Binding theory and levels of representation
Under the assumption that the binding theory consists of a set of filters which determine the wellformedness of binding for some level (or levels) of representation, an empirical issue concerning which levels are relevant arises. In most discussions it has been assumed that Principles A, B, and C apply as a unit. This may not be correct as will be discussed below.
For the purposes of discussion, we will assume the following levels of representation.
(1)
The mapping from S-structure to LF establishes quantifier/variable structures via a rule of Quantifier Raising (see May 1977) and perhaps some mechanism for reconstructing phrase-markers (see Chomsky 1981) as will be discussed below. The mapping from LF to LF’ (where “LF” stands for “logical form”) yields predication-structures via coindexing (as discussed in Chomsky 1982a:fn.11). Given these four levels, we may now consider what evidence might bear on the application of each of the three binding principles at each level. If a principle can apply at more than one level (as suggested in Chomsky 1981, 1982a), then the number of potential binding theories is greatly increased, as is the difficulty of identifying the correct theory. Moreover, some of the evidence that bears on the issue seems to lead to paradoxes. The following discussion is offered as an illustration.
Chomsky 1981 addresses the question of levels with respect to Principle C. He notes that the rules of Quantifier Raising and scope assignment to wh-phrases in situ will map the S-structures of the sentences in (2) onto the corresponding LF structures in (3), (irrelevant details omitted).
(2)
(3)
he c-commands John in (2), but not in (3). Since the pronoun and name are obligatorily disjoint in reference—which would be predicted if Principle C applies to (2) (i.e., at S structure rather than LF), examples like (2) provide evidence that Principle C holds at S structure.
There is also evidence that Principle C does not hold at S-structure. Consider the pair of sentences in (4).
(4)
While John and he may be construed as coreferential in (4a), they must be construed as disjoint in reference in (4b). This difference in interpretation appears to be systematic, based on the distinction between the relative clause in (4a) vs. the sentential complement in (4b). Given that the sentential complement, but not the relative clause, subcategorizes the noun report, it may be that the subcategorization domain is accessed in the interpretation of the variable bound by the WH-phrase. Presumably this access does not extend to relative clauses, which fall outside the subcategorization domain of the nouns they modify.
The interpretation of (4a) follows from the assumption that Principle C holds at S structure. Unfortunately, this assumption gives the wrong result for the interpretation of (4b), where obligatory disjoint reference holds between John and he as if the pronoun c commanded the name. At S-structure, the pronoun does not c-command the name in (4b), and therefore Principle C will not account for the only possible interpretation. One way to reinstate the appropriate c-command relationship between the name and the pronoun is to assume that some sort of reconstruction applies between S-structure and LF. That is, in LF the sentential complement of the noun (but, crucially, not the relative clause) would occur in the object position of submit. Some support for a process of reconstruction as sketched comes from other cases of WH-movement where Principles A and B are involved.
(5)
(6)
(5a–b) show that disjoint reference holds between the pronoun and the name in spite of WH-movement, which destroys the c-command relation that held between them at D-structure. (5a) is presumably a violation of Principle C, whereas (5b) violates Principle B. However (5d), in contrast to (5e), shows SSC (or TSC) effects—i.e., disjoint reference between him and John is blocked by the presence of the intervening subject Mary. (6c) indicates the same effects with respect to bound anaphora, where proper binding of each other by they is blocked by the intervening subject Mary. In other words, Mary functions as an accessible subject for the anaphor in spite of the fact that c-command between Mary and each other does not hold at S-structure. These paradigms would be accounted for under the assumption that Principles A and B apply at some level of representation in which reconstruction has applied to S-structure.
It should be noted at this juncture that the notion of reconstruction is at best problematic, as discussed in Higginbotham 1980: section 6 and Higginbotham 1983: section 3. The major difficulty involves the analysis for the LF-representation of sentences like (7).
(7)
Which book about which pianist did she read?
The pronoun she cannot be coindexed with the variable bound by which pianist in LF. This would follow under Principle C given that variables (including WH-trace) have the status of names and that the pronoun c-commands the relevant variable in LF. The idea is to assimilate structures like (7) to those of (8), which contain names instead of a variables.
(8)
She read a book about Ingrid Haebler.
To do this, (7) would have to have an LF representation like (9).
(9)
Yet the status of the term [x about y] in (9) is questionable on both syntactic and semantic grounds.
The above discussion seems to suggest that the notion of c-command may not be sufficient for binding theory when extended to Ā-binding as in the case of WH-movement and Quantifier Raising. As we have seen, the issues surrounding this question are complex and will require substantial clarification. The question of levels of application for the binding theory depends in part on the prior question of the sufficiency of c-command
. Before concluding this section, let us examine some additional evidence, which does not involve the question of the sufficiency of c-command, but instead bears on the question of whether Principles A, B, and C apply as a module at the same level(s) of representation. The evidence concerns equative constructions as discussed in section 2 with respect to the notion ‘accessible SUBJECT’.
(10)
In order to determine that those is not an accessible SUBJECT for the anaphor herself, those must be coindexed with NPj. Yet given this coindexing, we might ask why (10) is not in violation of Principle C. One possible answer would be that Principle C applies at a level of representation prior to the coindexing of those and NPj. Suppose that this coindexing results from a rule of predication which maps LF onto LF’ Then Principle A must apply at a different level of representation than Principle C. Given (10) we might assume that Principle A applies at LF’, whereas Principle C applies at LF or S-structure, depending on the sufficiency of c-command for Ā-binding. In this way there may be some motivation for distinguishing Principle C from Principles A and B. So far, however, there appears to be no motivation for separating Principles A and B in terms of the level(s) of representation to which they apply.
