Multiple-Choice Synonym and Analogy Problems

Title: Multiple-Choice Synonym and Analogy Problems

Research Question: Can different merging rules for combining probability distributions improve the accuracy of solutions to multiple-choice synonym and analogy problems?

Methodology: The researchers developed three merging rules - mixture rule, logarithmic rule, and a novel product rule - to combine the output of separate modules that produce probability distributions for multiple-choice synonym and analogy problems. They applied these rules to two problems commonly used to assess human mastery of lexical semantics: synonym questions and analogy questions. They set parameters for the merging rules using a maximum likelihood approach on a collection of training instances.

Results: All three merging rules resulted in ensembles that were more accurate than any of their component modules. The differences among the three rules were not statistically significant, but it was suggestive that the popular mixture rule was not the best rule for either of the two problems.

Implications: The study suggests that ensemble methods, which combine the output of successful, separately developed modules, can improve the accuracy of solutions to multiple-choice synonym and analogy problems. The results also indicate that the popular mixture rule, while commonly used, may not be the best choice for these types of problems. The product rule, introduced in this study, could potentially offer better results.

Link to Article: https://arxiv.org/abs/0309035v1 Authors: arXiv ID: 0309035v1