Automatic interpretation of loosely encoded input

摘要

Knowledge-based systems are often brittle when given unanticipated input, i.e. assertions or queries that misalign with the ontology of the knowledge base. We call such misalignments “loose speak”. We found that loose speak occurs frequently in interactions with knowledge-based systems, but with such regularity that it often can be interpreted and corrected algorithmically. We also found that the common types of loose speak, such as metonymy and noun-noun compounds, have a common root cause. We created a Loose-Speak Interpreter and evaluated it with a variety of empirical studies in different domains and tasks. We found that a single, parsimonious algorithm successfully interpreted numerous manifestations of loose speak with an average precision of 98% and an average recall of 90%.