Previous work has shown that there is a major vocabulary barrier for new or intermittent users of computer systems. The barrier can be substantially lowered with a rich, empirically defined, frequency weighted index. This paper discusses experience with an adaptive technique for constructing such an index. In addition to being an easy way for system designers to collect the necessary data, an adaptive system has the additional advantage that data is collected from real users in real situations, not in some laboratory approximation. Implementation considerations, preliminary results and future theoretical directions are discussed.