Live data from Hacker News

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

gwern.net

381–390 of 404 posts

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

#381
I feel like this is one of those "of course" papers that people will dismiss, but it is good in science to confirm via orthogonal methods things that ought to be true[0]. All evidence points to language evolving for the purpose of communication, so of course its primary function is communication. It was only later co-opted for use in thought.

[0] I like the analogy of a subway network. Something like a star topology is unreliable because you have one problem and a whole swath of the city is inaccessible. But if you start building secondary and tertiary links between stations, you get robustness. Even if there's a problem or two you can still get anywhere.

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

#382
There's an interesting debate here -- see this tweet. https://x.com/ylecun/status/1769137037959913690. Cool to see an actual article on it as opposed to speculation from personal experience.

Now for some speculation from personal experience. Some people leverage language more for internal thought and some people have other intermediate representations. I'm personally more of a language thinker except I have a specific avenue for math/programming that is highly optimized. I imagine many use language to solve math problems (and no way is right or wrong).

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

#383
Sounds reasonable. Animals (including humans) construct mental maps of their environment without language, which is an important ability for everything from seasonal migration to location of reliable food sources. For social animals, simple vocal phrases like 'follow me to food' (e.g. in parrots, crows, etc.) would be secondary to the construction of that mental map, but do confer a large survival advantage on their social group (group selection in evolutionary terms).

If so, this has some interesting implications for LLMs and AGI. First, LLMs don't seem to be thinking or reasoning if they're only doing next-word-prediction based on a large training set - so a step towards real AGI might be a system that had an internal map of reality that it was constantly updating with new sensory information, with the LLM as a layer above it that allowed it to communicate the relevant details of that internal map to others?

Personally, every time I 'think in words' it always begins with 'how can this be explained' and the actual thinking that precedes that is done with images, diagrams, graphs, etc.

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

#384
This is surprising to me because when I learn something, I usually find that trying to speak it out helps a lot of thinking about it clearly. Put it another way, the effort to turn something into language helps forming the thought about it. Thus I don't know whether counts as a counter argument.

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

#385
post #367

I'm bilingual, people keep asking me what language I think 'in', and I don't think I think _in_ a language. For lack of a better way of describing it, I rather think in concepts, sometimes the voice in my head likes to play along by imagining the words representing those concepts, sometimes in English, sometimes in Romanian, but the words are not necessary. In dev words, I guess I think in structs and then optionally…

I'm bilingual and I can switch my permanent internal dialogue to one language or another, and It happens automatically if I'm actively using it for a few minutes. This afternoon that I've been on HN for a while I'll think in English, probably until wife comes. In dev words, I could say the same as you, but maybe the serialize phase is not optional.

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

#386

This would be sort of a death blow to much of postmodernism, no? "Reality is structured like a language and the free play of symbols reigns over all"-style thinking was very popular at one time, but this would rule against it.

No. Language stays a framework for modelling. Reality stays available to us only in form of models (pictures, sounds, ideas, measures, texts, videos, this text and what you imagine based in the text). The claim "Reality is structured like a language and the free play of symbols reigns over all" holds regardless of us thinking in symbols, words, pictures, concepts, cliches, emotions or _other models_. We have access only to symbols (not reality they represent) thus we "stuck" with this postmodernism discovery for observable future.

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

#387
post #367

I'm bilingual, people keep asking me what language I think 'in', and I don't think I think _in_ a language. For lack of a better way of describing it, I rather think in concepts, sometimes the voice in my head likes to play along by imagining the words representing those concepts, sometimes in English, sometimes in Romanian, but the words are not necessary. In dev words, I guess I think in structs and then optionally…

I'm bilingual (having learned English relatively late - in school), and i think in both languages. I think my mind language depends on which one I'm immersed currently (work/browsing internet: English, real life meetings: my native one). I have very long internal monologues in both languages picked unconsciously/randomly, and sometimes switching.

Another fun quirk is sometimes I remember that I've read a sentence, I can "quote" it, but I'm not sure if I've read it in my native tongue or in English. I guess my mind sortes them in a language independent way sometimes?

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

#388
post #367

I'm bilingual, people keep asking me what language I think 'in', and I don't think I think _in_ a language. For lack of a better way of describing it, I rather think in concepts, sometimes the voice in my head likes to play along by imagining the words representing those concepts, sometimes in English, sometimes in Romanian, but the words are not necessary. In dev words, I guess I think in structs and then optionally…

I am not bilingual but rather can speak fluently in some foreign language. However, I constantly find myself thinking in the foreign language I know, especially for topics I learned from reading in foreign books.

Another simpler example is when I count I usually count in my native language, even in a specific dialect when I was taught very young!

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

#389
Some people use language in thought more than others. Some don't use language at all. It's certainly time to exit the notion that language is necessary for complex thought.

Thinking the Way Animals Do: Unique insights from a person with a singular understanding. https://www.grandin.com/references/thinking.animals.html

> A horse trainer once said to me, "Animals don't think, they just make associations." I responded to that by saying, "If making associations is not thinking, then I would have to conclude that I do not think."

>I have no language-based thoughts at all. My thoughts are in pictures, like videotapes in my mind. When I recall something from my memory, I see only pictures. I used to think that everybody thought this way until I started talking to people on how they thought. I learned that there is a whole continuum of thinking styles, from totally visual thinkers like me, to the totally verbal thinkers. Artists, engineers, and good animal trainers are often highly visual thinkers, and accountants, bankers, and people who trade in the futures market tend to be highly verbal thinkers with few pictures in their minds.

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

#390
Was just remembering some advice I received from a counsellor years ago: "thoughts lead to actions, actions lead to emotions". That never fully resonated with me, because if thoughts as we understand them are the origin of everything, we should have a lot more control of ourselves than we actually do.

Now I tend to think that both thoughts and emotions have the same origin: some sort of internal representation of the world around us that we do not directly control. Thoughts, whether visual or auditory or symbolic, arise from this "model". I'd like to think of this model as not consisting anything that we would recognise directly as thoughts or emotions or even memories. It may be a "machine language" layer that has no conscious representation. I wouldn't call it subconscious thought because that would imply a thought being thought, only without conscious awareness.

Post reply on HN