For a moment, forget all about language classrooms and textbooks and teachers. How does a person naturally learn a language? By exposure, exposure, exposure. We learn language by being immersed in the culture where it is spoken. We hear millions of sentences spoken, and each time, our subconscious mind associates something to the sentence. The powerful subconscious mind works constantly to find patterns in everything, and so it discovers the hidden patterns in the sentences we've heard, and thus the first fruits of language begin to blossom within us. As we grow in the language, we can parse and understand new and exotic sentences we've never heard before, just by comparing them to sentences we have heard before. Connecting the dots, if you will. (Read more on this "connecting the dots" phenomenon in my article, How The Mind Learns)
If you want to learn a language naturally, with the kind of learning that produces perfect native fluency, why would you do anything other than soak up millions of sentences? Oh, right, because it takes years and years to learn a language that way. Well, it doesn't have to. And it doesn't, if you use the cutting-edge language-learning technology of Sentence Mining.
Sentence Mining makes use of a fairly recent technological innovation, the spaced repetition system, or SRS. An SRS is a computer flashcard program (like Anki or Mnemosyne, both free) that lets you make and review flashcards. The thing about an SRS, though, is that when you review a flashcard, you rate how well you understood it. The program uses your ratings to determine when it's best to show you the flashcard. In this way, the old process of reviewing flashcards gets super-optimized. But wait, aren't flashcards just for rote memorization, the sort of thing which is entirely unnatural for language learning? No! Flashcards have another use, having nothing to do with rote memorization, and that use is Sentence Mining.
Sentence Mining is the process of making flashcards, where the "Question" side of the card is a sentence in the target language. To "answer" the card, you don't try to recall anything from memory, instead you merely try to comprehend the sentence, by any means necessary. It doesn't matter whether you parse the sentence using your knowledge of grammar and vocabulary, or whether you just somehow know the sentence's meaning because you're familiar with the book it came from, or whether you memorize the sentence's meaning outright. All that matters is that, somehow, you comprehend the sentence. If you comprehend the sentence, rate it based on how easily you comprehended it. If not, fail it (rate it "Again" in Anki, or "0" in Mnemosyne).
The "Mining" part of "Sentence Mining" is the part where you go forth into the world and collect these sentences. Every book, every video game, every website, every textbook, every movie in the target language, is a goldmine waiting to provide you with linguistic fluency.
WHY DOES SENTENCE MINING WORK?
Think of the language processing center of your brain as a jungle. When you encounter a sentence, the sentence gets launched into the jungle and follows the path of least resistance, and wherever it ends up, that's how you understand the sentence. If you don't know anything about the language, then the necessary paths have not been blazed. The sentence crashes into a tree, and you "understand" it as freakin' gibberish.
As you review sentences you can comprehend, you carefully push them through the jungle to the right destination so that you understand them correctly. As you push them along, they gradually wear paths through the thick foliage. The more sentences you comprehend, the deeper and more sophisticated the paths become.
The cool thing is that the paths blazed by the sentences in your deck, work just as well when you encounter entirely new, exotic sentences you've never seen before.
Eventually, you've comprehended so many sentences, that the "paths" in the jungle have evolved into an intricate modern superhighway. Now, whatever sentence you launch into the jungle, it'll easily, quickly get to the right destination. Congratulations, you're fluent at reading in your target language!
HOW TO CHOOSE WHICH SENTENCES TO MINE
Any sentence which you can comprehend, is good. When you get to an intermediate or advanced point, though, that'll be too many sentences for a busy important person like you, so we'll need to weed them down.
First, identify a "sentence target". A sentence target can be:
* a particular vocabulary item
* a particular type of verb conjugation
* a particular grammar rule
* a particular idiom
* a proper noun (a place name or person's name)
* a particular character (especially in Chinese/Japanese)
* anything else you want to learn
Once you've chosen a sentence target, the goal is to find a sentence which you could understand, if you understood the sentence target. For example, if your target is a particular vocab word, look for a sentence in which you already know all the other words, and you know the grammar, etc. The only thing you haven't already learned is the target.
It's important that you be able to at least know whether or not you comprehended the sentence. There are two primary ways to test your comprehension: 1) a translation which comes along with the sentence, or 2) prior knowledge of the context, e.g. the sentence comes from the target language version of a book you've already read in your native language.
When you're first starting the language, the requirement that you be able to test your comprehension, really limits the sentences you can use. That's why when I'm just beginning to study a new language, the bulk of my sentences come from textbooks (where they have English translations provided) or from translations of things I already know in English (a personal favorite of mine is to sentence-mine the target language versions of old video game classics like Final Fantasy 7 or Chrono Trigger).
If you can't find any sentences meeting the criteria, then you're not ready for that sentence target yet. When I was first studying Japanese, this happened to me frequently. If this happens, take a look at the sentences you did find for the target, and ask why they failed. For example, say I want to learn a tough Japanese vocabulary word about finance. All the sentences I find for it, they all involve other tough financial vocabulary. I can see whether any of those words are more appropriate sentence targets. Sometimes, you might have to go multiple levels deep! I know I did, but as a result I have a very good grasp of those vocabulary words, so much deeper than just memorizing an English translation.
In the beginning stages of the language, feel free to make some direct vocabulary cards as a "crutch". After all, if you know absolutely no vocabulary, then it's gonna be awfully hard to find some sentences you can understand! (Although, if the language is closely related to your native one, you'll enjoy tons of obvious cognates) Later on, as you're getting better at the language, you can gradually delete the rote memorization vocab cards. When you memorize a vocabulary word by rote, you're not really learning it, since "translation" is really only "approximation". But some rote-memorized vocab words will give you the foothold you need to start comprehending sentences (without having to spend several years, like a baby does).
Another alternative for vocab items if the sentences are too hard, is the picture card. With a picture card, one side is a picture (say a picture of a rabbit) and the other side is the word. Nowhere does your native language get involved. The point of learning a new language is not to be able to see a rabbit, and think "Rabbit. The Japanese translation is Usagi." The point is to see a rabbit, and right away think "Usagi."
HOW MANY SENTENCES TO COLLECT?
The more the merrier, but again, you're a busy important person, so let's optimize! What I do is grab a small handful of sentences that "hit" my desired sentence target. Usually, that's enough. If it's not enough, you'll discover it pretty soon, when you notice that in your daily reviews, you keep failing those cards. If you make that discovery, you can go back and get more sentences. This actually leads to an interesting phenomenon where your flashcard deck becomes easier when you add more cards.
Bear in mind that however many cards you mine for the target right now, the list of cards which hit that target, will always grow, as long as you study the language. For example, say I want to learn the French word "escargot". I mine 5 sentences using that vocabulary. A week later, I'm mining a totally different vocabulary word, and one of the sentences for that word just happens to contain "escargot". Now I have 6 sentences for escargot! In this way, in my Japanese deck I have literally hundreds of sentences for some of the more common words, by now.
Also bear in mind, the sentences aren't the sole way you'll be learning the sentence target. Depending on its commonness, you should also be encountering it in your actual usage of the language (in conversations or movies or songs), where it goes unmined. Having sentences in your deck just optimizes the heck out of the learning process, but if for some word you just can't find enough sentences, it's still not the end of the world.
WHAT IF THERE ARE NO SENTENCES?
This is interesting. Let's say someone studying English as a second language wanted to learn the word "antidisestablishmentarianism". But, what's this, there are no sentences anywhere that really use this word, besides using it as an example of a long word! In that case, that's a strong indication that this word isn't really important, and learning it is a waste of time (except, in this case, possibly, as an example of a long word).
For certain vocabulary words, there's an exception to the value of sentences. For example, if I wanted to memorize all the dinosaur names in Japanese, it'd be better to just rote memorize the list (excepting the more interesting ones like the T-rex and the pterodactyl). Sure, it'd just be a meaningless list of words, but then again, the dinosaur names are a meaningless list of words to me in English! But really, if a word is an exception to the value of sentences, almost by definition it's not worth studying. (At least, no more so than any other trivial information, and I'm sure if you're gonna study trivia, you can find something more fun than lists of words)
WHAT TO PUT ON THE ANSWER SIDE OF THE FLASHCARD
Here are some things you can put on the answer side of a flashcard:
* An English (or whatever language) translation
* Contextual notes (See this article)
* Pronunciation notes
* An audio file of the sentence being spoken by a native speaker
* A picture illustration
* Any combination of the above
* Or, one of my favorites: Nothing!
Yeah that's right, it's perfectly okay to make a flashcard with a blank answer side. (On Anki, though, you'll have to go into the model preferences and specifically enable blanks.. they're disabled by default.. what the hell, Anki?) After all, the goal with Sentence Mined flashcards is not to memorize what's on the other side of the card, but to comprehend the sentence written on the question side.
Basically, the whole point of the answer side is for you to know whether or not you did, indeed, comprehend the sentence correctly.
VARIATIONS ON SENTENCE COMPREHENSION
Besides comprehending the sentence, here are some other types of flashcards you can make with sentence mining.
Pronunciation: The point is to read the question side of the card aloud. The answer side contains any pronunciation notes you need (or an actual audio file) to check yourself.
Writing: If the target language has different characters than your mother tongue, you might make cards where the question side indicates a sentence, but written in Romanized form (or kana, or something), and the answer side is the real sentence. The objective when reviewing the card is to write the sentence based on the Romanized version. This variation is especially good for Chinese/Japanese.
Copying: An easier version of "writing", in this variation the question side is the sentence, and the answer side is also the sentence. Your objective is simply to copy it onto paper by hand. Rate yourself based on how fast and easy the actual mechanics of writing are, how good your handwriting is, etc.
Shadowing: The "question" side is an audio file, and your objective is simply to speak along. The question side might also include the sentence in text format.
Listening Comprehension: Like the default sentence card, but instead of a text sentence on the question side, there's an audio file on the question side.
Transcription: The question side is an audio file, and the objective is to write the sentence on paper. Feel free to replay the audio file as needed.
One variation you do not want to do is the reverse-sentence card. This would be where the question side is an English sentence (or whatever your native tongue) and the answer side is the sentence in the target language. If you tried cards like these, you'd just end up memorizing the answer sides (with great effort) and wouldn't really learn anything you could generalize beyond that one single lone sentence. This sort of process would be very unnatural, because babies never translate.
If you use different variations of sentences in the same flashcard deck, make sure you can tell which kind of card each card is, before flipping it over.
SENTENCE MINING IN ACTION
I'm doing a 30-day challenge to teach myself as much French as I can in several hours a day for 30 days. Not that I'm particularly interested in French, but rather I'm interested in Earthian, the collective body of all languages spoken on Earth. The project is called: The French Revolution. You can read there about my own sentence mining efforts. The whole project writeup is packed to the gills with all sorts of other interesting language observations as I learn so much about learning languages.
OTHER COOL ARTICLES
Drilling Flashcards Without Music. I was shocked when I found out how much faster I got my scheduled reviews done!
Autodidact: Be A Self-Teacher. Well, you don't have much choice, because Sentence Mining is so elite and cutting edge, you won't find it in any classrooms yet. When you become a good self-teacher, learning things feels almost too easy, it's like playing the game of life with cheat codes on.
Irregular Verbs In Japanese. Japanese is one of the most regular, structured, logical languages in the world. Find out just how few irregular verbs it has, and wish that you'd studied Japanese in high school instead of Spanish!
Sentence Method Cause And Effect (outside link). Fellow blogger Tibul talks a little about his own experience with Sentence Mining.
If you want to learn a language naturally, with the kind of learning that produces perfect native fluency, why would you do anything other than soak up millions of sentences? Oh, right, because it takes years and years to learn a language that way. Well, it doesn't have to. And it doesn't, if you use the cutting-edge language-learning technology of Sentence Mining.
Sentence Mining makes use of a fairly recent technological innovation, the spaced repetition system, or SRS. An SRS is a computer flashcard program (like Anki or Mnemosyne, both free) that lets you make and review flashcards. The thing about an SRS, though, is that when you review a flashcard, you rate how well you understood it. The program uses your ratings to determine when it's best to show you the flashcard. In this way, the old process of reviewing flashcards gets super-optimized. But wait, aren't flashcards just for rote memorization, the sort of thing which is entirely unnatural for language learning? No! Flashcards have another use, having nothing to do with rote memorization, and that use is Sentence Mining.
Sentence Mining is the process of making flashcards, where the "Question" side of the card is a sentence in the target language. To "answer" the card, you don't try to recall anything from memory, instead you merely try to comprehend the sentence, by any means necessary. It doesn't matter whether you parse the sentence using your knowledge of grammar and vocabulary, or whether you just somehow know the sentence's meaning because you're familiar with the book it came from, or whether you memorize the sentence's meaning outright. All that matters is that, somehow, you comprehend the sentence. If you comprehend the sentence, rate it based on how easily you comprehended it. If not, fail it (rate it "Again" in Anki, or "0" in Mnemosyne).
The "Mining" part of "Sentence Mining" is the part where you go forth into the world and collect these sentences. Every book, every video game, every website, every textbook, every movie in the target language, is a goldmine waiting to provide you with linguistic fluency.
WHY DOES SENTENCE MINING WORK?
Think of the language processing center of your brain as a jungle. When you encounter a sentence, the sentence gets launched into the jungle and follows the path of least resistance, and wherever it ends up, that's how you understand the sentence. If you don't know anything about the language, then the necessary paths have not been blazed. The sentence crashes into a tree, and you "understand" it as freakin' gibberish.
As you review sentences you can comprehend, you carefully push them through the jungle to the right destination so that you understand them correctly. As you push them along, they gradually wear paths through the thick foliage. The more sentences you comprehend, the deeper and more sophisticated the paths become.
The cool thing is that the paths blazed by the sentences in your deck, work just as well when you encounter entirely new, exotic sentences you've never seen before.
Eventually, you've comprehended so many sentences, that the "paths" in the jungle have evolved into an intricate modern superhighway. Now, whatever sentence you launch into the jungle, it'll easily, quickly get to the right destination. Congratulations, you're fluent at reading in your target language!
HOW TO CHOOSE WHICH SENTENCES TO MINE
Any sentence which you can comprehend, is good. When you get to an intermediate or advanced point, though, that'll be too many sentences for a busy important person like you, so we'll need to weed them down.
First, identify a "sentence target". A sentence target can be:
* a particular vocabulary item
* a particular type of verb conjugation
* a particular grammar rule
* a particular idiom
* a proper noun (a place name or person's name)
* a particular character (especially in Chinese/Japanese)
* anything else you want to learn
Once you've chosen a sentence target, the goal is to find a sentence which you could understand, if you understood the sentence target. For example, if your target is a particular vocab word, look for a sentence in which you already know all the other words, and you know the grammar, etc. The only thing you haven't already learned is the target.
It's important that you be able to at least know whether or not you comprehended the sentence. There are two primary ways to test your comprehension: 1) a translation which comes along with the sentence, or 2) prior knowledge of the context, e.g. the sentence comes from the target language version of a book you've already read in your native language.
When you're first starting the language, the requirement that you be able to test your comprehension, really limits the sentences you can use. That's why when I'm just beginning to study a new language, the bulk of my sentences come from textbooks (where they have English translations provided) or from translations of things I already know in English (a personal favorite of mine is to sentence-mine the target language versions of old video game classics like Final Fantasy 7 or Chrono Trigger).
If you can't find any sentences meeting the criteria, then you're not ready for that sentence target yet. When I was first studying Japanese, this happened to me frequently. If this happens, take a look at the sentences you did find for the target, and ask why they failed. For example, say I want to learn a tough Japanese vocabulary word about finance. All the sentences I find for it, they all involve other tough financial vocabulary. I can see whether any of those words are more appropriate sentence targets. Sometimes, you might have to go multiple levels deep! I know I did, but as a result I have a very good grasp of those vocabulary words, so much deeper than just memorizing an English translation.
In the beginning stages of the language, feel free to make some direct vocabulary cards as a "crutch". After all, if you know absolutely no vocabulary, then it's gonna be awfully hard to find some sentences you can understand! (Although, if the language is closely related to your native one, you'll enjoy tons of obvious cognates) Later on, as you're getting better at the language, you can gradually delete the rote memorization vocab cards. When you memorize a vocabulary word by rote, you're not really learning it, since "translation" is really only "approximation". But some rote-memorized vocab words will give you the foothold you need to start comprehending sentences (without having to spend several years, like a baby does).
Another alternative for vocab items if the sentences are too hard, is the picture card. With a picture card, one side is a picture (say a picture of a rabbit) and the other side is the word. Nowhere does your native language get involved. The point of learning a new language is not to be able to see a rabbit, and think "Rabbit. The Japanese translation is Usagi." The point is to see a rabbit, and right away think "Usagi."
HOW MANY SENTENCES TO COLLECT?
The more the merrier, but again, you're a busy important person, so let's optimize! What I do is grab a small handful of sentences that "hit" my desired sentence target. Usually, that's enough. If it's not enough, you'll discover it pretty soon, when you notice that in your daily reviews, you keep failing those cards. If you make that discovery, you can go back and get more sentences. This actually leads to an interesting phenomenon where your flashcard deck becomes easier when you add more cards.
Bear in mind that however many cards you mine for the target right now, the list of cards which hit that target, will always grow, as long as you study the language. For example, say I want to learn the French word "escargot". I mine 5 sentences using that vocabulary. A week later, I'm mining a totally different vocabulary word, and one of the sentences for that word just happens to contain "escargot". Now I have 6 sentences for escargot! In this way, in my Japanese deck I have literally hundreds of sentences for some of the more common words, by now.
Also bear in mind, the sentences aren't the sole way you'll be learning the sentence target. Depending on its commonness, you should also be encountering it in your actual usage of the language (in conversations or movies or songs), where it goes unmined. Having sentences in your deck just optimizes the heck out of the learning process, but if for some word you just can't find enough sentences, it's still not the end of the world.
WHAT IF THERE ARE NO SENTENCES?
This is interesting. Let's say someone studying English as a second language wanted to learn the word "antidisestablishmentarianism". But, what's this, there are no sentences anywhere that really use this word, besides using it as an example of a long word! In that case, that's a strong indication that this word isn't really important, and learning it is a waste of time (except, in this case, possibly, as an example of a long word).
For certain vocabulary words, there's an exception to the value of sentences. For example, if I wanted to memorize all the dinosaur names in Japanese, it'd be better to just rote memorize the list (excepting the more interesting ones like the T-rex and the pterodactyl). Sure, it'd just be a meaningless list of words, but then again, the dinosaur names are a meaningless list of words to me in English! But really, if a word is an exception to the value of sentences, almost by definition it's not worth studying. (At least, no more so than any other trivial information, and I'm sure if you're gonna study trivia, you can find something more fun than lists of words)
WHAT TO PUT ON THE ANSWER SIDE OF THE FLASHCARD
Here are some things you can put on the answer side of a flashcard:
* An English (or whatever language) translation
* Contextual notes (See this article)
* Pronunciation notes
* An audio file of the sentence being spoken by a native speaker
* A picture illustration
* Any combination of the above
* Or, one of my favorites: Nothing!
Yeah that's right, it's perfectly okay to make a flashcard with a blank answer side. (On Anki, though, you'll have to go into the model preferences and specifically enable blanks.. they're disabled by default.. what the hell, Anki?) After all, the goal with Sentence Mined flashcards is not to memorize what's on the other side of the card, but to comprehend the sentence written on the question side.
Basically, the whole point of the answer side is for you to know whether or not you did, indeed, comprehend the sentence correctly.
VARIATIONS ON SENTENCE COMPREHENSION
Besides comprehending the sentence, here are some other types of flashcards you can make with sentence mining.
Pronunciation: The point is to read the question side of the card aloud. The answer side contains any pronunciation notes you need (or an actual audio file) to check yourself.
Writing: If the target language has different characters than your mother tongue, you might make cards where the question side indicates a sentence, but written in Romanized form (or kana, or something), and the answer side is the real sentence. The objective when reviewing the card is to write the sentence based on the Romanized version. This variation is especially good for Chinese/Japanese.
Copying: An easier version of "writing", in this variation the question side is the sentence, and the answer side is also the sentence. Your objective is simply to copy it onto paper by hand. Rate yourself based on how fast and easy the actual mechanics of writing are, how good your handwriting is, etc.
Shadowing: The "question" side is an audio file, and your objective is simply to speak along. The question side might also include the sentence in text format.
Listening Comprehension: Like the default sentence card, but instead of a text sentence on the question side, there's an audio file on the question side.
Transcription: The question side is an audio file, and the objective is to write the sentence on paper. Feel free to replay the audio file as needed.
One variation you do not want to do is the reverse-sentence card. This would be where the question side is an English sentence (or whatever your native tongue) and the answer side is the sentence in the target language. If you tried cards like these, you'd just end up memorizing the answer sides (with great effort) and wouldn't really learn anything you could generalize beyond that one single lone sentence. This sort of process would be very unnatural, because babies never translate.
If you use different variations of sentences in the same flashcard deck, make sure you can tell which kind of card each card is, before flipping it over.
SENTENCE MINING IN ACTION
I'm doing a 30-day challenge to teach myself as much French as I can in several hours a day for 30 days. Not that I'm particularly interested in French, but rather I'm interested in Earthian, the collective body of all languages spoken on Earth. The project is called: The French Revolution. You can read there about my own sentence mining efforts. The whole project writeup is packed to the gills with all sorts of other interesting language observations as I learn so much about learning languages.
OTHER COOL ARTICLES
Drilling Flashcards Without Music. I was shocked when I found out how much faster I got my scheduled reviews done!
Autodidact: Be A Self-Teacher. Well, you don't have much choice, because Sentence Mining is so elite and cutting edge, you won't find it in any classrooms yet. When you become a good self-teacher, learning things feels almost too easy, it's like playing the game of life with cheat codes on.
Irregular Verbs In Japanese. Japanese is one of the most regular, structured, logical languages in the world. Find out just how few irregular verbs it has, and wish that you'd studied Japanese in high school instead of Spanish!
Sentence Method Cause And Effect (outside link). Fellow blogger Tibul talks a little about his own experience with Sentence Mining.
4 comments:
Good article. It has cleared up a lot of parts of sentence minning.
Thank you!
Do you put your sentences in the same SRS file as your individual vocabulary words? Let's say, for example, that the word you are learning in German is "mörgen" (to like). In Anki, or any SRS, would you have four fields: one for the german word, one for the english definition, one for the german example sentence, and (potentially) one for that sentence's translations? Or do you have separate files for each of vocabulary and sentences? I ask this particularly in light of a comment you made in your 30-day French Revolution about failure card repeat times, where you noticed that failed cards repeating during the session made sense for individual vocabulary words, but not for sentences. Any advice?
I am trying this whole sentence mining strategy and I think it will work out pretty well. I have a question though. I use text-to-speech software that allows me to hear the sentences that I mine when I flip the card over to the backside (using Anki). I know that the software is not perfect, but it gives a general idea of what the pronunciation should sound like. However, the liaison is not perfect and the rhythm of the speech (rises and falls in pitches through phrases) is not even close to being right. I am quite familiar with French and can pick out these errors, I just use it to hear the pronunciation of certain words. Nonetheless, I feel that this can still be detrimental to my learning of French. What do you think?
thanks for this article. This, combined with your "french revolution" series has remotivated me to try and increase the number of cards i put in anki. Presently i only add about 10 sentences per day, but i'm going to try experimenting with small bursts of 100 new cards per day for short periods.
Post a Comment