SHRDLU Natural Language Understanding Program Developed by Terry Winograd at MIT
In 1970, Terry Winograd at the Massachusetts Institute of Technology completed SHRDLU, a natural language understanding program that allowed a user to converse in English about a simulated world of coloured blocks, demonstrating that a computer could parse and respond to complex grammatical instructions within a constrained domain.
Background
Most early natural language programs took a brute-force approach to understanding English: they matched patterns of words against a list of templates and returned canned responses. Joseph Weizenbaum’s ELIZA, built at MIT a few years earlier, is the best-known example. It could mimic a therapist convincingly enough to fool people in short exchanges, but it understood nothing. It just reflected words back.
The deeper problem was that understanding a sentence requires more than parsing grammar. You need to know what the sentence is about. “Put the red block on the one that is taller” only makes sense if you know which blocks exist, which is red, and what “taller” refers to in the current situation. No program at the time could connect grammar to world knowledge in a reliable way. Natural language research and knowledge representation were largely separate problems, and nobody had found a clean way to run them together.
What happened
Terry Winograd, working on his doctoral research at the Massachusetts Institute of Technology, built a program he called SHRDLU. The name came from a linotype keyboard sequence, chosen without particular meaning. What the program did was quite specific: it maintained a simulated tabletop scene of coloured blocks, pyramids and boxes, and let a user type instructions and questions in plain English.
SHRDLU used Michael Halliday’s systemic grammar, a theory that treats grammar as a set of choices a speaker makes depending on what they want to communicate, rather than a fixed set of rewrite rules. Winograd implemented this as a parsing procedure that worked alongside a model of the blocks world, so that grammatical analysis and factual knowledge informed each other as the sentence was read. When you typed “Find a block which is taller than the one you are holding and put it in the box,” the program worked out what “the one you are holding” referred to, checked whether any block was tall enough, and either moved it or explained why it could not. It also answered questions about what it had just done and why.
The system ran on a PDP-6 computer at MIT and was demonstrated in 1970. Winograd described it fully in a paper published in Cognitive Psychology in 1972, which became one of the most widely read documents in early computational linguistics. The dialogue transcripts published with the paper showed exchanges of genuine complexity, with anaphora (words like “it” and “one” that refer back to something said earlier), ellipsis, and multi-step reasoning all handled correctly within the blocks world. Within that world, the performance was striking. Outside it, the program knew nothing at all.
Why it mattered
SHRDLU showed that a computer program could parse and respond to syntactically complex English sentences by combining grammatical analysis with a semantic model of a small, well-defined world, the 'blocks world'. The work demonstrated both the power and the limits of symbolic, knowledge-intensive approaches to language understanding: fluency within the microworld did not generalise beyond it, a lesson that shaped debate about the scope of natural language processing for the following two decades. Winograd's 1972 monograph became a standard reference for researchers in computational linguistics and artificial intelligence.
People
Organisations
Massachusetts Institute of Technology
Sources
- SHRDLU, Terry Winograd's project page.Stanford University HCI Group (Winograd personal archive).Primary source
- Understanding natural language, Terry Winograd, Cognitive Psychology, Vol. 3, No. 1, 1972.Elsevier / Academic Press via Cognitive Psychology journal.Primary source
- Understanding Natural Language (book), Terry Winograd, Academic Press, 1972.ACM Digital Library record for Academic Press edition.Secondary
Cite this page
AI Achievements. (1970). SHRDLU Natural Language Understanding Program Developed by Terry Winograd at MIT. Retrieved 2026-08-22, from https://achievements.ai/milestone/shrdlu-computer-program-developed-terry
@misc{achievements_shrdlu_computer_program_developed_terry,
title = {SHRDLU Natural Language Understanding Program Developed by Terry Winograd at MIT},
author = {{AI Achievements}},
year = {1970},
url = {https://achievements.ai/milestone/shrdlu-computer-program-developed-terry}
}