Simplification of Swedish Text by Monolingual Machine Translation Transformation Rules for Simplified Swedish

Sammanfattning: Discussions on the concept of ”Klarspråk” and ”Lättläst” are major topics in Sweden at the moment with efforts towards increasing readability of government organization information in the project ”Begriplig Text”. In the field of natural language processing, a lot of efforts have gone into text simplification. This thesis examines the possibility of using hand crafted transfer and generation rules to increase the readability of text through monolingual machine translation. Transfer and generation rules were developed that handle Swedish grammatical structures including explicitivity and word order. The data used in this thesis consist of manually collected data sets from government organization websites and Uppsala University. The translation results were evaluated using manual evaluation and the automatic readability evaluation methods LIX, OVIX and nominal ratio (NR). Test results show that the developed rules can produce successful transformation on text although not significantly altering the automatic evaluation scores. The results from the manual evaluation show that a higher readability can be achieved with the developed rules although more tests on bigger data sets are needed in order to fully evaluate all rules developed in this thesis.