Noise Reduction Experiments in Machine Translation

@inproceedings{Okita2009NoiseRE,
  title={Noise Reduction Experiments in Machine Translation},
  author={Tsuyoshi Okita and Sudip Kumar Naskar and Andy Way},
  year={2009}
}
Noise Reduction Experiments The first experiment of noise reduction (or outlier reduction) [5] is in sentence level. We train our model based on our parallel corpus. Then, we remove all the training data whose distance from the decision plane is +∞ under a given similarity measure. In other words, we decrease the complexity of parallel corpus by selecting sentences in training data since IBM Model 4 seems not expressive enough for a given parallel corpus, i.e. it would need more complex model… CONTINUE READING