Pragmatic Quality Assessment for Automatically Extracted Data

Automatically extracted data is rarely “clean” with respect to pragmatic (real-world) constraints—which thus hinders applications that depend on quality data. We proffer a solution to detecting pragmatic constraint violations that works via a declarative and semantically enabled constraint-violation checker. In conjunction with an ensemble of automated information extractors, the implemented prototype checks both hard and soft constraints—respectively those that are satisfied or not and those that are satisfied probabilistically with respect to a threshold. An experimental evaluation shows that the constraint checker identifies semantic errors with high precision and recall and that pragmatic error identification can improve results.