ABSTRACT
Recent experimental work in psycholinguistics has led to rather contradictory results, with different experimental paradigms pointing to mutually incompatible parsing algorithms and even contradictory parsing architectures. This chapter tries to provide supplementary criteria using evidence from linguistic and computation theory that can help us to narrow the range of possible human language comprehension systems that can be submitted to more effective experimental test. The paper discusses three algorithms and two general parsing architectures (ATN and LR(k) parsers) that are compatible with results from reading comprehension and recall studies first discussed by Wanner and Maratsos (1978). We show that only the LR(k) parser using an analogue of the successive cyclic movement analysis can both predict Wanner and Maratsos’ data and explain why natural languages are universally governed by a linguistic restriction known as the subjacency constraint. Along the way, we provide general criteria for distinguishing when a mechanism explains a body of experimental data from those cases where the mechanism is merely compatible with a class of empirical results.
