The Design And Implementation Of Natural Language Interface For Highmoralmap Via Semantic Parsing

A variety of Web-based map applications develop rapidly in recent years, and they have gained wide application and popularity in our daily life. However, there is relatively little research on natural language interface of map operation. By taking Highmoralmap which is widely used as the research object, this thesis carry out a systematic pilot study of implementing the natural language interfaces for Highmoralmap via semantic parsing method. The main content of this thesis includes the following three aspects:1. The first natural language interface corpus and formal meaning representation based on Highmoralmap API by means of function encapsulation are constructed and developed, which lays a solid research foundation for the semantic parsing system.2. On the basis of the developed corpus, a semantic parsing algorithm with feature-word based rule matching method is implemented, and then a transform-based algorithm which can learn rules automatically from the training corpus is designed. Experimental results show that the rule set for semantic parsing can be obtained automatically by the transform-based algorithm and does not require any manual intervention, while producing a comparable performance with the artificially constructed set of rules. Therefore, the transform-based algorithm can be more effectively applied to larger datasets.3. Finally, a visual website with the proposed semantic parsing algorithms for natural language interfaces on Highmoralmap is designed and implemented. The visual website is beneficial for users to operate Highmoralmap, and can provide a basis for further enriching and expanding the corpus in the way of human-centered computing.