A Minimum Error Weighting Combination Strategy for Chinese Semantic Role Labeling

Many Semantic Role Labeling (SRL) combination strategies have been proposed and tested on English SRL task. But little is known about how much Chinese SRL can benefit from system combination. And existing combination strategies trust each individual system’s output with the same confidence when merging them into a pool of candidates. In our approach, we assign different weights to different system outputs, and add a weighted merging stage to the conventional SRL combination architecture. We also propose a method to obtain an appropriate weight for each system’s output by minimizing some error function on the development set. We have evaluated our strategy on Chinese Proposition Bank data set. With our minimum error weighting strategy, the F1 score of the combined result achieves 80.45%, which is 1.12% higher than baseline combination method’s result, and 4.90% higher than the best individual system’s result.