The catalog: a flexible data structure for magnetic tape

The files of data used in linguistic research differ from those found in other research applications in at least three important ways: (1) they are larger, (2) they have more structure, and (3) they have more different kinds of information. These are, of course, all simplifications but not gross ones. It is true that the files that must be maintained by a large insurance company or by the patent office are so large as to pose very special problems, but the uses to which the files are to be put are fairly well understood and their format and organization is not usually subject to drastic and unexpected change. It is also true that the data from a bubble chamber is interesting only if collected in vast quantities, but this is not the only respect in which a bubble chamber is a special kind of tool. A typical linguistic job will bring together a number of files, each very large by the standards of everyday computing: a body of text, a dictionary and a grammar for example. The grammar, if it is anything but a very simple one, will contain a large number of elementary items of information of different kinds, each related to others in a number of different ways. This is what it means to say that the file has a lot of structure. The dictionary may also contain grammatical codes which may consist of characters from one of the languages represented in the dictionary or may be something altogether different. If the dictionary contains alternatives to which probabilities are assigned, then these will presumably be in the form of floating-point numbers. This is what it is like for a file to contain different kinds of information.