The language brain matters more for programming than the math brain? (2020)

Original Article ↗ Hacker News Discussion ↗

Study scope, metrics, and skepticism

  • The underlying experiment was small (42 enrolled, 36 completed) and short (≈7.5 hours of Codecademy Python).
  • “Math” in the paper was operationalized as numeracy (everyday arithmetic/probability), not higher math (algebra, discrete math, logic, etc.).
  • Reported variance in learning outcomes: general fluid reasoning + working memory ≈34%, language aptitude ≈17%, EEG beta/gamma ≈10%, numeracy ≈2%.
  • Raw correlations for numeracy vs language aptitude were actually similar; a stepwise regression then assigns almost all unique variance to language, which several commenters call statistically fragile / p-hacking.
  • Tasks partly tested speed at progressing through English-language lessons, so better readers may simply finish more material, inflating “language” effects.
  • Many argue the article’s headline (“language brain vs math brain”) overstates and misrepresents what the paper actually shows.

What counts as “math” and how it relates to programming

  • Long argument over definitions:
    • Some treat programming as a direct instance of mathematics (formal logic, functions, relations, set theory, lambda calculus, PL semantics).
    • Others reserve “math” for explicit calculation and formal proof, and see most industrial coding as engineering plus communication with only light math.
  • Several note that relational databases, SQL, graphs, state machines, types, etc. are explicitly mathematical whether or not programmers perceive them that way.
  • A recurring complaint: schooling equates math with arithmetic, while real math is about abstraction, structure, and proof—much closer to good program design.

Language, humanities, and real-world programming

  • Many experienced developers report strong verbal skills (high verbal test scores, love of reading, foreign languages, writing) but mediocre traditional math, yet long successful programming careers.
  • Fast reading, clear naming, documentation, and the ability to rephrase and explain problems are described as central day‑to‑day skills.
  • Several anecdotes: GRE verbal better predicted CS PhD success than GRE math once math scores saturated; early English or Latin strength correlating with good programming; English majors historically feeding into software.
  • Some push back with stereotypes about “humanities coders” producing poor code, but others call this bias and selection effect.

Math-heavy CS vs everyday software engineering

  • Distinction repeatedly drawn between:
    • Theoretical CS / ML / graphics / crypto, which genuinely demand advanced math.
    • Typical application/web/business software, which rarely needs more than logic, basic algebra, and perhaps some probability.
  • Many say university CS math requirements were a barrier and rarely used later, while others insist those abstract courses sharpen the kind of reasoning that underlies both good algorithms and good program structure.

Other themes

  • Debate over whether recursion is “fundamentally linguistic” vs a general formal concept.
  • Observations that LLMs can write plausible code yet still struggle with deeper reasoning are used both to argue for and against a “language-centered” view of programming.
  • Several commenters suggest the real drivers are general problem‑solving ability, abstraction, and working memory, with language and math as overlapping ways to exercise those capacities.