Is extensive or intensive reading faster for vocabulary acquisition?

[PeterBormann]

Hi guys,

Sorry, no time for longer online discussions right now.
But just a hint:

why would they stick at all.
using ultra-reading while listening?

• If you can’t guess the words from context
  (i.e., you don’t understand the meaning of the
  words / word groups)

and

• (even if you understand the words / word
  groups, but) they don’t occur often enough,

not much of the new vocabulary will and can stick.

However, that’s to be expected, right?

If you apply this RWL technique in the right way (selected material,
right number of unknown words in a text - say: ca. 10-15 percent of unknown words or less, varying audio speed, etc.) with the appropriate
" enhancements" (re-listening and use operations à la L1->L2 based on LingQ-to-Anki,
etc.), then it works quite well.

I’m applying it for Dutch at the moment, and I’m very satisfied with the results so far…

That said:

1. This technique doesn’t work with every text genre, esp. not with complex
   literary / poetic, scientific and philosophical texts, and it doesn’t work
   with texts where there are many unknown words (say, more than 10-15 %
   unknown words) either. You can’t rush through such texts…

2. My main criticism reg. (ultra)reading-while-listening is this:
   It doesn’t do much for use-acquisition. But that seems to be the
   problem of all input “purist” approaches (AJATT, MIA, Refold, etc.).

At least, in my experience, their efficiency is far too low (it’s almost like trying to get better at sprinting by jogging longer and longer distances at a slow pace).

Therefore, you need more use-operations (see the RWL “enhancements” mentioned above) to counter that…

Nice Sunday
Peter

PS -
BTW, no one should rely exclusively on “guessing from context” as a vocabulary acquisition strategy when applying (U-)RWL…

[Gigusek]

I try to mix it up. Sometimes I’ll read intensively, sometimes extensively.

When you read extensively you can also mark words that you’ll process later and/or import to e.g. Anki, that way you’ll still take care of the unknown words but not stop reading to do that (in fact it’s probably even more productive to do it in bulk like this).

My personal preference when targeting increasing my vocab (when you say acquire you’ll encounter some hardcore CI-hypothesis fans, just a headsup ) I prefer what you described as the semi-intensive reading for the majority of words, and intensive for the most difficult ones. I typically don’t lose time on seeing the entire sentence translated when I know what it means, but when the word just “doesn’t make sense” I might take some time on looking into it.

But it’s also worth doing some extensive reading just for the sake of it, for enjoyment, and to not go crazy and feel like using a dictionary is synonymous with the activity of reading. It also gives you a chance to train the mental muscle of letting go of ambiguity and accepting that what you come across might not be 100% clear to you.

[nfera]

@S.I

If unknown words are equal to the unknown in the sense that they’re flying by with the same speed and intensity of the attention payed to them, why would they stick at all. Don’t know how it works for those who manage to acquire vocab through this technique.

This technique may work with a very high known word ratio, because then you really pay attention to the unknown word. And if you are at a high level, you can follow the story in the background, as your main focus is trying to quickly work out the meaning of the new word. But, yeah, I think the ability to control your pace would also solve this issue, which is only possible without the audiobook (or pausing it all the time). Alternatively, slowing down the audio to allow for more time with unknown words. Fast audio is definitely great for drilling words you are somewhat familiar with though.

@davideroccato

Unfortunately, many of these things are very difficult to measure because there are too many variables to consider, otherwise it would be nice to know what it would be to crunch 10M words fast and 8M much slower on the same amount of time. Who would come out with the best overall language management? Difficult to say but I root for the slower.

@davideroccato There are indeed many variables. The comparison for me would more be like reading while listening on LingQ at ~55 wpm, reading while listening on YouTube with Language Reactor at ~150 wpm (I just tested it with the Harry Potter audiobook on YouTube with auto-subs at 1.3x, but having to repeat sentences every now and again) or true extensive reading while listening at 1.65x/~240 wpm. In this case, using my reading while listening speeds, we would be talking about 10M words read as extensive reading while listening, ~6M with Language Reactor and ~2.5M with LingQ. As mentioned, the LingQ reading speed is so slow because of the amount of time it requires to get a decent definition (which often includes pausing the audio to open up a dictionary and write a definition). With the translation under the subtitle on Language Reactor (aka bilingual text), you can merely glance down often to get a good definition, without ever having to click anything, hence the increased reading speed. Which of these three options would you back?

One thing I would like to mention with my above experience of the Harry Potter series is that perhaps my analysis was flawed, as I may not have been able to recognise many of the words in list format, but I could recognise/guess more unknown words in context. So perhaps this means they are partially in the process of being learnt, but just not well enough yet to be recognised in list format, devoid of any context, which generally requires a high familiarity of a word.

@PeterBormann What kind of wpm/wph speed are you looking at for your extensive reading while listening?

[GMelillo]

@S.I

If unknown words are equal to the known ones in the sense that they’re flying by with the same speed and intensity of the attention paid to them, why would they stick at all. Don’t know how it works for those who manage to acquire vocab through this technique.

Exactly. I don’t think is what @nfera was thinking, but sometimes I’m left with the impression that people believe that reading increases vocabulary because you are just passing words through your head, and somehow, without you having any subjective sense of constructing meaning, or drawing inferences, you will learn those words.

Maybe there is some black box subconscious mechanism at work for some people, but in my experience when I learn a word in context it is because the context is giving me enough information to infer the meaning, or at least infer something about the word. If I can infer absolutely nothing about the meaning of the word - if I don’t have some sense of what that word at least could mean in that sentence. even very vaguely - I don’t seem to learn much from simply reading it.

So again, one way or another, whether clicking on the dictionary or drawing your own inference, there is some subjective arrival at some sort of meaning of the word.

There is a very narrow subset of inferences in context that are so effortless and so immediate that it seem as if I just learned the word for free. But this this is like winning the lottery. Most of the time I’m mentally engaged in the process of reading, and I’m aware, at some level, that I am drawing inferences.

Personally I think that drawing my own inferences makes a stronger connection, so I try to do that as much as possible, but while the connection is stronger when I draw my own inference, it’s not really that much stronger, especially if do some on the spot reinforcement, e.g. by re-reading the sentence with the new word out loud in a meaningful way, than when using a dictionary.

@S.I

But as the natural (even more than)-SRS it works fine and deserves to be among the best tools for making the language learning process to be more engaging.

Exactly. And cycling through repeated readings of books is also a great way to guarantee that you reinforce new words. The pattern of read → two months later run the flashcards for a few days → re-read or listen or read while listening works for me.

I think that lots of reading is good because reinforcement in some sort of meaningful context is better than SRS-ing a dictionary definition. This might be why some people also swear by sentence mining or close deletion for vocab acquisition. If I could pull out a few sentences around every Lingq and review just that (I think Kaufmann said he sometimes does something like that - skipping ahead in old lessons to lingqs) that would probably be marginally more efficient. Exporting lingq-ed sentences to a SRS is something similar. It’s all kind of the same thing.

[PeterBormann]

Hi nfera,

what kind of wpm/wph speed are you looking at for
your extensive reading while listening?

Depends on the language level:

• A1-B1: audio speed 1.x (however, I also use “warm up” approaches such as grammar light and
  various SRS for the most freq. words / sentences) → there is nothing “ultra” or “extensive” at this level.
• B1 or B1-B2 (for reaching B2-C1 / C1): audio speed ca. 1.25 - 2x (depending on the difficulty of the book, podcast, YT vid, and Netflix series) → the “ultra” variant with increased speed and extensive reading starts here
• C1 and higher: audio speed ca. 1.5-2x (U-RWL or only reading / listening)

Texts:

• No fiction before a C1 level
• Everyday language (many dialogues) / shorter (unknown) non fiction / longer (known) fiction

But, of course, I use several “enhancements” (always: re-listening and L1->L2 SRSing) The specific mix of tools / approaches depends on the closeness / distance of the L2 and the language level, though.

Sorry, that’s all the time I have…

Nice week
Peter

PS -
Reg. your low-freq word / collocation problem mentioned above
That’s def. a job for an SRS (LingQ-to-Anki / Migaku-Anki / Anki and Co alone).

If you import the text first into LingQ, mark the low-freq. words / collocations
and then learn it by means of an SRS, it’s a straightforward process.

PPS -
For your new vocab problem reg. extensive reading
It’s best to jump to the unknown words / collocs / sentences first and mark them in LingQ.
Only after that, it makes sense to apply RWL / URWL.

Re-listening and L2-L1 SRSing help with long-term memorization (extensive reading alone
does not - at least, if the occurrence of this vocab is <= ca. 20). But, of course, using
this vocab is even better

[PeterBormann]

Hi GMelillo,

Well said!

Only one point:
We should distinguish here between two kinds of acquisition:

• Recognition-acquisition
• Use-acquisition

Reading a lot helps primarily with the recognition, but less with the use of vocab.

The inefficient part of “input-only” approaches comes into play when they want to tackle
the use-acquisition problem mainly by relying on more and more and more… recognition
operations…

Have a nice day
Peter

[nfera]

@PeterBormann So you go through the text first, lingQing New Words (by pressing the right arrow on the browser, I’m guessing), then read while listen to the lesson? You don’t read while listening, pausing the audio whenever you have to open up a dictionary / edit a definition? What effective wpm/wph average are you looking at? Yesterday, I see for Dutch you read 5,190 words. So assuming you did your two Pomodoros of 25 minute = 50 minutes, so 5,190 / 50 = 103 wpm for your semi-intensive reading (of 10-15% New Words for a 6k length lesson).

[S.I]

@PeterBormann

• If you can’t guess the words from context
  (i.e., you don’t understand the meaning of the
  words / word groups)

and

• (even if you understand the words / word
  groups, but) they don’t occur often enough,

Even if I do can guess them from context (almost every time), but they don’t occur often enough, I’ll more likely just stick with their more basic synonyms. To find such a material with just the right number of new words repeated the right number of times is kind of utopic. Sometimes I just happen to find myself having read something like that coincidentally and yeah, it works. But, most of the time, I have to deal with various range of unknown words, because factoring the time I’d have to spend searching for ideal materials with just right percentage of unknown words would be over-optimisation.

@PeterBormann

If you apply this RWL technique in the right way (selected material,
right number of unknown words in a text - say: ca. 10-15 percent of unknown words or less, varying audio speed, etc.) with the appropriate
" enhancements" (re-listening and use operations à la L1->L2 based on LingQ-to-Anki,
etc.), then it works quite well

And this is more suited fot the nfera’s semi-intensive term.

I’ve been doing this and still do, but what exactly is ultra- about this? Even if you increase the speed, but still need to stop and spend a lot of time on the traditional looking up the words and SRS-ing, re-reading and re-listening.

@GMelillo

black box subconscious mechanism at work

With the label on it: “I JUST DON"T GIVE A F****”

[PeterBormann]

Here are some quick answers:

@nfera

go through the text first,

At your level (around B2 / B2-C1):
I first jump to the unknown words and
mark them (usually phrases / collocations).
But I don’t read the entire text.

Then biz as usual…

You don’t read while listening, pausing the audio whenever you have to open up a dictionary / edit a definition?
Usually I never pause.

“So assuming you did your two Pomodoros of 25 minute = 50 minutes,”
In Dutch, I’m still at a <= B1 level. Therefore, the procedure is as follows:

1. RWL: audio speed 1-1.5x (the Dutch podcast is too slow for my taste)
2. Re-listening without reading while doing other things.

Rinse and repeat for ca. 25-60 min daily - depending on my time budget.
(plus SRS: later in the afternoon / evening - if possible)

A “purer” RWL / U-RWL would be:

• two full Pomodoro sessions à 25 min: RWL
  (re-listening 1-2x in the afternoon / evening / next morning).

However, I don’t have the time for that right now…

[davideroccato]

@nfera @PeterBormann These metrics 1x, 1.5x etc. have no meaning from a measuring point of view. How do you calculate wpm/wph? That would be a more serious metric to consider and compare.

The problem is that everything is depending on the speed of the speakers and they change all the time.

I try to fine tune it by listening to stay more or less at the same speed (considering LingQ doesn’t allow to have many speed choices and it is not even consistent on different OS) but how do you calculate for each LingQ lesson how many wpm do you process?

[PeterBormann]

Hi S.I.,

@S.I

still need to stop

You don’t pause. See my answer to @nfera:
At a B2 (or higher level), you either jump to the unknown words first and mark them
or guess them from context (ideally).

SRS-ing
… useful for 2 use cases:

• High freq vocab at the beg. levels
• Low freq words / collocations from B2-C1 onwards.
  (U)RWL is not (!) sufficient here bc. low freq words don’t
  occur often enough.

However, I use them for L1->L2 testing all the time.
But using them is even better…

re-reading
Usually not necessary.

re-listening
Definitely: repetitio mater studiorum est.
And you can do this while doing other automated
activities.

Have to go now…

Nice day
Peter

[PeterBormann]

@davideroccato

a more serious metric to consider

Hi Davide,

LingQ gives you the number of words read in a (U)RWL study session.
Normally, I write them down on my LingQ profile (for Dutch - I also did that
in Spanish and Port. in 2021 / 22, but not anymore) and can relate them to the
time spent.

Herzliche Grüße
Peter

[davideroccato]

@GMelillo

Exactly. And cycling through repeated readings of books is also a great way to guarantee that you reinforce new words. The pattern of read → two months later run the flashcards for a few days → re-read or listen or read while listening works for me.

@GMelillo This is something I have started to understand with this method as well. I don’t do the flashcards because I want to simplify the process but I came out with the conclusion the read/listening again the same book over and over after few months is a good way to reprocess everything and convert many yellow words by a natural way. That conversion is very strong. Here “chatGPT style” could come in handy with flashcards or stories created with missing words, in the same subject of the book read.

Another thing I’ve been considering by other people, and something I do randomly for now, it is to read/listening the same book in many different languages.

[nfera]

@davideroccato

How do you calculate wpm/wph?

@davideroccato If it’s a book as extensive reading, you get the total number of words of the book then divide it by the length of the audiobook. Eg. 239k/27 hrs = 8.8k wph. To get the wpm, you divide it by 60. = 147 wpm. If you are listening at speed, then you multiple this by the speed. 147 x 1.65 = 242 wpm

For measuring my LingQ speed, you start a stopwatch on your phone, do your lesson, then stop the stopwatch, when you are finished. Check your LingQ stats to see how many words you read. Eg. 3,000 words read in 30 minutes = 3000 / 30 = 100 wpm

For checking my speed on Language Reactor, I do the same thing as LingQ with the stopwatch, as I import the YouTube video into LingQ afterwards to add 1x listened and 1x read.

P.S. @PeterBormann Yeah, I’m at an upper intermediate level with Italian, so it’s not really compatible to compare your reading speed on Dutch, as an upper beginner/lower intermediate.

[davideroccato]

@PeterBormann

LingQ gives you the number of words read in a (U)RWL study session.
Normally, I write them down on my LingQ profile (for Dutch - I also did that
in Spanish and Port. in 2021 / 22, but not anymore) and can relate them to the
time spent.

@PeterBormann LingQ’s number of word read in unreliable if you want to seriously calculate and understand a system about it.

But the problem is that, with your concept, for example, and saying someone “should” read at 1.3x or 1.5x is too vague.
It would be more precise to say, you should read at 150 wpm or 180 wpm and so on.
But these figures should be calculated, by the system/software, before starting the lesson. The software should calculate the speed of the speaker and give me back the exact speed I need to listen to, with consistency, before I start the lesson. So, if I read only 1 page, I will know that I’m reading at that precise speed.

I constantly switch from one language to another and I leave pages half done to continue the next day. LingQ is very unreliable on how it calculates words read in all those situations.
I know that you try to do your best on how you calculate and measure your read words BUT then when you output that to other people as a method, it becomes not so easy to put in place.

If you tell me to read at 1.5x it doesn’t make sense because the speaker can be slow or fast depending on the lesson.
If you tell me to read at 150wpm I should have a system that guarantees me before I’m reading that I will be at 150wpm.

I’m just telling you this so that you can consider it if you are building/creating anything about it. Or you are probably considering it already.

@nfera: thanks for the process, the same answer applies here.

[PeterBormann]

@davideroccato

It would be more precise to say,

Final word for today
"It would be more precise to say, "
“Cum grano salis” is enough.
And usually I don’t care about words read / word known or any other stats.
Gimme two Pomodoros a day (or rather a whole routine: 2 Pomodoros → re-listening → SRSing or other “use” operations). The rest takes care of itself - in the long run.

[davideroccato]

@PeterBormann yes, but that is for you, I know that. I thought you were creating a method about Ultra reading/listening. Or a book or a software about it. Did I have the wrong impression?

[nfera]

@davideroccato

But the problem is that, with your concept, for example, and saying someone “should” read at 1.3x or 1.5x is too vague.
It would be more precise to say, you should read at 150 wpm or 180 wpm and so on.

@davideroccato It’s not really necessary, because normal speech is at a pretty standard speed. It’s generally around ~150 wpm. An extra few words more or less doesn’t really change much. There’s a big difference between 300 wpm and 100 wpm in listening speed, but very little between 145 for a slow talker and 155 for a faster talker.

LingQ’s word counting system is not inaccurate. I agree the automatic counting system as you are reading doesn’t work if you are jumping back and forwards between pages all the time, but if you aren’t, it works completely fine. At the end, you can adjust it to equal 1x. Counting how many words are in a lesson is a pretty simple 1 + 1 thing. You can see how many words are in a lesson on the desktop. You think this count is wrong?

Edit: the issue of reading speed comes when you are doing all the extra faff of not reading while listening. Eg. pausing the audio to check the dictionary, pre-reading a text like @PeterBormann does, or skipping back on Language Reactor when I missed a word. This is what really reduces the actual amount you read per hour / effective reading speed. Sure, I may have the audio on 1.25x or whatever, but all the stopping to open the dictionary, means that at the end of the hour, I actually only read ~3,300 words (i.e. at 55 wpm).

[davideroccato]

@nfera

It’s not really necessary, because normal speech is at a pretty standard speed. It’s generally around ~150 wpm. An extra few words more or less doesn’t really change much. There’s a big difference between 300 wpm and 100 wpm in listening speed, but very little between 145 for a slow talker and 155 for a faster talker.

@nfera so, you are saying that if I listen to an audiobook talked very slow where I can easily choose 2X and a Spanish fast speaker where I could struggle at 1.1x, the fork would be only 145/155 wpm?

[nfera]

@davideroccato Aside from beginner content, which is deliberately very slow, most native speakers speak at around the same speed. Obviously, it may differ across languages/cultures, because like a word in Chinese is different than in, say, Italian, but I imagine in Chinese most people speak around the same speed amongst each other. We are not talking about the one old man down the street, who speaks incredible slow, but rather most people. Most of your friends probably speak around the same speed. If someone is reading an audiobook at 100 wpm, they would get fired…

Obviously, @PeterBormann’s advice for listening on increased speed is not meant to be taken as “you always have to listen at this particular speed and if you don’t, you’re a failure.” There are many things, which affect at what speed you can listen to the speaker at, such as: familiarity with that voice/accent, how clear they speak, how clear the audio is, familiarity with the subject matter, etc. You just try to listen at a higher speed. Obviously, you adjust the exact speed for every individual circumstance, whether it be 1.25x or 1.30x. If you say to listen at 175 wpm, people don’t understand you. They do understand 1.25x though.