mathnlp

Does representing all of mathematics with one operator, eml(x,y)=exp(x)−ln(y), help GNNs learn symbolic reasoning? A preregistered study measured honestly at 250k-expression scale — nulls and one published gate fail included.

Typed text-to-AST parsing, symbolic verification, and selective prediction for executable quantitative clinical constraints.