The neuron isn’t really picking out any semantic content at all but is firing on tokens that occur at a specific periodic position in the document (i.e. it’s acting as a position‐detector rather than detecting any particular word or topic).