Multi-Label Transfer Learning for Multi-Relational Semantic Similarity

The semantic relations between two short texts can be defined in multiple ways. Yet, all the systems to date designed to capture such relations target one relation at a time. We propose a novel multi-label transfer learning approach to jointly learn the information provided by the multiple annotations, rather than treating them as separate tasks. Not only does this approach outperform the traditional multi-task learning approach, it also achieves state-of-the-art performance on the SICK Entailment task and all but one dimensions of the Human Activity Phrase dataset.