Symbolic Probabilistic Reasoning for Narratives
Hannaneh Hajishirzi, Erik T. Mueller · 2011
We present a framework to represent and reason about narratives that combines logical and probabilistic representations of commonsense knowledge. Unlike most natural language understanding systems which merely extract facts or semantic roles, our system builds probabilistic representations of the temporal sequence of world states and events implied by a narrative. We use probabilistic transitions to represent ambiguities and uncertainties in the narrative sentences. We present exact and approximate reasoning algorithms that take a representation of a narrative, derive all possible or the most likely interpretations of the narrative, and answer probabilistic queries by marginalizing over these interpretations. In our experiments, we show that our representation together with reasoning algorithms enables semantic understanding of narratives and answering probabilistic questions whose responses are not contained in the narrative. We report our results for two domains of Robocup soccer commentaries and a children story with focus on spatial contexts. To this end, we apply natural language processing (NLP) tools together with statistical approaches over common sense knowledge bases to represent a narrative in our framework. 1