Automated Extraction of Swedish Neologisms using a Temporally

This thesis presents an automated system for extracting neologisms using machine learning approaches. The neologisms are extracted from a large temporally annotated corpus containing newspaper articles and blog posts. We find that our system is different from much of the previous research on neologism extraction and justify these differences by relating it to current research in evolutionary linguistics. Our main contribution is a system which can incorporate a larger amount of features than any previous system. Our approach also enables multiple annotators to provide feedback to the system and affect the way the system ranks words in regard to “newness”. In addition, our system is capable of assisting users manually extracting neologisms by providing statistics regarding word usage over time. We also present and analyze words annotated by twelve anonymous annotators in regard to their “newness” and are able to draw conclusions on how language users perceive neologisms.