Semantic parsing as an energy minimization problem

Language understanding is not just a matter of knowing the language but, to a considerable degree, it is a matter of logical inference using world knowledge. A pure linguistic approach cannot interpret any natural language utterances; it can only constrain their interpretation, and the rest must be left to semantic parsing. To discriminate sense in any semantic parsing, a reader should consider a diversity of information, including syntactic tags and restriction, word frequencies, collocations, semantic context and role-related expectations. However, current approaches make use of only small subsets of this information. In this paper, we show how semantic parsing can be formulated as a sequence of processes in which multiple sources of knowledge are incorporated. A resolution-based inference procedure is shown to determine semantic meanings from the analyzed utterances. This is done on the basis that language technology has to employ test-efficient solutions with respect to the phenomena occurring in real-world language understanding