Language for Granted - Automatic Evalutation of the Language Used in Grant Applications
Peeter Sällström Randsalu · 2007
Abstract This thesis examines whether it is possible to ascertain a measurable difference between granted and refused grant applications with automatic methods. A corpus of project descriptions from the Swedish Science Council was examined with different classification techniques and using different linguistic features. A Bayes classifier is shown to be a good predictor for this type of problem and the number of prepositional phrases in a document is shown to be a good attribute for classification. The results show that there does exist a statistically measurable linguistic difference between granted and refused applications. Sammanfattning Denna uppsats undersoker om det gar att, med automatiska metoder, mata en skillnad mellan godkanda och avslagna bidragsansokningar. En korpus med projektbeskrivningar fran Vetenskapsradet undersoktes med olika klassifikationstekniker for olika lingvistiska sardrag. Det visar sig att en Naive Bayes-klassificerare fungerar bra for denna slags problem och ocksa att antalet prepositionsfraser i ett dokument ar en bra utgangspunkt for klassificering. Resultaten visar slutligen att det finns en statistiskt matbar spraklig skillnad mellan godkanda och avslagna ansokningar.