Paper: Post-hoc Manipulations of Vector Space Models with Application to Semantic Role Labeling

ACL ID W14-1501
Title Post-hoc Manipulations of Vector Space Models with Application to Semantic Role Labeling
Venue Continuous Vector Space Models and their Compositionality
Session
Year 2014
Authors

In this paper, we introduce several vector space manipulation methods that are ap- plied to trained vector space models in a post-hoc fashion, and present an applica- tion of these techniques in semantic role labeling for Finnish and English. Specifi- cally, we show that the vectors can be cir- cularly shifted to encode syntactic infor- mation and subsequently averaged to pro- duce representations of predicate senses and arguments. Further, we show that it is possible to effectively learn a linear trans- formation between the vector representa- tions of predicates and their arguments, within the same vector space.