Lapis app: a segmentation engine that understands grammar, dictionary lookups, similar kanji, homonyms, synonyms, SRS

I’m liking Lapis so far, it’s helped me learn and figure out a few new grammar points I’ve seen or heard around. The only thing for me is the SRS part. I have it set to appear 52 in a year - so once a week because I’m not sure what to do exactly with it. I like how WK has things you get correct show up in short intervals, and then extends it as your continue getting them correct. I would find it easier if there is a way to set it up to do it that way

1 Like

That’s exactly how Lapis SRS does it too. It is spaced properly (just like WK/Anki), but it just uses different settings for that. Retirement interval specifies the max review interval, after which the card will be considered learned. Reps count tells us how many times you want to review the card throughout. Level intervals gets calculated from that.

If you’re not sure you can really just leave the default settings, it should be good enough (subject to change though).

BTW, setting reps count to 8 and retirement interval to 120 days will give you intervals very similar to that of WK (but the shortest unit we have is a day, so it’s still a bit different). I wouldn’t recommend this for sentence cards (WK’s intervals work very well for kanji), but if you’re used to it you can use it.


We’ll have a simulation graph in the future so the settings will be easier to visualize.

3 Likes

Thanks for letting us know! Feel free to keep it up :+1:

6 Likes

Awesome project, seriously. Just tried it out and it works really well. As of now, the site is already plenty awesome. What kind of financing do you have in mind for the future? Donation-based? Subscription-based? One time payment?

Also, there’s one thing I’m missing badly at the moment: An option to reverse cards. Like, for example, the card I imagine has this on the front:

Build the following sentence using an i-Adjective:
“Building that was not tall.”

And this on the backside:

⾼くなかったビル.

The input could either be converted using Microsofts IME or an input similar to WK. But without that option, it’s impossible to learn using the grammar yourself. Or am I missing something out and this is already possible?

2 Likes

Thanks!

Right now it’s donation based (Patreon). We want to avoid having a subscriptions model, but we’ll have to go back to it if Patreon based approaches don’t pan out.

What we have is designed with input based learning in mind, and without having to write answers yourself (target → native).
So the process I imagine if I want to learn a grammar point: I input a sentence that includes that grammar. I select that grammar in selection mode (so the grammar doc is easily accessible from the back of the card when I need it). In review, I read the sentence and I either understand it or recheck the grammar doc. So it’s comprehension based. The example you outlined is output/creation based.

I know it’s my personal approach. Some people might prefer the output/creation based example you outlined above, but I don’t feel like this makes sense to implement in Lapis. The value of the Lapis SRS is in its integration with segmentation and all the other features. It’s really geared more towards comprehension based cards. Does that make sense?

You could offer donations via PayPal or Bank Transfer additionally. I don’t know about others, but I like giving one-time payments that are usually more than I would pay monthly :).

Does make sense :). I think it’s still safe to say that both approaches help you with different objectives. Your approach is geared towards being able to understand Japanese while my approach is geared towards being able to speak Japanese and both are equally important in my opinion. It’s of course up to you what you want to implement, but I do hope you reconsider the idea, should more people other than me ask for it ;).

3 Likes

We want to prepare a one-time option too, but PayPal doesn’t work in the country I’m staying in. We’re looking for alternatives (once we do we’ll be updating the lapisapp.com landing page, thanks).

Definitely. I can foresee this being a different mode, just need to think of a way that sits well within what we have. But to understand you more, what you’re saying is you really want a simple Front/Back both being just text and you answer it by typing it out, and maybe with the addition of the grammar doc we have too in the back as a reference?

2 Likes

Will gladly donate once it’s available ;).

I’m not sure if we’re talking about the same thing. What I imagined is more like an Anki flash card / WaniKani itself, where you only see the front (in this case “Building that was not tall.”) and you have to type in the answer before you get to see the supposed answer and you can decide yourself / let the program decide if your answer is correct. And if you get it wrong or want to double-check your answer, you use the grammar doc to go through it step by step.

1 Like

Yeah, I meant desk when I said deck.

That’s the gist of it. The desks almost do what I want, but not really since clicking the “Segment this sentence” on a desk item just segments it without allowing you to select any additional words to include.

Yeah that’s a major difference in use. Personally, I have ADHD so I do a bunch of prep work to minimize the amount of task switching I have to do. Any time I drop what I’m doing, there’s a good chance I’ll go off task.

Prep work here would be putting the paragraphs I’m going to be reading into Lapis ahead of time. If there was something confusing, I would switch windows from the book to Lapis, and the confusing part would have been prepped already (aside from selecting stuff for inclusion).

Any paragraphs that weren’t confusing would be deleted from the desk after the reading session. Anything that was confusing would have the confusing parts selected and then inputted into the SRS.

Hope that makes sense

4 Likes

That’s exactly what I understood too :smile:

Sounds good as a different mode of SRSing. I would still have said other SRS systems might fit better, but since Lapis has the grammar docs and potential for including a context sentence to segment, there is value in implementing it. I’ll keep this in mind. There’s a huge backlog and a big list of priorities though so it might take a while. Thanks for the feedback!

There’s a “Copy sentence” action now so you’ll at least be able to copy paste the sentence directly to segmentation page.

Oh I see what you mean. Check my next comment for an update on desks (renamed to boards), I think the new update gets you closer to what you want. But one difference still is that (if I’m understanding right) you want to put a whole paragraph as is, without splitting it to sentences first, and you’re imagining a flow in which Lapis walks you through the sentences, and you either add them to SRS or ignore them. Am I understanding right?

A heads up that “Desks” was renamed to “Boards”. The old name was just asking for trouble, it’s too close to “Decks” that even I got them mixed a couple times while implementing it.

The new name fits well. A board is a temporary space where you pin stuff you’re interested in to.

Also, you can now pin an example sentence to a board:

5 Likes

Thanks for considering it! :slight_smile:

1 Like

I don’t particularly mind using entire paragraphs as inputs. The biggest part of my request was just the ability to add inclusions after adding inputs to the board so that I could handle the work in batches.

The current update to boards does more or less what I want. Although it would be nice to be able to modify inclusions by clicking the “segment this sentence” button on the board, rather than having to go > edit > update inclusions > update from current selection for each.

This is also a random observation but the segmenter is unfortunately getting annihilated by hiragana right now:


See: ほんとう、みんな as well as some hiragana spellings of kanji (続ける、許して)
You may or may not be aware of this, just thought I’d point it out in case.

1 Like

Good to know. And makes sense, will see if it’s possible to make it easier.

This is expected as of now. Here’s some more details why:

There’s a big segmentation overhaul being worked on that should take accuracy to another level. There’s a possibility I’ll be able to left the kana restriction then.

2 Likes

Initial implementation of an SRS simulator is now done. You can reach it from the deck settings.

Run a simulation to anticipate how the review load will change with time. You can tweak some fields as you approach the desired ideal review load you want.

4 Likes

Here’s another segmentation issue I ran into in Lapis.
(Please let me know if these are useful or if I should just wait for the big overhaul you mentioned in the previous post before mentioning additional ones)

眠たげなその子を見下ろす。

Lapis seems to have trouble figuring out this is 眠たい + げ grammar construct (seemingly / showing signs of being sleepy). It actually takes たげ and believes it comes from てあげ. It marks 眠 red and doesn’t give me a box for it:

4 Likes

So the OP doesn’t really succinctly explain what Lapis is, just that you are introducing a project called Lapis. Maybe someone can either confirm my understanding or correct my lack of understanding.

The general Idea is that when I come across sentences during my studies that I don’t understand, I copy + paste them into lapis, Right? This isn’t a program like Wanikani or Bunpro that you work through or level up in. Am I understanding that correctly?

Maybe I’m just daft, but for whatever reason I didn’t come away from reading the OP with a good understanding of how Lapis works, just that it does Segmentation, Dictionary Lookups and has a built in SRS, that are utilized in…some way.

Perhaps if I just made an account and poked around in it I would understand, but I’m tired and don’t feel like it at the time of posting this. Again, apologies for the silly question.

2 Likes

Most of these will be taken care of after the big update, but feel free to continue reporting them. They sometimes help me find edge cases.

As for this one in particular, it’s just a matter of registering the げ grammar (since this is a straightforward one and kinda common, I registered it, should work after I deploy, thanks). As of now, we have ~200 grammar points registered. That’s a very small number, but there’s a higher priority on finishing the segmentation overhaul before we get started on registering new grammar. This does mean segmentation is limited until then, but since this is still beta it’s OK.

1 Like

No worries. Yeah logging in and trying it out would probably tell you all there is to know. I think “just that it does Segmentation, Dictionary Lookups and has a built in SRS, that are utilized in…some way” is as good a takeaway as any without trying it out. (Just in case you missed it, there are features intros hidden in a collapse, click on them to expand)

There’s a segmentation tour in-app when you login too. There’s also a knowledge base, it’s still a work in progress.

Correct. There’s no levels, it doesn’t have a predetermined course similar to wanikani or bunpro.

Think of Lapis as a combination of a dictionary, a grammar reference, and an SRS. You enter a sentence, it shows you all the words/the constructs/the grammar, and provides a super easy way to create an SRS card out of the result. There are more nice stuff around this, such as a database of example sentences smartly indexed by grammar points, vocab sheets, etc.

The goal is that Lapis becomes most of what you need throughout the language journey, especially after someone starts reading and listening to native material.

Heya, I absolutely love Lapis so far, it’s interface is quite intuitive to me and the information displayed is clear as water.

Might I suggest also advertising Lapis on the Bunpro forums? I’m sure people there are the perfect market for it.

2 Likes