Living with Machines is a 'radically collaborative' project at the intersection of history and data science, in partnership with a national library, data science / AI institute and universities. Crowdsourcing has a dual function within the project. It produces tangible outcomes such as data that feeds into linguistic and other research questions or can be used to train machine learning algorithms, and provides less tangible public participation outputs including new and enhanced relationships, skills and experience for participants, researchers working with cultural heritage collections. Positioned as a platform for public engagement with the research questions and primary sources that underpin the project, the crowdsourcing tasks are designed to expose some of the processes of data science and digital history to participants, while the project benefits from their questions and insights as they contribute. The tasks designed for the project must both be enjoyable and meaningful enough to attract participants, while also delivering data to the quality required for the computational linguistic processes it supports. While voluntary crowdsourcing and computational linguistics are well-developed fields with established methods and epistemologies, they typically have different timelines and ways of developing datasets from source records. Working within the affordances of Zooniverse, the platform chosen for these tasks, adds further complexity. This paper discusses how the team managed to align goals and methods to produce outcomes that met the needs of both disciplines, despite these constraints. It considers how well the outcomes met various metrics for success, and reflects on lessons learnt.