Semantic relations are a much debated part of the semantic analysis of texts. Entities (actions, objects, attributes, etc) interact in certain ways, and the description of their interaction must be captured somehow in the representation of the meaning of the text units that encompas them (words, phrases, clauses, multi-clause sentences and larger text units).

In this presentation we will first set the background for two experiments involving the detection of semantic relations. We will show what semantic relations are for us, how we approach them and on what levels. We will then present methods of automatically assigning semantic relations in base noun phrases and between morphologically related words. We use different machine learning tools to select from a set of possible indicators, established by manual analysis, the ones that are actually useful in determining the type of interaction between pairs of entities.

These experiments have given us interesting insight into the nature of semantic relations and into the usefulness of some popular lexical resources and machine learning techniques.