One of the reasons it took a long time to get computers to understand human speech is that parsing syntax and looking up words in the dictionary gave bizarre interpretations, as the literal meaning is often not what people mean.

Judge as of 2026-10-11 against the history of natural language processing and speech/language understanding. TRUE: historical and technical accounts show that computer understanding of human language took a long time to achieve, and that rule-based approaches (syntactic parsing plus dictionary/lexical lookup) were a major obstacle because they produced odd or wrong interpretations, largely because people often mean something other than the literal meaning (ambiguity, idiom, implicature, pragmatics). MOSTLY TRUE: the account is broadly right but the evidence assigns the delay mainly to a related cause the claim leaves out (e.g. acoustic speech recognition, lack of data or computing power), with the literal-versus-intended-meaning problem a real but secondary factor. MIXED: the sources give roughly equal weight to the stated explanation and to unrelated causes. MOSTLY FALSE: the long delay is real, but the sources show that literal-interpretation problems played only a minor role. FALSE: the sources show that parsing and lexical lookup did not produce such misreadings, or that the slow progress had nothing to do with them.

Set before the research began.

1. E1 Views of Text-Meaning in Computational Linguistics: Past, present, and future Graeme Hirst, Department of Computer Science, University of Toronto

   “The computer’s job was taken to be figuring out what it is that the user “really wants” from the “literal meaning” of what they say; for example, I’d like a beer, said to a domestic robot, means Bring me a beer, and do it right now.”

2. E2 [PDF] A demonstration of the nonfeasibility of fully automatic high quality translation Yehoshua Bar-Hillel (reprinted in ACL Anthology / MT Archive; originally Advances in Computers, vol. 1, 1960) · 1960

   “(Notice, e.g., that the very same— fictitious!—thesaurus approach for English-to-French translation that would correctly render pen by "plume" in the sentence The pen was in the inkstand would incorrectly render pen by "plume" in the sentence The inkstand was in the pen.)”

3. E3 COMPUTATIONAL LINGUISTICS (chapter 32) Graeme Hirst, Oxford University Press handbook chapter (proof)

   “Rather, what was rejected was the idea that deep analysis, especially formal semantics, and complex representations of knowledge of the world are a necessary or desirable part of the solution of any practical problem in language processing.”

4. E4 MINDS 2006-2007 Report of the Speech Understanding Working Group: Historical Development and Future Directions in Speech Recognition and Understanding NIST / U.S. Government MINDS workshop series (authors: Baker, Deng, Khudanpur, Lee, Glass, Morgan)

   “A primary reason language models in current ASR systems are not portable across tasks even within the same language or dialect is that they lack linguistic sophistication – they cannot consistently distinguish meaningful sentences from meaningless ones, nor grammatical from ungrammatical ones.”

E1 and E2 show that inferring intended meaning was treated as a core task and that a lookup approach could give wrong readings. But no evidence item says why computer language understanding took so long or how much each cause mattered, and E3 and E4 do not point to a different main cause, so the criteria cannot be applied.