Takahashi and Kishino [TK91] investigated understanding the Japanese Kana manual alphabet (consisting of 46 signs) using a VPL DataGlove. They then analysed the data using ``principal components''. This method isolates the important factors using a statistical method, by finding the attributes that vary the least in a given set of instances, and then using a set of rules based on these instances.
Based on this, they built up a table that designated the positions of individual fingers and joints that would indicate a a particular handshape. By then matching the handshape against what is essentially this template, the handshape of the person is matched, and the class is decided on.
They found they could successfully interpret 30 of the 46 signs, while the remaining 16 could not be reliably identified, due to a variety of constraints, such as the fact that they were moving gestures and that sufficient distinction could not be made in situations where fingertips touched.