Rigorously articulated conversational studies (based on George Kelly's personal construct theory) of the reading of complete texts raise new questions about the cognitive processes by which meaning is attributed to the printed word. Four generative models are examined: probabilistic, phrase structure, transformational, and semantic. No single model provides a universal explanation, although each illuminates certain parts of the conversational results. A "capacity for process" model, in which a variety of individually specific "programs" can be run, is outlined, suggesting that each reader operates more than one specific model. This superordinate model facilitates learning-to-learn conversations in which reading effectiveness changes significantly. (Author)