The central task for a Machine Reader is integrating information acquired from text with the machine's existing knowledge. Direct Memory Access Parsing (DMAP) is a machine reading approach that leverages existing knowledge and performs integration in the early stages of parsing natural language text. DMAP treats machine reading, fundamentally, as a task of knowledge recognition, and creates new knowledge structures and instances only when it cannot map the input text to existing knowledge. A goal of the research is to be able to use existing knowledge to improve the acquisition and integration of new knowledge, from (simplified) natural language. DMAP's understanding is driven by memory structures, and it maps immediately and directly to existing knowledge. This provides an opportunity to experiment with and evaluate methods for using existing knowledge (both semantic and episodic) to facilitate machine reading. We present the basic architecture of a DMAP implementation, three experiments to leverage existing episodic memory, and the implications of the experiments on future research.