Learning to Translate: A Statistical and Computational Analysis

Table 5

Performance obtained training the regressor on 80% of the data and testing on 20%. This process has been iterated 1,000 times. Experiments have been performed independently on the Europarl and Giga corpus dataset.

Data	𝛼LM	𝛼TM	Estimation Error
Europarl	0.0102±2×10^−4	0.0134±2×10^−4	5.45×10^−5±14×10^−4
GigaCorpus	0.0092±7×10^−5	0.0126±7×10^−5	−5.57×10^−5±2×10^−4