ii-i4


concepts which co-occur with similar concepts.   For example, if one is

see~ng to identify that 1'aeroplane'1 and "aircraft't represent the same

concept, one might deduce this not from the fact that they co-occur in

the same document, but that they each separately co-occur with words

such as "fusilage", "propeller", "aileron11, and the like.  The SMART system

m~kes provision for iterated concept-concept correlation to any depth.

    The specifications affecting this procedure are as follows:


    CONOON n   the nun~ber "n" is the nun~ber of iterations of the
               concept-concept correlation process.   "CONCON 1"
               specifies simple first-order correlation;

    ~DECC a    a is either "COS" or "OVLAP" to specify cosine or over-
               lap correlation for the first concept-concept correlation;

    ~D~C a     a is either COS or OVLAP to specify the correlation mode
               for concept-concept correlations after the first corre-
               lation, i.e., for the second and following iterations;

    CUTCC x    x is a number between 0 and 1 (0.6 is typical) to specify
               the cutoff for the first concept-concept correlation;

    CUT2C x    x is a nuzi~ber between 0 and 1 specifying the cutoff for
               correlations after the first;

    CO~~IN n   n specifies the lowest concept number which will be
               correlated. This specification is to be used primarily
               with null dictionaries prepared by THES (see part 6.1)
               in which the concepts are arranged by frequency, thus m~aing
               it possible to specify the lowest frequency word to be
               correlated;

    CONMAX n   n specifies the highest concept number which will be corre-
               lated. CO~~IN and CONMAX are used because the statistical
               procedures are not accurate for words which occur either
               very rarely or very frequently;