I saw this on the anki mail list (http://groups.google.com/group/ankisrs/b...d018eb1a7#).

http://www.babylonia-ti.ch/BABY207/PDF/mondria.pdf

Jan-Arjen Mondria Wrote:Summary: myths, facts, and didactic suggestions
In this article, I have tried to show that the facts about vocabulary acquisition are sometimes just a bit different or more complex than some myths would have us believe. Finally, by way of summary, I have set side by side the seven myths and facts, each case supplemented with a didactic suggestion (see Table 4).

Looks like a lot of double-speak to get a paper written:

Quote:Therefore, the conclusion should be
that the idea that the knowledge of a
couple of thousand words takes you
far is a myth. The fact is that this
knowledge takes you a long way, but
not far enough.

The myth wasn't that it took you 'far enough'. It was that it takes you 'far'. Which is the say as 'a long way'.

"Word lists are of limited value." - That's not a myth. It's a fact. Nothing is of 'unlimited value'. You can't learn Japanese just from word lists, so they are, by nature, limited.

The next part might be correct, or might not. There isn't any definitive proof.

Learning in context... His argument is that 'If you only provide one sentence, it might confuse the learner when another sentence is used' and 'It takes time to create sentences.' Well, the second one is stupid... There are plenty of existing sentences in the word. As for the first, the solution is simple: More context. It's like basting a turkey with a thimble of water and then saying "Nope, that doesn't work. Give up on that."

I'm not even reading the rest. It's obviously the didn't put any actual thought into it and merely regurgitated some crap they read on the net just so they could get a paper written.

This is another topic on someone else's thoughts on learning vocab.
http://forum.koohii.com/showthread.php?tid=5415

Anyhow,

Quote:Some words occur far more often than other ones. Consequently, knowing a relatively small number of words takes you far.

Information theory says all the information, or at least most of it, is contained in rare items. Common words like for, is, and the don't contain much information.

Quote:Research, however, has shown that learning words in semantic sets is not the best option. On the contrary, related words are more easily confused (interference) and learning them takes considerably more time than learning unrelated words.

I personally found this to be a problem only sometimes.
Like when Nakama is trying to teach you 借りる, 貸す, 返す in the same word list. They have similar meanings and all start with the same kana, making them easily confused.

Quote:Words whose meanings have been inferred from context are retained better.

I think they have the wrong myth here. It's important to be able to infer meanings when reading/listening because it saves you from having to use a dictionary.

Quote:Words learned productively are retained better.

Here, his data is flawed. A productive test is a harder test, so of course scores were lower.

We actually don't seem to have had a predictive model for semantic information of words in texts using information theory, comparing notions of entropy and self-attraction and frequency distribution and other terms I know nothing of, until pretty recently: http://forum.koohii.com/showthread.php?tid=3792

The most informative words are clustered in domains (I believe novels = 1-3000 word domains) where the most common words in those domains are the most informative, or somesuch.

As an aside, I think many people who have problems with sentences and end up arguing against learning words in context are simply not really learning words in context in the first place, they were performing more of a painstaking surface memorization of the sentence as a whole, a rookie mistake, easy to overlook and attribute to a problematization of contextual learning.

nest0r Wrote:We actually don't seem to have had a predictive model for semantic information of words in texts using information theory, comparing notions of entropy and self-attraction and frequency distribution and other terms I know nothing of, until pretty recently: http://forum.koohii.com/showthread.php?tid=3792

The most informative words are clustered in domains (I believe novels = 1-3000 word domains) where the most common words in those domains are the most informative, or somesuch.

As an aside, I think many people who have problems with sentences and end up arguing against learning words in context are simply not really learning words in context in the first place, they were performing more of a painstaking surface memorization of the sentence as a whole, a rookie mistake, easy to overlook and attribute to a problematization of contextual learning.

Could you explain to me how to avoid surface memorization?

sikieiki Wrote:Could you explain to me how to avoid surface memorization?

i+1

Multiple sentences

sikieiki Wrote:

nest0r Wrote:We actually don't seem to have had a predictive model for semantic information of words in texts using information theory, comparing notions of entropy and self-attraction and frequency distribution and other terms I know nothing of, until pretty recently: http://forum.koohii.com/showthread.php?tid=3792

The most informative words are clustered in domains (I believe novels = 1-3000 word domains) where the most common words in those domains are the most informative, or somesuch.

As an aside, I think many people who have problems with sentences and end up arguing against learning words in context are simply not really learning words in context in the first place, they were performing more of a painstaking surface memorization of the sentence as a whole, a rookie mistake, easy to overlook and attribute to a problematization of contextual learning.

Could you explain to me how to avoid surface memorization?

When I first started doing sentences, I used Mangajin/Japanese the Manga Way as a template: http://books.google.com/books?id=xaXukH7...&q&f=false (sans ローマ字, instead it was kana-only form below the kanji-kana form, and translation/grammar notes below that, each section divided by brackets)

It's just a matter of mentally deconstructing it into its constituent parts, and maintaining that awareness after parsing the information even while re-constructing it, so each time you see the sentence you're putting it together rather than trying to retrieve the whole sentence, and you continue doing so until it's second nature. By the time I had done a relative handful of sentences, I had the method down and started combining with individual word cards and other types of sentences and different kinds of focus. To me, at the core it's all the same, it's just a matter of how you arrange things into complementary components and integrate with real world use, whatever it might be.

Mondria provides Figure 1 (apparently from one of Paul Nation's works), and gives it only a passing mention as though it speaks for itself, when it really doesn't, and doesn't prove anything.
There are several blanks in the passage whose meaning could be fairly accurately guessed from the context. There are others that, while not clear enough to reveal their meaning conclusively, could at least lead to a general notion of their meaning, to be supported in further reading.

nest0r Wrote:

sikieiki Wrote:

nest0r Wrote:We actually don't seem to have had a predictive model for semantic information of words in texts using information theory, comparing notions of entropy and self-attraction and frequency distribution and other terms I know nothing of, until pretty recently: http://forum.koohii.com/showthread.php?tid=3792

The most informative words are clustered in domains (I believe novels = 1-3000 word domains) where the most common words in those domains are the most informative, or somesuch.

As an aside, I think many people who have problems with sentences and end up arguing against learning words in context are simply not really learning words in context in the first place, they were performing more of a painstaking surface memorization of the sentence as a whole, a rookie mistake, easy to overlook and attribute to a problematization of contextual learning.

Could you explain to me how to avoid surface memorization?

When I first started doing sentences, I used Mangajin/Japanese the Manga Way as a template: http://books.google.com/books?id=xaXukH7...&q&f=false (sans ローマ字, instead it was kana-only form below the kanji-kana form, and translation/grammar notes below that, each section divided by brackets)

It's just a matter of mentally deconstructing it into its constituent parts, and maintaining that awareness after parsing the information even while re-constructing it, so each time you see the sentence you're putting it together rather than trying to retrieve the whole sentence, and you continue doing so until it's second nature. By the time I had done a relative handful of sentences, I had the method down and started combining with individual word cards and other types of sentences and different kinds of focus. To me, at the core it's all the same, it's just a matter of how you arrange things into complementary components and integrate with real world use, whatever it might be.

It is impossible for me to do this. After hearing 円安で日本企業の輸出が増えた a thousand times over, after I hear 円安で I know what the rest of the sentence is and can not forget it. I would be lucky to understand 輸出 by itself or in another sentence.

Sounds like you're allowing the aural/prosodic aspect of the sentence as a whole and perhaps the English translation to carry the weight of your reviews.

Invest effort into mentally deconstructing the sentence, backwards and forwards, etc. (the backwards-->forwards thing was the only thing I picked up from Pimsleur). Take your time looking up the definitions, processing the sounds, writing out the words once or twice mentally or physically, subvocalizing, speaking, forcing yourself to actually read the sentence piece by piece as the sound plays/as you subvocalize, understanding how the elements fit together. It sounds like a lot, but having streamlined sentences that are systematically arranged at the onset (i.e. smart.fm/kore) really helps. Likewise with the 'i+1' mentality some have (though I prefer i+x).

I don't know how fast other people are with their reviews, but for me, I breeze through sentences really easily, always did, except when I first started out, because I tried doing all the KO2001 sentences as audio dictation cards--listening only, then writing them out, then flipping and reading the sentence. Once I realized the value of multiple modalities and how much my own efforts per card mattered, I completed those foundational sentence sets as recognition sentences. ;p