Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data

9 Some possible counterarguments

1. How did Bender and Koller answer “there is so much form out there, surely that is enough”?
   Quantity doesn’t matter to the principle: the octopus may watch forever. People keep generating new intents about evolving worlds, so producing apt responses without meaning would require memorizing infinitely many stimulus–response pairs.
   • Bender & Koller, 2020, Climbing towards NLU (Proceedings of the 58th Annual Meeting of ACL), §9, pp. 5192–5193
2. How did Bender and Koller answer “BERT improves meaning-related tasks, so it must have learned meaning”?
   It has probably learned something about meaning, as syntax and semantic similarity capture something about it: a useful but incomplete reflection of meaning in form. LMs learn that reflection, not meaning itself.
   • Bender & Koller, 2020, Climbing towards NLU (Proceedings of the 58th Annual Meeting of ACL), §9, pp. 5192–5193