Cross-Language Portation of Natural Language Database Interfaces by Using Machine Learning

In this paper we present results from the portation of a natural language interface architecture from German to Japanese. One of the main obstacles to the efficient use of multilingual environments is usually the high amount of required manual knowledge engineering. Therefore, we apply techniques from machine learning to automate the acquisition of linguistic knowledge. In our interface architecture a prototype-based machine learning algorithm replaces an elaborate semantic analysis component. The learning task is to select the correct command class based on semantic features extracted from the user input. We recorded the user interaction from the German source application and used this data as training set for the machine learning module. The algorithm successfully abstracted from language-specific phenomena and produced linguistic knowledge at the semantic level, which could be used to classify Japanese input. The complete interface architecture was implemented by using deductive object-oriented database technology so that we could develop an integrated environment for the machine learning module, the natural language interface, and the database application.