The parsing of natural language is the product of dense interactions among various comprehension processes. We believe that traditional models have greatly underestimated the richness of these interactions. We propose a model for low-level parsing which is massively parallel, highly distributed, and highly connected. The model suggests a solution to the problem of word sense disambiguation which is psychologically plausible and computationally feasible. The paper outlines the general connectionist paradigm followed by a brief description of a three-level network to do parsing. Finally we trace through an example to illustrate the functioning of the model.