# Terry Winograd

> 1946– · Computer Scientist
>
> **Recorded contribution:** SHRDLU natural language system; influenced HCI; advised Larry Page

## How to use this dossier

Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory.

## 1. Historical orientation

Terry Winograd (born 1946) built SHRDLU as a doctoral project at MIT around 1968–1970. In a simulated blocks world, users issued typed commands and questions; the system combined parsing, semantic interpretation, planning, and a database of objects and actions. Its fluent conversation looked impressive precisely because the domain was tiny and closed. Winograd later criticized assumptions of symbolic AI and, with Fernando Flores, emphasized language as situated action and design; at Stanford he became a major human-computer-interaction educator. Advising a famous founder is incidental. His historical arc—from SHRDLU to HCI—teaches how benchmark success depends on world boundaries.

## 2. The problem inherited

Natural-language systems could manipulate syntax but lacked enough contextual knowledge to connect words with objects, actions, references, and plans.

## 3. The central contribution

Winograd integrated language and planning inside a deliberately constrained microworld, then used its limitations to rethink the relation among computation, language, and human practice.

## 4. Reconstruct the mechanism

1. Parse an utterance into grammatical structure and map phrases to candidate blocks-world objects and predicates.
2. Resolve references using dialogue history, visible objects, type constraints, and the small world model.
3. Translate a command into goals and plan legal robot-arm actions while updating the symbolic database.
4. Generate an answer or clarification from the resulting state, explanation trace, and conversational context.

## 5. What changed downstream

- SHRDLU became a landmark demonstration of integrated natural-language understanding in a microworld.
- Its limitations informed skepticism toward scaling closed symbolic domains to ordinary life.
- Winograd's later work helped establish human-centered design and HCI at Stanford.

## 6. Attribution, limits, and uncertainty

- SHRDLU understood a hand-built toy world, not unrestricted language or general physical/social reality.
- Fluent answers can result from tight domain constraints without robust common sense.
- Later philosophical conclusions are influential interpretations, not a proof that symbolic methods are never useful.

## 7. Reconstruction lab

Build a five-object text world supporting move, stack, ask-location, and pronoun reference. Add one ambiguous command and one real-world request, then classify which assumptions make the in-domain exchange possible.

## 8. Evidence trail

- [Procedures as a Representation for Data in a Computer Program for Understanding Natural Language](https://hci.stanford.edu/winograd/shrdlu/) — MIT
- [Terry Winograd](https://hci.stanford.edu/winograd/) — Stanford University

---

*Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*