Rendering Morphosyntactic Features of Legal Spanish Judgments Using Neural and Statistical Machine Translation

Abstract

Addressing machine translation (MT) in the legal context, this chapter compares Spanish-to-English neural MT (NMT) and statistical MT (SMT) output from the same MT provider during these different paradigm periods. The chapter focuses on translation renditions of various morphosyntactic features originating from a morpho-syntactically complex text of judgment summaries issued by the Supreme Court of Spain. One the one hand, the chapter evaluates NMT and SMT translation solutions of frequent morphosyntactic features revealed by means of corpus analysis software in the areas of verb and subject order and active and passive voice. On the other, a set of complex sentences is analysed to reveal how NMT and SMT translations of these and other morphosyntactic features might fare under more dispersed or broader contextual conditions. NMT appears to provide more adequate solutions in the case of most of the individual morphosyntactic features analysed, as well as a tendency to be more consistently reliable from a grammatical equivalence perspective. SMT, however, may provide more peculiar or contextually desirable solutions in certain cases, but these solutions cannot be relied upon as much as the syntactically thorough or grammatically equivalent approaches provided by NMT on a more consistent basis.