Abstract
A uniform data structure is defined to represent sentence components (words) for a parallel processor of the natural language. The basic notation of rough sets is used for the representation of sentence elements in the computer, to exhibit their context and to select proper meanings. An approximation space reflecting the context of any word in a sentence is defined by induction from the most general meanings of words adjoining that word in the graph model of the sentence and from a contextual knowledge base. The contextual knowledge base specifies the meaning of each word with respect to the most general words expected to adjoin that word in the graph model of the sentence.