Learning Keyboard Layout

1. Colemak

Roughly three years ago I decided to learn the Colemak keyboard layout. I did that without moving around the physical keys of my keyboard, i.e., I learned to type it “blind” from the get-go (other than the advantage for typing speed, it also makes it easier to use qwerty keyboards away from home). For this task, I made unidirectional anki cards. The front (“question”) side consisted of a single letter or punctuation mark, while the back side consisted of a photo of either my left or right hand, with a single finger highlighted. The highlighted finger is the finger that is responsible for typing that character. This, incidentally, served a double role. First, of course, it taught me the location of the keys (only roughly, of course, as there are three rows, as well as some fingers which are responsible for more than one column), but second, it also taught me “proper fingering”. I think my fingering might have receded to something less than optimal since, but still today my Colemak fingering is much better than my qwerty’s.

Another key part of this deck was the adjustment of the review intervals. If one goes to “options”, one can see “steps” under the “New Cards” and “Lapses” tabs. These are the intervals in minutes of the next reviews of correctly answered new and young(?) cards. I believe the default is “1 10” (I personally use “1 60” for my own language decks), meaning that after you answer a card correctly for the first time it will show 1 minute later, then 10 minutes later if answered correctly two times in succession, and then the next day on the third success.

With a keyboard layout we want not just to remember where the keys are, but recall them fast. My approach was thus: first, when reviewing, I didn’t let myself a long time to think. If I’m thinking for two or three seconds, that’s already too slow––– I’d flip the card and mark it wrong. Second, I changed the “steps” to “1 2 6 10 12 14 15 16 20 60 240” (though I’ve never practiced long enough for the longer intervals to really matter; I practiced until the deck had no cards to show me, which probably happened when all cards reached a 6 or 10 minute interval) which made for a frenzied reviewing experience, adequate for the task at hand.

I don’t remember now how big of a part the anki deck served in this learning, but I suppose it was very helpful. I do remember going through an excruciating phase of typing words very slowly, but I got passed that. Either way, I use Colemak until today, typing at a speed of roughly 80-90 WPM4, so the transition was all in all successful.

2. Stenotype

I have had a dormant stenotype deck for a while, but recently my older computer keyboard broke down and I got a new one with ah N-key rollover which allowed me, together with the open source Plover, to commence with studying steno and which made the anki deck relevant.

I’ll try to make the description here as fathomable to people who had never heard of stenotype in their life before, but my main concern is to explain what I’m doing to people who might be already engaged in learning to steno.

A stenotype is a chorded keyboard. You press more than one key at a time to produce a punctuation mark, a syllable, a whole word or even a whole expression. Ignoring the numericals for the moment, it means that as opposed to the 34*2=68 possible inputs of a regular keyboard (34 keys combined or not with the shift key), in steno there are 2^22 = 4,194,304 possible inputs (actually somewhat less, but this is just to illustrate the general order of magnitude). The challenge, therefore, is to learn which keys to simultaneously press for a given desired output, to be able to recall this combination quickly, and have the digital dexterity to correctly execute the mechanical motion.

As you might imagine, in comparison to  “regular keyboards”, the learning curve is much longer5, but the fruits of the labour include a much faster typing speed6 achieved with a much decreased movement of the fingers.

This is still work in progress, and I’ll return to update this section.

The individual keys of the steno keyboard are assigned individual letters (as well as a “star” key), and there are letters which are assigned key combinations (for example, the combination of the left “t” and “k” is assigned to a “d”). Many letters are represented twice (or even more with the case of “s”) in the keyboard, once on the left and once on the right side (or hand).

Before studying any steno theory at all, I’ve created an initial deck with many similarities to my Colemak deck. I use the same intervals and try to answer cards quickly. The front side of the cards contain a letter and a side (left/right), as well as sounds (short and long vowels, diphthongs). For single key letters, the back side contains a picture from the set used for the Colemak deck, representing a finger, as well as a text that indicates “up” or “down”, corresponding to one of the two rows of the steno keyboard. Letters with assigned combination of keys have a back side with a chart of the steno keys with the relevant keys highlighted.

After “”mastering”” these, I proceeded to learn steno theory through an online guide. This guide contains link to simple online practices, and my experience with them, especially the simpler ones, was very rewarding. I believe the preluding anking was very beneficial. First, it divided a task into two separate tasks, rendering both simpler to learn –– particularly since I’m typing it “blind” on a qwerty layout: one task is finding the letters on the keyboard, while the second task is choosing letters to form a word. The triviality of anking makes the first task rather effortless; one can learn the layout even without actively trying to remember. The difficulty of the second task –– the online practice accompanying the guide –– is stripped from the difficulty of searching for keys (and from the uncertainty that I’m hitting the wrong letters, since I’m blind) allowing me to concentrate on learning a “single thing” instead of a multipule of them, making it easier to learn. First, trivially,  because the learner part of me is tasked with only learning one thing. Second, because I do the practice faster (since finding the keys I’m picking is faster), meaning I can practice more “pick the right letters/keys” at any given period. The latter also means that I’m able to work up my speed of word-typing, unhindered by the speed of finding the right keys.

The speed of finding a key is not zero, and it “dosen’t need to be” so. However, I seem to see an opportunity for an expansion of the “low level” steno anki deck. Let’s say that the time for me to figure out where to move a single finger to input a letter is x. Naively, if this is the only thing I know, it would take me y*x time to figure out where to move my fingers to type y letters. This is “fine” with a qwery keyboard where keys are hit sequentially. However, it seems like much typing speed can be gained by learning to conjure finger movements to hit an arbitrary amount of keys at the same speed that one conjures a movement to hit one key, namely x7. I have already studied letters that are represented as a combination of keys which I presume to be able to conjure at an equal speed as single-keyed letters. The expansion would involve adding cards with multiple letters at the front. Given the way syllables/words are constructed with steno, I think I’ll restrict letter combinations to either side of the keyboard, and not both.

As there are 7 keys on the left side, we have 7 choose 2 or 21 2-key combinations, five of which are single letters. There are 35 3- and 4-key combinations, 21 5-key combinations, 7 6-key combinations and, of course, a single 7-key combinations. In total, counting all from 1-key “combination” to 7, there are 127 –––– a reasonable amount to memorize. The math is trickier for the right side with the pinky responsible for 4 keys, but it would be less than 254 cards. In total (and ignoring combinations including the star key) it would be less than 381 cards. As there’s no way around it, I’ll learn it either by sheer orthodox practice or through anki, I think it’s a worthwhile pursuit. The real hurdle would be to actually generate the cards, and hopefully I’ll find some way to automate it (with the hope that automating this would take less time than manually doing it). Either way, I suppose this deck could in the future also assist other steno-learners, and therefore a worthwhile investment.

For after my introductory steno theory learning, an anki deck I’ve downloaded with the 3000 most frequent English words is waiting for me to learn (and adjust to my own preferences). I’ll probably go for it after I am pretty solid with the letter-combination deck. This words-deck, if I apply to it a similar speed-standard as for the letters so that I don’t just learn to recall the “typing” but recall them fast, should –– I believe –– greatly push me forward in typing speed. First, by giving me “applied” practice of the theory, and second, simply by making me proficient in typing these words which together should account for a very high percentage of the words in any text. (à la Zipf’s law)

4. Based on some online tests. I think I might be typing generally somewhat faster, that is, when I’m not copying a text in a manner that necessitates me to either look at the source text or look at what I’m writing, but not both (my eyes don’t go this way).

5. Somewhat analogously speaking. Does a qwerty keyboard have a learning curve at all? The characters are printed on top of the corresponding keys. One could compare it to “blind typing” on a qwerty, or talk about the time it takes to acquire a certain WPM typing speed, which is in the general sense problematic as well. With steno there are expressions you might not know how to type, making certain texts to greatly slow you down by forcing you to spell the expressions, or even slow you down infinitely long (0 WPM) had you somehow skipped to learn how to spell single letters. But I’m just wasting words here: this “theoretical consideration” is of no importance to anyone, I’d imagine.

6. That is, an “average stenographer” types at a significantly faster rate than world champion qwertiest. This statements might hide the fact that most students don’t finish steno school. What this means is not so clear; the pressures of school can be very different than the inherent difficulty of what they teach. The way I see it, “average stenography” would be the level of someone who actively works on improving the skill until a plateau in progress. I think studying steno is a long and not a particularly inherently enjoyable journey, but a very feasible one as long as one steadily moved forward.

7. If you need some motivation here, imagine a well trained pianist. Presumably, the pianist can play repeatedly a single chord at the same speed that she plays a single key (with a single finger).

Learning German

I have been interested for a while in the ability to learn a foreign language through Anki; not by using it as an assisting factor, but as the core teacher. I have been interested in learning languages the way native speakers learn it, i.e., by a massive exposure to content rather than through learning prescriptive grammatical rules. One might think of it as a “data-driven approach”.

I’m still working on how to do it the best way. My initial plan was to go through these phases: first, learn a significant amount of words so as to acquire the “building blocks” necessary for the next phase. Second, “learn” (rote memorise) a significant amount of sentences. To be clear, this involves a deck of cards with a Germans sentence on one side and the translation on the other. And third, engage in German “media”: articles, books, movies and so on. At this phase Anki would only be used to maintain previously acquired knowledge, and to rote-memorise examples that pop-up whose understanding is not included in or not generalized from the thitherto deck.

Vocabulary memorisation. My most enduring German vocabulary deck was created by taking a pdf version of the Langenscheidt English-German dictionary  of most frequent German words. It had about 4,000 entries which I manually (and for quite a long period) added into Anki. Later these were supplemented by other words I have encountered here and there. So far I have logged 527 hours into reviewing it and I currently no longer review it. The experience of studying form it had taught me several things:

1. There’s an “emergent learning” going on.

My deck cards had errors that were introduced by one of at least these two ways. First, there were errors in my source material. My pdf was not an official document, but a scan. I don’t know how exactly it was done, but the impression I got was that the conversion tool (the pdf did not contain a “picture” of the scanned book, but a text. That is, presumably, during the pdf creation the pictures of the scanned book pages were converted into text) was attuned to English. While it didn’t get all the German unique orthography wrong, it quite often did. Umlauts were often dropped (e.g. the pdf would have “Marz” instead of “März”) and eszetts (“ß”) were often substituted with one of a plethora of possible Latin letters. I have intercepted some of these errors during the process of ankization (creating Anki cards), but some passed through. Also, presumably it is possible that there were errors in the original scanned book.

Second, there were errors that arose during my ankization, mostly in the form of typos. These included letter substitutions (one real example is my flashcard for (political) “Party”; my German side read “Partel” instead of “Partei”. I attribute this error’s very long endurance to its similarity to other correct German words such as “Viertel” and “Achtel”, both of which were in the same deck) and letter permutations (e.g. “geschreiben” instead of “geschrieben”). I can’t remember if such errors occurred, but it’s possible I have input the wrong gender for some words (e.g. that I put “m” instead of “f”).

Ostensibly, I was merely learning word pairs (German word – translated word), such that the learning of one card/ word-pair should not in any way interact with the learning of another card. However, German morphology is not random. There were different lexical patterns in my deck, such as general word structures, as well as a relationship between word genders and their form (for example, words ending in “-schaft”, generally equivalent to the English “-ship” (e.g. “relationship), are always feminine). It wasn’t something I had expected to happen, but I realized at some point that I have learned some of these patterns, which is why I say that there was an “emergent learning”. What I set to learn was word associations, but in addition I learned something that transcended these.

I came to realize this after not infrequently I’d come upon a card during anking (reviewing of Anki cards), get a sense that it is wrong, and after checking it, find that I was right in my suspicion. This intuition(!) of mine did not come from “meeting” the German word “out in the open” and finding out that I was wrong about what I had learned. Rather, based on the rest of the deck, that word would strike me as profoundly irregular and exceptional, defying the pattern set by others words. This phenomenon suggest that sufficiently big Anki language decks could be rather resilient to error, as one can detect these “outliers” during the process of learning.

To drive the point home, I’d say I find this noteworthy because had I been a “perfect learner” I would have learned all my cards together with their mistakes. Instead, my mind simplified things in the process, the data going in went through a “compression” if you will so that instead of learning so many letters for each verb, for example, my mind recognized reoccurring forms and simply learned for each verb its form and a few letters (not all) necessary to construct the verb. That there are such verb forms in German is not secret –– grammar distinguishes between “weak” and “strong” verbs.  The interesting thing is that they were not encoded in the data I was taking in, but I learned them nonetheless (though implicitely. I still cannot tell you what the distinction is) and therefore could detect errors in my cards.

2. “Shortcutting”

I have seen more than one “guide” on the internet that admonishes against using any non-target language in decks, advocating instead the making of “purely target language” (German in my case) decks by pairing simple words with illustrations, and more complex words with word definitions in the target language, the rationale being –– if I remember correctly –– that by doing so one avoids forcing the brain into performing the acrobatic feat of going through another language every time a word is being recalled. The assumption here is that after learning word pairs, whenever I want to say “Huhn”, my mind would have to first conjure the word “chicken” from the idea of chicken and then recall the German association “Huhn”, instead of jumping directly from the idea of chicken to the German word.

I have not found this to be the case. Most of my “interaction with German” has been through my Anki decks, and yet ––– and this was particularly the case during my very intensive vocabulary deck learning period ––– sometimes while speaking English and picking my words as I go I’d recall a word in German immediately, but have to search for a longer or shorter while for the English word. These were well rehearsed words, but one cannot attribute the effect only to the frequency of exposure, since every time I was exposed to the German word I was also exposed to the English word ––– due to the nature of the word-pairs ––– and, in addition, I was more likely to encounter the English word outside of the deck context, which would suggest that I should have been better at recalling the English rather than the German word. With regards to the exposure, I could speculate that my “learning mind” paid more attention to the German rather than the English word (which by itself would suggest that it “doesn’t matter” if one uses non-target language in the cards), making the German more “salient” to my mind, creating an effect that is akin to being exposed more often to the German. Otherwise, there are few other “minor effects” that I speculate could have been at play here but which I find not to be of particular interest and so I won’t elaborate on them here. However, for the curious mind I’ll offer one hypothesis about one particular word: “Topf”.

Still today, the word “Topf” suggests itself to my mind quicker than the English word through which I have learned it the idea of a “pot” (the saucepan kind). In fact, this effect is so strong that I believe I conjure the English word by “translating” the German one every time I want to use it. This is very speculative, but I believe that this is why it is the case: First, I think there might be a onomatopoeic effect. Purely by its sound, “Topf” suggests to me more strongly the household item than the word “pot”. More significantly, however, I believe, the word “Topf” is unambivalent, while the word “pot” has many different meanings, only one of which is “saucepan”1. It’s as if by “smearing” its meaning over many different concepts, it had lost some of the strength of its association with this one particular item. In addition ––– as comical as it sounds like ––– the word “Topf” contains the English word “top”, which to me is very strongly associated with the way one uses a pot: you put a pot on top of the stovetop, and you put the lid on top of the pot.

Regardless of this curious phenomenon, I think the mind creates “shortcuts”  where they are necessary, the same way it does when one learns to play the piano, for example. Upon learning to play a composition, a beginner might have to think about putting one finger on the C key, another finger on the E flat key, and another one on the G key every time she encounters the C minor chord in the score. With experience she’ll be able to “just play the chord” without thinking about its constituent keys, and even play common (and not so common) chord progressions. Similarly, with practice I think the mind “abandons the scaffolding” and jumps immediately from the concept to the target language word.

I don’t necessarily advice against using a pure target language deck. However, if you are at a point where you need to decide which path to take, I’ll offer the next words, based partially on common sense and partially on speculations and limited experience. I’ll address specifically the issue of using illustrations as the back side of the target language word; there’s no problem in using a target language definition as the back side for learning of “more advanced” words ––– I have used such a deck very successfully with English.

First of all, it takes significantly(!) more work to construct a deck that uses images, as oppose to only words. First, you need to find an adequate picture, and in addition you need to manually copy it into the “add card” window. Can you imagine doing this for 100 cards? 1000? 4000? If you use only words you can copy pairs into a single text file and then import it wholesale into a deck (I imagine one might implement a tool that allows to import a list with words and images. If you ever do it, I hope it’s for a better cause than for language vocabulary learning). Further still, the internet is bountiful with word lists waiting for you to grab and import them, to say nothing of existing shared decks2. There are also shared decks with picture-word cards, but how useful do you think it would be to use them, such decks made by other people? Is that a picture of a “dish”? “meal”? “plate”? “spaghetti”? And nouns are relatively trivial ––– imagine illustrations of verbs: “receiving”? “giving”? “trade”? “friendship”? And how are you going to know if the picture is a verb or a noun (yes, there’s a trivial solution to this, but it probably would not be implemented in the deck you’d download)? You’ll end up looking the word in the dictionary anyway, and thenceforward according to this naive theory you’ll arguably have to first think of the translated word first, lengthening the path your mind has to make whenever you recall the word (e.g. concept -> picture -> English word -> German word), effectively only aggravating the original problem rather than solving it. This problem, by the way, might also exist in decks you create yourself. What if you neglect a deck and start to forget the “captions” of your pictures? And how did it work in the first place? Since you had to dig up the pictures to include them in the deck, when you’re reviewing the cards aren’t you first recalling the word you were “translating” into a picture when you see the picture, before trying to recall the target language word? I don’t know, I’ve never practiced with such a deck, but I speculate.

Sentence memorization. While my vocabulary deck above made me good at recognizing a lot of the words I have encountered, as well as impress German speaking friends, it did little to advance my articulation or comprehension of sentences of a structure more complex than the very basic kind ––– and my German remained roughly at the “party trick” level. What prompted me to get out of the rut and move to memorizing sentences (as planned, but never scheduled) was a comment of my friend (a Spaniard who had learn German after high school) who said I could have spoken a considerably better German had I invested those 500+ hours in more orthodox ways.

As you’ll see now, I didn’t quite recede to orthodox ways of studying, but I did move on. Retrospectively, I’m not sure how valuable it was to have spent that time on vocabulary learning, in particular as I was supplementing my deck with less common “found words” (which takes time to rehearse). I think there probably was some merit in the study as a whole, but in the name of urgency I could have better spent my time by proceeding-on from vocabulary learning instead of ever expanding the vocabulary deck. I think it might be helpful to have an “advanced vocabulary deck”, but mostly after the general basics of the language ––– syntax, grammar and so on ––– had been already mastered (indeed, this is what I had done with English, to much benefit).

As far as my study of German served a dual role of acquiring the language and of a running a “studying methodology experiment”, at this point I had waned off my initial “purist” approach3. I went through an English online “German grammar review”, whose example sentences were the basis of my new sentences deck. Some, by the way, turned to be erroneous; don’t use non-native speakers’ language guides, especially if they don’t reside in a “target language country” –– even if they are professors of the language –– unless you have a good reason to trust them. Other than mining the sentences I did make an effort to learn the “prescriptive rules” of the language. I suppose that that review was not entirely comprehensive, but otherwise my assumption is that if one memorises enough examples of “every sentence structure”, one can get a good basic command of a language (and then a more comprehensive vocabulary learning is the next step, whether through exposure to the language, or through rote memorisation).

Currently I’m at this stage, and I’m yet to be able to say that I’m a speaker. It’s hard for me to assess how good my German is currently; some texts are fine for me to read, while others (for example, all kinds of letters from banks and so no) intimidate me. I’m doing fine in simple interactions, such as with shopkeepers, but perhaps it’s not much of a standard by which to judge. I think I’m at a good point at which I could move on to “immerse myself” in the language.

I can make two remarks, however, about sentence-decks. First, while there is an “emergent phenomenon” of the kind experienced with vocabulary decks, it’s harder to take advantage of it. There are cases when several cards suggest that another card is wrong, but sometimes it is not the case ––– things get more complex once we step from patterns of words to patterns of sentences. The problem is that It’s harder to check the correctness of a sentence ––– one cannot just look it up in the dictionary. However, if one verifies somehow (through a German speaking friend, for example) that the seemingly wrong sentence is indeed correct, it’s a good indication that the deck needs more examples of the singular card’s case, since the point is learning to generalize from the deck, and that card proved you that your generalization is not good. Without more example, one might learn a “over-generalizing/simplified” rule.

I perhaps took a wrong approach, as well. My deck contains bidirectional cards (German -> English and English -> German), but I think it might have been better to have only German -> English cards, at least for a (long) beginning. Having the English -> German ones is of course beneficial for learning expression but:

1. It takes considerably longer time to review them ––– obviously it’s easier to learn the meaning of sentences than to learn how to form the German translations. I think had I put this expression aside I could have amassed and rehearsed a much larger (solely German -> English) deck, which would have accelerated my comprehension.

2. Somewhat related to an issue above; it is the case that several different English sentences are the correct translation of the German one. One can easily recognize when one was correct deciphering the meaning of a German sentence, even if the back English sentence is not identical to the one one conjured. It’s difficult doing so in the other way around. If the German sentence I “thought up” is not identical to the one I had on the back of the card (for example, a word is switched by a synonym, or a different word order was used), I have no way to know whether my sentence is correct or not, and I can only guess. This, of course, is a problem, since the role of Anki is precisely to be the verifyer. If I was wrong but I thought I was right, or if I was right while I thought I might have been wrong, I learned something wrong. Problematic. This problem does not occur with a deck of exclusively target language -> “source language” cards.

Otherwise, all in all, I think the sentence deck is progressing me in the language, but this is still an experiment on the go!

1. I have learned only during the course of writing this paragraph that the word is “saucepan” and not “sauce pan”. Beside that, I have also learned only now that it seems like “saucepan” refers specifically to a pot with one long handle, less so to a “short-eared pot”. I think this effect (specifically with regards to “Topf”) wouldn’t have occurred had I used a “Topf / saucepan” card rather than a “Topf / pot” card; “saucepan” conjures a stronger impression on my mind than “pot”, which, again,  I think, relates to the multiplicity of meanings of the word “pot”. That being said, it still seems like the “danger” of learning target language words through another language is not real. After all, I learned the association between the idea of the household item and the word “Topf” without strengthening the association between the same item and the word “pot”.
Also, as a side note, I’ll mention that “saucepan” is not a word I have used very much. If I use it more often today, it is solely due to this whole story with “Topf”.

2. I found many shared decks to be problematic for me for different reasons, and hitherto I’ve never used one extensively myself. But I imagine these are added regularly to the anki website, and that an adequate deck can often be found. Especially ones that were generated semi-automatically by their creators tend to have a larger or lesser amount of errors, but it might be worth it at times to use those instead of creating one from scratch.

3. I laud such experiments such as Michal Ryszard Wojcik’s Norwegian experiment, through which he had taught himself Norwegian solely through Norwegian material. Further, I’m ready to believe that this approach can be very effective, but I suspect that the effectiveness might step from the way he had engaged with the language, and that non-target material would not have detracted from progress, and probably accelerated it. However, I can imagine I’m wrong at this point.
Either way, despite the appeal of such a method to my taste, at present I believe the quickest (least time consuming/ most flexible timewise) way to learn a language is through a combination of “pure language immersion” and “supervised learning”.