Fuzzy semantics for a natural-like language defined over a world of blocks

Winograd used a toy world called BLOCKS for his famous "natural language understanding" system SHRDLU. In this world all distinct objects are equally distinct: a red-orange block is no closer to being a red block than to being a black block; and a rather large block is no closer to being a large block than to being a small block. This paper presents an APL package which gives a fuzzy set semantics for a limited English-like language (in which red-orange is near to red and so on), in a manner which is convenient for <u>reference</u>: the system accepts instructions in the form of simple indicative sentences, and demonstrates its "understanding" by indicating the block which is the intended referent. The work is based on the principle of maximum meaningfulness" and is robust, in the sense that small changes in the instructions or in the state of the world do not disturb it.