Evolutionary Language Understanding

"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."--Bloomsbury Publishing. Cover Half-title Title Copyright Contents Acknowledgements 1 Prologue 2 The Origin of the Task The Need for Parsing The Compilation Metaphor The Messiness of Real-Life Language Probabilistic Approaches Notes 3 Simulating Evolution Stochastic Optimization Optimization by Simulated Annealing Practical Versus Ideal Algorithms A Pilot Experiment The Pilot Algorithms First Results Notes 4 An Early System Parsing for Speech Recognition The Structure of an Early System Plausibility v. Probability: The Mk 2 Evaluation Metric Word-Ancestor Assessment A Sample of Mk 2 Output Improving the Language Model Notes 5 The Current System: Goals and Standards Outline of APRIL3 Software Engineering Input Text The SUSANNE Analytic Scheme Limitations of the Analytic Tradition Logical Structure The Emergence of Standards Notes 6 The Current System: Front End and Parsing Algorithms Word-Hypothesis Lattices The Word Tagger Terminal Node Evaluation The Parse Annealer The Solution Space Initialization Move-Set Move Generation Notes 7 The Current System: Language Model The Structure of the Model Estimating Probabilities from Sparse Data An Example of the Good-Turing Technique Prototype Optimization Results from the Language-Model Annealer Notes 8 Parallel Tree Optimization The Need for Speed Parallel Processing An Algorithm for Optimizing Trees Implementation of the Algorithm Experimental Results A Provisional Conclusion Notes 9 Epilogue Note References Index