Abstract
The current endeavour focuses on the notion of positive versus negative polarity preferences of verbs for their direct objects. We observed verbs with a relatively clear positive or negative polarity preference (called polar), as well as cases of verbs where positive and negative polarity preference is balanced (called bi-polar). These polarity preferences of verbs are induced on the basis of a large dependency-parsed corpus by means of statistical measures and a lexicon of manually curated prior noun polarities. Given (learned) polar verbs, the contextual polarity of their direct objects can be derived. We reached a lower bound of 81.97% and an upper bound of 93.34% precision in these experiments. The polarity of a noun was predicted by the majority vote of the verbs that take that noun as its direct object in our corpus. In a second experimental setting,1 we also considered the role of neutral nouns co-occurring with these verbs. We found that the induction of the (tripartite) prior polarity of nouns can be achieved with a precision of 75.97%.
Klenner, Manfred