Feature Engineering and Post-Processing for Temporal Expression Recognition Using Conditional Random Fields

We present the results of feature engineering and post-processing experiments conducted on a temporal expression recognition task. The former explores the use of different kinds of tagging schemes and of exploiting a list of core temporal expressions during training. The latter is concerned with the use of this list for post-processing the output of a system based on conditional random fields. We find that the incorporation of knowledge sources both for training and post-processing improves recall, while the use of extended tagging schemes may help to offset the (mildly) negative impact on precision. Each of these approaches addresses a different aspect of the overall recognition performance. Taken separately, the impact on the overall performance is low, but by combining the approaches we achieve both high precision and high recall scores.