Tomoshi’s dictionary data — entries, definitions, usage notes, pronunciation and the rest — is updated separately from the app itself: the app downloads a new data pack in the background, and it takes effect the next time you start up, with no need to upgrade the app. This page records what each data update changes that you can actually feel while looking words up.

A data pack’s version is its release date; you can see the version you have on the app’s About page.

2026-10-13

  • Simplified Chinese glosses reviewed sense by sense for the 41,082 most common entries; 2,660 rewritten. For every entry in the top 40,000 by conversational frequency plus every JLPT-tagged entry, each sense’s Chinese gloss was checked against its JMdict English gloss by one model; glosses it rejected went to a second, independent model, and only those both models rejected were rewritten by a third and reviewed again. Typical fixes: two senses collapsed into the same word (スイング’s boxing “swing” and jazz “swing” both read 摇摆), glosses that were too narrow (つける was missing the “to add to a bill” sense), and kanji read at face value (押し目 was glossed as “weak; falling” when it means a dip during a rise).
  • Traditional Chinese glosses retranslated into Taiwan usage from the English senses (41,077 entries). They had been a mechanical conversion of the Simplified text and kept mainland vocabulary (日元, 软件, 账户 …). This batch was rewritten by a model from the JMdict English senses and the reviewed Simplified glosses, then spot-checked; the remaining entries stay converted, with two systematic character fixes (賬→帳, 日元→日圓).
  • Chinese example-sentence translations checked for 17,501 sentences; 718 rewritten, 1,039 moved to the right sense, 81 wrongly attached sentences removed. Both mistranslations (配る’s 商品として rendered as “prize”) and sentences filed under the wrong sense (つける’s 警官はその男をつけた sat under “place under guard” rather than “to tail”) were handled only where two models agreed; sentences attached to the wrong entry altogether (少女 しょうじょ examples under 乙女) were removed.
  • Chinese meanings rewritten for 388 common kanji; Japanese definitions rewritten for 379 common words. Meaning lists for JLPT / jōyō kanji were reviewed character by character (恋 no longer lists “lover”, 遂 drops a misread “suicide”, 雰 gains “atmosphere”); model-written Japanese definitions of frequent words were checked line by line (背脂 is no longer “meat”, リクライニング is no longer a piece of furniture).
  • Transitive/intransitive pairs corrected: 被る／被せる removed (both are transitive), 細まる–細める added; when a verb has several counterparts the typical pairing now comes first. Sense notes of the “esp. 〈spelling〉” kind were retranslated by spelling / reading.
  • The nine Simplified synonym-contrast cards mentioned on 10-07 (標 glossed as a shrine rope, etc.) ship in this release, with the Traditional version in sync. All of the above remains AI-generated and reviewed by a separate model; report problems with the “Send feedback” button in the app.

2026-10-07

  • All English synonym-contrast notes rewritten (13,368 cards). In the English edition, the one-line note next to each synonym, near-synonym and antonym — how that word differs from the headword — had been translated from the old Simplified Chinese cards, and most of it no longer matched after the 2026-10-06 rewrite. The notes were written afresh from the new Simplified cards by one model, reviewed line by line by a separate model, sent to a third model for rework when they failed (up to three rounds), and checked by hand where disagreement remained. They follow the same brief as the Chinese cards: scope, typical collocations, register and whether the words are interchangeable. Only 5,381 cards had an up-to-date English note in the previous release; 13,503 do now.
  • This release updates the English data only; the Simplified and Traditional Chinese data are identical to 2026-10-06, and the handful of Chinese cards corrected since then (e.g. 標 wrongly glossed as a sacred rope) will ship with the next update. Contrast content remains AI-generated and reviewed by a separate model; use the “Send feedback” button in the app to report a problem.

2026-10-06

  • All synonym-contrast notes rewritten (13,365 cards, Simplified Chinese). The one-line note next to each synonym, near-synonym and antonym on an entry page — how that word differs from the headword — was written afresh by the latest model, reviewed line by line by a separate model, sent back for rework when it failed, and refereed by a third model where the two disagreed. The new notes are more concrete: scope, typical collocations, register and whether the words are interchangeable, instead of restating the gloss.
  • About 18,000 invalid synonym/antonym links removed. Some automatically selected relations did not hold (e.g. ショーツ listed as an antonym of Ｔバック), about 5,000 pointed at a homograph (e.g. クオータ “quota” linked to クォーター “quarter”) and have been re-pointed, and about 10,000 “synonym” links were downgraded to “similar”. The affected cards hide the removed words.
  • All Chinese reverse-lookup contrast cards rewritten (9,299 cards) with the same write → review → rework pipeline; 221 far-fetched candidates were dropped and 1,188 homograph mix-ups in the Chinese reverse index corrected.
  • All per-pattern grammar “easily confused” notes rewritten (3,721 notes, in Simplified Chinese, Traditional Chinese and English). Every confusable pattern now has a note (17 were missing); 21 pairs of duplicate grammar entries for the same expression were merged (the list goes from 707 to 686; old links still open); the entry for つつ had been written as つ〜つ by mistake and is corrected with new examples.
  • Sense notes now have Chinese translations (5,180 notes). JMdict’s English sense notes (e.g. “usu. in the negative”) are shown in Chinese in the Chinese editions once the app supports it.
  • The Traditional Chinese synonym-contrast and reverse-lookup cards were re-translated from the new Simplified text. The English synonym-contrast notes are not updated in this release; they will follow in a later data update. All contrast content remains AI-generated and reviewed by a separate model; use the “Send feedback” button in the app to report a problem.

2026-09-26

  • All 707 grammar entries rewritten (Simplified Chinese, Traditional Chinese and English). Meanings are now listed sense by sense, each pointing to its example sentence; formation is given as one rule per line, with contractions, past forms and other add-on rules on lines of their own; “usage” covers nuance and context, while common learner errors are listed separately under “watch out”. Every entry was written afresh by the latest model, then reviewed line by line by a separate model with a third model as referee, and sent back for rework until it passed. Current app versions show the new text right away; after the next app update the grammar card lays it out as senses, formulas and cautions.
  • 31 grammar example sentences corrected. Some examples did not actually use the pattern in question (for instance ないでください standing in for ないで), others were unnatural or ungrammatical (偉いぶって should be 偉ぶって); they have been rewritten, with translations in all three languages and cloud pronunciation updated to match.
  • Grammar explanations remain AI-generated content, reviewed by a separate model; anything that looks wrong can be reported with the “Send feedback” button in the app.

2026-09-02

  • “Which word to use” cards for English lookup grow from 2,644 to 9,941. When one English word (say discuss or great) maps to several Japanese words, only the ~2,600 most common English words used to have a card explaining how to choose; the cards now cover around 10,000 common words, every one reviewed by a separate model.
  • Fixed cards whose notes were attached to the wrong Japanese word. In the last pack some cards (great, for one) showed the note about 偉い under 偉大. Each note is now bound to the exact word it describes, and every card was regenerated that way.
  • Better ordering and matching in English lookup. Common words come first: loud now leads with the everyday 煩い (うるさい) instead of the rare 高らか. Archaic and rarely used senses no longer outrank everyday words (they are kept, and labelled, only when an English word has no other match); qualified glosses such as 碌に, which is only used in negative sentences, are no longer offered as plain equivalents of well; words usually written in kana are shown in kana; and duplicate candidates for the same Japanese word have been merged. About one in ten English words gets a different top result as a result.

2026-09-01

  • The “easily confused” notes on grammar cards now have Traditional Chinese and English versions. After yesterday’s English pack went out it turned out that the notes in the “easily confused patterns” section of a grammar card (both the summary and the point-by-point comparisons) were still showing in Simplified Chinese under the Traditional Chinese and English interfaces. All 707 grammar points now have Taiwanese Traditional and English notes of their own (707 each, every one reviewed by a separate model), in the same language as the rest of the card.

2026-08-31

  • New English language pack. English has been added as a dictionary language: with the English pack you can look Japanese up from English directly — type eat and you get 食べる (close to 100,000 reverse-lookup index entries, covering around 12,000 common English words) — plus 2,644 “which word to use” cards explaining how to choose between the several Japanese renderings of one English word. Near-synonym notes, spelling notes, transitive/intransitive explanations and grammar cards all come in English too. Pick it on first launch, or under “Dictionary language” in Settings.
  • Grammar explanations now come in Traditional Chinese and English. The explanations for all 707 grammar points used to show in Simplified Chinese under the Traditional Chinese and English interfaces; they are now written in Taiwanese Traditional and in English prose (with the example translations following suit), all reviewed by a separate model.
  • Spelling notes rewritten in full (both Chinese variants). The spelling notes for 35,615 entries have been rewritten by a new method: the old ones tended to take things for granted on questions like which spelling is older or whether it comes from a classical source, and errors of that kind are far rarer after the rewrite. The Traditional Chinese version has been rewritten alongside.
  • Near-synonym notes reviewed one by one. All 14,382 near-synonym cards have been gone over, and factual errors were corrected in 4,232 of them (wrong readings, meanings explained backwards, register attached to the wrong word and the like). The Traditional Chinese version is now translated from the reviewed Simplified version and localised to Taiwanese usage, replacing the older independently written version, which had a higher error rate.
  • Transitive/intransitive and grammar explanation fixes. All 655 transitive/intransitive cards and 707 grammar explanations were corrected under separate review (including a wrong intransitive partner for 泡立つ, and 10 problems with grammar examples and readings), and a batch of previously mismatched or missing transitive/intransitive pairs has been corrected and filled in.
  • 292 entries have gained a Chinese definition. They previously had only an English gloss to go on.

2026-08-20

  • Kanji have been regraded by JLPT level. Each character’s level used to be inferred from the earliest level at which it appeared in a word — 曜日 and 電話 are N5 words, so 曜 and 電 were tagged N5, which left 439 characters at N5, nothing like the roughly 80 in the usual exam material, while N1 was down to 445. The levels now come from the widely used table of the same lineage as jisho.org’s: N5 80 / N4 169 / N3 370 / N2 374 / N1 1237. The “JLPT kanji” page in the reference tables and the level badges on kanji cards have been updated to match. The few characters missing from that table (分 among them) have been given levels from the pre-reform official syllabus.

  • The JLPT level of 41 words has been corrected. The N5 and N4 word lists were checked against the pre-reform official test standard and other sources, and three kinds of problem were fixed:

    • Beginner words like だけ, 〜達, 〜匹, 〜枚, すぐ and 〜屋 had been tagged N1; they are back at N5/N4.
    • 円, 時, 父, 分（ふん）and どんな have been moved back to N5 from harder levels.
    • A whole missing category of greetings (おはようございます, さよなら, ただいま, どういたしまして…) and some overlooked basics (顔, ない, 頑張る, 専門, 正月…) have been added.

    JLPT levels remain a community estimate, not official data.

2026-08-12

  • A word is no longer both a synonym and an antonym of itself. Looking up やはり put やっぱり in the synonym column and the antonym column at once — when it is simply the colloquial form of the same word. Synonyms come from the word database and antonyms are generated separately, so the two sides each said their piece with nothing to catch a collision. All 60 conflicts have now been reviewed pair by pair and the untenable side removed:
    • Pairs that were synonymous to begin with — 矢っ張り／矢張り, キッチン／台所, 看護婦／看護師, 海抜／標高 — have lost the antonym marking.
    • Pairs that really are opposites — 収賄／贈賄, 入金／払い, 持っていく／持ってくる — have lost the synonym marking.
    • Pairs that are neither, such as 社長／専務, 少年／青年 and 戦術／戦略, have lost both.
  • 50 near-synonym cards have been rewritten to match. Where an entry lost a synonym above, its notes still referred to a word no longer in the list; they have been regenerated from the new synonym groups.

2026-08-05

  • 484 Traditional Chinese usage notes have been rewritten. These notes used to be converted straight from Simplified Chinese, which squeezed the sentences into a telegraphic shorthand — 「動作一同預設用一緒に，正式用共に」 — that does not read like Chinese. They are now written out properly in Taiwanese usage: 「『一同』在日語中依對象與語感有別：單純一起行動多用『一緒に』，強調共同立場或書面語用『共に』……」. This covers the rendering notes used by Chinese-to-Japanese lookup, the kanji-variant notes and the transitive/intransitive notes. It only affects what you see with the interface language set to Traditional Chinese; nothing changes for Simplified Chinese.

2026-08-04

  • Furigana on example sentences no longer belongs to the wrong entry. When a kanji word has several readings, the furigana on an example now follows the entry you are actually looking at. Previously, looking up 開く（あく）showed the example 「ドアが開く」 annotated with the reading of the other entry, 開く（ひらく）. This update corrects 165 examples and 14 transitive/intransitive examples, among them:
    • ドアが開く → あ (was ひら); お腹が空く → す (was あ); 身が汚れる → けが (was よご); 豆が弾ける → はじ (was ひ); 髪を逆立てる → さかだ (was さかた)
  • 4 mismatched transitive/intransitive pairs fixed. These entries had been paired with the entry next door because the kanji spelling is the same:
    • the transitive partner of 下る（くだる）is now 下す（くだす）rather than 下ろす（おろす）, with the examples still 判決が下る／判決を下す;
    • the partner of 極まる（きわまる）is now 極める（きわめる）rather than 決める（きめる）;
    • the 入る（いる）↔ 入れる pairing has been withdrawn — its example 「気が入る」 actually uses the other entry, 入る（はいる）, and that pairing already existed;
    • the examples for 埋める（うずめる）↔ 埋まる have been swapped for sentences that really use that reading, 顔を枕に埋める／顔が枕に埋まる (the old example 「穴を埋める」 is うめる).

2026-08-03

  • Chinese-to-Japanese lookup now finds set phrases. Looking Japanese up from Chinese used to give you word-for-word equivalents only, which left gaps wherever the natural way to say something is a fixed phrase. Looking up 头疼 gave you 頭痛 alone, while the more idiomatic 頭が痛い and 頭痛の種 were nowhere to be found. Around 1,200 Chinese words have gained expressions of this kind, and their “choosing between renderings” notes have been updated to say which to use when.
    • For example: 头疼 → 頭痛 / 頭が痛い / 頭痛の種; 生气 → 怒る / むかつく / 怒り / 腹が立つ; 丢脸 → 恥ずかしい / 面目ない / 見っともない / 恥をかく
    • A further 17 Chinese words that previously returned nothing at all now have Japanese equivalents.

2026-08-02

  • The data behind “Common attachments” has been filled in. The attachment-pattern table in the reference tables was missing its accompanying data, which arrives with this pack; 235 of the attachment notes were rechecked and corrected in the process.