AI-based language models like ChatGPT have recently astonished with impressive results on
computational Natural Language Processing tasks. However, the prevailing models only possess limited
capability of incorporating context by only leveraging contextual information contained in the
processed text itself, whilst neglecting external context like the current situation. Such contextual
information, however, is key for correctly interpreting natural communication. For example, how would
we interpret the tweet Im expecting that Erika will cause major trouble...? Without any further
information, we will probably assume that the entity Erika refers to a person of that name. But what
if we now read on the news ticker that cyclone Erika is expected to cause major damage in our area
tonight? Based on this situational context, we will thus adapt our interpretation accordingly, now
assuming that the tweet is rather referring to this anticipated cyclone event.
The research project SITCON (Situational Context Representations) thus aims to investigate the
question of how to systematically leverage such situational context within computational information
fusion systems. This necessitates developing solutions for modeling such external, dynamic situations,
and exploiting this contextual information for informing the information processing pipeline
accordingly. To practically examine this problem, the application domain of crisis computing, seeking
to leverage state-of-the-art Information Technology and Communication tools for improving crisis
management, is foreseen as tangible evaluation testbed. Concretely, these evaluations aim to study
how the automated, situative interpretation of thousands of social media messages (like eye-witness
observations) in real-time can support emergency responders situation awareness during time-critical
crisis situations.