Stylistic Experiments for Information Retrieval

Information retrieval systems are built to handle texts as topical items: texts are tabulated by occurrence frequencies of content words in them, under the assumption that text topic is reasonably well modeled by content word occurrence. But texts have several interesting characteristics beyond topic. The experiments described in this text investigate {\em stylistic variation}. Roughly put, style is the difference between two ways of saying the same thing --- and systematic stylistic variation can be used to characterize the {\em genre} of documents. These experiments investigate if stylistic information is distinguishable using simple language engineering methods, and if in that case this type of information can be used to improve information retrieval systems. A first set of experiments shows that simple measures of stylistic variation can be used to distinguish genres from each other quite adequately; how well depends on what the genres in question are. A second set of experiments evaluates the utility of stylistic measures for the purposes of information retrieval, to identify common characteristics of relevant and non-relevant documents. The conclusion is that the requests for information as typically expressed to retrieval systems are too terse and inspecific for non-topical information to improve retrieval results. Systems for information access need to be designed from the beginning to handle richer information about the texts and documents at hand: information about stylistic variation cannot easily be added to an existing system. A third set of experiments explores how an interactive system can be designed to incorporate stylistic information in the interface between user and system. These experiments resulted in the design an interface for categorizing retrieval results by genre, and displaying the retrieval results using this categorization. This interface is integrated into a prototype for retrieving information from the World Wide Web.