Human-centered, augmented machine translation: analysing user experience, quality and productivity in interactive post-editing vs traditional post-editing

Recent language technology developments have disrupted the translation profession. Traditionally, the research and industry focus has been on using more computational power and training larger language models, often neglecting the users of such technology. To date, the goal of technology development has been the creation of an intelligent agent that emulates human behaviour to increase automation. As a response, a novel technology design framework has gained a foothold recently: human-centered artificial intelligence (HCAI), where, instead of human replacement, the aim is to align the development of tools with user values, preferences, and needs, at the same time as enhancing human capabilities and performance. If applied to machine translation (MT), we can talk about human-centered, augmented MT (HCAMT). This shift, moving from emulation to empowerment, places humans at the center of AI/language technology. This paper considers the analysis of machine translation user experience (MTUX) to be a way to foster HCAMT. To demonstrate this, we conduct a longitudinal user study with 11 professional translators in the English-Spanish language combination to analyse the effects of traditional post-editing (TPE) and interactive post-editing (IPE) on MTUX, translation quality, and productivity. MTUX results suggest that translators prefer IPE to TPE because they feel more in control in this newer form of translator-computer interaction and more empowered in their work with MT. Productivity results suggest that translators working with IPE achieve statistically significantly higher productivity than when working with TPE. Quality results indicate that translators offer more fluent translations in IPE, and equally adequate translations in both post-editing modalities. All these results allow for reflection on the potential adoption of IPE as a more human-centred MT modality for post-editing, one that empowers users, who have been increasingly reluctant to interact with MT post-editing in industry workflows. The paper also lays the basis for exploring HCAMT with MT users beyond professional translators, opening the door to more inclusive and diverse user-centred MT research.