This method is designed around my specific learning situation. It will not and cannot suit everyone else’s. So in order to make this post as useful as possible to as many people as possible, my goal will not to be to convince everyone that they should use my method, but to try my best to explain why I think it works well to solve my specific learning obstacles. My hopes are that by discussing the thought process behind the method other learners might be able to find certain useful bits of information that they could adapt to their specific situation and language goals.
My post will be divided into 4 sections.
The first section will describe the method in ideal terms. This first section will ignore any practical issues of implementing the method in favor of describing how the method works free of practical limitations. My practical way of generating my method is limited by my skill set, time and resources, so I think it will be helpful to describe what my method would look like if those limits didn’t exist.
The second section will dive into the practical side of implementing the method and some of the deficiencies with my implementation. I’ve made several compromises to make creating the method more time efficient. For example my method uses lots of machine translation and audio, a compromise some learners may not be interested in making. In order to make this post as useful as possible to everyone I think its useful to discuss the reasons for the various compromises I made, and other learners can judge if any of those compromises would be acceptable to them, or think of alternatives.
The third section will briefly outline a few variations of the method. These aren’t methods I’m using, but ideas I’ve had that would make different trade offs. Some learners who don’t want to make the same compromises I’ve made, may find some of these variations more applicable to their specific learning situation and goals.
The fourth section will be a review of the many methods I’ve used in the past. I’ve tried to build my current method based on the strengths of different learning materials I’ve used and things I’ve learned from them. This will try and explain my thought process behind why I built my current learning method the way I have. If someone doesn’t think the specifics of my method will be of any use to them, this section might have some useful ideas. For me this is the most important section as I don’t want to encourage the use of any single learning method, but rather encourage learners to think about how well a method is working and how learning can be made more efficient.
1. The method under ideal circumstances.
Fű zöld. - Grass is green.
Az udvar zöld. - The yard is green.
Az udvar füves. - The yard is grass.
Fű az udvaron. - Grass is in the yard.
A fű az udvaron zöld. - The grass in the yard is green.
Zöld gyűjt napfény. - Green collects sunlight.
Fű gyűjt napfény. - Grass collects sunlight.
Az udvar gyűjt napfény. - The yard collects sunlight.
Fény jön a nap. - Light comes from the sun.
Napfény jön a nap. - Sunlight comes from the sun.
Egy jó színe zöld . - A good color is green.
A zöld jó szín gyűjtésére napfény. - Green is good color for collecting sunlight.
Fű gyűjt napfény is, mert ez a zöld. - Grass collects sunlight well because it is green.
Az udvar gyűjt napfény is, mert a fű. - The yard collects sunlight well because it is grass.
Now I haven’t tried this specific card out as a Hungarian beginner, but I would bet that someone with very little, possible no knowledge of Hungarian could quickly learn all the words here with this kind of presentation. They might not understand all the grammatical points, and it might not even be 100% grammatically correct, but I bet they would understand the vocabulary at a rate much faster than single word cards, or a course like Assimil could teach them.
The above picture is what my method looks like. Basically it is an anki deck with a picture and audio on the front, and the written L2 and L1 translation on the back. Each card has several sentences on it. There are several principles I try to keep in mind when building these cards. None of these principles is essential to my learning method, but each one seems to increase the efficiency of it. Each card doesn’t have to be perfect, but it seems that the more principles each card contains the faster and more efficient learning will be.
- The sentences should relate to the picture in some way. Not every sentence has to relate directly to the picture, but the picture should give you clues as to what the sentences are about. Its more important for the first few sentences relate more literally to the picture, but from there you can stray a bit from the picture if it keeps your story interesting.
- The sentences should follow a logical progression. You should try to tell a story, or have the sentences follow some sort of progression that makes it easier for you to predict and recall which sentence will follow next.
- The sentences should be in N+1 increments of difficulty. Both sentence patterns and vocab count as +1 especially for opaque languages. New words should be introduced in familiar sentence patterns, new sentence patterns should be introduced with only familiar words.
- Rework the same vocab enough and you can hop over N+1. You don’t want to stray too far away, but if you have 0% comprehension of a language try re-working the same vocab in a lot of different ways. A kind of magic happens when you hear the same words re-worked in different sentence patterns. Your brain grabs onto the repeating sounds and suddenly can figure out their role in the various sentences. Even if the first few sentences aren’t N+1 difficulty from your starting knowledge, re-work the same words enough and your brain can figure it out.
-If you really have 0 starting knowledge of an opaque language and no idea how it works, start with simple sentences using proper nouns and known facts. Sentences like “Asia is bigger than Europe” or “Tom and Jim are friends” should be partially comprehensible in almost any language, and will make you accustomed to basic sentence patterns in languages that act much differently from your native one.
- Cards should rely on as little knowledge from other cards or previous learning as possible. Most courses have too steep a learning curve, particularly for opaque languages, and this makes learning much less efficient. You could learn 5000+ words just with “see spot run” type sentences if you were so inclined. You probably will eventually want to make more complicated cards to write more interesting sentences, but simpler is almost always more efficient. For the first few hundred cards, I would try and make the most difficult sentence in the card learnable from the other sentences in the card for someone with 0 exposure to the language. It is helpful to have the same words on multiple cards, just don't rely on previous cards having taught you the word. Have it so you can learn it from the new card in case you forgot. You can study more cards per day if they are independent of each other. Many anki sentence decks limit how many new cards you can add per day because they grow more difficult and it can take days before the old cards are consolidated enough for you to easily learn the new ones. If your cards don’t require any pre-existing knowledge to learn them, then your only limitations are time and interest.
-Even though you’ll be writing a lot of “See spot run” type of sentences they can still be funny and tell stories. The more memorable a story is the easier it will be for you to remember it. You can do stuff like. “The boy likes hats. There is a new hat shop. The boy likes new hats. He likes purple hats. He only wears purple hats. The shop only has blue hats. The boy hates blue hats. The boy hates the new hat shop. The shop burns down. The boy burns the hat shop down. “
-Some sentences should demonstrate the function of the words. For example if you look at the first three sentences of the example card.
Fű zöld. - Grass is green.
Az udvar zöld. - The yard is green.
Az udvar füves. - The yard is grass.
These are useful for introducing you to the key sounds of the sentences, but they give little clue to what the meaning of each word is. Even with the picture you won’t necessarily be able to guess which word means what. That’s why it’s important for later sentences like
Egy jó színe zöld - A good color is green.
Grass or a yard cannot be a good color, so you know that zöld means green, and when you replay the dialog that will help you understand the above 3 sentences. Not all sentences have to demonstrate the function of words, but a few are essential. If your sentences are all like
“Jim has a house, a car, and two children. Mary has a house, a car, and two children. Tom has a house, a car, and two children.”
It doesn’t matter how many times you repeat it, you’re not going to learn the list of words efficiently because nothing in the sentence gives clues to what they mean. Its better to have stuff like.
“Mary drives her car. Tom is at his house. Mary is driving her car to Tom’s House. Mary has an accident. Mary runs over Tom’s children. Mary accidentally runs over Tom’s two children.”
The second grouping is more meaningful and will be easier to understand than the listing of words in the first grouping.
2. Practical implementation, trade offs, compromises and deficiencies
So obviously the above described course is nice, but one has to find a way to generate the material efficiently. If it takes too long to build a deck like that, then it becomes more efficient just to use a pre-made course. I personally use machine translations and machine generated audio. This is because I study several opaque languages and I’m in no hurry to produce correct output, I only want to increase my comprehension. I type in English because it’s faster to come up with amusing stories that way and I can start learning a language knowing nothing about it. I can also translate the same deck into multiple languages, so I only have to build it once and then I can re-use it for all my future learning. I build the deck in Anki in English, export it to excel to quickly copy paste everything to google translate, import it back to Anki, and use the AwesomeTTSadd-on to rapidly add Google translate audio to all the cards.
This tradeoff might become irrelevant for someone with a different skill set/resources or based on future availability of materials. If in the future there’s more content online and its easy to build these kinds of cards by mining native materials, then one might not have to use machine translation. Someone with better computer programming skills than me might be able to design a program able to mine native sentences to build cards like this. Or someone might be willing to pay a native speaker to write these kinds of sentences for them and record audio. Machine translation is an acceptable compromise for me based on my skill set and resources for my goals, but other people might find other ways to build decks like the one I described above or choose different compromises.
Simple sentences are easier to translate. I recommended this earlier for building vocab, but it’s also a good cautionary measure against bad machine translations. Simple sentences are easier for machines to translate. If one wants to reduce possible bad habits machine translations form, you could use this method learn vocabulary rapidly, and then use a more traditional course to learn more complex sentence patterns.
Machine audio is monotonous. I’ve noticed that because machine audio doesn’t know what words to emphasize it can be more difficult to understand the meaning of a sentence. In my experience the positive factors outweigh this deficiency, but it definitely is not ideal.
3. Some possible variations
I realize that for most learners the use of machine translation/audio and the time spent building ones own course might not be acceptable compromises. Some of these principles can be applied to existing materials to make them more efficient.
For instance a big deficiency in most courses, and materials is that they don’t build up at a proper N+1 difficult increment for opaque languages. One way around this is to use machine translated sentences to re-work new vocab in simple sentences. I’ve tried it on Subs2SRS cards, and if you re-work new vocab in simple sentences from machine translations, it rapidly increases the speed at which you can learn the complex sentences. The machine translation here is a useful bridge, and you can delete the machine translation/audio as soon as you readily understand the native sentence.
You could also use this to review course material without having to type in the full course lessons. For example you could study Assimil normally with 1 lesson per day, and then add 1 lesson to anki each day along with some pictures of key vocab and anything notes you want to remember, and you’ll probably be able to more efficiently memorize all of the course material that way reviewing through anki.
4 - Some History of my Methods and Course Reviews
So I got most of the ideas for this method by experimenting with lots of other methods and mixing and matching them. Not all my ideas were practical to apply to my specific method. If none of the above has any appeal to your learning style, hopefully there will be some interesting ideas in here. Again I’d like to repeat that I think the mindset of analyzing how well a method is working and mixing or modifying methods to increase learning efficiency is way more important than the specific details of any single method.
Listening-Reading – One of the reasons I think Listening-Reading works so well for some is that it throws so many sentences at you, you can’t help but find ones useful for your level of learning. My theory behind how listening-reading works is that you learn from useful sentences and ignore difficult ones. Because you learned something from the first pass, more sentences will become appropriate to your level on subsequent passes. Most courses don’t give nearly enough exposure or repetition for content to really sink in. Listening-Reading gives a lot more exposure, but in my experience Listening-Reading with opaque languages, while possible, results in a lot of time being wasted listening to sentences that are undecipherable. Listening-Reading also rewards lots of time commitment. Because the book isn't designed for learning, the longer you do Listening-Reading the more words you’ll re-encounter before they leave your working memory. You also can’t reliably review these words unless you come up with a system of writing them down and interrupt the process. If you’re doing pure Listening-Reading you have to do it for long portions of the day over a long period of time to make it more efficient at low levels. It’s a very useful transition to native materials, but I don’t think the time requirements are practical enough to make it a useful way of learning opaque language from scratch. It can be a fun way to mix native materials into your routine at a low level though. And Listening-Reading is a great stepping stone to extensive reading.
Sentence Decks – My method is at its core a sentence deck. But I have several problems with existing sentence decks. Most often sentence decks put very little effort into the learning curve, and even the best ones that do consider this, don’t do it in the most efficient way possible. Often with sentence decks I find I need more exposure to a word for it to really consolidate. If I add a new sentence and it only re-works familiar vocabulary, it will make me learn all other related sentence cards faster. If it introduces new vocabulary, it adds difficulty and energy needed to learn them, and may not be simple enough for me to learn efficiently and easily. I may have to pick apart and analyze the sentence to discover how all the new vocab functions. So there's not usually a reliable correlation between number of "new" cards per day and how easier or harder my reviews get. A new card might make reviewing easier or harder if the sentence is picked at random, and usually its harder. Also even sentence decks designed for an efficient learning curve, slow you down by requiring knowledge to build up. Often your pace is limited by how fast the information consolidates. As I previously mentioned you can learn more vocabulary faster if the cards don't build off each other so much. So I like the idea of sentence decks in principle, but organization and learning curve is essential to it being efficient, and there's still room for improvement.
Courses vs Anki – Courses often want to teach more than they have room for. So the efficiency of the method often goes down as they try to cram more content in later lessons, and this is even worse for opaque languages. Courses can also be cumbersome to review if lessons are long. Anki makes reviewing more efficient and almost any course would ideally be made more efficient if turned into an Anki deck. The problem is the loss of time converting a course into an Anki deck. Putting sound and images into an Anki deck costs time and transcribing from a physical book is the worst. Language courses are often so inefficient at teaching opaque languages, that building an anki deck for them will usually save a lot of time in the long run. If the deck is re-useable for multiple languages this is especially true. I’m sure if you gave one French learner an Assimil course and another French learner an Assimil course converted into an Anki deck, the French leaner with the Anki version would learn faster. But if the Anki learner had to manually type out all the lessons and build the Anki deck by themselves, the person working directly from the course would probably learn faster in the long run since they didn't waste time preparing materials.
Some specific course comments…
Assimil – Learning French with Assimil was an enjoyable and smooth experience. Learning Russian with Assimil took a bit more determination and review, but eventually I learned a good amount of Russian from it. Trying to learn Hungarian from Assimil was near impossible. Assimil’s memorable dialogs are well worth imitating, they made remembering content much easier. But their learning curve isn’t suited to opaque languages, and they often introduce too much new vocabulary alongside complex new grammar rules and sentence patterns. In some of the modern courses when they do try to make opaque languages “easier” they do it by reducing the content in the early lessons. Learning new words requires the same amount of exposure regardless of your level. Early dialogs shouldn’t aim for as a few sentences as possible, but rather should aim to introduce new words at a gradual pace and re-work them in simple sentence patterns.
Teach Yourself Beginner’s Hindi – I’ve never run across another teach yourself course like it, but Teach Yourself Beginner’s Hindi does something brilliant, that more language courses should emulate. It’s a “beginner’s” course, but the dialogs are way longer than most normal courses, only the sentence patterns are "beginner". As far as I can tell this is the only course that acknowledges that beginners need as much exposure as any other students and gives you enough exposure to simple sentence patterns for them to really sink in the same way that happens with extensive reading of books among higher level learners. Though it still has some limitations of other teach yourself courses like having lots of L1 instructions on the audio, and not being long enough to continue this method of teaching up into intermediate. But the method is sound for the content it covers.
Linguaphone – One of my favorite features of linguaphone is the use of pictures in its early lessons. For the first few lessons, you can look at the picture for the lesson and genuinely figure out a lot of the dialog just by analyzing the picture. However eventually it runs into the same limitation as all other courses in that it wants to take you to a high level and sacrifices efficiency for new vocab. Each lesson introduces way too much new vocab, and as the dialogs get longer, the picture is less helpful in deciphering them. Part of my inspiration for my first decks, was to make something like a 500 lesson linguaphone, where every lesson was decipherable from the corresponding image, and each dialog as funny as an Assimil dialog.
FSI – FSI gets the learning curve for opaque languages right where other courses go off the rails. FSI Hungarian saved me, by introducing me to Hungarian sentence patterns at a proper pace, where Assimil charged ahead and leaved the learner behind. It also helped me learn that if you repeat the same sentence pattern enough substituting a single word each time, the meaning of an opaque sentence can suddenly become clear. It doesn’t make the individual words being substituted easy to remember, but you understand how different words can take on a meaning by being plugged into a certain spot of a sentence pattern. Its deficiencies is that its long lessons are unwieldy for review, and its not as funny as Assimil, nor does it give good audio or visual clues like other courses to help you decipher audio. I tried mining sentences from FSI for Anki decks and it was a very effective method, and I've gotten many positive comments on Ankiweb for my FSI decks. But I think Anki decks can go much further and faster than these applying some ideas from other courses.
Berlitz Basic Course – This is a brilliantly designed course, in that its 100% native audio and it uses sound cues and known facts, and re-occurring characters and jokes to make its dialog 100% comprehensible, and it repeats words enough that you have time to get used to their sound and decipher them from context. Its brilliant, except that it’s only available for languages that are very transparent to English learners, and doesn’t go as far as courses like Assimil. Still I always wanted a Berlitz equivalent for opaque languages and that provided a lot of inspiration and motivation for my current method.
French in Action – Similar to Berlitz Basic Course, it’s a brilliant method, if only it took you further and was available for more opaque languages.
Book 2 – Book2 was one of my favorite places for mining native audio for sentence decks after FSI. I wanted for a long time to create something that could function like a Berlitz or French in Action for opaque languages, and Book2 showed me the path. A lot of early dialogs for Book2 use proper nouns and known facts, and these were often very transparent no matter how opaque the target language. That made it much easier to acclimate myself to opaque languages at very low levels. I realized that one could learn opaque languages from scratch as smoothly as effortlessly as I learned French, if the material was designed appropriately.
Gabriel Wyrner’s Anki Method – My Anki deck implements several ideas from his method. When I discovered his method, I was using sentence decks with audio on the front and translation on the back mined from various courses. What I took away from his method was that it was worthwhile to write your own cards, because writing your own anki cards creates a memory in your brain that makes it much easier to learn that card later. Where I part ways with his method, is that for one he brute memorizes single words rather than full sentences. For another I think he overly cautious about avoiding himself from translations. I agree with him that understanding from context does seem faster than understanding from translation. But I don’t think that seeing a translation harms your understanding in any way. I think translations can result in a “partial learning” where you can associate a sound with seeing an L1 translation, but not fully recognize it on its own. This can be a helpful stepping stone to understanding a word purely from context, and a very helpful backup since finding ways to learn 100% from context can be maddeningly difficult. So my mindset is try to learn from context, but then check a translation to see what you missed, hide the translation and try again.
Anyways that was my long rambling way of trying to discuss my current learning method. I'm still working on improving it, but it seemed like I hit a new milestone recently, so I wanted to try and compile all my ideas together. I'd be curious to know if anyone has other ideas about language efficiency or building faster materials, or thinks of some alternative ways to implement some of these methods.
