Normalizing Chinese temporal expressions with multi-label classification

Temporal expression contains crucial temporal information in texts. Understanding its temporal semantics is important in many NLP applications, such as information extraction, document summarization and question answering. Temporal expression normalization involves mapping from the different classes of expressions to the values of certain temporal attributes. A temporal expression may belong to one or more classes, but each class of expressions normally shares the same mapping procedure. In this paper, we explore the possibility of applying multi-label classification techniques in the context of temporal expression normalization. More specifically, two models, named independent binary classification model and compared binary classification model, are evaluated, compared and analyzed. Once the possible class(es) of a temporal expression is determined, the corresponding mapping rules are called to transform it into the corresponding attribute(s). Experiments on a substantiate data collection show that, based on the result of machine learning classification, the performance of temporal expression normalization is comparable with that of deliberate rule set.