- 69 -

One-place, single-term hierarchies
        By far the most difficult device to establish was that involving hierarchical
linkage. Two major (and connected) problems arose. Firstly, the arbitrary and
somewhat artificial restriction implicit in the need to place each term in one hier-
archy only. This arose inevitably from dealing with single terms and meant that the
assistance normally given to definition by context was absent. Secondly, the prob-
lem of interpreting the prolixities and ambiguities of the natural language index vocab-
ulary in terms of this particular type of controlled vocabulary.

Problems of hierarchy
        The term hierarchy as normally used in indexing can mean one of three dif-
ferent things:
1) A generic hierarchy; i. e. , a system of subordinating some terms to others
whereby only terms which reflect the relationship of being kinds of a thing are
subordinated to that thing. Other relations are excluded. But the basis for the for-
mation of the species may or may not be a ,fundamental' characteristic.
2) A strict genus/species hierarchy, differing from (1) in that it is confined to the
use of ,fundamental' characteristics; e.g. 0 Methane could not be subordinated to
Fuel (as it is in the test schedules) since a fundamental definition of Methane does
not require characterization by this attribute. A parallel has been drawn by Gardin
(Ref.26) with the distinction between paradigmatic and syntagmatic relations, the
former reflecting permanent or necessary relations and the latter temporary or
contingent ones. A modern faceted classification uses both types of hierarchy in that
the same term might appear in two or more different facets according to its status
(as a product, an agent, etc. ) and not be confined to the facet where it ,fundamentally'
belongs.
 3) A hierarchy which includes generic and non-generic elements; i.e. , one which
 subordinates some terms to others regardless of the relation involved, so long as
 the subordinated term can be seen to belong to some category or facet of the 'con-
 raining' class, e.g., the subordination under a term of its properties, parts, pro-
 cesses, etc., as well as its kinds. This situation is typical of nearly all existing
 library classifications.
         Reasons why (3) should be treated as a separate device (,non-generic hierarchical
 linkage'} have already been given and are not considered here. In choosing between
 (1} and (2) for single-term hierarchies, logic seemed to suggest that (2} be chosen;
 for if each term may go in one hierarchy only it is arguable that that one place should
 at least reflect the most essential characteristics of the thing represented. On the
 other hand, the practical purposes of hierarchy in indexing would sometimes be ill-
 served by such an arrangement. This purpose is to provide for each term a set of
 class-mates standing in the same relation (of Thing/Kind)to the containing class
 and thus facilitate the expansion, or contraction, of any given class by the inclusion
 or exclusion, of some or all of these helpfully related neighbouring terms; and
 ~elpfully' here depends on the subject context.
          If the terms are relegated to a ,fundamental' or 'common' category, these help-
  ful relations tend to become tenuous; e.g., if the term Upper is located in a highly