The neuron lights up on uncommon, content-bearing tokens—especially proper names, acronyms, step/figure/section numbers, and technical terms—i.e. the “important” or domain-specific words in a passage.