Nest0r's methods

From Learning Japanese Wiki (RtKWiki)

Jump to: navigation, search

Contents

Overview

Strategies

Tutorials

For now, I'll begin with some step by step instructions on various strategies I use.

New cards in Anki

I have devised a method for studying new and failed cards in Anki based on spaced retrieval research, primarily the publications by Karpicke & Roediger.

This tutorial will be based on how I do my single-word vocabulary cards. You can apply this technique to any card type, but I will use my own layout here as a concrete example.

Using Rikaisan in conjunction with the shared Anki plugin called Realtime Import, both tools created by the genius philanthropist forum member cb4960, I create cards while reading Japanese text in Firefox simply by running the mouse over a given word and hitting a keyboard key, as long as a deck is open in Anki. I have the aforementioned single-word vocabulary deck, which I call my extemporaneous vocabulary deck, open in this case.

Once a batch of new cards is gathered and waiting in Anki, for example 25 new cards, we can begin.

  1. In Anki, go to Settings → Deck Properties → Advanced and the bottom 2 checkboxes should be empty; otherwise disable Per-day scheduling and Show failed cards early. Set Initial button 3 interval to 0.8-.95 days.

  2. In Study Options under the New Cards tab make sure you have New Cards/Day set to a high enough number to accommodate your desired amount, set Display Order to Show new cards in random order and Show new cards before reviews. Under the Reviews tab set Max Failed Cards to 0, and set Display Order to Review cards in order due, Show failed cards at end. No timeboxing with this method, for new/failed cards.

  3. If you have more than your desired number of new cards for this session, go to the Card Browser by hitting Ctrl+F and search for “is:new” by typing (if you're neurotic, feel free to type it without looking as soon as the browser opens), then suspend the excess new cards, for now. Select those 25 new cards left over and tag them all “Limbo”. If you have failed cards, temporarily suspend the “Limbo” cards, then clear your failed cards with the method described in steps 7-8 for studying failed cards, then unsuspend the “Limbo” cards.

  4. Now then, you should have “Remaining: 0 0 25” at the bottom of the main Anki window, if you have 25 new cards you want to study. Start studying by hitting the Review button from the Study Options menu. Flip the card so you have all of the information on the Front and Back displayed for your perusal. Use your own study methods.

    I use mnemonics and multiple senses such as audio and images and writing out the words. That is to say, I use quick mental tricks to chunk the information and/or make it meaningful to me for that particular study session, I listen to the audio, subvocalize it, speak it aloud, and write out the word with a stylus and touchscreen device (pencil and paper also works).

    For my single-word vocabulary cards, all of which are tagged ‘vocabulary’ in this deck, I use the Audio Download shared plugin(s) to instantly download JDIC audio for the words at the press of a button. For this first review process here in Step 4, I study the cards, adding audio as I go. To cycle through the cards I hit Failed after studying each one. This doesn't matter as we will reset the cards as new in the next step.

  5. Open the Card Browser by hitting Ctrl+F, and type “is:failed”. Select the cards you just studied (which should be the only cards displayed since we cleared the failed cards in Step 3), and under Actions, Reschedule as new cards.

  6. That was the study session of this phase. Now for the testing session. Wait 3-10 minutes, then test yourself on the new cards, by hitting the Review button from the Study Options menu and letting Anki show you the Front of the card, in my case the kanji word and audio file and/or kana or furigana, and given this multimodal cue, you have to retrieve the meaning. Make sure to always flip the card and look at the Back for feedback, even when you pass the card. In my case this includes the complete sentence the word came from, the definition, and the URL. Pass and fail the cards depending on whether you got them right. If you pass, it doesn't matter which button you hit, you'll be resetting again.

  7. Go to Study Options to avoid a glitch I have detected where sometimes audio for the displayed card for review in addition to the audio in the preview will play when you preview a card, open the Card Browser by hitting Ctrl+F and type “is:failed”, then hit F2 to preview each failed card so you can restudy it from there.

  8. When you've finished restudying each failed card, close the Card Browser, wait a few minutes, then review the Failed cards. It doesn't matter what button you pass them with as we're resetting soon, but fail the cards you get wrong and repeat Steps 7-8, restudying with F2 preview and then testing the Failed cards, winnowing them down to 0 Failed, until you've passed all of the new cards, 25 of them in this example, one time.

  9. Open the Card Browser when this study/testing batch's “criterion level” has been reached (all of the cards successfully tested once), display cards tagged “Limbo”, select them, Actions, and Reschedule them as new cards. Under Actions → Delete Tags, remove the “Limbo” tag.

  10. Come back in 3-4 hours and review these now new cards again. Follow the same criterion level of one pass each, restudying failed cards in the card browser with F2 preview (I recommend doing all failed cards this way, as long as you have enough of them to make it worthwhile, like at least 5). This time when you pass them, grade them with initial button 3, which is set to .8-.95. The idea is that sometime in the next 18-24 hours you want to review them again. The next day/night when you review them, you might want to grade them with button 4, because the small interval we set for initial button 3 will temporarily shorten that day's intervals. After this, review the cards as you have always done, spacing them days then weeks then months apart.

What is key here is the initial learning period with three study/testing phases separated by 3-4 hours and then 18-24 hours, wherein you study and test until you reach criterion level, having passed all cards once per study/test phase.

‘Extemporaneous’ Vocabulary Deck Layout

On my deck layout for the tutorial, plus some short notes on other topics, which I will expand in the future.

  • First, I like to disable suspend leeches and set leech threshold to 3 failures, in Settings → Deck Properties → Advanced. The 3 failures is a number I plucked from Karpicke & Roediger's research that I found reasonable, but this way I get reminders that a card is a leech, rather than having it suspended. What I tend to do is add an image to the Front of cards I fail 3 times, or sooner, with the shared Anki plugin, Image Download.

  • On the Front, I have the kanji word, the kana word, the audio, and the image. On the Back I have the definition, the sentence, and the URL, which is set up in the Layout like so, to open Firefox when I click it: <a href="{{{URL}}}">{{URL}}</a>

  • My Rikaisan settings for the Realtime Import plugin are as follows (under the Clipboard & Save tab in Rikaisan's options): Anki tags: vocabulary; Save format: $d$t$r$t$n$t$s$t$h$t$u$t$b; Field names: Kanji Kana Definition Sentence Highlighted URL Cloze

  • I place the multisensory information on the Front in order to make the cue robustly transfer-appropriate, and because this allows me to modulate the informativeness of cues, such as by adding images to more difficult cards, or unveiling them incrementally. So for cues in this deck, I think in terms of modalities. For targets, I think in terms of types. That is, the focus of the card, which here is a word's meaning, which I also modulate by refining that understanding with the extra contextual information, should I choose.

  • The other sense is tactile, using muscle memory, through handwriting. This increases recognition and recall of information. I will link to references and explain further later.

  • Subvocalization, or more generally when you articulate language in your mind, is a process that I think should always precede speaking aloud in a learning session, which I will also explain in the future.

  • Mnemonics, using relational strategies, elaborating information, and chunking, are all important skills that studies show greatly enhance memorization. It is ideal for language learning. They don't have to be highly structured and long-lasting; most of mine I improvise in an instant, then forget. What's important is that they help encode the information per encounter.

  • On feedback (flipping the card and looking at the back), studies show that feedback, even when you got the answer right, improves recall. Part of the reason why is that retrieving a memory renders it plastic, so you can strengthen it further.

Extracting collocations from electronic texts with AntConc

Collocations are very valuable for language learners. I may explain more in detail in the future. For now, suffice to say that if you have electronic texts in Japanese that you wish to use as study material, that in addition to extracting words (using MorphMan, also to be explained later), you can extract chunks of the target language that co-occur with a high frequency.

Go here: http://www.antlab.sci.waseda.ac.jp/software.html and download the relevant AntConc file. If it's the Windows .exe, open it (no installation that I recall), and:

  1. Global Settings→Language Encodings→UTF-8 or Shift_JIS

  2. File→Export Settings→Default name (antconc_settings_320.ant) in same folder as .exe so that when the program starts it uses this and you don't have to keep selecting the language. You can always import settings files also.

  3. File→Open File(s)/Open Dir.

  4. With those files showing up in a column to your left, go to the Clusters tab.

  5. Check ‘N-Grams’ box, and the Clusters tab will now say N-grams instead of Clusters.

  6. Set N-Gram Size to Min. Size 1 and Max. Size 20 (For some reason you have to scroll up/down rather than type numbers.)

  7. Min. N-Gram Frequency = 4

    (Those above size and especially frequency settings can vary, but I find the size works best that way due to how Japanese is processed in AntConc. Probably want much lower for English or somewhat lower for larger numbers [like triple digits] of Japanese texts.)

  8. Sort by Word or Frequency. You can always re-sort by clicking Sort. Sorting by Word is useful because of how n-grams are processed: http://www.antlab.sci.waseda.ac.jp/software/AntConc_Help/N-grams/Overview_(N-grams).htm (One set of n-grams for ‘This is a pen’ would be ‘This/This is/This is a/This is a pen’, so word-sorting lets you skim more easily to that final line.)

  9. Click Start!

  • You can copy the results manually by clicking slightly to the side on the same line, or click directly for Concordance, or File→Save Output to Text File. You can also duplicate and undock the results window by clicking Save Window.
  • Also, because we're using default n-gram settings, it's only processing words, not punctuation, numbers, symbols, etc. You can enable those in the Global Settings but it skews the results quite a bit so I just leave it and when I see results that have gaps in them indicating where those tokens (question mark, number, etc.) used to be, I click it to see the context in the Concordance tab. In those cases you'll get no hits because it uses that punctuation-removed line as a search query, so you'll need to select just a section of it and re-search, e.g. if you click だけど それは in the search results it'll take you to the Concordance tab and say No Hits, but just re-search for だけど or それは and sort by 1L or 2R or whathaveyou.
  • For text files that are formatted with Aozora or HTML markup, you might use Aozora Remover here, to clean them up for AntConc processing: Community Tools. You might wish to use the Aozora Gaiji Replacer from there also.

AntConc's author is thinking of incorporating ‘concgrams’ into a later version. Concgrams account for all the constituency and positional variation, e.g. ‘government expenditure’ can be ‘government's own expenditure’ or ‘expenditure of the government’, but normally this wouldn't get picked up by linear-only searches. There's some tools that do this automatically for texts, but they seem to be hard to get. You can manually achieve a similar effect in AntConc by searching for ‘government’ and adding the context word (in Advanced settings) ‘expenditure’ with a horizon (another instance of selecting the span, this time within how many words to the L or R of the search term).

Links

Books

This is a simply written book I discovered thanks to a video that RevTK forum user Thora posted. It is a good primer for various learning improvement concepts, particularly the concept of multisensory integration.

Papers

An accessible paper written by Paul Nation, a well-known linguist, on principles of language learning. Very useful for its well-rounded summary of different elements that are important for learners, based on actual research on what works.

Abstract: The activities in a language course can be classified into the four strands of meaning- focused input, meaning-focused output, language-focused learning and fluency development. In a well designed course there should be an even balance of these strands with roughly equal amounts of time given to each strand. The research evidence for the strands draws on the input hypothesis and learning from extensive reading, the output hypothesis, research on form-focused instruction, and the development of speaking and reading fluency. The paper concludes with 10 principles based largely on the four strands. The strands framework and the principles provide a basis for managing innovation in language courses.

Another accessible paper by respected researchers summarizing principles of learning, including spaced retrieval.

Abstract: Recent advances in memory research suggest methods that can be applied to enhance educational practices. We outline four principles of memory improvement that have emerged from research: 1) process material actively, 2) practice retrieval, 3) use distributed practice, and 4) use metamemory. Our discussion of each principle describes current experimental research underlying the principle and explains how people can take advantage of the principle to improve their learning. The techniques that we suggest are designed to increase efficiency—that is, to allow a person to learn more, in the same unit of study time, than someone using less efficient memory strategies. A common thread uniting all four principles is that people learn best when they are active participants in their own learning.

Resources

Tools

Speculations

Back to top

Retrieved from "http://rtkwiki.koohii.com/wiki/Nest0r%27s_methods"