Modelling the ziji Blocking Effect and Constraining Bound Variable Derivations in MC-TAG with Delayed Locality

MC-TAG derivations using delayed locality (i.e. non set-local composition) have been used to model long distance binding in natural language. In this paper, I explore the possibility within Synchronous TAG that tree sets composing via delayed locality may have many possible syntactic derivations for a single semantic form. Using the Mandarin (Chinese) anaphor ziji as a test case, I show that the blocking effect which is described in other gener-ative frameworks as a constraint on covert movement can be modelled as arising from constraints across possible derivation families for a given tree set. Further observing the interaction of ziji with other bound variables in the language, I propose that anti-locality effects in variable binding arise when a more specialized bound variable is available. This has consequences for the definition of a typology of bound variables. Variable binding in Synchronous TAG (STAG) has been modelled as a phenomenon which exploits delayed locality. Building on the definition of delayed-local derivations presented in Chiang and Scheffler (2008), bound variables can be modelled as multi-component sets (MCSs) in both the syntax and semantics. As formulated in Storoshenko and Han (2015), Figure 2 shows that the bound variable obligatorily composes non-locally, with the α components substituting into the argument position of the predicate selecting the variable, while the β components compose directly with the antecedent. This is most crucial on the semantics side, where the variable's auxiliary tree composes with the " scope part " of a STAG generalized quantifier, as in Figure 3. The bound variable carries a function from e,t to e,t which accomplishes the variable binding, in a form similar to Büring's (2005) Binder Index Evaluation Rule. On the syntax side, the β component acts as solely an agreement check. The English bound variable pronoun her is lexically specified for third person feminine ϕ-features; in this case there is no gender specification on a student, so the combination goes through. For nominals with inherent gender specification, for example every boy, the resulting feature clash would block the derivation. In this way, agreement is treated as a purely syntactic phenomenon, with no manifestation in the semantics. As written, these trees could compose into a sentence such as A student loves her father.