Common use of Automatic Evaluation Clause in Contracts

Automatic Evaluation. The systems have been evaluated using a set of state-of-the-art automatic metrics, namely BLEU, NIST, TER and GTM. We report also the size of the vocabulary of the test set and the amount of out of vocabulary (OOV) words. We have a number of systems according to the data used for training and tuning and the sentence aligner used to split this data: • The system v0 is trained and tuned on Europarl, considered to be a general-domain corpus. This is the baseline system to which we will compare our domain-specific systems. • Systems v1 are trained on the union of Europarl and the domain-specific data and tuned on domain-specific data. • Finally, systems v2 are trained and tuned on domain-specific data only. Regarding the sentence aligner used, we have two sets of systems. Europarl sentence splitter is used for systems Europarl while SSplit is used for systems Ss. Evaluation results are shown in Tab. 5-4.