Binding theory and empty categories
In GB it is assumed that the binding theory applies to empty categories as well as to lexical NP-types, and therefore provides a typology of empty categories similar to that of lexical NPs (see Chomsky 1981, 1982a). A critical examination of this assumption follows.
To a large extent this view is based on the observation of earlier analyses (e.g., Chomsky 1976) that the opacity effects for lexical anaphors in the S-paradigm extend to NP-trace and PRO. It was assumed that NP-trace and bound PRO could be considered as empty category analogues to lexical anaphors and that their distribution would then follow from the same principles that determined the distribution of lexical anaphors (i.e. the SSC and TSC of the earliest accounts and their descendents).
This analysis is not supported by the opacity effects in the NP-paradigm since neither NP-trace nor PRO may occur as arguments of nominals.
(1)
(1a) cannot be interpreted as ‘John read his own books’ and (1b) is not a wellformed variant of John’s book was reviewed. (1b) is independently excluded by the Case Filter (Chomsky 1980, 1981) since the NP containing the lexical item book is not marked for case by the passive predicate reviewed. To exclude PRO from a variety of positions (e.g., in (1a)), Chomsky (1981) analyzes PRO as a pronominal anaphor and therefore subject to both Principle A and Principle B. In this analysis the binding domains for both principles are stated in terms of governing category, where specification of a governor is crucial. Thus PRO as a pronominal anaphor must be both bound and free in its governing category if it has one. The conclusion then is that PRO can only occur in ungoverned positions—i.e., subject of gerunds and infinitivals where does not apply (cf. Aoun and Sportiche 1983).
Given that the binding domains for Principles A and B are distinct with respect to the NP-paradigm as discussed above, it no longer follows as a theorem of the binding theory that PRO must be ungoverned. Specifically, the nonoccurrence of PRO in NP is not explained by the binding theory. Furthermore, given the formulations of Principles A and B above, the binding theory now accounts for the relevant cases of PRO binding, given in (2), as follows.
(2)
(2a) is excluded because PRO-binding violates Principle B (although it satisfies Principle A). (2b) is illformed because PRO is free in the domain of an accessible SUBJECT (=AGRi) and therefore violates Principle A, though it satisfies Principle B. (2c) is illformed for the same reason, where the accessible SUBJECT here is the syntactic subject Mary. This analysis of PRO as a pronominal anaphor excludes the illformed cases, but is only partially relevant for the wellformed (2d). With respect to Principle A, the matrix S is the binding domain for PRO. Thus PRO satisfies Principle A in (2d). Since is a barrier to government, PRO is ungoverned in (2d) and therefore neither satisfies nor violates Principle B. If PRO were governed (i.e. if had occurred), then PRO-binding in (2d) would violate Principle B. If a lexical pronoun himi were substituted for PRO; in (2d), then to avoid the Case Filter the lexical pronoun must be marked for case, hence governed and thus in violation of Principle B. For (2d) then it is irrelevant that PRO is analyzed as a pronominal, in contrast to (2a) where it is crucial.
This leaves the problem of the nonoccurrence of PRO in NP. If we take it as an axiom of the theory that PRO cannot occur in a governed position, then the full distribution of PRO (including non-control PRO, which doesn’t behave like an anaphor in any event) is accounted for. Of course this eliminates the motivation for analyzing control PRO as a pronominal anaphor. Its distribution would be accounted for by the above-mentioned PRO-axiom and the lexical property of particular verbs which allow (or require) control structure complements (e.g., persuade (obligatory control) vs. expect (optional control)— see Jackendoff 1972).
(3)
(4)
As with lexical anaphors, the subject of an infinitival complement is transparent to binding from the matrix clause.
In the NP-paradigm, however, NP-trace is generally disallowed, as illustrated by a comparison of illformed examples like (5) and (6) with the corresponding wellformed examples (7) and (8).
(5)
(6)
(7)
(8)
The ungrammaticality of (5) is predicted within case theory since the NP pictures of e, being governed by the passive participle, will not be assigned case—in violation of the Case Filter, which requires that a lexical NP be case-marked. The ungrammaticality of (6) follows from the theory of predicate/argument structure, which prohibits a single NP from being assigned more than one argument function in a sentence. In (6), John is assigned the argument function of subject of discovered and in addition the argument function of object of the nominal pictures via trace-binding, in violation of the condition on the uniqueness of argument assignment. Because case theory and the theory of predicate/argument structure independently predict the nonoccurrence of NP-trace in NPs, the NP-paradigm for NP-trace provides no motivation one way or the other for considering NP-trace as an anaphor.
Another instance where NP-trace does not pattern like a lexical anaphor involves binding across non-θ-subjects. While it appears that an anaphor may be free in the domain of an accessible non-θ-subject, a NP-trace may not (as noted in Freidin and Harbert 1983).
(9)
(9) cannot be explained as a violation of the Case Filter or conditions on predicate/argument structure. It could be excluded under Principle A if NP-trace is analyzed as an anaphor with the proviso that nonlexical anaphors are not sensitive to the distinction between θ-subjects and non-θ-subjects.
NP-trace is unlike a lexical anaphor in yet another way It does not bear a θ-role independently of its binder. While a lexical anaphor receives its reference from its binder, the NP-trace transmits a θ-role to its binder. Reference is not at issue. Suppose that for NP-trace the domain of accessible SUBJECT does not involve the θ/non-θ distinction as indicated in (9). Then the binding domain for anaphors with respect to the S-paradigm could be reformulated in terms of θ-domain since both formulations are equivalent with respect to the S-paradigm. (Recall that the NP-paradigm is not relevant for the formulation of binding domains with respect to NP-trace). Thus a θ-role can only be transmitted within a θ-domain, just as reference can only be transmitted within a θ domain (defined in terms of accessible θ-SUBJECT). NP-trace, like a lexical anaphor, may not be free in its binding domain.
The distinction between NP-trace and WH-trace is based on the different properties of these traces in wellformed structures. NP-trace is always caseless, whereas a corresponding WH-trace (i.e. in an NP-position) is always casemarked (see Lasnik and Freidin 1981). There is also a distinction between the binders of the two types of trace. NP-trace is always bound by an NP in a grammatical function (GF) position (e.g. subject), while WH-trace is always bound by a phase in COMP, a non-grammatical function position. In the GB framework, a GF position is designated as an A position; and a position, as an A-position (where “A” stands for “argument”). NP trace is properly bound when it is A-bound in its binding domain. WH-trace is properly bound when it is Ā-bound subject to certain locality requirements (e.g., Subjacency see Chomsky 1977b, 1981). Given that PRO is ungoverned, then it too, like NP-trace, is a caseless empty category.
The case-marked/caseless distinction is therefore sufficient to differentiate empty categories that function like anaphors from WH-trace. This functional interpretation of empty categories contrasts with a derivational interpretation in which an NP-trace is a trace resulting from movement between A-positions, as opposed to a WH-trace, which results from moving a WH-phrase to an Ā-position (COMP).
Before exploring the functional analysis of empty categories, let us first determine the status of WH-trace with respect to the binding theory. The binding theory as presented above is a theory of A-binding. NP types are determined on the basis of behavior with respect to Principles A, B, and C as applied to A-binding. To determine the status of WH-trace, we consider its behavior with respect to A-binding. The simplest case is given in (10) where the WH-trace (ei) is A-bound by the pronoun he.
(10)
The coindexing in (10) expresses the interpretation “which person is such that that person saw himself?”. Because the question who did he see? has no such interpretation, this representation must be excluded. The question cannot be asked with intended coreference between the subject and object of see. To express this intended coreference, the question must be formulated as who saw himself?. Therefore a WH-trace does not behave like a lexical anaphor. The example does not distinguish between the two remaining possibilities for the status of WH-trace—as a name or a pronoun—since both must be A-free in S.
The two possibilities can be distinguished in the subject position of a finite sentential complement, where a pronoun may be bound but a name must be free. The construction which distinguishes between the binding of pronouns and names shows that WH-trace behaves like a name rather than a pronoun. Consider the following paradigm.
(11)
In (11.a.i) and (11.b.i) the pronoun is free in this binding domain, thereby satisfying Principle B. In (11.a.ii) the name John is bound, and therefore this representation violates Principle C. Similarly, intended coreference between the subjects of say and won in (11.b.ii) is not possible, in contrast to (11.b.i). Assuming that the WH-trace has the status of a name, (11.b.ii) is excluded under Principle C as well.
So far, we have empty categories that behave like anaphor s (NP-trace and control PRO) and names (WH-trace) with respect to the binding theory. Whether there are empty categories that behave like pronouns remains to be determined. If there are, then the typology of empty categories would seem to mirror the typology of lexical NPs (with perhaps the exception of pleonastic elements—though see Chomsky 1982a). Evidence for empty categories behaving like lexical pronouns is open to interpretation. Putting aside the question of whether they exist in English (see again Chomsky 1982a), let us consider some evidence from Spanish.
In Spanish, empty categories occurring in the subject of finite clauses must be differentiated. The empty categories in (12) illustrates the standard paradigm for NP trace with raising verbs.
(12)
In (13) however the empty categories cannot be analyzed as NP-traces (or PRO since they are in governed positions).
(13)
In (13a) the empty category must be coindexed with an antecedent, whereas in (13b) it cannot, and in (13c) coindexing is optional. Thus only in (13c) does the empty category exhibit the same behavior as a lexical pronoun. The obligatory coreference in (13a) and the obligatory noncoreference in (13b) might be ascribed to lexical properties of the matrix verb dejar and querer, just as verbs of obligatory control in English impose (as a lexical property) coreference between their subject (or object) and the subject of an infinitival complement. In any event, the empty categories in (13) are distinct from NP-trace because they do not transmit a θ-role to their binder. As noted above, they are also different from PRO in that they occur in a governed position. Following standard practice, we will refer to such empty categories as pro (read “small pro”).
Under this analysis, empty categories can be distinguished in terms of the properties of their binder and the structural position they occupy.
(14)
(14) translates into the following feature matrix.
(15)
Given this feature analysis of empty categories, one question that arises is why are there only four empty category types when sixteen are possible.31 For example, is there an empty category like WH-trace with respect to its binder but ungoverned and caseless? Presumably the answer depends on what empty category types are compatible with the theory of grammar. Thus an empty category which is case-marked must be governed since case-marking occurs only under government. This rules out the possibility of an empty category which is case-marked but ungoverned.
In the current GB framework then, some properties take precedence over others. The problem is to determine which properties of empty categories are basic and which can be predicted by the various principles of grammar. It is by no means obvious how this is to be established. Any one feature in (15) could be eliminated and the remaining three would still distinguish between the four empty category types. If two features are eliminated, then the four empty categories could be distinguished only with the two features in (16).
(16)
This suggests that a functional determination of empty categories minimally requires reference to at least one property of the binder and one property of the empty category itself.
The functional determination of empty categories has some intriguing consequences for the analysis of representations containing empty categories. Consider (17) for example.
(17)
There are two possible derivations of (17), either by movement—in which case ei is a NP-trace, or by base generation—in which case the empty category is either PRO or pro. Functionally, ei is governed, case-marked, and A-bound by an antecedent with an independent θ-role. Therefore ei does not function as a trace (see (15)) regardless of how it is derived. Rather, it is functionally pro and hence a pronominal with respect to the binding theory. Therefore (17) violates Principle B since pro is not free in its binding category.
The alternative derivational analysis of the empty category in (17) requires two different explanations for the ungrammatically of the example. Trace-binding here would be ruled out by the prohibition against nonunique θ-role assignment, assuming that John would be assigned a second θ-role via trace-binding. PRO-binding would be excluded because PRO would be governed in violation of the PRO-axiom; and pro binding would violate Principle B as discussed above.
It might appear that this analysis allows us to derive the condition on uniqueness of θ role assignment via the functional determination of empty categories. The full derivation does not go through however. In (18), the empty category is functionally determined as pro, but would not violate Principle B.
(18)
Since the empty category is both governed and case-marked, it cannot be analyzed as PRO; nor can it be analyzed as NP-trace because it is case-marked and has a binder with an independent θ-role. Functionally this empty category is pro. However, in English the distribution of pro is limited to parasitic gap constructions, as in (19) where ei is a variable (WH-trace) which licences the parasitic gap (=pro) (see Chomsky 1982a, for discussion).
(19)
Note that ei does not bind because it does not c-command We might therefore assume that whatever accounts for the limited distribution of pro in English will also account for both (17) and (18) above, without recourse to Principle B, and independently of the functional determination of empty categories.
Nonetheless, the functional determination of empty categories is available within the GB framework for the analysis of derived structure. Whether it is empirically or conceptually better motivated than the derivational account of empty categories remains to be determined. What is clear is that, under the derivational determination, the binding theory can be used to identify empty category types (e.g., NP-trace as an anaphor), whereas binding theory cannot be used in this way under the functional determination.
