Language is primarily a tool for communication rather than thought [pdf]

Sure. But when you do have it, would it be fair to say it does shape your thoughts, no matter how small?

You're confusing the object language (human natural language) and the meta-language (formal grammars - ala Chomsky).

Formal grammar theory relies on recursive formulae to define what is a a language (formally set of sentences), natural language itself does not use/require recursion as understood in the same way. Some natural languages support grammatical features like sub-clauses but that's only "recursive" in the sense that attempts to describe such features (using formal language grammars) use recursion when in fact and in practice nesting sub-clauses do not "go on forever".

An analogy with computer programming languages - programming languages exist that do NOT support recursion (pre-Algol) but nonetheless require recursion to define their formal grammar. For example, a programming language can support limitless nested lexical scoping (loosely analogous to sub-clauses in natural language) WITHOUT supporting recursion (which requires a stack and managing frames).

For some programming languages (functional for example), recursion is such a key feature that it's inconceivable that the language could be useful without it.

This is not the case for natural languages. Most of the people I interact with daily have English as a second language and rarely or never use sub-clauses (which are not recursive but are often used to claim that language is recursive) so from practical experience, I know that English as a communication tool looses very little expressive power when users do not use sub-clauses.

Not OP but I also don't think with an internal monologue most of the time. For me it's often more like mentally manipulating abstract shapes or quantities and trying to make them fit together. When I'm writing software I'm literally thinking about pointers and bytes etc, not thinking about the words "pointer" and "byte". This is highlighted by the fact that after intense programming sessions I have dreamed about code, like I am computer memory and I'm being allocated by a memory manager or something.

Sometimes I do explicitly think with an internal monologue, though. Like if I'm debugging something I'll sometimes narrate what the program is doing in words. Also if I'm trying to figure out how some event happened I'll try to tell a story in my mind. It helps then as it forces me to serialise things.

When I read there is sometimes an internal monologue. When I write, there isn't. I don't talk like this. I think it's quite clear sometimes when people write with an internal monologue as their text reads like speech (which can be a good or bad thing, depending on its purpose).

I can think of that too. But it's not the same. A crude analogy would be to working memory. "Buying books without reading them, as a phenomenon" takes half a dozen or more slots in memory. "Tsundoku" is one slot.

That makes sense given the analogous LLM, which doesn't 'think' via the individual or even sum of the tokens it produces, but in-between in the hidden layers. The output - or actualization of the token - is simply just the form of communication used to convey the 'thought'.

I guess you could say its inappropriate of an analogy since the LLM must use generated tokens to feed into itself in order to 'think' over time beyond a single token. But I would argue that's simply because we enforce the communication method.

One does not need language to have a known concept for "objects fall to the ground when let go". Babies learn this without language. They do not have words for the particular concepts such as "object", "fall", "ground", but as concepts they are learned all the same.

For thought, all there really is, are concepts, or forms in platonic terms. In our case, learned patterns in the world which correspond to particular sense perceptions. Imagination, or the latent space between the forms, is our initial (and only, I'd argue) mode of thinking. These forms, encoding both logical and sensory - of which they are inseparably bound - can be strung together to form thought. They agree to a type of logic, affording the capacity to extract meaning out of the world.

It is an embodied logic, not boolean algebra. The logic is grounded in the ground, the earth underneath your feet. It is grounded in the sun, in its heat and light. It is grounded in the logos, or the intelligible unfolding of the universe. The intelligibility is afforded because of its own nature - that it is built up out of real patterns, patterns all the way down. But it also requires the appearance of the patterns in our sense perceptions in order to be 'seen' and thus known.

As such, logic of this sort is found to be based off of the learned patterns of how the world appears to us so as to make it intelligible, affording our ability to live within it. Then you get sensory being mapped into grounded logic and vise versa, so as to experience the unfolding of the logic with imagination, and I suppose the affective state changes it produces are then fed back into the logic. So emotions seem necessary in order to 'react' to sense perceptions, so that the logic is grounded in caring about what happens, like a mother does with a child. Otherwise there is just appearance, which has no qualia in and of itself, and so the patterns that can be discovered in the world have no ground.

Determining the necessity of the ground of appearances is probably what will determine if we can have AGI or not. We'd like to think computation alone will get us there, or even embodied cognition, but without the transcendentals, its dimensionality of knowing will always exclude this ground, so it cannot know what patterns are most relevant or salient, so it cannot perform relevance realization and overcome combinatorial explosion. So AI cannot be a temporal embodied agent, since the logos of the world is incomprehensible, except through itself, of which we Beings are made out of. We rely on lower order logic, such as math, for our AI. It is how we make sense of the real patterns we see in the world, but they are an abstraction over the thing itself. So a machine is confined by boolean algebra, while we are confined by the logos itself. Sounds nice anyhow.