Research On Natural Language Interface To Structured Data

Nowadays, the information on the Internet is complicated and overloaded. Moreover, the format of the information is various, and the quality of that is uneven. Among them one kind of the most high-quality data is structured data , including the relational database, the ontology, and the XML database. The structured data of great value in use on the Internet is abundant, and the amount of it keeps on increasing.The traditional inquiry of structured data usually requires the users to be familiar with some specific formal query language (such as SQL) and the structure of the data, and then construct the inquiry sentences according to their information requirement. While the natural language interface to structured data allows users inquire the structured data by natural language without requiring users be familiar with query language and data construction. Therefore, it is a kind of more friendly way of information inquiry, and immensely promotes the usability of structured data. Thus, the research of the natural language interface to structured data is of great practical significance. Meanwhile, the natural language interface to structured data is namely the question answering track basing on structured data, and the QA is a hot topic in the fields of natural language processing and information retrieval, it can be seen that the natural language interface to structured data is also of great research value.The format of structured data is different, with a result that its corresponding natural language interface technology is different. This paper mainly researches on the natural language interface to two kinds of most common structured data which are the relational database and the Semantic Web.As for the natural language interface to relational database, two kinds of methods which are basing on shortest path and basing on sequence tagging are put forward. While the method of shortest path is of too much simpleness and inflexibility, it is ineffective. The method of sequence tagging has better effect, but it needs us to annotate large amount of corpus.On the conclusion of the experience and lesson drawn from the experiment of the natural language interface to relational database, the semantic web ontology is chosen to complete the natural language interface experiment. The method which we propose with the semantic web ontology and based on entity-relation path search achieves the best result.Experiments show that, not only on the relational database, but also on the ontology, the method put forward in the paper could construct a better natural language interface system. By contrast, as a kind of the data format of the natural language interface, the semantic web is a better choice.