One tool fits all? The role of tool type and language proficiency in translation tool-assisted second language writing processes

When we write texts in a language that is not our first, we often feel uncertain about how to express certain ideas in that language. In these moments, many writers turn to translation tools such as bilingual dictionaries (e.g., WordReference.com) or machine translation (MT) platforms (e.g., Google Translate) for help. Over the past decades, these tools have not only become more diverse but also more automated. For example, while traditional bilingual dictionaries only provide translations for individual words, MT tools can generate translations of entire sentences or even paragraphs at once. At the centre of this dissertation is the question how access to more automated translation tools, such as MT tools, affects writers’ cognitive processes (e.g., the extent to which they cognitively process the translations in the tool output or plan the content of what they will write next). In addition, it explores how the writer’s language proficiency shapes these effects (e.g., whether the effect of the tool on the cognitive processing of the translations differs between beginner or more advanced learners of the language). To address these questions, I collected data from a group of Dutch-speaking learners of Swedish with varying proficiency. Each learner wrote two texts in Swedish with access to an MT tool and another two with access to an online bilingual dictionary. While they wrote, I captured their keystrokes, computer screen, and eye movements on the screen. From this rich dataset, I derived measures that allowed me to make inferences about the effect of the type of tool available to the writer on their cognitive processes and the potential role of the writer’s Swedish language proficiency in this relationship (e.g., how long the writer’s gaze stays fixed on the translations, or how long the writer pauses at locations in the text that are associated with specific cognitive processes). Through statistical analysis, this dissertation shows that learners’ cognitive processes during writing are affected by the type of tool available to them and, by extension, the level of automation that the tool offers. For example, learners entered more words into the MT tool, which suggests that they outsourced more language-related cognitive processes when they had access to the more automated tool. However, their gaze also stayed fixed for longer on the MT translations, which suggests that they were more cognitively engaged when processing the output of the more automated tool. They also paused longer at locations in the text associated with planning processes and looked at text further away from the word they were writing during pauses. This suggests that access to more automated tools makes writers focus less on micro-level, language-related problems and helps them to engage with their writing process at a higher cognitive level. The writer’s proficiency played a limited but noteworthy role in shaping how the type of tool available influenced the writer’s cognitive processes. For example, the effect of MT access on how far away in the text the learners looked during pauses was larger for the less proficient.