Semantic index assignment

Conventional performance criteria for communication networks do not take into account the semantics of the data to be communicated. For example, (word) error rates treat errors between semantically similar words (car and automobile) and semantically distant words (car and computer) equally. In reality, the meaning of the message is distorted much less when automobile is recovered instead of computer when the intended message is car. In order to correctly address the performance of a semantic system, a new performance criterion is necessary that takes into account the semantic similarities between recovered words. We study in this paper the index assignment problem with a source that produces semantic messages to develop a better understanding of how their meanings affect the semantic error performance in a noisy communication network, and in particular for networks with queries. To this end, we utilize the semantic distances based on lexical taxonomies as a distortion measure in a communication system. Our findings indicate the need for development of semantics-aware physical systems that allow for better integration of human factors and intelligence within complex systems design.