The main idea of this paper is to exhibit in a data file a symmetry between objects and attributes by storing data as a huge binary matrix, where the rows correspond to objects and the columns to the values that an attribute might assume. The symmetry is desired because retrieval questions ask for particular objects, as well as for objects obeying certain conditions on their attributes. Since the range of some attributes is rather large, methods for condensation are discussed, e.g., storing for an integer the powers of 2 (which of course just amounts to storing the integer in the usual way), and storing large character strings.

Since the matrix of binary values still may be very sparse, a method of condensing its columns is proposed: A fixed number c of binary digits is compressed into a zero if all of them are zero and into one, otherwise. This may be done repeatedly. An optimum value of c is computed.

The method seems rather useless for several reasons. It ignores completely that the binary matrix must be stored in a linear storage; thus a retrieval in one or the other direction amounts more or less to a sequential search, which supposedly was to be avoided by this method. The comparison with inverted files is incomplete since bit strings can be, and have been, used for indicating the keys of objects. Also, no mention is made of the difficulties of finding the attribute values of an object if the data are stored by column (possibly compressed), or updating compressed bit strings.