CleverDeck Blog

Flashcards aren’t very sexy – especially when talking about the bleeding edge of language learning tech. In this post, we wanted to write about some of the decisions and research that went into CleverDeck and show you that, despite the association with dusty old index cards, we think we’ve put together the coolest flashcards in the history of flashcards.


Learning a Language vs. Learning Vocabulary

Learning a language is a tricky business. Immensely rewarding, potentially career advancing, and a lot of fun – but challenging in a way that learning a new math concept or dance move isn’t. It’s more like learning piano or body building. It takes daily practice, years of commitment, and there will always be room to improve. What feels like slow progress in the beginning leads most people to sell their piano, let their gym membership expire, and give up on learning a language before the real rewards show themselves.

This isn’t to dissuade anyone from becoming the ripped, Japanese-speaking, concert pianist of their dreams. It’s just that when we talk about learning a language – and specifically about how software can help accomplish that – it’s important to accept it as a complex task that takes a lot of focused work. Learning a language requires a multi-faceted approach: one that includes speaking, listening, grammar instruction, reading. . . and memorizing lots of new words (we’ll get to that). Don’t let anyone tell you otherwise. Rosetta Stone has made billions promising that their magic yellow box would make learning a new language as effortless as learning your first language as a child.

Given that learning a language is such a daunting marriage of different skills, why would we build an application that only addresses one part of it? Simply, narrowing our focus from learning a language to learning vocabulary let us focus on building a solution that fits that particular problem really well. We think we know how to build a really good application that helps people learn lots of words; we’re not so sure what the perfect holistic language learning application looks like (but probably something like if you combine this and this and this and this. . . and this, of course).

CleverDeck isn’t going to make you fluent in anything if it’s the only thing you ever use to learn a language – we don’t think any single tool will. But we think it’s going to make your life dramatically easier as part of your set of tools for learning your new adopted language.


In Defense of Flashcards

Rote memorization is out of fashion these days and that’s a good thing. In the context of language learning, this has led to a lot of emphasis on acquiring vocabulary through exposure – that by simply listening and speaking enough, we can learn new words automatically. Unfortunately, while learning new words from context is an extremely good strategy, research shows that vocabulary should be explicitly taught and learned. It will not develop fast enough through simple exposure to the spoken and written language.

As much as we might wish for there to be a magical way to learn thousands of words through osmosis, it turns out that simply testing ourselves on information we want to remember isn’t a bad approach (better than just studying lists, anyway). Even some of the dogmas in the language learning community, like always using a monolingual dictionary and learning new words using already known L2 vocabulary, don’t seem to be supported by the research. What is clear is that building a substantial vocabulary quickly can be immensely useful in pattern recognition and grammatical understanding that aids in every other part of language learning. If we were to pick only one part of language learning to help solve, it seems vocabulary acquisition isn’t a bad one to start with.

So, an index card with a word on the front and translation on the back doesn’t seem like a terrible approach, if not old-fashioned. Certainly much better than just relying on exposure - and certainly better than getting blue in the face discussing language learning pedagogy all day instead of actually learning. But this is the 21st century, and we walk around carrying powerful computers in our pockets all day. Surely we can leverage this fact to improve on the old index-card-and-pen model.


Spaced Repetition

Making a big stack of flashcards and reviewing them every day is one approach. Research on memory retention shows us we can do so much better, though.

It turns out that our ability to remember something follows a predicable pattern. The first time you learn a new piece of information, you can expect to remember it for a certain amount of time. Let’s say one day, for the sake of example.

If you review that particular information again, you expect to remember it longer – two days. If you review it again, you’ll remember it even longer – four days. If you review it yet again, maybe you’ll remember it for fifteen days. The amount of time you can remember increases exponentially with each review – but only on one condition (here comes the cool part):

Research shows that we can optimize our memory retention by reviewing information just when we’re about to forget it. This is great news. Not only can we spend much less time reviewing information, but we get results that far exceed the naive approach of reviewing every day.

Of course, the tricky part is knowing when you’re about to forget something. That’s where an SRS (spaced repetition system) application like CleverDeck comes in handy – by modeling the patterns of memory failure described in research, CleverDeck predicts when you will forget words and alerts you to review them at the optimal time.

But we can still do better. By telling the application whether or not we remember a word that it has decided to show us, it can adjust it’s internal model to match our real-time feedback: it can show us words we don’t remember sooner than those that we do.

SRS applications can be arbitrarily sensitive to feedback and encode a wide range of confidence reports (so, you could tell the application how well you remember something with a rating from 0-6, for example). Based on our own user testing, we decided to build three confidence levels into CleverDeck: I forgot, I barely remember, and I immediately remembered it. While more confidence levels could have made the system more sensitive, we also took into account what psychologists call decision fatigue. Having too many choices leads to cognitive exhaustion and can be counter-productive – deciding between I remembered it pretty easily and I remembered it easily for every card did, in fact, make the app unpleasant to use for most people.

Much has been written about spaced repetition, so check out some of these other sources if you’re interested.


How Many Words?

So, you’ve chosen a cool spaced repetition application (hopefully CleverDeck) to manage your vocabulary learning and can’t wait to get started. Now you just need some words to learn.

The best way to find new words to learn is to encounter them in the wild – either from a grammar book, newspaper, TV show, or personal conversation. The personalized context is extremely helpful to memory retention. CleverDeck let’s you create your own cards, so you can always add a word you find yourself.

But we realize that most language learners can’t commit full-time to learning. We wanted to pre-populate CleverDeck with enough words so that those of us that can’t (or don’t want to) manage and create our own lists could simply use the app and learn enough useful words to reach basic fluency. Something you could just pull out of your pocket and use each day to learn all the basic vocabulary you need to know in a language in an efficient and optimized way. We warned of viewing any learning tool as magical earlier, but this idea gets us excited and is why we built CleverDeck.

How many words do you need to reach fluency? The answer depends on the language and corpus (subtitles vs. newspapers, speaking vs. reading) used in research, but studies show that just 100 words will make up about 50% of everything you will ever hear and that with only 1000 words, a learner can comprehend about 72% of what is written and spoken around them. The benefit of learning more words decreases exponentially – so learning another 1000 words results in a 7% increase in comprehension, and adding yet another 1000 brings us only another 5%.  Based on our own survey of the available research, we found 3000 words to be the sweet spot, most importantly because a 3000 word vocabulary seems to be enough to confidently discern the meaning of new words from their context. It also corresponds well with being able to pass the highest levels of the Common European Framework (CEFR).


Which Words?

Okay – so 3000 words. Which ones do we learn?

In deciding which 3000 words to build into CleverDeck, we found that a hands-on approach combining the expertise of native speakers, teachers, and multiple sources produced the best results.

Simply including the 3000 most common words of a language isn’t a good approach. Finding frequency lists is easy, but they include a huge number of articles and derived words. Many prepositions and articles don’t make good flashcards, and we don’t want multiple flashcards for different verb conjugations (for example). Frequency lists were helpful in making sure no important words were overlooked, but in general, frequency isn’t a great measure of “usefulness.” As a language learner, I want to learn the word for classroom, but it’s very unlikely that classroom is a high frequency word in any language.

Defining categories and choosing important words within those domains allowed us to build a list that was diverse, comprehensive, and useful to many different user profiles. By allowing users to prioritize entire categories, we made sure that a businessman, student, and artist would all be able to find the basic vocabulary most useful to them.

We also cross-checked our own lists with those in popular grammar books and user-created lists hosted around the web.

Finally, we included a handful of essential phrases as the very first items in our lists so that CleverDeck would be useful to tourists and travelers, as well as hard-core language learners.

Of course, no matter how hard we try, we know we are going to find words that we wish we’d included. We’ll add them to later updates as we find them, but users can always create their owns cards for any words not included by default.


Adding Sensory Cues and Context

Each time we encounter a new word in real life, whether in our native or target language, it’s surrounded by an array of sights, sounds, feelings, and contextual cues. Studies show that we learn most effectively when one or more of our senses are engaged – and each additional sense translates into greater comprehension and memory gains.  That’s why we included audio and hand-selected imagery to every card we bundled with CleverDeck. By seeing and hearing words, your memory benefits greatly over just reading. CleverDeck also includes different study modes that test on different senses, like a mode that only presents the audio and a mode that requires the user to type the answer.

But we can still do better than just adding additional sensory information. Context also dramatically improves memory retention. In the absence of discovering a new word in the wild yourself, we tried to recreate the experience of context in two ways.

First, we implemented thematic clustering within our lists. Words are grouped into categories (such as politics, music, and business) as discussed above, but these categories are themselves broken into smaller chunks that comprise the 3000 word list. In cases where words did not have an obvious categorization, we tried to place them next to groupings that share some cognitive relationship. According to schema theory, this clustering helps our brains make more immediate connections with new information that aids in memory retention.

Second, every word included in CleverDeck is accompanied by an example sentence. This not only provides the contextual benefits described above, but helps reinforce and introduce other vocabulary in context, aids grammatical pattern recognition, and resolves semantic ambiguities that may arise in translation. We also attempted to show verbs as transitive or intransitive and reveal their object markers or prepositions where applicable.


Manufacturing Motivation

Few of us can maintain the glassy-eyed inspiration and fervor with which we first decide to learn a new language forever. That’s perfectly normal and okay, but we built some game mechanics into CleverDeck to hopefully help convince your brain to stay engaged for the long haul.

Remembering cards correctly awards points. You can spend these points to unlock “prizes” unique to each CleverDeck (so, the French version has French prizes, and so on). There’s also a mastery bar that allows you to visualize your progress.

That’s the basics, but we thought we could take some hints from the addictive features of some mobile games to do better. Our brains love exponential growth and random reinforcement, so point values increase by unpredictable, exponentially increasing intervals (depending on mastery). You can multiply all point rewards by building a multiplier that increases if you access the application every day.

Finally, CleverDeck uses your phone’s notifications to remind and motivate you to review your cards each day. (Of course, you can adjust this or turn it off altogether.)

Due to the fundamental nature of the tasks, it’s unlikely mankind is ever going to create a learning application that is as crippling addictive as Candy Crush. But research shows that our brains love these gimmicks, even when it is conscious of them. We hope we’ve put them to good use.


Being Smart About Getting Smart

So, hopefully we’ve made the case that CleverDeck is more than just digital index cards. We’ve taken what we could from the available research, didn’t take shortcuts on putting together content, and applied our own decades of accumulated experience learning languages in building what we hope will be extremely useful to the language learning community. It’s been a labor of love, to be sure. We look forward to releasing more languages soon, adding to the ones we already have, and continuing to improve CleverDeck to make it as useful as we possibly can.