Topic Maps are means for representing sophisticated, conceptual indexes of any information collection for the purpose of semantic information integration. To properly fulfil this purpose, the generation of Topic Maps has to base on a solid theory. This paper proposes the Observation Principle as the theoretical fundament of a future scientific discipline Topic Maps Engineering. SemanticTalk generates sophisticated, conceptual indexes of speech streams in real-time. Reflecting the Observation Principle, this paper describes how these indexes are created, how they are represented as Topic Maps and how they can be used for semantic information integration purposes.