Spoken Language Understanding in dialogue systems, using a 2-layer Markov Logic Network: improving semantic accuracy

Ivan Meza-Ruiz and Sebastian Riedel and Oliver Lemon


We describe a two layer Markov Logic Network (MLN) model for the Spoken Language Understanding (SLU) task in dialogue systems. We augment the set of features used in Meza-Ruiz et al. (2008) with the help of off-the-shelf resources. We show that this setup increases the performance of the previous MLN models, which also outperform the state-of-the art .Hidden Vector State. (HVS) model of He and Young 2006. In particular the 2 layer approach produces more accurate sets of slot-values for user utterances (9% improvement).


Paper (PDF)