Computation with doubling constituents
In the biolinguistics program, syntax has access to only limited and minimal operations such as Merge (and Move) plus local (to Phase) probe-goal, Chomsky (2000, 2001). Under this blueprint, the challenge for implementation is twofold: (1) to show that such a system operating locally can systematically derive complex facts, and (2) that computation, i.e. the selection of the appropriate operation is locally (and perhaps globally) efficient. To this end, available mechanisms may be prioritized to create an unambiguous instruction stream, e.g. Merge over Move. In this paper, we develop a computationally verified implementation of classic Binding Theory facts, including pronoun and anaphor asymmetries for mono- and multi-clausal sentences, possessive DPs, and picture DPs. We take as our starting point Chomsky’s probe-goal system plus Kayne’s (2002) doubling constituent proposal for pronoun-antecedent coreference relations. Within this framework, maintaining Phase locality between probe and goal forces independent licensing of an r-expression from its pronominal component via an operation of Last Resort theta Merge. We maintain that this Last Resort variant of Merge maintains the computational efficiency of the probe-goal system in that it operates precisely at the limit of probe-goal search domain and it does not introduce any additional choice points into the instruction stream.