Applying Deep Language Understanding to Open Text: Lessons Learned

AbstractHuman-level natural language understanding (NLU) of open text is far beyond the current state of the art. In practice, if deep NLU is attempted at all, it is within narrow domains. We report a program of R&D on cognitively modeled NLU that works toward depth and breadth of processing simultaneously. The current contribution describes lessons learned – scientifically and methodologically – from an exercise in applying deep NLU to open-domain texts. An overarching lesson was that although learning to compute sentence-level semantics seems like a natural step toward computing full, context-sensitive, semantic and pragmatic meaning, corpus evidence underscores just how infrequently semantics can be cleanly separated from pragmatics. We conclude that a more comprehensive methodology for automatic example selection and result validation is needed as prerequisite for success in developing NLU applications operating on open text.

Return to previous page