What got lost in online machine translations? Effects on Aspect and Passivization from a literary corpus

Résumé En

Aspect is a crucial ingredient of temporality, which reflects the speaker's internal perspective on a given situation (Ayoun 2013). This article compares human translations and online machine translations (MT) with the purpose of describing how the distinction between perfective and imperfective aspect is realized in Spanish when translating from English. Our results come from 1.6 million-word Social Sciences corpus from which 82 perfective get-passive constructions (e.g., I got saved) were elicited. Although the general pattern is for atelic predicates (states and activities) to occur with the imperfect tense and for telic predicates (accomplishments and achievements) to occur with the preterite, in Spanish all the aspectual predicates can be expressed with preterite and imperfect, depending on what the speaker wants to convey (Montrul & Slabakova 2003: 357). The results show that Spanish imperfect readings were largely preserved in human translations compared to online machine translations. This study discusses the machines' overextension of the preterite as a systemic regularity that bring MT's outputs close to L2 learners' grammars.