Background: Forums in massive open online courses (MOOCs) enable written exchanges on course content; hence, they can potentially facilitate learners' cognitive engagement. Given the myriad of MOOC forum messages, this engagement is commonly analysed automatically through the linguistic features of the messages. Assessing linguistic features of learners' forum messages involves consideration of the learning tasks. MOOC forum discussion tasks, however, have not been previously considered. Objective and Method: This study explores the effects of MOOC forum discussion tasks on learners' cognitive engagement. Based on the structure of observed learning outcomes (SOLO) taxonomy, we manually annotate distinct levels of cognitive engagement encouraged in forum discussion tasks and displayed by learners in messages starting discussions (i.e., thread starters). We study the linguistic features of thread starters in relation to the pedagogical design of the discussion tasks. Additionally, we use random-forest modelling to identify the linguistic and task-related features that help to categorise learners' cognitive engagement according to SOLO levels. Results: Manual analysis showed that learners' thread starters mainly reflect surface SOLO levels and include few academic words and cohesive language. Random-forest modelling showed that these linguistic features, together with the SOLO levels encouraged in the discussion tasks, played an important role in identification of learners' cognitive engagement. Major Takeaways: Our results highlight the importance of the pedagogical design of MOOC forum tasks in helping learners engage cognitively. Our study also contributes to the empirical evidence that learners' linguistic choices can afford insights into the quality of their cognitive engagement.