Can machine translate dialogue acts: evidence from translating dialogues from English to Arabic

Abstract

Machine translation advances translation quality at morphological, syntactic, and semantic levels. The pragmatic level of machine translation is also evolving, but challenges remain due to cultural and contextual issues on the one hand and machine translation deficiencies on the other. While computational studies have made strides in automating translation tasks, linguistic-oriented research in this area remains sparse. In response to this gap, this study seeks to assess the effectiveness of Neural Machine Translation, as exemplified by Google Translate, in translating dialogue acts inherent in natural English conversations into Arabic, drawing upon Austin's theory of speech acts and leveraging a corpus of authentic sources1. Our findings highlight certain challenges in the machine’s identification of the performative functions of the utterances in conversations, viz. directives, expressives and representatives. Such challenges emanate from specific linguistic features of English conversations (e.g., idiomatic expressions, polysemous words, and deixis) and the lack of contextual information in everyday discourse. These challenges ultimately impede the faithful representation of speakers' intentions in the translated output.