FOA Home
As their central role in WordNet suggests, the hierarchic BT/NT or
hypernymy relations are especially important. Because of the constant
analytic pressure of the Western intellectual tradition, topics continue
to be refined into smaller topics, which the next generation of scholars
immediately refine further. This has led to a wide range of
classification taxonomies associated with various social groups of
scholars and scientists.
Figure (figure) shows several different
sources from which indexing information might be obtained. The most
important single source of subject indexing is the Library of Congress
(LoC). This indexing system is the basis of most large libraries because
its indexing scheme covers all disciplines. For exactly the same reason,
however, the indexers at LoC have far too much to do, and the resulting
indices are admittedly crude. Partly as a response to the lack of
adequate indexing structures, various professional groups have developed
their own taxonomies to help organize the information within their
particular technical specialty. For example, the Association of
Computing Machinery has developed an ACM
Computing Reviews Classification for use by its {\em Computing
Reviews} publication [REF396] . This
taxonomy is much more specific, and therefore more widely used by
computer specialists. It lacks, however, many of the advantages of the
LoC system. In general, the keywords are assigned by the authors rather
than by trained librarians. The system is rarely if ever integrated into
the operations of libraries. And while the indexing structure is much
more refined than that of the LoC, it is still too crude for most
research currently going on in any one sub-specialty. This has caused
some practitioners in various sub-specialties to develop their own
extensions. For example, David Waltz was commissioned by Scientific
Data-Link to extend the ACM's {\em Computing Reviews} taxonomy for the
sub-specialty of artificial intelligence (AI) [REF331] . Waltz's extension is extremely
refined and helpful to AI practitioners. At the same time, it is even
more {\em ad hoc}, its ``sponsoring institution'' has less impact, and
consequently it is even less well accepted within libraries.
All three of
these indexing systems are examples of TOP-DOWN KNOWLEDGE
STRUCTURES. That is, they are developed by various social
institutions as prescriptive languages used to represent the consensus
opinion as to how information is to be organized. Such ``consensual''
knowledge structures are critical if individuals are to {\em share}
information. Each indexing system represents a compromise between
increased scope and diminished resolution. Increased scope brings along
with it broader acceptance and adherence. These advantages are bought at
the expense of acceptance by users actively involved in technical
specialties.
The central role of hierarchic BT/NT relations in organizing
keyword vocabularies should make us especially concerned with a precise
semantics for this relationship. Most would agree that if {A} is a
broader-term than {\tt B}, then {\tt B} ``is a'' {\tt A}. But as
knowledge engineers within AI have known for a long time, the ubiquitous
{\tt IS\_A} relation admits a number of interpretations which can
support much different types of inference [REF55] [REF97] . In general, the BT/NT relation
seems to correspond most closely to the ``a kind of'' implication
relating predicates {\tt A} and {\tt B}: (\forall x) B(x) \rightarrow
A(x)
Earlier generations of Internet users participated in the
construciton of the extensive UseNet
hierarchy of discussion boards, on topics from {\tt
ALT.SEX.FETISH.ROBOTS} to {\tt COMP.SYS.MAC.OOP.TCL}; see Section §7.4.5 for an example of the use of this
hierarchy in text classification tasks. These days the most widely know
taxonomies are WWW DIRECTORIES such as Yahoo! , where employees of this company
have constructed a hierarchy, primarily of places to spend money. One of
the most exciting recent developments is the development of
\defn{collaborative classification} efforts such as the Open Directory Project (DMOZ)
which involve large communities of experts, each working on making sense
within their own are of expertise.
Top of Page
Taxonomies