Trace Theory
The notion “trace” assumed here is defined as an empty category—formally [β e], where “β” denotes an arbitrary category of the base and “e” stands for the identity element. For example, [NP e] is an NP trace, [PP e] is a PP trace, and so on.
As empty categories, traces have two distinct sources. They may be transformationally generated by movement operations, which leave an empty category (identical to the moved category) at the site from which the movement occurred. This follows in part from the structure-preserving hypothesis (Emonds (1976))—see Chomsky (1977b) and Freidin (1977) for discussion. Traces may also be base generated, which follows from the optionality of base rules (cf. Chomsky (1965) and Emonds (1976))—assuming a convention that an unexpanded category β is equivalent to [β e]. It is assumed here that all categories in base structures are indexed, including base-generated traces.6 However, coindexing is only a result of movement or construal rules—e.g., the rule that assigns a reflexive pronoun an antecedent (see Chomsky (1977a)).
The derivation of “truncated” passives illustrates both sources.
(1)
The base structure (1b) (details omitted) of sentence (1a) undergoes NP Preposing, generating (1c). The trace in (1b) is base-generated; the one in (1c) is transformationally generated. As in (1), base-generated trace is not coindexed with any lexically filled category, whereas transformationally-generated trace is coindexed by definition of movement. The difference in indexing serves to distinguish between the two types of trace throughout a derivation. They are subject to different surface structure conditions, as will be discussed below.
The relation between a moved phrase and its trace is interpreted here as that of bound anaphora, following Chomsky (1977a; 1977b) and Fiengo (1977). This interpretation allows us to capture the striking generalization that traces and bound anaphors (e.g., reflexive pronouns and reciprocals (each other)) are subject to exactly the same binding conditions (Chomsky (1976)). For example, traces, like reflexive pronouns, must be bound to an antecedent in S—in contrast to regular pronouns (he, she, etc.), which are essentially free in reference (see Lasnik (1976) and Chomsky (1976)). Binding can be formally expressed in terms of coindexing of categories. Under this interpretation, the distinction between base-generated and transformationally-generated trace translates as unbound vs. bound anaphor.
In general, bound anaphora must meet the following conditions on binding:
(2)
PB accounts for the following facts concerning the occurrence of bound anaphors like reflexives, reciprocals, and traces in surface structure.
(3)
The examples in (5b) contain “bound anaphors” that are not bound to an antecedent, thus violating (3a). In (5c) the examples contain bound anaphora where the antecedent does not c-command the anaphor, thus violating (3b). Only the examples in (5a) are well formed with respect to PB.
With this preliminary sketch of trace theory in mind, we turn now to the analysis of strict cyclicity. Within the framework of trace theory, the derivation of (4) involves (6), which yields (4) as a surface structure.
(4)
(5)
(6)
The mapping of (5c) onto (5d) via Wh Movement violates the SCC. It also results in the erasure of a bound trace, ti in COMP2 . Recalling that an NP trace is formally [NPi e] we see that the effect of trace erasure is the replacement of the identity element “e” with lexical material and, more importantly, a change of index on the category.
A prohibition against erasing bound traces would also exclude the mapping from (5c) to (5d).
(6)
The TEP guarantees that once a trace is bound, it is inaccessible to rules and therefore remains invariant throughout the rest of the derivation. Compared with the SCC, it provides an alternative account of why (5) results in the misgeneration that makes no reference to notions like “cyclic domain” and “stage of the cycle.”
The TEP (or its equivalent) is essential for a unified theory of semantic interpretation under trace theory (see Chomsky (1976)). For example, the TEP excludes the following derivation that occurs within a single cyclic S domain.
(7)
The mapping from (7b) onto (7c) violates the TEP. The structure (7c) that results from the violation of the TEP is uninterpretable at the level of surface structure, the level at which all interpretation occurs within the unified theory. Given only (7c), there is no way to interpret NPi as the indirect object of the predicate was given. Within trace theory, the fact that (7c) does not receive the correct interpretation—if any at all—constitutes independent empirical motivation for the TEP (that is, aside from violations of the SCC it covers).
(8)
The derivation (8) also yields (5) as a surface structure of (5). Therefore, another condition is needed to prevent the misgeneration of (5); that is, a condition that excludes (8) as a possible derivation. (8) shows that neither the SCC nor the TEP is sufficient by itself to account for the misgeneration of (5).
