The current generation of sentiment analysis
systems is limited in their real-world applicability because they
cannot detect utterances that implicitly carry positive or negative
sentiment. We present early stage research ideas to address this
inability with the development of a dynamic triple store of events
associated with their prototypical sentiment.