Humanist Discussion Group, Vol. 39, No. 84.
Department of Digital Humanities, University of Cologne
Hosted by DH-Cologne
www.dhhumanist.org
Submit to: humanist@dhhumanist.org
[1] From: Gabriel Egan <mail@gabrielegan.com>
Subject: Re: [Humanist] 39.82: repetition vs intelligence: on LLMs (157)
[2] From: James Rovira <jamesrovira@gmail.com>
Subject: Re: [Humanist] 39.82: repetition vs intelligence: on LLMs (43)
--[1]------------------------------------------------------------------------
Date: 2025-07-17 12:34:01+00:00
From: Gabriel Egan <mail@gabrielegan.com>
Subject: Re: [Humanist] 39.82: repetition vs intelligence: on LLMs
Dear Humanists
There's rather a lot to respond to in
Tim Smithers's latest 9,750-word
posting on why Large Language Models
are not realy modelling human meaning.
I suspect that if anybody here wants
to know my response to Smithers's
posting, they'll want something
shorter than that. I believe that
an important part of where Smithers
goes wrong occurs in the first
2000 words of his posting, in
sections 1 and 2 about how human
language works. So, for this reply
I'll confine myself to just that
part.
Smithers writes:
> It's important in what follows to remember
> that the letters of the alphabets we use
> to write with are symbols for sounds.
No, people who are deaf from birth have
no trouble understanding letters and
alphabets, without ever having heard
a sound.
> They [the letters of the alphabet] are
> not symbols for bits of meaning: the
> meanings, to us as readers and writers
> of words, are not built from the
> letters we use to write them with.
> It is the sound of words that is
> built from the letters we write
> them with, and read them from.
If written meaning were primarily a
matter of sound then people who have
never heard a sound would not find it
so easy to make meaning using the
same written system as those who
can hear sounds.
> Writing systems are not semantic
> representation building systems.
> Writing systems are made of multiple
> systems of symbols needed to
> present well enough in written
> form the sound of spoken
> languaging.
Again, this insistence that sound is
primary in the construction of meaning
runs counter to the everyday experiences
of people who have never heard a sound.
> To build a LLM . . . we must remove
> [from the training data] all non-text
> material, images, pictures, photographs,
> diagrams, etc, and rip out all font
> selection and typographical formatting
> instructions; all the setting of
> the text on the page or screen;
> everything we put there to make
> the text readable and understandable
> by us readers.
This account is giving too much credit
to the font selection and typographical
formatting in the creation of meaning.
An LLM is perfectly capable of knowing
that "Just do it" is the slogan of the
Nike shoe brand despite not having the
benefit of learning from the accompanying
'checkmark' logo, the font, the all-caps
typographical styling, and so on. We
might note that the Humanist list postings
I receive have no font selection or
typographical formatting, unless we
count spaces and line-endings used to
make paragraph breaks as formatting.
By Smithers's model, he and I should
not be able to understand each other
in this styling-free medium.
> Good typography is not just decoration.
> It's needed for good accurate reading
> and understanding of what was written
> . . . .
This is demonstrably untrue. For all I
know Smithers could be reading my words
in a monospaced typeface printout from
a teleprinter, and for all he knows I
could be reading his in Braille.
In fact, I once supervised a student
whose only access to Shakespeare's
writing was in the form of Braille. I
found her accuracy of reading and
understanding of the play to be well
above the class average. She would
be insulted by Smithers's remarks.
> If LLMs really deal in words and
> meanings, why is all this needed
> aid [the typography] to reading
> the text first removed [in the
> training of an LLM]?
The premise here is mistaken: what
Smithers calls the "needed aid" is
not needed. Any deaf or blind
student of language could easily
debunk the model of language
presented by Smithers.
> Or, are we all now to believe
> that good typographical design
> plays no part in real language
> understanding from writing?
Yes, we should believe that. All
the evidence from people who have
no access to that typographical
design is against Smithers's model
of language.
To summarize then. Smithers claims:
i) that meaning is essentially a matter
of sound not writing. Classicists
and post-structuralists will notice
that this is Plato's attitude in
'Phaedrus', which as Jacques Derrida
pointed out is a thereby self-defeating
piece of written argument.
ii) that typography and layout are
essential in the construction of
written meaning.
I have not yet worked through the rest
of Smithers's posting, but since he
spends 2000 words on the above I assume
that points (i) and (ii) are important
to his overall view of the limitations
of Large Language Models. My aim in
these remarks is to persuade Humanists
that Smithers is wrong about (i) and (ii).
Regards
Gabriel Egan
--[2]------------------------------------------------------------------------
Date: 2025-07-17 11:44:58+00:00
From: James Rovira <jamesrovira@gmail.com>
Subject: Re: [Humanist] 39.82: repetition vs intelligence: on LLMs
Much appreciation for Tim's most recent post. That's exactly the point at which
I wanted to respond to Gabriel myself - the assumption, without any support,
that computer data of any kind holds "meaning" to the computer itself,
especially text display.
Tim's post could be a great jumping off point for a discussion of Derrida and
his deconstruction of the primacy of speech over writing. but there's no need to
go there right now.
My response is much simpler.
Words have meaning to human beings because they are associated with sounds and
feelings and experiences and memory and printed text. They are visceral, or at
least can be. They are also associated with relationships in different social
contexts, and so on, and all of this together combines to associate words with
personal identity.
And that is the point, at least one point and possibly the most important point,
at which words have meaning for human beings. And while we don't think only in
words, we think a great deal using words. Our thoughts are often comprised of
words.
Of course, different kinds of meaning are possible and we should discuss that
too, but I'm not going to right now. However, much of what I just said even
applies to working out math problems or conducting scientific experiments or
writing a computer program.
Words to a computer are just pixels. They are a display option. They don't have
any of those associations at all. As a result, words have no meaning to
computers. They don't understand them in any way comparable to human beings.
Computers do not think in words. They think, if we can use that term, in high
and low voltage states that are transformed into display patterns that human
beings ascribe meaning to. But that doesn't mean the computer itself does.
So using the language of meaning and understanding in relationship to words and
computers is a very bad form of equivocation. Yes, it is a fantasy. You have to
not be thinking in any detail at all about how computers work and how the human
mind works to think that they're equivalent or even comparable in any meaningful
way. It's the kind of claim you make to people who think that pro wrestling is
real, or if you want to write some ridiculous movie script along the lines of
the most recent Mission Impossible movie or The Matrix
Jim R
_______________________________________________
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
