What Makes Sense? : Searching for Strong WSD Predictors in Croatian
Nikola Bakarić, Jasmina Njavro, Nikola Ljubešić · Repozitorij Filozofskog fakulteta u Zagrebu' at University of Zagreb (University of Zagreb) · 2007
The goal of this research was to investigate and determine position of strong predictors for word sense disambiguation of Croatian nouns.Research was conducted using supervised learning methods and a corpus of around 70 million words.We have concluded that words in the immediate vicinity of an observed lexeme (1-5 words left and right) have the highest discriminative power.We have also measured the applicability and accuracy of the one-senseper-discourse method and found it to be very successful as well as the impact of sentence boundaries which proved not to be a good criterion for selecting strong predictors.