Finding a "right" way to study English → Japanese

I’d like to plan my own English to Japanese SRS, but I need your help.

I’ve made several Anki decks, but they’ve had issues that I couldn’t make work. I’ve tried Kamesame and KaniWani, which are wonderful tools, but they don’t quite solve the problems that I’ve faced.

Here’s the first problem: speaking isn’t thinking of one specific word.

I have dyspraxia, which means that even in English, I’ll try to think of a word and my brain makes that clicking sound like when you try to start a car with a dying battery. I’ve found that reversing the polarity of WaniKani isn’t enough to get practice in going from English to Japanese. Every SRS tool I’ve ever used (even the ones I’ve made) center on thinking of the specific Japanese word on the backside of the card.

Both English and Japanese both have a lot of near-synonyms with slippery/nuanced differences. Thus, this exercise gets murky fast. It devolves into a guessing game.

So that brings me to my main design principle. Like WaniKani, English clues should have multiple answers in Japanese. The best I’ve found is weblio, but if anyone has a better one to use, I’m open to suggestions (Jisho doesn’t quite do the trick for this task).

Let’s say I made a card where the English prompt was “threat.” To get this correct, I’d have to type in one of these words (via weblio): 脅し、脅迫、脅威、脅威となるもの、兆し、恐れ。

For a while, I thought this was what I was looking for. I’d go to weblio, type this stuff in on each side, and I’d be off to the races. Then I realized there’s a second problem: Japanese and English use words in multiple contexts that don’t always overlap.

Let’s go back to “threat” up there. 「脅迫」 means “threat” in the sense of, “Do what I want you to do or I’ll hurt you.” But 「脅威」 means “threat” in the sense of a menace, like how Magneto is a “threat” to the X-Men (whenever he’s not a hero or a baby or a young amnesiac clone or whatever). I wouldn’t want to use one word where I mean the other. And this is an easy mistake to make with a whole lot of words. I speak from experience with a lot of native speakers who finally beg me to just speak English.

Then a huge number of English words raise a third problem: English often uses the same words when Japanese doesn’t.

That’s why intransitive/transitive verb pairings are such a pain in the ass for us English-speaking Japanese learners. This also raises a number of problems if my English prompt is a word like “present,” which has multiple meanings across parts of speech. As you can see, the weblio entry is way more info than would be useful to write on a single flashcard.

The logical solution would be to make the English prompt more specific, right? Maybe make one card where the prompt is “present (gift)” and another is “present (now)” and another is “to present.” But that’s not how we think in English. I use the word “present” when it occurs to me to use the word “present.” I don’t think, “Here’s the context in which I’m using this word and how will I now put that concept in Japanese?”

That raises the fourth problem: if I make the English prompt too specific, I’m right back at the same guessing game that I’m trying to avoid.

Well, that’s where I’m at now. I want to make a tool that isn’t just guessing one specific Japanese word per flashcard, but I still have no idea how to go about doing that. If you have any insight or ideas, I’d love to hear them. From my decades of experience, I believe efficiently studying Japanese production is an open problem. Maybe we can find some solutions!


I think you’re looking for solutions where there really isn’t a good one. SRS isn’t some perfect all encompassing method to learn a language. SRS, if anything, is actually meant to be more of a secondary tool and a support to actual exposure in the language. Likewise, I think practicing production through SRS is secondary to actual production. In fact, I don’t and never did it and don’t know of anyone I look up to language ability wise who has ever done it and recommends it. I think its simply not necessary and another thing on the list of “You feel like its helping more than it is”.

So in your endeavor to find the “right” way to study output, I’m curious how you arrived at the idea that SRS should be a part of it to begin with? If I’m being entirely honest, I think you’ve got it wrong fundamentally.


I have to agree with this. If you have already recognized that there is rarely a 1 to 1 between Japanese and English, then why would you think SRS would be the best way to move forward in improving your skill? I think the best way to learn Japanese is not by cross referencing it with English, but to learn it as is and in context. Sure you can pick up words from Wanikani and other places, and you can look them up on Jisho to confirm an idea of what a word is. But overall you are better off learning usable Japanese by watching/listening/speaking. There are lots of words I know how to use correctly in Japanese but only have a loose idea of what they mean in English and thats because I gained them through exposure.
It is commendable that you are trying to find an organized and methodical way to expand your vocabulary, but it does seem that you are making it harder on yourself than it needs to be. Sorry if my advice is unhelpful. I truly wish you the best.


I agree with what others have said, I’m not sure if you want to do this.
If you really just wanted a solution, then it sounds like this could be at least partially resolved by adding more context to the front of the card (moving from word → word to sentence → sentence).


Magneto is a threat to the X-men.


Magneto is a threat to the X-men in Japanese
other example sentences using 脅威

I’ve found that my En → Ja cards tend to be some of the weakest, largely for the reasons you’ve mentioned. What I’ve started doing instead is having Ja → En sentences with a single word bolded to represent it as the focus of that card.


ボブさんは日本へ行った あとで 病気になった。


Mr. Bob became ill after he went to Japan.

a bunch of explanations / context on how あとで is used (from a grammar dict).

For the most part these sentences are not ones I create (as I might make mistakes), but instead taken from other resources such as manga I read or a grammar dictionary - often with small modifications like changing character names to be more generic so I don’t just rote memorise the character name → back.

I think studying these Ja → En sentences still helps reinforce the En → Ja usage, even though it doesn’t directly answer you question.


I’m in agreement with @Vanilla in that SRS is a poor substitute for producing Japanese in a real situation, which ideally would involve getting feedback from expert (typically native) speakers as you go. If your objective is to find an easy way to not produce the wrong Japanese words when you need them then I think you’re out of luck with this idea.

However, if it’s the research that drives you then creating an accurate English / Japanese ontology from scratch would be a colossal effort that would outlive you in all honesty. You’d also need a lot of input and commitment from native Japanese speakers for it to be accurate. uses data from the JMdict/EDICT project (sample entry) which has a number of “senses” for each word and then a list of “glosses” in various languages for that sense. It may well be that the raw data would serve you better than what you see on the Jisho site.

It seems to me that cross-referencing the senses (different Japanese meanings of a particular word) against the glosses (words / phrases in other languages for each sense) could be a starting point for your project.

1 Like

I think you’re discovering why many languages (not just Japanese) are taught by way of sentence patterns, gradually progressing from simple (“This is a cat”) to complex (“Tyger Tyger burning bright,/In the forests of the night: / What immortal hand or eye, / Dare frame thy fearful symmetry?” – William Blake.)

The simpler the sentence, the more amenable it is to basic pattern matching. The more complex, the more it requires “native-like” understanding of how the language “works” and the less SRS-friendly it is.


For output practice, grinding conversation is one choice but I like to practice output by writing more (w/ native feedback if possible), which is true for grammar too (get rid of the English). One can take their time to gather thoughts on new vocab/grammar and if practiced enough, it will come through in speaking eventually for me. The repertoire I need for output is must smaller than for reading, so I gravitate to what I favor saying (or just can say) in writing rather than all these possibilities which are not always so essential (though unavoidable for comprehension). I’m also finding favorite studied sentences in SRS are also helping with output even though it is not an Eng->Jp exercise. The context seems to get memorized easier and this is just naturally happening so something I’m trying keep up with as long as it is beneficial probably because it is closer to a Jp->Jp study scenario w/ surrounding Japanese in a repetitive study format (again, getting rid of English whenever possible). I don’t live in Japan so my excuse for output is not a daily basis unless I find ways. I find just general reading, I don’t get this effect as easily. I’m not able to just memorize a new word or expression on first pass or even repeated just on input alone and have ready for output in long term memory unless with some repeated exposure. I don’t claim to be a great speaker, I’m not where I want to be but this is what has helped me recently.

I use Kitsun Eng->Jp with my regular Jp->mean/reading reviews which I can either set with the same review pile or w/ sibling delay. It sort of eliminates the synonym problems I had with KW or KS, and I just make a judgement call whether I want to practice the Jp output further or not. A single platform is much better IMO too because it’s a harder threshold but much better longterm for later SRS stages I find…I don’t know if this is how Anki is used, probably can do the same if not already. I don’t really like to dwell on it though, if I got a synonym correct but not the exact card backside, I might pass it as that was just my choice unless I want to practice that answer more because I was clueless…it’s a judgement call. But it’s only tool to help with normal Jp meaning/reading rather than actual output as a step towards that point I feel. But the Japanese I actually need for output is much narrower and I don’t want any English in my thought process, I’m not expecting too much from this process. Like others said already, I wouldn’t expect too much from this process.

Yeah I don’t think this is machine-do-able. It’s the age-old problem of “grading for understanding” vs. “grading for the specific answer the key writer was thinking of.”

You could either, I guess, give almost a dictionary long description of the word, with explicit (intransitive/transitive) and (noun, verb, suru-verb, adverb, etc), or show a natural English sentence using the concept you were looking for, highlighted with what the user was supposed to translate into Japanese as a single word. You’re still going to have synonym problems.

I mean, one example I just ran into on WK was “prayer”. First problem: is that prayer as in “a prayer” or as in “the concept of praying”. Both are nouns. Ok I can figure that out from the example sentences (which is what they’re there for). But now to ask the question in reverse, how do you ask me for the concept vs. the thing? You really are going to have to have the same solution, example sentences.


I can’t deny that you’re all correct.

What I was hoping to do was use SRS to accelerate the my production practice in the same way that it’s accelerated my recognition practice. And since I’m a teacher, I’m always interested in ways to make learning easier, both professionally and personally. But it seems there really is no feasible way to do that here.

Ah, well. I’m not in a level/situation where I need production at the moment, so it can wait anyway. I appreciate your input, everyone!


If just for yourself, the highlighted example sentences idea might work.

…up until you start context-association-rote-remembering which answers go with which sentences instead of actually thinking through the answer. That’s a problem i run into with bunpro. Once I’ve missed a sentence a few times I start just remembering what I’m supposed to type when I see it. I’m not sure I’m learning the grammar at that point, only reflex typing.

A large enough question bank, though, it might work.


This is my basic issue with Bunpro. If they had a only a handful of all of basic sentence structures but with only the grammar point different then I think it would have had much more potential. E.g.

  • The cat pushed the ball
  • The cat pushes the ball
  • The cat didn’t push the ball
  • The cat isn’t pushing the ball
  • The cat wanted to push the ball
  • The cat didn’t want to push the ball

It would be less interesting but it would have stopped users being able to avoid the intended process by being able to remember the sentences.


Didn’t you make your own system for that purpose???

That’s a really good idea, actually. Cat, I want YOU to push the ball. Thank you for pushing the ball, Cat. You’d really only need a very small vocabulary to illustrate most of the grammar points (plus particles, the grammar point itself like より or ほうがいい, etc.)

I’m still not sure how you would auto-grade that without someone having to type in thousands of sentences and the correct answers, but it would isolate the grammar from the vocabulary much better.


Well… Yes I did! I suspect I’ll be thinking today how much effort it would be to repurpose it.


I think trying to get this kind of grammar tutorial across and get users producing at the same time is probably too hard. I’m wondering if it could have an easier multiple choice mode as well as having to type in the blank(s).

  • Test for correct grammar but near misses.
  • Basic grammatic errors.
  • Correct grammar but substantially different meanings (easier mode).



This is roughly what I’m doing with my sentence cards.

I take sentences from manga/grammar-dict/anime and change the names and objects to be more anonymous to make it harder for me to just pattern match(e.g. “oh it mentions Sasuke, so it has to be one of those 2 cards I added from Naruto”).

I’ve been finding that for many cards I’m now kind of skipping over names in my head and just coming up with a translation like “blah blah got sick after going to Japan”, so I might even change names & objects to some filler values (e.g. X, Y, Z).


Like a bauss.

Like a double-bauss.

Little off topic but curious

Have you ever tried learning tools designed for dyslexic Japanese people?

Very very curious to know your experience with those. (If those tools even exist)

First, minor point, Japanese is full of words that have multiple meanings when translated to English. It’s a many-to-many problem (as is basically any real language translation). 高い, for example, meaning tall, but also expensive. And Japanese is famous for its homophones.

As for your main question, I tend to agree with previous posters. SRS is not an ideal tool for production related tasks in general. There are a few areas where I’ve found it helpful:

  1. Conjugation and other grammar. Often, there’s one right answer, so you don’t have the one-to-many issue. Also, the advantage of this over drills is you can narrow in on the ones that cause you difficulty naturally by SRS. Drills are probably still needed early on, though.
  2. Repeat after me exercises. Repeating set phrases after pre-recorded speech at increasing intervals can really help develop the mouth feel of speaking a language. Often production is as hampered by the physical vocal dexterity as much as by the mental blanks. Having confidence in the one can help with the other. If you choose the right set phrases that are flexible and adaptable can allow you to communicate even if you can’t find the right word. And there’s something to be said for a nice long set phrase that gives your brain time to ing catch up with your mouth and remember what you were wanting to actually say! :wink:

Beyond those, for more general cases, have you considered using images rather than text as your clue? I use images a lot in my Anki decks.

Images can be endlessly unique and nuanced (you can put additional textual context on the back of the card if you feel you need it) so you should never have the “present” problem. The difference between the gift wrapped in a bow vs the guy at a white board with a laser pointer should be obvious.

Another aspect of images is it relies less on your L1. Doing Image → Japanese or Japanese → Image exercises eliminates the translation problem entirely, which causes many hiccups in production for new learners.

In my vocabulary decks, which have both English gloss and images, I often find I think of the image and the meaning first and struggle to think of the English word or phrase. I know what the word means, but don’t need the English crutch. This is occasionally problematic in my WK reviews where my decks overlap, since WK wants a specific phrase and I can only think of the meaning and image.

Images do take time to pick and choose, but that time is actually valuable in and of itself. The time you spend thinking about the word and its context while doing google image searches to find the right image is a great way to get the word and context into your head in the first place. So it’s not ‘wasted’ time by any means.

But yeah, that works really well for me, but I’m a highly visual thinker. I don’t know enough about dyspraxia to know if that’ll make a difference here? Heck, it may short circuit the short circuit for all I know, so at least you’ve got one less word to cause your brain to ‘click’.

Good luck. I have my own mental issues that make learning another language a real struggle, so I appreciate the problems you face, even if not the specifics. Let us know how it goes, whatever you decide to do. I’d love the data point for when I’m evaluating new methods to circumvent my own issues.


I’d love a system like this for comprehension drills. A few sentence structures, but fill in different elements as needed. Sentences don’t even need to really make ‘sense’ in the real world. “The building doesn’t want to eat the walrus” is just as good for grammar as “The cat doesn’t want to push the ball”. So you could just have patterns that have lots of ‘fills’, pick random nouns and verbs and build (practically) infinite sentences for practicing common grammatical constructs.