210 LANGUAGE AND THE INTERNET
Eventhemostbasicsemanticcriteriaaremissingfromtheheavily
frequency-dominated information retrieval techniques currently
used by search engines. All such engines incorporate an element
of encyclopedic classification into their procedure, but this is only
a small part of the answer to the question of how to implement
relevance. Any search-engine assistant needs to supplement its en-
cyclopedic perspective by a semantic one. The typical problem can
be illustrated by the worddepression, which if typed into the search
bo xof a search engine will produce a mi xed bag of hits in which its
senses within psychiatry, geography, and economics are not distin-
guished (nor, of course, less widespread uses, such as in glassware
and literature). The experience of trawling through a load of irrel-
evant hits before finding one which relates to the context of our
enquiry is universal. The solution is obvious: to give the user the
choice of which context to select.^23 The user is asked on screen:
‘Do you meandepression(economics) ordepression(psychiatry)
ordepression(geography).. .?’ Once the choice is made, the soft-
ware then searches for only those hits relevant to the selection. The
procedure sounds simple, but it is not, for the notion of context
has to be formalized and the results incorporated into the software.
But what is the semantic basis of a domain such as economics or
psychiatry, or of any of their relevant sub-domains? Which lexi-
cal items are the ‘key’ ones to be searched for, and how are they
organized? The task goes well beyond scrutinizing the items listed
in a dictionary or thesaurus. These can provide a starting-point,
but the alphabetical organization of a dictionary and the uncon-
trolled conceptual clustering of a thesaurus lack the kind of sharp
semantic focus required. In linguistics, several notions have been
developed to provide such a focus – such as the recognition of
lexemes (as opposed to words), semantic fields, sense relations,
and the componential analysis of lexical meanings.^24 They are not
unproblematic, but they do have considerable potential for ap-
plication in such computer-mediated situations as Web-searching
(^23) This is the procedure used in theALFIE[‘A Lexical Filter Internet Enquirer’] project: see
24 Crystal (1997d).
For semantic fields see Crystal (1997a: 104).