The reminder comes back every day, and so does the number on the icon Only one reminder was ever waiting in the queue. It went off in the evening, put the number on the app icon — and after that there was nothing left to go off, so the next day brought neither a reminder nor a number. Reaching the goal made it worse: clearing the queue for today cleared tomorrow along with it, so a day of practice switched the reminders off until you next opened the app. A week of them stands in the queue now, one for each day, each carrying the count that will be true that evening. The session bar is full on the last question The bar counted the questions you had closed, not the one in front of you: it stood empty while you answered the first question and stopped a step short at the last, showing six of seven with the seventh on screen. It counts the question you are looking at now, so the first one starts it and the last one fills it. VoiceOver says what a role row will do Level rows and the sister-app cards learned this in the last release; the rows that matter most did not. A role on a level, a role in the particle list, a weak role and a weak pair each read their name and their percentage in one breath and stopped there — so a row that starts a session sounded exactly like a row that reports a number. Every one of them says now where a tap leads. Reduce Motion turns the animations off Readings appearing above kanji, a translation sliding into place and a section folding away all moved regardless of what the system had been asked for. Reduce Motion in accessibility settings now stops all four: everything still opens and closes, it simply stops moving. Two first-level sentences used grammar from three levels up Both asked about が, and both did it with ~てくれる, the giving-and-receiving form this app introduces much later. One of them read “the person next door helped me”, where the “me” came from a word a beginner has no way to find in the Japanese sentence. Both sentences say the same thing now without that form, and the particle being asked about is the same as before.

The same question does not come back the same day One session never repeated a question, but the next one knew nothing about the last: three sessions in an evening drew three times from the same pool, and the same sentence could come up in all three. Sessions now remember what the day already asked and put those questions at the back of the queue. They come back only when nothing unseen is left, so a long day of practice still gets full-length sessions — it just stops asking twice. The role introduction no longer gives away the next question Meeting a role for the first time opens with three of its sentences shown complete, particle in place. The question right after drew from the whole role, so it often asked one of those three — four times out of five for the smallest role, one in three for the largest. That measures what you read a minute ago, not whether you know the role. Those three sentences go to the back of the queue now. They still come back, they are good sentences — just not first. Pronunciation plays with the ringer switch off, and no longer cuts your music Tapping the speaker on a muted phone played nothing at all, and there was no way to tell that the switch on the side of the phone was the reason. It plays now. The other half of the same fix: the first tap used to silence whatever was playing in the background and never brought it back. One question marked correct Japanese as wrong In 「この階でエレベーター＿降ります」 the variant で was offered as a wrong answer, but エレベーターで降ります is ordinary Japanese for going down by elevator — so anyone who picked it got a red mark for a correct sentence. That variant is gone. The catalog rule it broke is the one this app rests on: a particle that makes a correct sentence with a different meaning is not a wrong answer, it is a second one. が: one role, and it now says out loud that it covers two things The role held two different kinds of sentence — someone alive doing something (犬が走る, the dog runs) and something simply happening on its own (雨が降る, it is raining) — and its old name, “who does it, or what it happens to”, described the second kind as something being done to. Rain falling, flowers blooming and a meeting starting are not things anything is done to. The role is called “who acts, or what just happens” now, and its explanation shows both kinds side by side. A role description pointed at a role two levels away The explanation of が ended with “inside a clause that describes a noun が is needed too — that has its own role here, right after this one”. It does not stand right after: it sits two levels further on, at the connections level. A beginner on the first level looked for it and did not find it. The sentence says where that role actually is now. VoiceOver says what a row will do, not only what it says Rows merged into one element read their name, their percentage and their lock in one breath and stopped there — the lock is a shape, and read aloud it did not say that a tap opens the purchase options. Level rows and the sister-app cards in Settings say where a tap leads now.

が inside a description has its own role now One role was teaching three different things at once. Ten of its twenty-five sentences had が inside a clause that describes a noun — 田中さんが作った料理 — and that is not a shade of the same thing: は cannot stand there at all, and putting it there does not change the tone, it breaks the sentence. Those ten now have their own role, at the connections level. The study day starts at the hour you picked — on every day of the year On the night the clocks change, the boundary moved from four in the morning to five. One day a year, quietly.

Kanji are drawn in a Japanese face again Nothing in the text said which language it was, so the system chose the font for kanji from the languages set on the phone — with English or Polish on top it reached for a Chinese face, where a number of characters are shaped differently. An app that teaches Japanese was showing a character other than the one it teaches, and that kind of mistake is invisible unless the right shape stands next to it. Every Japanese line says what it is now, and the reading printed above it stopped running out of room at the largest text sizes. The session bar moves when you answer There were two bars in a session and which one you saw depended on your text size. At the usual sizes it was the one in the title, a hundred and forty points wide — a session of twenty-five questions moved it by five and a half points an answer, which is a bar that looks the same after an answer as before it. One bar stands now, under the navigation bar, across the full width, at every text size. The answers stand out in the light theme In the light theme the card holding the answers was white on white, and the only thing telling it apart from the screen behind it was a hairline outline. It stands on a gray ground now — the same one the rest of the app uses — and the outline stays. The dark theme never had the problem. A number on the icon says how much is left today The reminder now also puts a number on the app icon: the same one it says itself, and the same one the button on the home screen shows — how many questions are left to today's goal. It goes when you reach the goal, and nothing appears at all while your streak is paused. One thing has to be said out loud: if you allowed notifications before this release, iOS never asked you about badges and will not ask again, so the number stays away until you turn it on. Settings says so now, and takes you there. The introduction comes back if you swipe it away The sheet explaining how the app works is shown once in a lifetime — to someone who has not answered anything yet — and it counted itself as read the moment it appeared, before anyone could read a line of it. Swiping it away in the first second closed it for good. It counts as read when it closes now, whichever way you close it. The purchase screen reads like the rest of the app Restoring a purchase, the privacy policy and the terms stood there as bare text, while the same three rows in Settings have carried icons since the first release. They carry the same three now. The buy button also says what it is to a screen reader while a purchase is going through — until now it read there as just a button. The app asks what you make of it — once, and after a session The rating prompt was written for the first release and never once called: the only way to it was a button in Settings. It comes up on its own now, after a finished session and never in the middle of a question, and only once you have practiced enough for the question to be a fair one. A session you answered nothing in does not count towards it.

Role names say what the role is, not what to call it Three roles carried a grammar term as their name — topic, subject, direct object. The name stands on the progress bar and in the feedback, where nobody looks anything up, so it now describes what the role does; the term stays, once, in the paragraph that explains it. Eight particle footers stopped repeating the line they stand under, and say what the particle covers as a whole — including the roles you have not opened yet. The wording read again, in both languages The app spelled the same word two ways on one screen and the next, and the sentences in the catalog did it too. English keeps one spelling now — the American one — and Polish keeps one way of addressing you. It is not a matter of taste: two forms of the same word on one screen read as a mistake, and a mistake in an app that teaches, teaches. Long role names no longer break off at large text sizes At the accessibility text sizes a role name ended in an ellipsis instead of wrapping to a second line — forty-one places across the catalog screens, counted from screenshots rather than by eye. A name that breaks off does not say what the role is, and it broke off for the people who need those screens most. It wraps now, and the percentage beside it stays where it was. Catalog search finds Polish names typed without the accents Searching the particle catalog compared what you typed letter for letter, so "material" did not find "materiał" and the list simply came back empty. Thirty-five of the fifty Polish role names and descriptions carry an accent, so most of the catalog could be reached only from a Polish keyboard. Either spelling finds it now. Restoring a purchase says that it is working The row dimmed while it worked, and the reason why stood beside it in a sentence longer than the line it had. The sentence is gone; a spinner stands there instead — on the purchase row and on the restore row, because the two dim together. A report names all four things it carries Writing in named two of them and left out the other two, while the promise that you can delete them before sending covered all four. It names all four now — the app version, the build, the system version and the device model — so what you delete is what you were told about.

Practising your weak spots is no longer paid Aiming a session at the roles you get wrong was behind the purchase, and that was the wrong line to draw: knowing where you are weak is worth nothing if you cannot act on it. It is open to everyone now. What the purchase buys stays what it always bought — the levels. "Mastered" finally means something you can check The word appeared on three screens and was worked out differently on each. It has one definition now — eighty per cent on every skill of that role, not an average — and the colour scale, the weak-spot list and the progress row all read from it. Some roles will move a notch; the number changed because the rule did, not because the answers did. "I already know this" silences a role A role you know from elsewhere no longer has to be earned question by question. Marking it writes in full mastery and takes it out of the queue — as a declaration, which is what it is, so nothing pretends it was practised. A role you have not seen in a long time comes back on its own Choosing questions by weakness alone had a blind spot: something answered correctly months ago sits at the top of the scale and never comes up again. Time counts too now, so a role that has been quiet for long enough returns before you notice it slipping. The buttons count from the material, not from your setting The start buttons printed the number from Settings, whatever the catalogue could actually give. Each one asks the pool first now, so the number on the button is the number of questions you will get — and when it comes out lower than you asked for, a triangle beside it says which reason is behind it. New roles arrive three at a time A session could open with a run of unfamiliar roles one after another, which is the fastest way to make a good session feel like a bad one. New material now enters in threes, and the rest of the session builds on what you have already seen. Answer time no longer scores you without saying so How fast you answered was quietly part of the score, and nothing on the screen said it. It is announced now, and the threshold is stated: under eight seconds counts as fast — eight, not the usual five, because these questions put a whole sentence in front of you before the answer. Statistics say which period they are showing The screen referred to "the switch above", and what stands above is a period selector, not a switch — the switches are in Settings. It names the period now, and spells out what "mastered" means in the same place, so the definition and the number sit together.

See how it has been going A new Statistics screen under Progress shows the last 7, 30 or 90 days: accuracy day by day, how much you answered, how your mastery moved, and which roles and confusable pairs you get wrong most often. All of it is worked out from answers already on your phone — nothing new is collected and nothing leaves the device. Hear any sentence A speaker button next to the answer and next to every example on a role screen reads the sentence in Japanese. It uses the reading the catalogue knows, not a guess at the kanji, and works with no network. If your device has no Japanese voice installed, the button is simply not there. A reminder in the evening Settings can now send you one notification at an hour you choose, and only on a day you have not finished — it says how many questions are left, the same number the home screen shows. Nothing arrives on a day you finished, and nothing at all while your streak is paused. How it works, in the app A new page in Settings explains what the bars measure, why the app is organised by role rather than by particle, and what each switch does. New installs see the first few topics on the first launch. Sessions can skip what you already know Every role now has two switches: “I already know this” and “Set aside”. Both stop the role coming up in sessions the app picks for you, while choosing it yourself still works — nothing disappears. “Set aside” also takes the role out of the level total, and the level says how many roles it is counting. “I already know this” counts as done when levels unlock, but leaves your mastery bar alone: that one is earned, not granted. Getting on something is its own role バスに乗る was filed under destinations, where the explanation talks about arriving somewhere — which says nothing about why it is not バスで. Boarding a vehicle is now a role of its own, with its own explanation of why で does not fit. Start one role over, not everything A role screen and a level screen can now clear just their own progress. Answers about that material leave your history and every bar they counted towards is worked out again, so the numbers stay honest. Your notes, stars and switches are left alone. Your own notes on any role You can write in your own words what trips you up on a role, and star the ones you want to find again. Notes travel with your progress backup, so they survive a move to a new phone. Where two particles both work Some sentences take either particle and mean different things, so the app has never asked about them — quietly. A role screen now says it out loud, showing the sentence with the gap and what it means each way. Saying nothing about a boundary teaches just as much as saying something, only what it teaches is not true. Close to a hundred sentences read again The whole catalogue went through three lenses: whether a native would write it that way, whether the translation says what the sentence says, and whether exactly one particle fits. Sentences where two answers worked have been replaced. You can see when a sister app has reviews waiting The family row in “Your Japanese” shows an orange mark when there is something to review over there.

You can see which level you are on Sessions from the home screen have always drawn on the level you are working through, and the app never said which one that was. It says so now, and the Levels tab marks it. A break no longer costs you your streak Going away? Pause your streak in Settings and days without practice stop breaking it. While it is paused the home screen says so, so a streak on hold never looks like one you earned. Today and yesterday, by name Recent sessions from the last two days are labelled instead of dated — counted by your study day, so a session at half past midnight still belongs to the evening before. The day starts at the hour you set Twice a year, on the days the clocks change, the study day began an hour early or an hour late. Your daily count and the limit on new roles now start at the hour you chose, on those two days as well.

Which way you mix a pair up A weak pair used to say that you confuse に with で. It now says which way round: whether you put に where で belongs, or the other way about. When the two are even it says nothing, because then there is nothing to say. Japanese read in Japanese On the progress bars VoiceOver read particles with the interface voice, which came out as noise. It now reads them in Japanese, by their reading — は as "wa", not "ha". Messages you can read A backup that will not load says why, in the language of the app, instead of showing the raw error. The purchase screen has also stopped speaking the language of the system rather than the one you chose. Larger type The "What's new" sheet and the accent colour swatches only laid out correctly at the default type size. The day turns over on its own Left open overnight, the app kept yesterday's daily counter until you quit it entirely. It now turns over by itself, at the hour you set. Interrupted sessions still count A session you were in the middle of when the app was closed no longer disappears from today's count or from your streak. Weak pairs in one session The three pairs you confuse most now have their own entry under Weak spots — one session drawn from all three, instead of a row for each. The mode is part of the full version. Your Japanese, in one place A new section on the Today screen shows grammar, conjugation and particles side by side, each with its weakest spot and a tick where the daily goal is already done. Rows for the family's other apps appear on their own once they are on the same device — no account, no sign-in, no network. Why not the one you picked After a mistake the app used to say why the right answer is right. It now also says why the one you picked does not fit this sentence — for every wrong option, not only for the pairs people confuse most. Questions put right Writing those explanations turned up questions with two right answers, and ten with none at all. Their sentences and options have been replaced.