This book concerns non-linguistic knowledge required to perform
computational natural language understanding (NLU). The main objective
of the book is to show that inference-based NLU has the potential for
practical large scale applications. First, an introduction to research
areas relevant for NLU is given. We review approaches to linguistic
meaning, explore knowledge resources, describe semantic parsers, and
compare two main forms of inference: deduction and abduction. In the
main part of the book, we propose an integrative knowledge base
combining lexical-semantic, ontological, and distributional knowledge. A
particular attention is payed to ensuring its consistency. We then
design a reasoning procedure able to make use of the large scale
knowledge base. We experiment both with a deduction-based NLU system and
with an abductive reasoner. For evaluation, we use three different NLU
tasks: recognizing textual entailment, semantic role labeling, and
interpretation of noun dependencies.