Pinker was among those who argued that predicting the next word is not how language works.

Judge as of 2023. True: dated statements by Pinker from before the claim (books, articles, talks or interviews) argue that language is not produced or understood mainly by predicting the next word from statistical word sequences (e.g. critiques of associationist, Markov or n-gram models, or of connectionist/LLM-style accounts of language), so he was among those making that argument. Mostly true: he clearly made the argument, but in a slightly different form (e.g. against word-chain or associative models generally rather than 'next-word prediction' in those words). Mixed: his earlier writing is ambiguous, criticising statistical models only in part while also accepting prediction as central. Mostly false: his past writing only touches on the issue in passing or makes a narrower point that the claim overstates. False: his record shows he had endorsed next-word prediction as how language works, or shows no such argument while showing his positions on the topic to differ.

Set before the research began.

1. E1 Words and rules stevenpinker.com (Steven Pinker's website) · 3 December 2021 Own record

   “A comparison of English to other languages shows that contrary to the connectionist account, default suffixation is not due to numerous regular words reinforcing a pattern in associative memory, but to a memory-independent, symbol-concatenating mental operation.”

2. E2 Will ChatGPT replace human writers? Pinker weighs in. Harvard Gazette (Harvard University) · 14 February 2023 Own record

   “We don’t need to be exposed to half a trillion words of text (which, at three words a second, eight hours a day, would take 15,000 years) in order to speak or to solve problems.”

3. E3 Steven Pinker - Wikipedia Wikipedia · 14 March 2002

   “Pinker argued that language depends on two things: the associative remembering of sounds and their meanings in words, and the use of rules to manipulate symbols for grammar.”

4. E4 Pinker/Chomsky Q&A from MIT150 Panel Language Log (University of Pennsylvania Linguistic Data Consortium blog) · 15 September 2026

   “uh probabilities don't play a big role.”

Pinker's published Words and Rules abstract (E1) argues against a connectionist, associative-memory account of inflection in favour of a symbol-concatenating rule mechanism. Wikipedia (E3) independently summarises his view that grammar depends on rules manipulating symbols rather than association alone. A panel transcript (E4) records him saying probabilities don't play a big role, though the event is undated. This shows he argued against associative/statistical models of language, but the evidence frames it as a critique of connectionist pattern association rather than next-word prediction specifically, which fits mostly true.