Proceedings of the Annual Meeting of the Cognitive Science Society

Meaning depends on context. This applies both in obviouscases like deictics or sarcasm as well as more subtle situationslike framing or persuasion. One key characteristic of context isthe identity of the participants in an interaction. Our interpre-tation of an utterance depends on a variety of factors such asour personal history, background knowledge, and our relation-ship to the source. While demographics allow us to capturesome of this variance, the relevance of specific demographicfactors varies across contexts. To address these challenges, weintroduce a method for combining demographics and contextinto situated demographic embeddings—mapping representa-tions onto a continuous space appropriate for the given domain.We further demonstrate how to make use of related externalresources so as to apply this approach in low-resource situa-tions. We show the resulting representations to be interpretableand consider domain-specific similarity. Finally, we show howthese representations can be incorporated to improve modelingof a real-world natural language understanding task.