Adaptive Parsing: Self-Extending Natural Language Interfaces

1. Introduction.- 1.1. Natural language interfaces.- 1.2. The dilemma.- 1.3. Towards a solution.- 1.4. The frequent user.- 1.5. Conclusions.- 1.6. Reader's guide.- 2. System Behavior in an Adaptive Environment.- 2.1. Foundations: Least-deviant-first parsing and MULTIPAR.- 2.2. The model.- 2.3. Assumptions and hypotheses.- 2.4. Adaptation vs. customizable and instructable interfaces.- 2.5. Prior research in adaptation.- 3. User Behavior in an Adaptive Environment.- 3.1. The behavioral hypotheses.- 3.2. The experimental condition.- 3.3. Control conditions.- 3.4. User profiles.- 3.5. Results and discussion.- 4. System Architecture and Knowledge Representation.- 4.1. Representing knowledge in CHAMP.- 4.2. Learning-related knowledge.- 4.3. Application-related knowledge.- 5. Understanding Non-Deviant Utterances.- 5.1. Segmentation.- 5.2. The Coalesce/Expand Cycle.- 5.3. A detailed example: the Coalesce/Expand Cycle for "Can- 89 cel the 3 p.m. speech research meeting on June.".- 6. Detecting and Recovering from Deviation.- 6.1. Parsing as least-deviant-first search.- 6.2. The cache.- 6.3. Error detection.- 6.4. A detailed example: Error detection during the parse of "Schedule a meeting at 3 pm June 7.".- 6.5. Error recovery.- 7. Resolution: Choosing Among Explanations.- 7.1. Converting meaning to action.- 7.2. Using default knowledge and inference.- 7.3. Interacting with the databases.- 7.4. Confirmation by effect.- 8. Adaptation and Generalization.- 8.1. Adapting to substitution deviations.- 8.2. Adapting to insertion deviations.- 8.3. Adapting to deletion deviations.- 8.4. Adapting to transposition deviations.- 8.5. A summary of adaptation in CHAMP.- 8.6. Controlling growth in the grammar through competition.- 8.7. The effects of adaptation on performance.- 9. Evaluating the Interface.- 9.1. CHAMP's performance on the hidden-operator data.- 9.2. CHAMP's performance in real user interactions.- 9.3. Summary of results.- 10. Critical Issues.- 10.1. The effect of kernel design on learning.- 10.2. The effect of kernel design on search.- 10.3. The effect of search constraints on system predictability.- 10.4. The lexical extension problem.- 10.5. Summary.- 11. Conclusions and Future Directions.- 11.1. Main results.- 11.2. Future directions.- References.