Using contextual information for machine translation evaluation