Automatic translation and corpus analysis in the generation of academic abstracts

Abstract

Machine translation (MT) has made huge strides in the last decades and it is increasingly applied in the ESL classroom, both for language learning and translation purposes. This paper wishes to explore whether and to what extent automatic translations performed by an MT platform and a Language Model (LM) tool can be effectively
integrated and/or post-edited by corpus evidence. The language pair considered for the analysis is Italian/English, and the source text is an abstract focusing on academic Italian as an L2. The corpus consulted is the ARC (Anthology Reference Corpus), available on Sketch Engine. In this case, the Italian abstract is accompanied by an official translation into English. Therefore, this paper compares MT- and LM-driven output with the official target text. Additionally, it investigates to what extent corpus consultation can be integrated into the post-editing process to produce a qualitatively acceptable text in the target language. The paper’s findings indicate the high reliability of corpus-driven post-editing, revealing alternative translation options. It also brings collocations to the fore, thereby foregrounding language patterns. In addition, corpus analysis helps address MT- and LM-driven shortcomings. The paper discusses how corpus-driven post-editing can be seamlessly incorporated into the translation process and into translation and language education.