488

LESSWRONG
Petrov Day
LW

487
Communication CulturesPhilosophy of LanguageScholarship & LearningSocial & Cultural DynamicsRationality
Frontpage

15

Listening Before Speaking

by Alice Blair
11th Aug 2025
3 min read
3

15

15

Listening Before Speaking
2Alek Westover
1Alice Blair
2cousin_it
New Comment
3 comments, sorted by
top scoring
Click to highlight new comments since: Today at 9:48 PM
[-]Alek Westover2mo2-1

A few thoughts:

  • Having conversations is often much more productive for resolving confusions than just listening to other people's conversations. It's pretty clear why this would be the case: if I'm confused about thing X, then this may trickle down to small confusions about lots of things Y1, Y2, Y3. Then, it's fairly likely that at least one of Y1,Y2,Y3 will come up in conversation, and if me and the conversation partner iteratively find cruxes, we'll quickly identify X and then the conversation partner can give me the arguments for X, at which point I may change my mind.

  • I would learn a lot more algebraic topology by talking to an algebraic topologist and asking them to define all their terms rather than sitting and "absorbing" a lecture. Ofc "listen to content that you can basically understand with a little bit of new stuff thrown in" is going to be more effective than this, but this still feels like a super silly way to learn algebraic topology.

  • "Speaking so that you sound like a native" is not the only goal of language learning (I'd argue that it's a relatively minor one). For instance, if I was to learn a new language my goal would be "be able to kind of basically communicate" rather than to be a native speaker. I'd guess that practicing conversation is much more efficient for picking this skill up, but would be interested if you have evidence to the contrary.

  • I kind of disagree with the suggestions in the "culture learning" section. I don't think it's good for epistemics for people to try to adopt the opinions of people on LW, or to think of this as a goal. If you're only willing to talk to people that share your opinions, then that is an echo-chamber. However, I agree that it's more efficient from my perspective for people to have first read some basic content about why people are worried about AI before talking to me about it (although I suspect that this is much less efficient from their perspective because I can adaptively choose what to talk about based on their demonstrated confusions).

Reply
[-]Alice Blair2mo10
  • This is about socially transmitted skills. Discourse norms are, but technical concepts like algebraic topology and whatever you're trying to figure out in the first bullet aren't.
  • Sure, you can have plenty of other goals, I'm not trying to refute that in any way.
  • People can have LessWrong culture while disagreeing with a lot of the things on here! This is about being familiar with the norms and knowing that the LessWrong ideas even exist at all, not getting people to believe them. The point about making mistakes that the Sequences caution against is a descriptive claim about an archetypical person who is new to the community, not a necessary condition.
Reply
[-]cousin_it2mo21

Yeah. I've found that a really fast way to get better at a language is to listen to an audiobook, pause every now and then to imitate a few words with the right accent, and the rest of the time kinda engage my vocal cords in an imaginary way along with what I hear ("kinesthetic motor imagery" which I recently wrote about). It doesn't take too much work, can be done for like an hour every day or more, and leads to pretty amazing progress even after one month.

Reply
Moderation Log
More from Alice Blair
View more
Curated and popular this week
3Comments
Communication CulturesPhilosophy of LanguageScholarship & LearningSocial & Cultural DynamicsRationality
Frontpage

Epistemic Status: anecdata and intuition

edited GPTl;dr: For socially transmittable skills that require learning lots of new category boundaries (languages, subcultures, etc.), a deliberate input-heavy output-light phase at the beginning reduces fossilized errors and speeds later fluency.

Language Learning

A friend of mine, let's call him Bob, learned English outside of his critical language acquisition period, the time early in one's life when fluently learning a language is practically guaranteed, relative to the difficulties people face later in life. Usually this would imply that Bob has some sort of foreign-sounding accent, possibly retaining some of the grammar and syntax of his native language instead of that of English.

Yet Bob speaks fluent, native-sounding General American English. He knows about as many words as the native speakers around him, with some small gaps. I argue that he's done this by mirroring not only the pattern that babies use to learn language, but also by mirroring a more general type of strategy for fluently learning radically new types of communication: Listening before Speaking.

Bob spent around 6 years consuming media in English before really ever speaking. I don't think this process needed to take this long, but this sort of scale seems approximately right. After getting a bit of runway to work with, he would watch English YouTube and TV shows without any subtitles, seeing how much he could understand just through what he already knew. I argue this did several things that were helpful:

  1. Give him time to understand the phonemes in English without immediately imposing his native set of phonemes onto them. When a native English speaker, call her Carol, first tries to learn Spanish, she might parse the rolled r as "sounds like the English r with some foreign accent", and then go to try and say "burrito" with the English r, having incorrectly assumed it's the same thing that she was already familiar with from English. Spending more time Listening allows a distinct concept boundary to form around the exact details of the Spanish rolled r: where in the mouth exactly is it? How many oscillations tend to occur in it? etc. Bob can now draw that concept boundary much more precisely than Carol can, and much more similarly to how a native speaker would, even if he's working entirely on intuition and has no explicit, crystallized knowledge about any of the linguistics involved.
  2. Force him to actually rely on his map of English, rather than his map of any other language. Without subtitles or translations, there's much less "ohhh, that's the word for [native word]" and much more "huh, why did the [English word] have to do [English word]?" Working entirely in the English frame of reference allows his model English to fail fast and update quickly. It also trains the muscle that he'll need later, where he exclusively thinks in the English frame of reference because it's mandatory to, such as real-time interactions.

He's now up to 3 languages fluently and is learning another currently.


Culture Learning

Sometimes, someone new shows up in the in-person rationalist scene. Call him Dan. Three types of thing can happen then:

  • He doesn't say much, and just sits and absorbs the conversation
  • He engages actively in the conversation
    • and what he's saying makes sense and sounds normal to the rest of the people
    • and he is obviously fumbling around and e.g. making weird inferences that The Sequences caution against. Many find these sorts to be annoying, although many are welcoming to the newcomer.

These correspond to

  • Listening before Speaking (Listening Phase)
  • Listening before Speaking (Speaking Phase)
  • Speaking before Listening

In the cases where Dan hasn't yet Listened, he also hasn't acculturated fully to the rationalist scene, but in one case this comes across as an issue in a well-kept garden, and in the other he's relatively harmless. 

Dan fits in when he Listens before Speaking, which usually happens by reading what he can from the rationalist material online. He learns the distinctions and concepts well ahead of actually trying to use them, and doesn't get tripped up by Dunning-Kruger. He has already absorbed some powers from the community for himself, and others can see that he is already One Of Them. 

Dans who fumble like this may be highly intelligent and well-read outside of rationality, but they can reliably be classified as not having Listened before Speaking. This is the sort of community we have, whether we like it or not, where learning our ways is in this sense like learning a new language.

Conclusion

In all of these cases, it took a lot of actual effort to get to a state of fluency. Bob's learning didn't come from passively absorbing media so much as hungrily dissecting it. Similarly, people often seek out rationality for their own ends, since after all there is something real worth learning rationality for. This is not the only way to learn things fluently: some people are phoneticians and can pronounce a new language correctly on the first try, some people have more of the Art of rationality within them by default. Still, this is a useful tactic, all else being equal.