Home About Subscribe Search Member Area

Humanist Discussion Group

< Back to Volume 32

Humanist Archives: March 4, 2019, 8 a.m. Humanist 32.506 - authorship attribution for automatically generated texts

                  Humanist Discussion Group, Vol. 32, No. 506.
            Department of Digital Humanities, King's College London
                   Hosted by King's Digital Lab
                Submit to: humanist@dhhumanist.org

        Date: 2019-03-03 11:56:47+00:00
        From: Ken Friedman 
        Subject: Fwd: Re: [Humanist] 32.503: authorship attribution for automatically generated texts?

Dear All,

Maurizio Lana’s question fascinated me. I don’t know the technical
answer to the question, but it raises intriguing philosophical issues.
Since it is in theory possible to identify authors using stylistic
analysis, I imagine that it should be possible to determine the
authorship of an AI-authored document provided that other examples of
text by the same AI author exist. What is less clear is this: if an AI
author can emulate human writing successfully, it’s difficult to know
whether that author can be identified as AI rather than human.

On the slightly amusing — and possibly irrelevant side — I discovered
this morning that Google Translate sometimes seems to make sentences
that are more sensible than what I originally wrote.

Yesterday, I got a version of the Nigerian Scam in Turkish. I wrote back
a total nonsense reply, meaningless words of nonsense syllables strung
together at random. The author — supposedly a lawyer named Stephan
Williams — replied as though I had written a serious answer, attempting
to move me to the next phase of the scam by stating conditions and
asking for my oath of confidentiality. (I learned this when I pasted his
text into Google Translate.)

Today, I replied with a nonsensical response — a real set of meaningless
sentences. Under the header  “salatalık çorbası” (Cucumber Soup), I
wrote, “This much so many the esteemed baboon has enjoyed. I salute your
tribulations, yet I must rebuke the splendor of your reply. It is with
sorrow but dignity that I decline this offer. Magnificent Stephen
Williams, you are the toenails of splendor. I thank you, but I withdraw
from consideration.”

Google Translate rendered this in Turkish as, “Bu kadar çok saygın
maymun çok hoşuna gitti. Bakirelerinizi selamlıyorum, ancak cevabınızın
ihtişamını azarlamalıyım. Bu teklifi reddetmek üzüntüsüm ama onurlu.
Muhteşem Stephen Williams, sen görkemli ayak tırnaklarını Teşekkür
ederim, ama dikkate alınmayacağım.”

When I translated this back in English to get an ideas of what the
imaginary Stephan Williams would read, Google gave me something that was
silly — but oddly enough, more grammatical than what I had written in
English before translation: “So many respectable monkeys enjoyed it. I
salute your virgins, but I must scold the grandeur of your answer. I'm
sorry to reject this offer, but it's honorable. Gorgeous Stephen
Williams, thank you to your splendid toenails, but you will not be ignored.”

The text generators that create false academic prose often resemble
sentences in real journal articles.

Who knows what’s next? As Shakespeare’s Miranda puts it in The Tempest
(Act 5, Scene 1),

“How beauteous mankind is!
O brave new world
That has such people in ’t!”


Ken Friedman, Ph.D., D.Sc. (hc), FDRS | Editor-in-Chief | 设计 She Ji.
The Journal of Design, Economics, and Innovation | Published by
Tongji University in Cooperation with Elsevier | URL:

Chair Professor of Design Innovation Studies | College of Design and
Innovation | Tongji University | Shanghai, China |||
Email ken.friedman.sheji@icloud.com
 | Academia
http://swinburne.academia.edu/KenFriedman | D&I http://tjdi.tongji.edu.cn

> On 2019Mar 3, at 09:01, Humanist  > wrote:

> some of us surely read the news about the OpenAI GPT software*
> which has a great ability to automatically generate texts of high
> quality and for this reason was not released in public by the
> developers, breaking from the normal practice of OpenAI which releases
> the full research to the public. at the above address links are
> available to the unpublished paper describing the software and to the a
> limited small version source code.
> so one thought arises: could an authorship attribution system/software
> be able to recognize such texts? or: how could an authorship attribution
> system/software be devised to be able to recognize such texts?
> if not, what else?


Unsubscribe at: http://dhhumanist.org/Restricted
List posts to: humanist@dhhumanist.org
List info and archives at at: http://dhhumanist.org
Listmember interface at: http://dhhumanist.org/Restricted/
Subscribe at: http://dhhumanist.org/membership_form.php

Editor: Willard McCarty (King's College London, U.K.; Western Sydney University, Australia)
Software designer: Malgosia Askanas (Mind-Crafts)

This site is maintained under a service level agreement by King's Digital Lab.