In D’Alessandro & Van Oostendorp (2020), we proposed Magnetic Grammar, a model of phonological competence in which a language’s segment inventory is characterised entirely by features that attract or reject other features within a segment. The paper referred to a Python implementation demonstrating the learning algorithm. This squib reports on a bug in that implementation and presents a corrected version, which leads to a small but theoretically meaningful refinement of the learning algorithm.
The corrected implementation, with Jupyter notebook in which one can learn about and test the updated theory, is available as an open-source repository at https://github.com/fonolog/MagneticGrammar. A short explanation is on LingBuzz. The original paper is here.