CoChat
Sign Up
Cited by 4,609Speech and dialogue systemsNatural Language Processing TechniquesAI-based Problem Solving and Planning

The DOI matches its Crossref record; 2 of the 2 references it cites were found in a registry or an index; none of the 2 sources its claims cite is checked for support yet.

ELIZA—A Computer Program For the Study of Natural Language Communication Between Man and Machine

Open access
Weizenbaum, Joseph·1966·Communications of the ACM
DOI 10.1145/365153.365168
Open this paper in CoChat — free accountOpen PDF

Abstract

Communication Between Man And Machine,JosEPh ~VEIZENBA UM Massach.usclls[nshl-ute qf Tcchnu[ogg,* Cambridge, Mass.ELIZA is a program operating within the MAC time-sharing system at MIT which makes certain kinds of natural language conversation between man and computer possible.Input sentences are analyzed on the basis of decomposition rules which are triggered by key words appearing in the input text.Responses are generated by reassembly rules associated with selected decomposition rules.The fundamental technical problems with which ELIZA is concerned are: (1) the identification of key words, (2) the discovery of minimal context, (3) the choice of appropriate transformations, (4) generation of responses in the absence of key words, and (5) the provision of an editing capability for ELIZA "scripts".A discussion of some psychological issues relevant to the ELIZA approach as well as of future developments concludes the paper.Work reported herein was supported (iu prot) by Project MAC, an _\-lIT research program sponsored by the Advm~ced Research Pr,,jects Aget,cy, l)~l)ar~met~t ~,1" l)ofense, under Office of Naval Research Contract Number Nora'-4102(01).

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

ELIZA is an early natural-language conversation program that scans input for keywords and applies associated decomposition and reassembly rules to generate responses. The paper describes its architecture, script mechanisms, and psychological implications, while stressing that ELIZA simulates understanding rather than possessing it.

Key findings

  • ELIZA selects the highest-ranked keyword found in an input text, then tries matching decomposition rules and applies an associated reassembly rule; if no keyword is found, it can use the reserved `NONE` category for content-free remarks or the `MEMORY` mechanism to reuse earlier transformed text.
  • Scripts are data rather than part of the program, so ELIZA is not tied to one recognition pattern or language; the paper states that scripts existed in Welsh and German as well as English.
  • Implementation details include a keyword dictionary using a `KEY` vector of 128 contiguous computer words, hashing keywords to a 7-bit integer; the largest dictionary attempted contained about 50 keywords, and no list in the `KEY` vector held more than two key list structures.
  • The discussion reports that the main serious scripts made ELIZA respond roughly like Rogerian psychotherapists, and that some subjects were hard to convince ELIZA was not human, which the paper calls a striking form of Turing’s test.

Methods

  • Design: ELIZA was implemented on MIT’s MAC time-sharing system and written in MAD-SLIP for the IBM 7091 as printed; users typed natural-language statements, and a double carriage return transferred control to ELIZA.
  • Data or sample: the paper uses illustrative conversations and script fragments, especially a Rogerian-style script, rather than a formal user study or quantitative dataset.
  • Main analysis: technical exposition of decomposition and reassembly rules, keyword ranking via a keystack, substitution rules, equivalence classes, `MEMORY` and `NONE`, and script editing through the MAC `ED` program, followed by discussion of psychological credibility and future directions.

Limitations

  • Stated: ELIZA discards its inputs except for a few transformed through the `MEMORY` machinery, lacks the capacity to store selected parts of input and draw valid conclusions, and has often aimed to conceal its lack of understanding.
  • Stated: the script contains specific hypotheses about the world despite claims that the psychiatric interview format avoids storing real-world information; the paper also says bounds on ELIZA’s extendability come from the program, not from any script.
  • Evident: no controlled evaluation or quantitative outcome data are reported; the evidence is illustrative transcripts and script examples, and the provided text is OCR-noisy and incomplete in places, making some technical details uncertain.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/365153.365168

Cites

2 of 2 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 of 2 cited sources checked · 0 supported · 0 flagged

Show all 6 claims

  • Probably not a substantial claim. 35
  • It may state two separate claims as one. 9
36 of 36

  • The paper relies on this without support.

    The MAC system parameters generally contribute so little to the overall response time that conversational interaction with the computer need never involve truly intolerable delays.

    “These latter parameters generally contribute so little to the overall response time that conversational interaction with the computer need never involve truly intolerable delays.”

  • The paper relies on this without support.

    The determination that a word is not in a dictionary may well require more computation (i.e., time) than the location of a word which is represented.

    “The iirst is serious in that the determination that a word is not in a dictionary may well require more computation (i.e., time) than the location of a word which is represented.”

  • The paper relies on this without support.

    Almost none of the words in any given sentence are represented in the keyword dictionary.

    “One stems from the fact that almost none of the words in any given sentence are repre.sented in the keyword dictionary.”

  • The paper relies on this without support.

    The transformation rule mechanism of SLIP permits tagging of words in a text and their subsequent recovery on the basis of one of their tags.

    “The transformation rule mechanism of SLIP is such that.it permits tagging of words in a text and their subsequent recovery on the basis of one of their tags.”

  • The paper relies on this without support.

    A human speaker will not defend his illusion that he is being understood against all odds.

    “But he will not defend his illusion (that he is being understood) against all odds.”

  • The paper relies on this without support.

    In human conversation a speaker makes certain, perhaps generous, assumptions about his conversational partner.

    “In tmnutn eonversat:on a speaker will make certain (perhaps generous) assmnl)lions about his conversational partner.”

  • The paper relies on this without support.

    As long as the partner's responses can be interpreted consistently with those assumptions, the speaker's image of his partner remains unchanged and undamaged.

    “As long as ~ it remains possible to interpret the latter's response~ consistently with those assumptions, the speaker's image of his partner remains unchanged, in particular, undanmged.”

  • The paper relies on this without support.

    Responses difficult to interpret consistently may enhance the image of the partner through additional rationalizations.

    “Responses wlfieh are difficult to so interpret m a y well result in an enhancement of the inmge of the partner, in additional rationalizations which then make more con> plicated interpretations of his responses reasonable.”

  • The paper relies on this without support.

    When rationalizations become too massive and self-contradictory, the entire image may crumble and be replaced.

    “When, however, such rationalizations become too inassive and even self-contradictory, the entire image may erumble and be replaeed by another (“He is not, after all, as s m a r t as I thought he was”).”

  • The paper relies on this without support.

    When the conversational partner is a machine, credibility may be substituted for plausibility.

    “When the conversational pa.rtner is a machine (the distinction between nmchine and prograin is here not useful) then the idea of credibility may weU be substituted for that. of pla.'uaibiStfl in the above.”

  • The paper relies on this without support.

    The crucial test of understanding is not the ability to continue a conversation but to draw valid conclusions from what one is being told.

    “The crucial test of understanding, as every teacher should know, is not the subject's ability to continue a conversation, but to draw valid conclusions h'om what. he is being told.”

  • The paper relies on this without support.

    Important decisions increasingly tend to be made in response to computer output.

    “I m p o r t a n t decisions increasingly tend to be made in response to computer output.”

  • The paper relies on this without support.

    The ultimately responsible human interpreter of “What the machine says” is constantly faced with the need to make credibility judgments.

    “The uhimately responsible human interpreter of “What the machine says” is, not unlike the correspondent with E H Z A , constantly faced with the need to make credibility judgments.”

  • The paper relies on this without support.

    The act of writing a script is a kind of programming act and has all the advantages of programming, most particularly that it clearly shows where the programmer's understanding and command of his subject leaves off.

    “The act of writing a script is a kind of programming act and has all the advantages of programming, most particularly that it clearly shows where the progranuner's understanding and con> mand of his subject leaves off.”

  • The paper relies on this without support.

    The human speaker contributes much to clothe ELIZA's responses in plausibility.

    “The human speaker will, a~ has been said, contribute much to clothe EI,IZA'S responses h: vestmen:s of plausibility.”
15 of 15

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Flagged first, then by citations

Show all 2 references

Research works printed but not linked: 4

  1. Gorn, S. Semiotic relationships in ambiguously stratified language systems. Paper presented at Int. Colloq. Algebraic Linguistics and Automatic Theory, Hebrew U. of Jerusalem, Aug. 1964.
  2. 10.5555/889266 · DOI
  3. Rogers, C. Client Centered Therapy: Current Practice, Implications and Theory. Houghton Mifflin, Boston, 1951.
  4. Yngve, J. COMIT Programming Manual. MIT Press, Cambridge, Mass., 1961.

Cited by

Sorted by citations

Items 101–200 of 4,609

Related

See all

Metadata

Authors
Weizenbaum, J.
Affiliations
Massachusetts Institute of Technology (US)
Corresponding author
Weizenbaum, Joseph
Venue
Communications of the ACM
Publisher
Association for Computing Machinery (ACM)
Published
1966-01-01
Volume
9
Issue
1
Pages
36-45
ISSN
  • 0001-0782 (ISSN-L)
  • 1557-7317
DOI
10.1145/365153.365168
OpenAlex
W4236521339
Cited by
4,452
References
2
Citations by year
2026: 450 · 2025: 576 · 2024: 524 · 2023: 465 · 2022: 298 · 2021: 310 · 2020: 245 · 2019: 231 · 2018: 197 · 2017: 103 · 2016: 109 · 2015: 63 · 2014: 100 · 2013: 75 · 2012: 44
FWCI
10.41
Type
journal article
Language
en
Open access
Licence
www.acm.org/publications/policies/copyright_policy#Background (vor)
Locations
Communications of the ACM (publishedVersion)
Funders
Office of Naval Research
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
  • Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
  • Physical Sciences › Computer Science › Artificial Intelligence › AI-based Problem Solving and Planning
Keywords
Eliza, natural language conversation, human-computer interaction, decomposition rules, response generation, script editing

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-28, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review