In the past I've approached this in different ways, either collaborating on a #psycholinguistic experiment on #ReferringExpression production (https://aclanthology.org/W17-3522/) or #crowdsourcing variations on an existing #corpus (https://davehowcroft.com/publication/2017-08_interspeech_extended-sparky-restaurant-corpus/).

At the moment, I'm working to collect #dialogue/s grounded in a given information source (a description of a museum exhibit) to develop a dataset for training #ScottishGaelic / #Gàidhlig #chatbots for #QuestionAnswering (https://blogs.ed.ac.uk/garg/2022/08/23/scottish-gaelic-chatbots-for-museum-exhibits/).