Alan Cartwright wrote:
>
> On particular problem I have being battling with in making software that
> will automatically search and code documents on the basis of their words
> (which I call content analysis) is the complexity of phrase structures.
> Just take a simple example like attitudes to "tea".
>
> In a text you may get all of the following
>
> I like tea
> I don't like tea
> I really like drinking tea
> I don't really like drinking tea.
> She doesn't like tea.
> I prefer tea.
>
> (I am currently drinking tea!)
These examples show the key problems on computer aided content analysis
(here called autocoding). Ambiguity and negation must be taken into
account, and there is software that can detect negation. Also
interactive coding is appropriate if search patterns are negated or
potentially ambiguous.
>
> A second approach is to try and do it structurally for instance search for
> I (actor) don't (negation) like(preference) drinking (act) tea(object). By
> building dictionaries of each of these categories and feeding these into
> the searches you can dramatically increase the reliability of the searching
> procedure. However with often several hundred alternative words in each
> category this creates computational problems and large databases.
Yes, the social sciences just must make use of the knowledge linguists
already have and implement that, e.g. Wordnet, a thesaurus that contains
also grammatical information on words.
The key principle of computer aided content analysis is that your search
patterns are valid indicators for the category you are looking for. The
more the category occurs, the more important it is, that another
implication of this approach. So that makes computer aided content
analysis useful for only a limited number of research designs, but e.g.
in mass communication agenda setting or news factors can be more or less
operationalised by single words. If these search patterns are
represented by their pronouns, of course these are not counted (yet),
another problem that has to be solved.
Harald
------------
Dr. Harald Klein
Social Science Consulting
Königseer Str. 9
98708 Gehren
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|