The group, which also runs the online version of the Merriam-Webster dictionary, is alleging that OpenAI misused its reference materials to train AI models.