Markov Logic Networks for Spoken Language Interpretation

In this paper, the use of Markov Logic Networks (MLN) is considered for application in spoken dialogue systems. In spoken dialogues information that can be represented in logical form is often not explicitly expressed, but has to be inferred from detected concepts. Often, it is necessary to perform inferences in presence of incomplete premises and to get results with an associated probability. MLNs can be used for this purpose even in cases in which other known techniques llike CRF or Bayesian Networks cannot be easily applied. Results on the inference of user goals from partially available information are presented using the annotated French Media corpus.