> 174 Czech adults listened to 5 minutes of Māori... They were then tested on new audio clips from either Māori or Malay ... and asked to say if they were hearing the same language as before or not. The test phrases were acoustically filtered to mimic speech heard in the womb. This preserved melody and rhythm, but removed the frequencies higher than 900 Hz which contain consonant and vowel detail.
> Listeners correctly distinguished the languages more often than not
Just to be clear what was actually tested. The conclusions seem pretty heroic to me given the extremely limited nature of 'language learning' that happened here.
I think this is a really interesting and counterintuitive suggestion that potentially makes a lot of sense. I'd never considered this. I always assumed the difference was just that babies' brains are different, but it's true....as an adult most language learning centers heavily on the written word, and that was not how we learned to speak our first languages. We learned to speak and listen, then to write. Adults do it mostly the other way around. Very interesting idea. I'll try this with my next language.
Anecdotally, I'll say I've been learning a second language for 20 years by myself in a place where that language isn't spoken by anyone, and I've made more progress in the past 2 years than the preceding 18. What I changed:
- 15 minutes on anki vocab flashcards every night
- reading books in that language daily instead of english (my native language) and making flashcards for every word or phrase I didn't know
- listening to videos and podcasts in that language aimed at native speakers, NOT learners, and watching them REPEATEDLY until I could understand as close as I could get to 100% of it without subtitles. I try to spend 20-40 minutes a day listening. I especially put on podcasts in the car, but I have to be at home on my pc to rewatch with subtitles or a transcript to get the parts I couldn't understand initially.
It takes a lot of patience. There's no way around that.
After studying my target language on and off over the years with an Anki deck and not getting too far, I started “immersing” in my target language in a similar manner in addition to Anki a little over a month ago. I listen to native podcasts and audio ripped from native TV shows, at least an hour per day but occasionally 2-3 hours, and while it’s not as active as your method (I usually listen while preparing meals and eating), the difference it’s made in this short time has been remarkable. Reading appropriate level native written content helps too.
I feel like I’m starting to get an actual grip on the language instead of just being able to get the broad strokes of what someone is saying by latching onto vocab I’ve memorized, and while I wouldn’t say I’m comfortable with trying to hold a conversation yet it’s already dramatically improved my ability to compose understandable sentences.
This study reminds me of the Automatic Language Growth method. I can't remember all the specific but it was created by some guy long ago in like the 70s or 60s and tested it in Thailand. Basically, it was an approach that had the students listen to guides talk for like about a year before they before ever speaking. According to what I heard, people who went through it, would be able to speak fluently, as though it was their first language, no accent or anything.
Sadly, there are not a lot of comprehensible input based learning courses out there as most focus on speaking early.
There are a number of programs that use a purely audio-lingual approach with translation, like Pimsleur or the <language>pod101 series. In both cases, when I tried them as a short term tourist, people were surprised how well they could understand me for the short time I’d been learning. These were tonal languages very few tourists tended to learn, so local expectations were low and just being comprehensible was a win.
Yeah I've actually done one of Pimsleur's courses for Mandarin. But that's not the same thing as it's just like most courses in that they have you speak early rather than focus on listening.
I had been learning my target language mostly from books (and then watching/listening to media for osmosis) for years before I discovered Pimsleur. I can confidently say that Pimsleur increased my skills by orders of magnitude. Being forced to 1) speak out loud, 2) create sentences under time pressure and 3) absorb language rules implicitly rather than explicitly was a major game changer.
My only disappointment is that Pimsleur only created 2 levels for my target language, so after I had repeated them a number of times I had to look elsewhere for other (less effective) materials. Recently, I've been using the ChatGPT Voice Assistant to try and approximate the Pimsleur experience - it works to some degree, and it's novel that I can ask it questions during learning, but it can't really offer a curriculum in the same way.
Pimsleur is very effective when just starting out a new language.
In general, I find that the most important thing to learn first with a new language is the pronunciation. Compared to learning thousands of words, it's not that difficult to learn to correctly produce the sounds of a foreign language. If you are able to pronounce things properly, native speakers will think much more highly of your abilities in the language (which can backfire when they assume you're fluent).
We've created an app with listening-first approach and added some very useful features on top of that, like showing a sentence when requested and saving unknown words for later practicing.
There are three exercise types, one is prerecorded conversation where you pick one side and record your pronounciation so you can compare it with a native speaker.
Really glad to see more apps that focus on sentence level speaking and listening. I wish the mobile landing page had more screenshots of the app since my target language is not yet supported.
Also, we're happy to give the app away for free to nonprofits who'd find a use for it for their clients. It is also very easy for us to add domain-specific custom content, should there be an ask for it. If interested, PM me, e-mail is in my profile.
I built a system for this! Or, well, comprehensible input more broadly - it mostly uses text. Dreaming Spanish sorta takes this more audio-based approach.
I do question whether it's helpful to focus on audio without text. They're focusing on melody and rhythm, and it seems that listening without subtitles is better for that, but that doesn't get you understanding, while listening to comprehensible input with subtitles lets you get melody, rhythm, and actual vocabulary and grammar at the same time. It also lets you stretch "comprehensible" a bit further, since you have an extra source of contextual input.
Babies also have a lot of context for what the words they're hearing actually mean. I suppose there's the "watch translated peppa pig" approach for that.
My project (https://nuenki.app) translates appropriate-difficulty sentences into your target language as you browse, so you casually pick it up over time through comprehensible input.
I'm learning German and I've been enjoying browsing and reading with Nuenki. It's a lot of fun to see a sentence in context and find out you are simply able to understand it, though sometimes I have to fight the urge to instinctively hover to reveal the original text. Thank you for this plugin!
Dreaming in Spanish itself is a good example of a program that doesn’t rely on text to make its audio comprehensible. Their lower level videos include a lot of context that make it possible to understand the speech without written words.
Steve from LingQ once mentioned in passing on a Vlog that they have done studies where they gave students vocabulary lists (plus a vocabulary list free control group) before giving them a reading passage to study in a foreign language and they found that the vocabulary list group had worse outcomes. Does anyone happen to know which study this is? It has always stuck with me but I’ve never actually been able to pin down the exact study he was talking about.
"A Comparative Study on the Effectiveness of Using Traditional and Contextualized Methods for Enhancing Learners’ Vocabulary Knowledge in an EFL Classroom" where those who were given traditional vocabulary lists performed worse.
Good find. The description of the actual experimental conditions is incredibly vague, but it sounds to me a bit like the control group only got the vocabulary list with definitions and didn't read the short story, whereas the experimental group only read the short story and didn't get any definitions. Then the experimental group was able to successfully complete ≈10 of 15 fill-in-the-blank questions vs the control groups ≈8.
That is an okay experiment to compare the two teaching methods, but doesn't address the question of what happens when you combine them by reading a text with an accompanying vocabulary list. I would be rather surprised if additional access to definitions actually hurt learning. Like, what are dictionaries for, then?
A lot of the article’s findings about written translations and written transcriptions reflect my own anecdotes as a Korean language student. Listening and reading are distinctly separate skills but educational materials sometimes try to couple them together as interchangeable parts.
I’ve been working on a spaced repetition system that focuses exclusively on listening and speaking skills. It is not quite spicy enough to remove ALL writing (shown at the end of the review), but quizzes are done blind without showing written transcriptions.
>Early studies have proposed that a putative “sensitive period” for acquiring the sound patterns of a language ends around age 6. Not coincidentally, this is the age when many children learn to read.
Separate but related, a plug for Glossika (www.glossika.com), which primarily uses a listening and speaking approach to language learning. (I'm just a happy user.)
Babies are actually pretty bad at learning language. It's just that it's the only thing they have to do. Adults are much better at learning language, but they never put any time into it.
This is spot on. As a baby, you typically have two affectionate caretakers who dedicate a significant amount of their time to teaching you language/words, and also shower you with love/praise/attention for every little bit of incremental progress. Also, you're in an immersion program: you often can't communicate your wants/needs until you've learned the language, which adds another layer of incentive.
As an adult, if you were sent to an immersion program in a foreign country with two full-time foreign tutors/caretakers who loved you, like legitimately loved you with all their hearts, you would pick up that new language pretty darn quick.
Exactly; the advantage of input-based approaches is not their speed. You'll learn a lot faster if you spend the time on flashcards, grammar drills, and studying.
The advantage is that it's easier, and often more pleasant, to integrate.
That said, babies do have some neuroplasticity advantages.
We should play to our strengths - babies are hopeless at grammar drills, and adults aren't as good as babies at neuroplasticity - by applying our brain to the problem.
I say that, while having built a comprehensible input tool (https://nuenki.app). But it's useful as a complement to that study, as you can more readily run a browser extension or listen to podcasts than devote your entire life to focused study.
I think that's patently wrong. Babies and children in general absorb languages much easier than adults do. See children of parents speaking two different languages. By the time they reach school age they are fluent in both languages, and in the same time they need to learn how to be humans at the same time. Just because it's not learning as you conceive it as an adult it does not mean it's not there.
This is a common misconception. It is extremely rare for any child to be fluent in any language by the time they start school.
There are different levels of fluency that language training professionals recognize, mostly standardized on the ILR scale [0].
There are 5 levels on this scale. What people commonly consider "fluent" is level 3. There are also midpoints recognized by a "+", e.g. most people who study language seriously but don't have any special aptitude for language training get to a 2+. It's an open question of serious debate and active research of whether it's possible to train anyone past 2+ to 3. Some rare individuals seem to have an innate aptitude for language acquisition and they seem to breeze through their training to 3 and then continue on from there. Everyone else seems to plateau at 2+.
3 is just the level of "can easily be understood as a foreigner by a native speaker". It's not native level of mastery. "So fluent you can fool a native speaker" is 5.
All of that is to say that children are most definitely not fluent by the time they reach school age. Most children starting school will be at a 0+ level. Particularly intelligent children might be at a 1. But they are definitely nowhere near 3.
And indeed, by the time most people graduate highschool, I would struggle to admit that they are much beyond 3. Hell, I'd even say a large portion of people graduating college don't have mastery of their native language.
So no, I don't think children have a particularly large advantage in learning language. Despite significant, continue training in language for the first 18 years of their lives, most aren't able to get very far beyond what any studious adult can learn in 2 years.
Another hackernews article where the title is clickbait and the experiment doesn't really honor the title implication.
Language learning is not easy nor effortless nor "just a few minutes per day" endeavor. Babies had 24/7 immersion and a completely different context.
You can do experiments and try certain things but the framing needs to be honest and not "we tried this thing in a very small environment and extrapolated like mad".
"Our research builds on previous studies, which have found that spelling can interfere with how learners pronounce individual vowels and consonants of a non-native language."
This matches very well with my own personal observations. Most language instruction focuses very heavily on written instruction, which can be damaging early on. When you're just starting out a new language, you should learn the sounds by mimicking native speakers, not by reading letters off a page. When reading, your preconceptions of how those strings of letters would be pronounced in your own native language can interfere with correct pronunciation.
Once you've got the pronunciation of the basic sounds of the new language down correctly, you can start using written materials.
The conclusion is extreme. Reading is too much of a benefit in language learning to give up on at all, it's the best source of the widest range of comprehensible native input you're going to find.
I learned my second language by reading out loud, and only going on if I was sure I was pronouncing the thing right. We have the internet now; if you think you're pronouncing something wrong in a common language, you can go on the internet and make sure you're pronouncing it right. [edit: I remember when people were using their phones to transcribe what they were saying in L2 in order to check their pronunciation; do people still do that?]
When reading out loud sometimes I would sing the words, because you can't come up with a convincing way to sing a sentence unless you understand what it means.
Orthography is of course going to be distracting in English or French (or Chinese) because they have terrible orthography when it comes to representing sounds. But you're going to want to be literate anyway, so eventually you're going to have to learn how to recognize but ignore the details of the words.
Subtitles are a problem because people shut their ears off while reading them (as the article claims to confirm.) But if you can understand the subtitles, you've improved your literacy and gained an understanding of that piece of media. Now turn them off, and listen while knowing what's being said. Shadow and repeat after the speakers. Check your pronunciation against them. Roll the words in your mouth while the meaning is rolling around in your head. Make sure you know and can repeat every word that's being said, and make sure you understand the grammar.
Reading, writing, speaking, and listening are all very different things. But I'm not just learning a language to speak to people, so every one if them is important. I will not understand the cultural references that people use in daily conversation to communicate meaning unless I read. I can't replace their entire upbringing in a language by chatting, even in a focused way, for a few years.
Learning a language as an adult is different than learning a language as a baby. As an adult, you have another language to compare everything to, and you already know how languages work. A baby will become extremely fluent in their first language before they ever learn to read. You already know how to read. Take advantage. If you want to work on your pronunciation, there are lots of ways, like chorusing, shadowing, reading out loud, Pimsleur and FSI give you lots of chances to compare your pronunciation to a model, etc.
Babies take 7-8 years to get good at a language. They're no faster than you.
My problem is how to overcome the translating in your head to being able to think in a language even simple things.
I was in Costa Rica and while all of the customer facing people spoke English, one guy who brought us water asked a simple question “desayuno o cena?” After thinking about it for a couple of seconds, I realized he was asking whether we wanted the breakfast or dinner menu. Even the Spanish I do know, it takes me way too long to understand it and speak it.
> 174 Czech adults listened to 5 minutes of Māori... They were then tested on new audio clips from either Māori or Malay ... and asked to say if they were hearing the same language as before or not. The test phrases were acoustically filtered to mimic speech heard in the womb. This preserved melody and rhythm, but removed the frequencies higher than 900 Hz which contain consonant and vowel detail.
> Listeners correctly distinguished the languages more often than not
Just to be clear what was actually tested. The conclusions seem pretty heroic to me given the extremely limited nature of 'language learning' that happened here.
I think this is a really interesting and counterintuitive suggestion that potentially makes a lot of sense. I'd never considered this. I always assumed the difference was just that babies' brains are different, but it's true....as an adult most language learning centers heavily on the written word, and that was not how we learned to speak our first languages. We learned to speak and listen, then to write. Adults do it mostly the other way around. Very interesting idea. I'll try this with my next language.
Anecdotally, I'll say I've been learning a second language for 20 years by myself in a place where that language isn't spoken by anyone, and I've made more progress in the past 2 years than the preceding 18. What I changed:
- 15 minutes on anki vocab flashcards every night
- reading books in that language daily instead of english (my native language) and making flashcards for every word or phrase I didn't know
- listening to videos and podcasts in that language aimed at native speakers, NOT learners, and watching them REPEATEDLY until I could understand as close as I could get to 100% of it without subtitles. I try to spend 20-40 minutes a day listening. I especially put on podcasts in the car, but I have to be at home on my pc to rewatch with subtitles or a transcript to get the parts I couldn't understand initially.
It takes a lot of patience. There's no way around that.
After studying my target language on and off over the years with an Anki deck and not getting too far, I started “immersing” in my target language in a similar manner in addition to Anki a little over a month ago. I listen to native podcasts and audio ripped from native TV shows, at least an hour per day but occasionally 2-3 hours, and while it’s not as active as your method (I usually listen while preparing meals and eating), the difference it’s made in this short time has been remarkable. Reading appropriate level native written content helps too.
I feel like I’m starting to get an actual grip on the language instead of just being able to get the broad strokes of what someone is saying by latching onto vocab I’ve memorized, and while I wouldn’t say I’m comfortable with trying to hold a conversation yet it’s already dramatically improved my ability to compose understandable sentences.
BTW Anki has a new algorithm thats vastly more effective: FSRS.
This is a good approach for reading research papers in a new field too.
This study reminds me of the Automatic Language Growth method. I can't remember all the specific but it was created by some guy long ago in like the 70s or 60s and tested it in Thailand. Basically, it was an approach that had the students listen to guides talk for like about a year before they before ever speaking. According to what I heard, people who went through it, would be able to speak fluently, as though it was their first language, no accent or anything.
Sadly, there are not a lot of comprehensible input based learning courses out there as most focus on speaking early.
There are a number of programs that use a purely audio-lingual approach with translation, like Pimsleur or the <language>pod101 series. In both cases, when I tried them as a short term tourist, people were surprised how well they could understand me for the short time I’d been learning. These were tonal languages very few tourists tended to learn, so local expectations were low and just being comprehensible was a win.
Yeah I've actually done one of Pimsleur's courses for Mandarin. But that's not the same thing as it's just like most courses in that they have you speak early rather than focus on listening.
I had been learning my target language mostly from books (and then watching/listening to media for osmosis) for years before I discovered Pimsleur. I can confidently say that Pimsleur increased my skills by orders of magnitude. Being forced to 1) speak out loud, 2) create sentences under time pressure and 3) absorb language rules implicitly rather than explicitly was a major game changer.
My only disappointment is that Pimsleur only created 2 levels for my target language, so after I had repeated them a number of times I had to look elsewhere for other (less effective) materials. Recently, I've been using the ChatGPT Voice Assistant to try and approximate the Pimsleur experience - it works to some degree, and it's novel that I can ask it questions during learning, but it can't really offer a curriculum in the same way.
Pimsleur is very effective when just starting out a new language.
In general, I find that the most important thing to learn first with a new language is the pronunciation. Compared to learning thousands of words, it's not that difficult to learn to correctly produce the sounds of a foreign language. If you are able to pronounce things properly, native speakers will think much more highly of your abilities in the language (which can backfire when they assume you're fluent).
That's what we do at Latudio! [0]
We've created an app with listening-first approach and added some very useful features on top of that, like showing a sentence when requested and saving unknown words for later practicing.
There are three exercise types, one is prerecorded conversation where you pick one side and record your pronounciation so you can compare it with a native speaker.
I'd be curious to hear what you think about it.
[0] https://www.latudio.com
Really glad to see more apps that focus on sentence level speaking and listening. I wish the mobile landing page had more screenshots of the app since my target language is not yet supported.
Thanks. For now, kindly navigate here: https://www.latudio.com/whats-inside
What's your target language, if you don't mind me asking?
Thank you! I’m learning Korean.
Wish we could help. Good luck!
Also, we're happy to give the app away for free to nonprofits who'd find a use for it for their clients. It is also very easy for us to add domain-specific custom content, should there be an ask for it. If interested, PM me, e-mail is in my profile.
This sounds like a validation of Krashen's comprehensible input [0].
[0] https://en.wikipedia.org/wiki/Input_hypothesis
I built a system for this! Or, well, comprehensible input more broadly - it mostly uses text. Dreaming Spanish sorta takes this more audio-based approach.
I do question whether it's helpful to focus on audio without text. They're focusing on melody and rhythm, and it seems that listening without subtitles is better for that, but that doesn't get you understanding, while listening to comprehensible input with subtitles lets you get melody, rhythm, and actual vocabulary and grammar at the same time. It also lets you stretch "comprehensible" a bit further, since you have an extra source of contextual input.
Babies also have a lot of context for what the words they're hearing actually mean. I suppose there's the "watch translated peppa pig" approach for that.
My project (https://nuenki.app) translates appropriate-difficulty sentences into your target language as you browse, so you casually pick it up over time through comprehensible input.
I'm learning German and I've been enjoying browsing and reading with Nuenki. It's a lot of fun to see a sentence in context and find out you are simply able to understand it, though sometimes I have to fight the urge to instinctively hover to reveal the original text. Thank you for this plugin!
Dreaming in Spanish itself is a good example of a program that doesn’t rely on text to make its audio comprehensible. Their lower level videos include a lot of context that make it possible to understand the speech without written words.
Steve from LingQ once mentioned in passing on a Vlog that they have done studies where they gave students vocabulary lists (plus a vocabulary list free control group) before giving them a reading passage to study in a foreign language and they found that the vocabulary list group had worse outcomes. Does anyone happen to know which study this is? It has always stuck with me but I’ve never actually been able to pin down the exact study he was talking about.
https://www.sciencedirect.com/science/article/pii/S187704281..., perhaps?
"A Comparative Study on the Effectiveness of Using Traditional and Contextualized Methods for Enhancing Learners’ Vocabulary Knowledge in an EFL Classroom" where those who were given traditional vocabulary lists performed worse.
Good find. The description of the actual experimental conditions is incredibly vague, but it sounds to me a bit like the control group only got the vocabulary list with definitions and didn't read the short story, whereas the experimental group only read the short story and didn't get any definitions. Then the experimental group was able to successfully complete ≈10 of 15 fill-in-the-blank questions vs the control groups ≈8.
That is an okay experiment to compare the two teaching methods, but doesn't address the question of what happens when you combine them by reading a text with an accompanying vocabulary list. I would be rather surprised if additional access to definitions actually hurt learning. Like, what are dictionaries for, then?
A lot of the article’s findings about written translations and written transcriptions reflect my own anecdotes as a Korean language student. Listening and reading are distinctly separate skills but educational materials sometimes try to couple them together as interchangeable parts.
I’ve been working on a spaced repetition system that focuses exclusively on listening and speaking skills. It is not quite spicy enough to remove ALL writing (shown at the end of the review), but quizzes are done blind without showing written transcriptions.
Feel free to reach out if you have feedback or ideas: https://koala.cards
>Early studies have proposed that a putative “sensitive period” for acquiring the sound patterns of a language ends around age 6. Not coincidentally, this is the age when many children learn to read.
Socrates vindicated
Separate but related, a plug for Glossika (www.glossika.com), which primarily uses a listening and speaking approach to language learning. (I'm just a happy user.)
Humans can speak before they can spell or read. That could be a way to learn a language.
The same thing is argued in "Fluent Forever"
Babies are actually pretty bad at learning language. It's just that it's the only thing they have to do. Adults are much better at learning language, but they never put any time into it.
This is spot on. As a baby, you typically have two affectionate caretakers who dedicate a significant amount of their time to teaching you language/words, and also shower you with love/praise/attention for every little bit of incremental progress. Also, you're in an immersion program: you often can't communicate your wants/needs until you've learned the language, which adds another layer of incentive.
As an adult, if you were sent to an immersion program in a foreign country with two full-time foreign tutors/caretakers who loved you, like legitimately loved you with all their hearts, you would pick up that new language pretty darn quick.
Exactly; the advantage of input-based approaches is not their speed. You'll learn a lot faster if you spend the time on flashcards, grammar drills, and studying.
The advantage is that it's easier, and often more pleasant, to integrate.
That said, babies do have some neuroplasticity advantages.
We should play to our strengths - babies are hopeless at grammar drills, and adults aren't as good as babies at neuroplasticity - by applying our brain to the problem.
I say that, while having built a comprehensible input tool (https://nuenki.app). But it's useful as a complement to that study, as you can more readily run a browser extension or listen to podcasts than devote your entire life to focused study.
I think that's patently wrong. Babies and children in general absorb languages much easier than adults do. See children of parents speaking two different languages. By the time they reach school age they are fluent in both languages, and in the same time they need to learn how to be humans at the same time. Just because it's not learning as you conceive it as an adult it does not mean it's not there.
This is a common misconception. It is extremely rare for any child to be fluent in any language by the time they start school.
There are different levels of fluency that language training professionals recognize, mostly standardized on the ILR scale [0].
There are 5 levels on this scale. What people commonly consider "fluent" is level 3. There are also midpoints recognized by a "+", e.g. most people who study language seriously but don't have any special aptitude for language training get to a 2+. It's an open question of serious debate and active research of whether it's possible to train anyone past 2+ to 3. Some rare individuals seem to have an innate aptitude for language acquisition and they seem to breeze through their training to 3 and then continue on from there. Everyone else seems to plateau at 2+.
3 is just the level of "can easily be understood as a foreigner by a native speaker". It's not native level of mastery. "So fluent you can fool a native speaker" is 5.
All of that is to say that children are most definitely not fluent by the time they reach school age. Most children starting school will be at a 0+ level. Particularly intelligent children might be at a 1. But they are definitely nowhere near 3.
And indeed, by the time most people graduate highschool, I would struggle to admit that they are much beyond 3. Hell, I'd even say a large portion of people graduating college don't have mastery of their native language.
So no, I don't think children have a particularly large advantage in learning language. Despite significant, continue training in language for the first 18 years of their lives, most aren't able to get very far beyond what any studious adult can learn in 2 years.
[0] https://en.m.wikipedia.org/wiki/ILR_scale#ILR_scale
Edits: damn glass keyboards making my own fluency look suspect
So is this a bit about comprehensible input?
Another hackernews article where the title is clickbait and the experiment doesn't really honor the title implication.
Language learning is not easy nor effortless nor "just a few minutes per day" endeavor. Babies had 24/7 immersion and a completely different context.
You can do experiments and try certain things but the framing needs to be honest and not "we tried this thing in a very small environment and extrapolated like mad".
"Our research builds on previous studies, which have found that spelling can interfere with how learners pronounce individual vowels and consonants of a non-native language."
This matches very well with my own personal observations. Most language instruction focuses very heavily on written instruction, which can be damaging early on. When you're just starting out a new language, you should learn the sounds by mimicking native speakers, not by reading letters off a page. When reading, your preconceptions of how those strings of letters would be pronounced in your own native language can interfere with correct pronunciation.
Once you've got the pronunciation of the basic sounds of the new language down correctly, you can start using written materials.
The conclusion is extreme. Reading is too much of a benefit in language learning to give up on at all, it's the best source of the widest range of comprehensible native input you're going to find.
I learned my second language by reading out loud, and only going on if I was sure I was pronouncing the thing right. We have the internet now; if you think you're pronouncing something wrong in a common language, you can go on the internet and make sure you're pronouncing it right. [edit: I remember when people were using their phones to transcribe what they were saying in L2 in order to check their pronunciation; do people still do that?]
When reading out loud sometimes I would sing the words, because you can't come up with a convincing way to sing a sentence unless you understand what it means.
Orthography is of course going to be distracting in English or French (or Chinese) because they have terrible orthography when it comes to representing sounds. But you're going to want to be literate anyway, so eventually you're going to have to learn how to recognize but ignore the details of the words.
Subtitles are a problem because people shut their ears off while reading them (as the article claims to confirm.) But if you can understand the subtitles, you've improved your literacy and gained an understanding of that piece of media. Now turn them off, and listen while knowing what's being said. Shadow and repeat after the speakers. Check your pronunciation against them. Roll the words in your mouth while the meaning is rolling around in your head. Make sure you know and can repeat every word that's being said, and make sure you understand the grammar.
Reading, writing, speaking, and listening are all very different things. But I'm not just learning a language to speak to people, so every one if them is important. I will not understand the cultural references that people use in daily conversation to communicate meaning unless I read. I can't replace their entire upbringing in a language by chatting, even in a focused way, for a few years.
Learning a language as an adult is different than learning a language as a baby. As an adult, you have another language to compare everything to, and you already know how languages work. A baby will become extremely fluent in their first language before they ever learn to read. You already know how to read. Take advantage. If you want to work on your pronunciation, there are lots of ways, like chorusing, shadowing, reading out loud, Pimsleur and FSI give you lots of chances to compare your pronunciation to a model, etc.
Babies take 7-8 years to get good at a language. They're no faster than you.
My problem is how to overcome the translating in your head to being able to think in a language even simple things.
I was in Costa Rica and while all of the customer facing people spoke English, one guy who brought us water asked a simple question “desayuno o cena?” After thinking about it for a couple of seconds, I realized he was asking whether we wanted the breakfast or dinner menu. Even the Spanish I do know, it takes me way too long to understand it and speak it.