Smoothing parameter estimation framework for IBM word alignment models
This work addresses data sparsity issues in word alignment for machine translation, offering an incremental improvement by extending existing smoothing techniques to enhance model robustness.
The paper tackles overfitting in IBM word alignment models due to sparse data by proposing a smoothing parameter estimation framework that generalizes additive smoothing, allowing per-word-pair customization and scaling based on data performance, and introduces a smoothened error count criterion that generally yields the best results.
IBM models are very important word alignment models in Machine Translation. Following the Maximum Likelihood Estimation principle to estimate their parameters, the models will easily overfit the training data when the data are sparse. While smoothing is a very popular solution in Language Model, there still lacks studies on smoothing for word alignment. In this paper, we propose a framework which generalizes the notable work Moore [2004] of applying additive smoothing to word alignment models. The framework allows developers to customize the smoothing amount for each pair of word. The added amount will be scaled appropriately by a common factor which reflects how much the framework trusts the adding strategy according to the performance on data. We also carefully examine various performance criteria and propose a smoothened version of the error count, which generally gives the best result.