Contextual Computing for Natural Language Processing
Dissertation von Robert Porzel (2010)
As one knows from personal experience, human-human communication works extremely well despite all of the challenges presented above, i.e. we can understand each other despite all of the ambiguities and underspecifications present in our utterances even at noisy cocktail parties. The amazing robustness of human-human communication is - at least in part - a result of our contextawareness and our corresponding pragmatic knowledge, both of which enable us to disambiguate and decontextualize our interlocutor’s utterances robustly even under noisy conditions. The work presented herein builds upon the recognition of the fact that computational approaches to any of the three aforementioned challenges can benefit from the inclusion of contextual information, real world knowledge and correspondingly reified contextual knowledge in order to recover the user’s intent from a given conversational input. The main aim, therefore, is to present a formal approach for explicating contextual information and pragmatic knowledge, that can be applied, employed and evaluated in natural language understanding systems.