG distinguishes those phenomena which reflect properties of UG from those which are idiosyncratic and language particular. Thus the phenomena of a particular language fall into one of two categories: core or periphery. Periphery phenomena require extra descriptive apparatus beyond that available from CG. For this reason they are considered to be marked, whereas the phenomena that fall under CG are considered to be unmarked (see van Riemsdijk (1978b) and Koster (1978c) for additional discussion).
The way in which CG is formulated determines the cut between core and periphery. Alternative formulations of a “finite clause” condition on binding provide an interesting example. One formulation, the Propositional Island Condition (PIC) as in (2a), is given in terms of the notion “finite clause,” while another, the Nominative Island Condition (NIC) as in (2b), is given in terms of the notion “nominative Case.”
So defined, the PIC and NIC have different empirical effects. In the case of complex NP’s, the NIC but not the PIC predicts that (3) is grammatical. (See Chomsky (1980) for discussion).
The anaphor each other is free in the embedded finite clause but is not in the nominative Case, though it is bound in the matrix clause. Under the NIC the kind of binding exemplified by (3) belongs to the core and is unmarked. Under the PIC it belongs to the periphery and is marked—i.e., it will require some modification of CG to account for the grammaticality of (3). There is more to say about this issue and we will return to it below.
The grammaticality of (3) has been used to motivate the formulation of the finite clause condition on binding in terms of Case. Thus the argument for the NIC over the PIC is based on greater coverage of data. By itself this argument is inconclusive because it is always possible, as we will argue below, that the grammaticality of sentences like (3) reflects marked phenomena which cannot properly be used as motivation for particular formulations of CG.
Another argument for formulating CG in terms of Case concerns the replacement of the *[NP to VP] filter of Chomsky and Lasnik (1977) with the *N filter of Chomsky (1980).
As discussed in Chomsky (1980), the properties of (4) stipulated in the unlessclause follow naturally from Case analysis. In particular, a more restricted version of the context “adjacent to and in the domain of [−N]” in (4) is independently needed to account for Case-assignment. The replacement of the *[NP to VP] filter with the *N filter eliminates the seemingly ad hoc unlessclause by absorbing it into rather natural and independently motivated conditions on Case-assignment, to be discussed below. This substitution, though not entirely without problems—see Anderson (1979, 4.3) for discussion—also clears up several technical problems with the *[NP to VP] filter, as noted in Chomsky (1980).
The context “adjacent to and in the domain of [−N]” stipulated in the unless-clause of (4) appears in Case theory in terms of the more restricted notion “governed” as it applies to the rules of Case-assignment (6).
As a first approximation, the notion “govern” may be defined for configurational languages as (7).
As it stands, the definition of “govern” (7) is too restrictive because it does not account for Case-marking in all grammatical constructions. For example, in English it does not account for Case-marking of a direct object NP across an indirect object NP in double object constructions or of lexical NP subjects of infinitival complements. If (7) is the correct definition for CG, then these constructions fall outside the core with respect to Case-assignment and so require special modification of core notions to account for their grammaticality. The fact that infinitival complements in most languages do not allow lexical subjects suggests that the grammaticality in English of infinitival complements with lexical subjects should fall outside the core. Note that grammatical exceptions constitute no particular problem given the logic of markedness.
Case-assignment as formulated in (6) applies to any NP in an appropriate context, the simplest hypothesis. In addition to lexical NP’s, which must be marked for Case given the *N filter analysis of [NP to VP] constructions, (6) also applies to empty NP’s—i.e., PRO, NP-trace (henceforth NP-e), and WH-trace (henceforth WH-e). Some empty NP’s must be marked for Case—more specifically, subject to (6c)—under the assumption that the NIC is the correct finite clause condition on binding for CG. The NIC accounts for the illformedness of the representations in (9) only if PRO and NP-e are marked nominative by (6c).
This argument does not extend to WH-e, but an argument can be constructed on the assumption that the *N filter (henceforth *N) is, like its predecessor *[NP to VP], a filter of the Phonology. (This assumption is examined below).
Then, from the organization of the grammar in (1), *N applies after Deletion. In particular, *N applies after deletion in COMP (see Chomsky and Lasnik (1977) and Chomsky (1980) for discussion of such deletions). Granting this, we can now give an argument that WH-e in an NP position must be subject to *N, in which case it also must be subject to Case-assignment. Note first that COMP in relative clauses like (10) is a position where deletion may properly occur.
Under the assumption that *N applies only to lexical NP’s, (11b) ought to be grammatical because the offending item, un-Case-marked who as in (11a), has been deleted.
Since (11b) is ungrammatical, we may conclude that WH-e is subject to *N.9 Assuming empty NP’s have no internal structure, *N must be reformulated as (12).
This filter (henceforth *NP) should not be confused with the *NP filter of Rouveret and Vergnaud (1980). It is worth noting that lexical NP and WH-e fall together as a natural class with respect to opacity (i.e., the PIC/NIC and (Specified) Subject Condition—see Freidin and Lasnik (1981)) as well as *NP.
Under this approach, as under the approach of Chomsky (1980), infinitival relatives such as a man to fix the sink are apparent counter-examples if derived by WH-movement, since the subject of the infinitival cannot be marked for Case. Williams (1980) argues, partly on the basis of this fact, that the subject in these constructions is actually PRO. Note that such an analysis is required for infinitival relatives such as a problem to solve.
A Wh-element could not have been fronted from both e-positions in (13).
If Case-assignment applies after movement rules ((I-b) of (1)) as we have indicated, then WH-phrases in COMP will not in general be marked for Case, though their traces in appropriately governed positions will. Presumably, Case-assignment cannot apply prior to movement rules because subjects of passives are assigned Case only after NP movement.10 Thus *NP will give the wrong result for WH-phrases in COMP without some further modification of the grammar. There are at least two solutions to this apparent problem. A WH-phrase in COMP could be exempted from *NP or assigned the Case of its trace in a governed position by a special rule. The discussion which follows is neutral between these two proposals.
The question arises as to where the Case-assignment rule applies with respect to the rest of the grammar. We will assume that it applies as a unitary process, the simplest hypothesis. As such, it must precede *NP and also the NIC, as well as follow movement. Thus in order to determine the ordering of Case-assignment, it will be useful to determine the ordering of *NP.
Based on an EQUI deletion analysis of nominals like (14), there is an argument that *NP must apply after Deletion.
In the EQUI analysis the underlying structure of (14) is (15).
(14) is derived from (15) by the application of EQUI, which deletes PRO-self (see Chomsky and Lasnik (1977, 468)), and deletion in COMP, which deletes the complementizer for. If *NP applies after Deletions, the derivation goes through without problems. However, if the filter applies before the Deletion component, then self, which is considered lexical by virtue of having the same distribution as other lexical NP’s, must be marked for Case. Otherwise *NP will designate (15) as illformed and (14) would have no source as a result.
The question then is how self in (15) is assigned Case under the *NP-Deletion ordering. It cannot be assigned Case by the nominal desire because N does not assign Case to NP’s that are in its complement. Thus (16) is blocked by *NP because Bernie is not marked for Case.
In contrast, Bernie in (17) is marked for Case by the complementizer for and therefore not subject to the filter.
Extending this analysis to (15), we could claim that self is marked for Case by for, making it immune to *NP. Then PRO-self and for would delete, yielding (14).
This analysis is not viable. If the complementizer for can delete after assigning Case to self in. (15), then there is no reason to expect it not to delete after assigning Case to Bernie in (17), thereby yielding the ungrammatical (16). Since, by hypothesis, Deletion follows *NP, the filter cannot be used to account for the ungrammaticality of (16), an unsatisfactory result.
In general it appears that when the complementizer for assigns Case to the subject of an infinitival complement, for cannot delete. As Chomsky notes:
The marked for-infinitive construction has the form (85) in the base:
Only [+ F] verbs [i.e. verbs which can assign Case across an infinitival clause boundary HL & RF] take such complements as (85), though of course (85) appears quite freely in other contexts. To accomodate these constructions within the present framework, we must assume that for assigns Case to NP in (85) (presumably, oblique Case by rule (68a) [our (6a), HL & RF]) and that when it does so it is undeletable. (1980, 30)
Citing the observation that the for-complementizer and the homophonous preposition share some properties, Chomsky goes on to propose the following implementation of the nondeletability of a Case-assigning for-complementizer.
Again following a suggestion of Vergnaud’s, let us assume that in the base, the complementizer for may or may not be assigned the feature [+ P] (i.e., assigned to the category Preposition). If it is assigned this feature, it is undeletable under the re cover ability condition [footnote deleted, HL & RF]. If for is not assigned [+ P], it will not assign Case and may delete, and in the complement to a [+ F] verb such as want, the subject of the infinitive will be assigned objective Case by the governing [+ F] verb.… (1980, 31)
An alternative account of the relationship between for-deletion and Case-assignment will be presented below.
The nondeletability of the for-complementizer when it has assigned Case accounts for why the illformed (16) is not derived from (17) by Deletion. This eliminates one problem for the EQUI analysis of (14) under the assumption that Deletion follows Case assignment and *NP, but creates another. If in (15) for assigns oblique Case to self (see fn. 5), then for cannot delete. As a result, the grammar will derive (18) and not (14) from (15).
The illformedness of (18) falls under the *[for-to] filter of Chomsky and Lasnik (1977). This account generalizes to WH-movement as well. The *[for-to] filter also excludes (19).
the WH-e in subject position is not Case-marked and therefore illformed under *NP. Alternatively, we could assume that the WH-trace is marked oblique and that the deletion of for is in violation of the recoverability condition.
As shown above, the principle governing the deletability of the for-complementizer contributes to a quite general account of (16)–(20). It creates problems for the EQUI analysis of (14), but only if *NP is assumed to apply before Deletion. In this way the EQUI analysis of (14) provides an empirical argument for considering *NP as a filter of the Phonology, in which case it follows Deletion as in (1).
We turn now to considerations relevant to the ordering of for-complementizer deletion and Case-assignment. The principle governing the deletability of the for-complementizer has interesting consequences for the Case analysis of [+ F] verbs like want, the verbs which can assign objective Case across a clause boundary. On the basis of examples like (21), we assume that want subcategorizes for for-infinitive constructions (see (85) above) and that the underlying structure of (22) is (23).
Given Chomsky’s principle governing for-complementizer deletion, it must be assumed that for in (23) does not assign Case to Mary and that Mary is assigned Case by want instead.
If Case-asignment applies before Deletion, then the context (24) is involved; but if it applies after Deletion, then the context (25) is.
In terms of the definition of “governed” in (7), the relationship between V and NP violates both the adjacency and sisterhood conditions in (24), but only the sisterhood condition in (25). Assuming Chomsky’s principle and the ordering Case-assignment before Deletion, we are committed to allowing want to assign Case to the subject of an infinitival complement in violation of both conditions defining the relation “govern.” In contrast, the assumption that Case-assignment follows Deletion allows us to maintain at least the adjacency condition for Case-assignment with [+ F] verbs. Structures like (26) provide further motivation for the latter assumption.
In (26) wants is neither adjacent to nor a sister of Mary. If want can assign Case to an infinitival subject in this context-as must be the case to account for (22) under the hypothesis that Case-assignment precedes Deletion, then we are faced with the problem of explaining why Case-assignment does not apply to Mary in (26). This problem does not arise under the converse hypothesis that Deletion precedes Case-assignment. Under the latter hypothesis, Caseassignment in (22) would not violate the adjacency condition and so the ill-formedness of (26) falls out from the general theory of Case.
There is however one class of cases, as in (27), where Case-assignment does apparently apply between a verb and a NP that are neither adjacent nor sisters.
( designates the zero morpheme complementizer of Chomsky and Lasnik (1977)). Unless the Wh-e’s in S are marked for Case, such constructions should be excluded by *NP. The fact that the questions in (27) appear to be grammatical suggests that once again we are confronting phenomena that fall outside the core. Recall that [NP to VP] structures with full lexical subjects are exceptional among the languages of the world, a fact which supports the analysis under which structures like (22) also fall outside the core. Yet compared with (22), the grammatical status of (27) seems to us significantly less certain.
It is worth noting that whereas exceptions to both the adjacency and sisterhood conditions seem to involve only Case-marking of WH-e, exceptions to just the sisterhood condition may involve lexical NP (as in (22)) as well.
The WH-e counterpart of (22) is (28).
We suspect that when lexical NP is assigned Case in violation of one or more of the core conditions, WH-e will also be Case-marked in that position, but not conversely. This suggests phenomena which might properly fall under a substantive theory of markedness.
Considerations relevant to a theory of markedness provide yet another argument in favor of ordering Case-assignment after Deletion rather than before. Once again let us consider the two ordering theories:
For theory A Chomsky’s principle governing the deletion of for-complementizer must be stipulated in order to account for (16), and (18)—(20). From this it follows that Case assignment must be able to apply across a complementizer in the derivation of (22) from (23). Thus Case-assignment in the derivation of (22) involves exceptions to both the adjacency and sisterhood conditions on government. Now if Case-assignment in (23) violates both conditions in assigning Case to lexical NP, then we have no principled way of accounting for the ill-formedness of (29) since presumably Case-assignment could apply to lexical subjects of the infinitival complements in these constructions.
Under theory A, there appears to be no principled way to distinguish the different possibilities for Case-marking of lexical NP and WH-e in terms of the core notion of government.
Considered in terms of the acquisition of a grammar, theory A is problematic. In acquiring the grammar of English, if the language learner must modify the Case assignment rule to allow for (22) as discussed above, then that modification would, presumably, allow in (27)–(29). How (29) would then be excluded is a mystery since negative evidence of this sort is not available in the corpus that in part determines the particular grammar acquired.
We could, of course, avoid this problem for theory A by claiming that modifications of CG are made on a case by case basis. If this is the correct way to proceed, then modifications of CG would be truly arbitrary and there would be hardly any motivation for a substantive theory of markedness.
In contrast to theory A, theory B allows for a clear distinction between (22) and (27) in terms of the adjacency and sisterhood conditions on government. Case-assignment violates only the latter in the derivation of (22), but both in the derivation of (27). Most significantly, under theory B, it follows automatically that if for assigns Case, it does not delete—since deleted for will not be present when Case-assignment applies.
Given the organization of CG in (1), ordering Case-assignment after Deletion has two important consequences. First, *NP must be ordered after Deletion regardless of how nominals like (14) are analyzed. Second, the NIC cannot be maintained as the relevant finite clause condition on binding since LF has no access to Case-marking.
As mentioned above, the empirical argument of Chomsky (1980) for the NIC is inconclusive. Though bound anaphors can often occur embedded in the subjects of finite clauses such as (30), freely referring pronouns can also occur in these positions as in (31).
Here John and him can freely co-refer, even though under the NIC, Disjoint Reference should apply (see the appendix of Chomsky (1980)). The question arises, then, as to which is the more “basic” phenomenon—the application of bound anaphora or the blocking of Disjoint Reference. In fact, the pronoun in (31) is apparently free in reference quite generally, while there is substantial variability about the acceptability of (30a and b). All else being equal, this substantially weakens the argument for the NIC.
An analysis of the distribution of lexical anaphors in complex NP’s provides an independent argument against the NIC and in favor of the PIC. Formulated as the NIC, the finite clause condition on binding in conjunction with the (Specified) Subject Condition (SSC, see Chomsky (1980)) allows a lexical anaphor in the subject of a finite clause to be bound by an antecedent outside that clause just in case the anaphor is not marked nominative. The relevant cases involve reflexive pronouns and reciprocals as in (30a and b) above. The PIC, in constrast, rules out (30a and b). It also accounts for (32), whereas the NIC does not.
Since the illformedness of structures like (32) should follow from CG, the NIC analysis requires the addition of another condition on binding. An obvious candidate is a stipulation that anaphors must be bound (see Freidin (1978)). This stipulation is redundant with respect to those cases already covered by the NIC and SSC—e.g., (33) and (34) respectively.
Given the PIC instead of the NIC, the stipulation is totally redundant. That is, the stipulation falls out as a theorem given the PIC and SSC as axioms (see Freidin and Lasnik (1981) for further discussion). This situation does not provide a decisive argument against the NIC in favor of the PIC, but it is suggestive.
Another argument against the NIC is provided by facts concerning variation across idiolects. If the NIC is the correct core principle, then we might expect a situation where structures like (30a and b) are grammatical in all idiolects. But if this situation does not hold, then we face the serious problem of explaining how the language learner constructs a grammar which excludes (30a and b) in the absence of the relevant evidence—i.e., negative evidence.
On the basis of an informal survey, we have found that there is in fact variation across idiolects concerning (30a and b). Interestingly, the variation in part distinguishes between reflexive pronouns and reciprocals. Thus four possible idiolects can be projected based on the grammaticality judgments concerning (30a and b).
Only (a), (c), and (d) were attested in our informal survey. This suggests that reciprocals may have a wider distribution than reflexives, perhaps another phenomenon to be covered by a theory of markedness.
Given that (30a and b) fall outside the core phenomena, it seems reasonable to suppose that idiolectal variation results from very specific modifications of CG which in turn result from conflicts between CG and the initial data for language acquisition. But then it should be possible that some corpus of initial data contains only the reflexive cases (30b) and not the reciprocal cases (30a). Given the existence of such a corpus, an auxiliary hypothesis is needed to account for the non-existence of idiolect (b). We assume that this hypothesis will be provided by a substantive theory of markedness—i.e., an empirical hypothesis about the limited range of variation among grammars.
