This book records a unique attempt over a ten-year period to use stochastic optimization in the natural language processing domain. Setting the work against the background of the logical rule-based approach, the author provides a context for understanding the differences in assumptions about the nature of language and cognition.