On Jul 20, 2025, at 5:20, Santafe <[email protected]> wrote:
This is a nice framework, Glen, even if one then has to do a lot of work to
find out whether there are good cites for some of the proposed themes.
It has had me thinking over the past day about the “alternation” between
Hofstadter-Lakoff, and whoever the Logicians of the moment are (Carnap-Quine or
Putnam or whoever).
One of the big themes that I assume would be behind the Hofststadter-Lakoff
position, and in different ways Damasio, would be this premise:
— Take some subset of things the brain does, which involve producing, or bringing
into current activity, some “that which is not”, where “not” is meant to indicate
“not currently streaming in through senses”, and also, up to the previous moment,
“not whatever was in the active role". My long clumsy phrase above is often
just called “memory”, though one could equally well regard it as “imagination” if
one thinks that imagination is a kind of synthetic or constructive manipulation of
the same primitives as memory.
— Suppose that the basic mechanism for that process in the last bullet is
resonance by some kind of content-similarity. So the novel produced thing
“which is not” is not identical to whatever was currently in the active role,
and can properly be called something “produced”, or “brought into the active
role”. But neither is it very far from, or free of, whatever was active that
led to its selection/production.
— The above content-resonance based program would be so different as to be nearly an
“opposite" from an address-based lookup, and in some idealized limit, the
address-based lookup is meant to provide complete independence between the address
and the content. The latter description of a machine process seems to overlap quite
heavily with the defining aim of logic and of the logical-system aspect of
mathematics (as characterized by Hilbert), in the sense that the symbols are
supposed to take on dynamics in their own isolated, synthetic world, without
dependence on “binding”, to such an extent that one can put aside even understanding
what binding is or how it is done, and still intend to make arguments about
properties of this synthetic domain.
Then suppose one had to make full operational systems out of all one primitive
or all the other. Or nearly so. I wouldn’t say the NN-based MLs are fully
content-similarity based, in the sense that there is a lot of structure there
that doesn’t rely on content similarity to take its form. It is what the
engineers fix as the design. Probably in brains that is also true to
considerable extents; Broca and Wernicke areas go into more-or-less
stereotypical places, and visual cortex already has a lot of organization
before there is anything for it to process. But brains might make much more
use of content-similarity to take their form and connectivity than ML systems
currently do. The kinds of problems Chuck Stevens used to worry about: how do
brains continuously function, while also growing, and seem to use the content
of their ongoing activity in essential ways as part of the directing input for
their growth?
I guess the above full operating system would look rather different from one
based on the von Neumann architecture as its central design paradigm.
But would I want to say that either then cross-cuts the other so strongly that
they are skew, that neither can be in any sense what the other is? I assume I
would not, and the reason would be the capacity for simulation.
People — and almost surely most of this fine-grained activity is going on in brains, so I want to claim that it is okay to focus the attention of a few sentences on what they do — do engage in deliberative activities (counting things out, working through logic puzzles along rule-system pathways, etc.), and even if we found that they used a nearly all-associative architecture to do it, that wouldn’t change the fact that at the end, there is a collection of states and events that carry the logical a-semantic tags faithfully. I would expect (after all, this is biology), that for some classes of symbol-like things that need to be used often in all people, the simulation hierarchy also gets hacked and tweaked a lot, to move its overall input-output function down to a lot more rigid and primitive level. Jackendoff’s “3-system” picture of message-passing phonology, grammar, and semantics seems to claim certain quite symbol-based programs working very fast and dense at low levels
in at least the first two of the three.
I imagine that this above fencing-of-views is conducted on something like this structure. One side says that we can identify primitives that are much simpler than the simulations they produce, with the latter being high-order syntheses from the former, and that therefore the primitives are “more fundamental”. As long as one knows that “more fundamental” is just a tag for the longer argument about “more primitive w.r.t. synthesis”, that can be okay. But if the simulation brings into existence something whose organization (deliberation with characteristics of logic and symbol-addressable content) has a compact description fully different-in-kind from that of the primitives, I don’t think one gets to deny that the new architecture has come into existence as a thing-in-itself in the world, even if it was by way of simulation that it was produced. I think my view here connects to your (Glen’s) earlier arguments that things really need to be produced to get credit for being
carried out. I have (in a paper that at this rate may never actually see the far side of a production process) that these symbolic things, even if just learned and used as deliberative sequences in private thought, have about the same artifact-status is the un-willed natural phenomena in the world, and different in nature from whatever our ongoing practice with, and experience of them is.
All kind of statements of the elementary, I guess, and things everybody in the
literature-conversation and here would already take as known and obvious, so
not addressing high-order questions, and thereby not interesting as well. But
maybe some terms for clearing underbrush? If they are not already wrong?
Eric
On Jul 19, 2025, at 8:13, glen <[email protected]> wrote:
So, what the hell, right? Below is Perplexity's response when queried for
background needed to start on an interdisciplinary program to do a competent
literature review for a theory of metaphor. Obviously I started with the 10
post corpus and expanded from there. I feel like computational linguistics
would at least partially cover what I said in the previous post. So to do a
good job, we need expertise in philosophy of language, theoretical and
computational linguistics, and neuroscience (and/or experimental psychology). I
use Perplexity because she's never yet lied to me about references (and the
others have lied to me … a LOT, the bastards). But these references are still
poor quality for the most part, especially [3], which is just a conversation
with ChatGPT. :face_with_rolling_eyes:
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Foundational Compatibility: Quine, Carnap, and the Roots of
Analogy/Metaphor
Quine and Carnap
* *Carnap*focused on the precision of language and the method of
“explication”—replacing imprecise or metaphorical notions with clear,
scientific language. To him, metaphors had psychological appeal but were
obstacles unless rendered into rigorous terms; meaning was considered in
relation to linguistic frameworks and their pragmatic use, not as a direct
window into cognition or the body[1].
* *Quine*rejected strict distinctions between analytic and synthetic (i.e.,
logically true versus empirically contingent), emphasizing/holism/: all
knowledge—including logic and language—is interconnected and susceptible to
revision. Metaphors like Neurath’s boat became models for how concepts are
interdependent within a web of belief, but not in the same embodied or
experiential way emphasized by cognitive linguists[1][2].
Hofstadter and Lakoff
* *Douglas Hofstadter*frames analogy as the “engine of cognition”—not merely
a linguistic or literary device but the fundamental way humans think and
reason. In his view, all concepts are “bundles of analogies,” and
analogy-making operates at every level of cognition, above and beyond
language[3][4].
* *George Lakoff*, building from cognitive linguistics, sees metaphor as
central to “embodied cognition.” Metaphors are not mere icing on language but
are structuring principles of thought itself, rooted in bodily experience and
sensory-motor systems (e.g., “Life is a Journey” shapes reasoning and
behavior)[5][6][7].
Synthesis and Limitations
* While Quine’s holistic web of knowledge and Carnap’s framework-based
philosophy help illuminate how conceptual systems allow meaning to emerge and
shift, neither can fully ground Lakoff’s or Hofstadter’s theories. Both
analytic philosophers were wary of metaphor as an explanatory tool—except in
circumscribed ways—whereas Lakoff and Hofstadter make metaphor and analogy
the/primary/fabric of cognition and meaning[3][5].
2. Necessity of Broader Disciplines: Psychiatry, Neuroscience, and
Linguistics
To robustly link logic/language (per Carnap and Quine) to contemporary accounts
of metaphor (as in Lakoff and Hofstadter), engagement with additional
disciplines is essential:
Cognitive Science and Neuroscience
* Neuroscience and experimental psychology have provided evidence that
metaphor and analogy-making are rooted in neural processes. For instance,
studies of right hemisphere processing and research on spatial and conceptual
metaphors suggest that metaphoric and literal language involve overlapping
brain circuits, challenging the older notion that metaphor is just secondary
“ornamentation”[8][9].
* Research supports Lakoff’s thesis that conceptual metaphors bridge physical
experience and abstract thought, showing mappings between embodied actions and
higher-level conceptual reasoning[5][8][9].
Linguistics
* The field of cognitive linguistics (where Lakoff’s work is foundational)
has demonstrated empirically that metaphor governs pervasive patterns in
language and thought, suggesting that the structure of logic in language is
itself metaphoric to a surprising degree[6][7].
* These insights go well beyond the analytic tradition’s focus on formal
language and highlight the “embodied mind” as a nexus for logic, language, and
metaphor.
Psychiatry and Psychology
* Psychiatry and cognitive psychology explore how metaphoric and analogical
reasoning underpins memory, identity, conceptual development, and even
psychopathology. This reinforces the view that metaphor is not epiphenomenal
but structurally constitutive of mind[9].
3. Comparative Table
Theory/Discipline How it Treats Metaphor/Analogy Foundation for
Metaphor-Centric Cognition?
Carnap Seeks to expunge metaphors via explication; language as constructed
framework Partial—offers logical clarity, not cognitive roots[1]
Quine Conceptual holism: metaphor as model for theory change; all knowledge
interconnected Partial—highlights systemic flexibility, not embodiment[1][2]
Hofstadter Analogy as the “core of cognition,” encompassing metaphor
No direct grounding in analytic philosophy—needs cognitive science[3][4]
Lakoff Metaphor as core structuring of thought, rooted in the body
Grounded in linguistics and neuroscience[5][6][9][7]
Linguistics/Psychiatry/Neuroscience Metaphor/analogy as built into both
brain and language, shaping cognition Essential—provide empirical and
theoretical tools[8][9][5]
4. Conclusion
* *Quine and Carnap provide partial philosophical foundations*for thinking
about how meaning and knowledge are systemic and context-dependent, but*neither
grounds metaphor or analogy as central to cognition in the way Lakoff and
Hofstadter envision*.
* *Interdisciplinary perspectives—especially from cognitive neuroscience,
linguistics, and psychology—are necessary to fully link the logic-language
discussions of analytic philosophy to the metaphor-based frameworks of Lakoff
and Hofstadter*.
* The logic of language, as Carnap explored, gains new grounding and
explanatory power when viewed through the lens of embodied, brain-based
metaphor and analogy highlighted by cognitive science and contemporary
linguistics[5][6][8][9][7].
[1]https://ejap.louisiana.edu/ejap/1997.spring/price976.html
[2]https://dornsife.usc.edu/scottsoames/wp-content/uploads/sites/191/2023/11/Quine_Carnap.pdf
[3]https://ditext.com/wordpress/2025/01/09/douglas-hofstadter-and-george-lakoff-on-analogies-and-metaphors/
[4]https://kobu.agency/the-lab-report/metaphors-look-beneath-the-surface/
[5]https://geometrymatters.com/cognition-and-the-embodiment-of-geometry-in-george-lakoffs-metaphors/
[6]https://en.wikipedia.org/wiki/Conceptual_metaphor
[7]https://terpconnect.umd.edu/~israel/lakoff-ConTheorMetaphor.pdf
[8]https://www.metaphorik.de/sites/www.metaphorik.de/files/journal-pdf/12_2007_goschler.pdf
[9]https://apcz.umk.pl/THS/article/view/ths.2002.002
[10]https://www.apperceptual.com/collected-quotes/analogy-and-metaphor
[11]https://worrydream.com/refs/Hofstadter_2001_-_Analogy_as_the_Core_of_Cognition.pdf
[12]https://www.reddit.com/r/cogsci/comments/b4z31/analogy_as_the_core_of_cognition_by_douglas_r/
[13]https://www.youtube.com/watch?v=n8m7lFQ3njk
[14]https://en.wikipedia.org/wiki/George_Lakoff
[15]https://www.jstor.org/stable/44084945
[16]http://mercercognitivepsychology.pbworks.com/w/page/61206434/Analogy Theory as the
Foundation for Cognition
<https://linkprotect.cudasvc.com/url?a=http%3a%2f%2fmercercognitivepsychology.pbworks.com%2fw%2fpage%2f61206434%2fAnalogy%2520Theory%2520as%2520the%2520Foundation%2520for%2520Cognition&c=E,1,Wk7rNEDEYeILsqwO7ULabw9kxPk6VK9HT1caenPm8zwMifUIxjtiFAsdF4IYSiuVi3rDQBz-2QzoPJxXst3n_gNqNJ2ftEiFthx3yqVjcsVERA,,&typo=1>
[17]https://plato.stanford.edu/entries/carnap/carnap-quine.html
[18]https://cs.uwaterloo.ca/~jhoey/teaching/cs886-affect/papers/LakoffJohnsonMetaphorsWeLiveBy.pdf
[19]https://www.sciencedirect.com/science/article/abs/pii/S1364661317301535
[20]https://en.wikipedia.org/wiki/Fluid_Concepts_and_Creative_Analogies
On 7/18/25 1:59 PM, glen wrote:
I dug out 10 posts that I think provide the corpus for this discussion. Attached. And I
may find the energy/desire to do some kind of work fleshing it out. But first, Steve's
invocation of "explanation" (either in the xAI sense or the science/knowledge
sense) reminded me of Melanie's article:
https://www.science.org/doi/10.1126/science.adt6140.
In contrasting Hofstadter's conception versus Lakoff's, and in light of our
(well, some of us) reification of LLMs as humans or humans as LLMs, the
question that consistently emerges is What is the relationship between
computation and body?
I've expressed my stance several times, I think. That formal definitions of "compute", like the one Lee proposed awhile back (by Soare) [⛧] that requires computation be definite, do not exist outside or apart from bodies of some kind. So in the context of both Dave's brain-computer and Eric's actual-formal evolution, my stance is not Platonic (or Popper's World 3 ... or whatever). Even/especially things like code/proofs executable as software have bodies. To some extent, if it can't be executed, then it's not True/real, hearkening back to "effective procedures" or somesuch. But going back to Eric's question on 7/16/25 6:19 PM about what work is done by the theory itself, assuming some of these abstractions (analogies, computable expressions) are schematic, we can make them less semantic/bound/definite by making them polysemous/multiply[bound|defined]. So in Eric's case, the terms in the/a logic of evolution can be unbound and rebound to a new context (and maybe tested for
inference and fidelity after the rebinding). Or in Dave's (and Steve's and Melanie's), unbind our "language" about people and rebind them to LLMs. Then as in ALife, rebind evolution and maybe even brain-computer (given models like Beliefs, Desires, and Intention or other forms of agency). I'm too ignorant to understand [un|re]binding in RNA Worlds - but I assume something similar could be done, as Eric seems to suggest.
So my answer to the relationship between computation and body, maybe resolving Hofstadter
vs Lakoff, lies in this [un|re]binding of the "logic". And where such
[un|re]binding fails, you can ratchet it back a bit. Maybe not *all* the terms in the
logic can be [un|re]bound, but *some* can. To be clear, I'm also talking about functions
being [un|re]bound/implemented, not merely atoms.
To me, that sort of program would lead to a methodologically useful theory of
analogy/metaphor. I feel like I've been infected with something like type theory in
saying this. I can't help but think there is a cadre of people already doing this work.
They just don't call it "theory of metaphor".
[⛧]
https://www.taylorfrancis.com/chapters/edit/10.4324/9780203100806-2/logic-modeling-logics-models-rudolph-lee
