Experience with ROBOT in 12 Commercial, Natural Language Data Base Query Applications

The ability to understand Natural Language has long been a goal of Artificial Intelligence Research, and it is still far from being solved, however, in the early 1970's the AI research techniques reached a point whereby certain applications became feasible for the first time. Since that time, several systems such as PLANES[1], LIFER[2], and ROBOT[3,4,5] have been built that have demonstrated that the current state of the art is sufficient for quite good natural language data base query. The implementation of the R0301 system has been geared for high performance and instaliaoility in accual real world environments. As such, it offers the AI research community some insight into the difficulties encountered when putting the current: AI technology in the hands of people in the real world. This paper discusses the unexpected linguistic and semantic difficulties encountered in the 12 commercial applications to which ROBOT has been applied during the last year and a half.