How to Look Up a Kanji You Don't Know: Hover, Handwriting, OCR, and Radical Lookup
Knowing how to look up a kanji you don't know is a routing problem, not a single tool choice. The right method depends on whether the character is selectable text, pixels on paper, or only inside your head.12 If you memorize one workflow, you lose time on every other source type. The useful habit is to identify the source first, then pick the matching path.
Overview
Kanji lookup methods split cleanly by the source you are reading. Each source gives the computer a different surface to read. Treating "kanji lookup methods" as one menu of equivalent tools is the mistake this article exists to correct.
The decision tree has three branches, each with its own fallback ladder. If you choose the wrong branch, a one-second hover can turn into a twenty-minute radical hunt.
Digital text is already Unicode and routes straight to a dictionary.12 Paper or screen-captured images are pixels and need optical character recognition (OCR) first.34 Partial-recognition lookups (you can picture the shape but have no source to capture) need a structural index: radical grid, multi-radical, or SKIP code.256
Why "look up a kanji" is really three problems
Lookup tools split by source because each source exposes a different surface for the computer to read. Digital text on a screen is already encoded as Unicode characters. The lookup problem is just routing the character to a dictionary.12
Handwritten or printed text on paper is pixels, not characters, so an OCR step is required before any dictionary can see it.34 Partial recognition means you can picture or partly trace the shape, but have neither text nor an image you can OCR. It requires a structural index: radical grid, multi-radical, or SKIP.256
The three branches are not interchangeable. A radical-grid query against text you could simply hover over wastes effort; a hover lookup against a paper book is impossible without a camera step in between.
The decision tree at a glance
The tree branches first by source, then by what works on that source. Each path has its own fallback. That is what makes the workflow a tree rather than a flat list.
Branch 1 (digital text) is fastest: use a hover-popup extension first, then copy-paste into Jisho as a fallback.172 Branch 2 (paper or image) needs a capture step before any dictionary can read the character.8934 Branch 3 (partial recognition) is the slowest path, but it always works, even with no internet.21056
What this article does not cover
Deep tool reviews of each browser extension, OCR app, and manga reader belong in their own canonical pages, so this article does not duplicate them. The same applies to radical theory, which has its own home elsewhere in the kanji subcategory. Here, components are named by what they look like, without arguing about which one is the "official" Kangxi radical.
Stroke-order rules for handwriting input also sit outside this article. The handwriting section below describes what the input does, not how to write the strokes; that explanation belongs in the stroke-order article.
Path 1: Digital text on a screen
When the kanji is on a screen and the text is selectable, the right choice is almost always a hover extension. Selectable means the cursor can highlight individual characters. It excludes text rasterised into an image or locked behind an extension-blocking viewer.
Hover lookup with Yomitan or 10ten Japanese Reader (preferred)
Yomitan describes itself as "a pop-up dictionary browser extension for language learning" and as the successor to Yomichan. It ships installers for Chrome, Firefox, and Microsoft Edge under the GPL-3.0 licence.111 The popup shows glosses for the matched word, kanji stroke-order diagrams, audio pronunciation, and Anki flashcard export through AnkiConnect. User-installed dictionary content, including EPWING, drives the lookup.1
10ten Japanese Reader (formerly Rikaichamp) is "a browser extension to translate Japanese by hovering over words." It runs on Firefox, Chrome, Edge, Safari (App Store), and Thunderbird, also under GPL-3.0.7 Compared with Yomitan, its main differences are pitch-accent display, automatic name-dictionary cross-reference, automatic year and unit conversions, and a localised UI in Japanese and Simplified Chinese.7
Both extensions resolve inflected forms automatically: hovering over a conjugated verb returns the dictionary form.17 That is why hover lookup outpaces copy-paste for running text. You do not have to manually strip okurigana to find the entry.
Old tutorials, screenshots, and Reddit threads still reference Yomichan, which was sunset by its original owner on 26 Feb 2023.11 The codebase was forked and renamed; learners following older instructions should install Yomitan from the official browser-store listings rather than chasing dead Yomichan links.111
Both tools are community projects, not commercial products. Downloads should come from the official browser stores or the project repositories, never from third-party mirrors.17
Copy and paste into Jisho when hover is unavailable
Jisho.org accepts kanji, kana, and rōmaji queries directly; pasting a single character into the search bar returns the kanji entry with readings, meanings, JLPT level, school grade, stroke count, and SKIP code.21213 The documentation lists search operators that narrow kanji results: #strokes:N, #grade:N, #jlpt-nN, #skip:P-S1-S2, and KANJIDIC2 indices are all valid.2
Copy-paste is the correct fallback whenever the source page blocks extensions. Common offenders include DRM-protected ebook readers, locked PDF viewers, web apps that disable text selection, and text rendered as an image inside an HTML element.
The copy-paste fallback is slow because it works one character at a time. For long-form digital reading, the hover extension is roughly an order of magnitude faster because it parses inflected forms automatically.12 Use Jisho copy-paste only when the page refuses to expose its text to the extension.
Worked example: a kanji inside a news article
Take the character 鬱 (utsu, "gloom; depression; luxuriant"), JLPT N1, 29 strokes, SKIP 4a25.1, junior-high jōyō.14 It is rare enough that even advanced learners may forget the reading. It is also dense enough that radical lookup is painful, which is exactly where hover lookup wins.
鬱蒼とした森の中を歩いた。15
"I walked through a thickly overgrown forest."
Hovering on 鬱蒼 in Yomitan or 10ten returns 鬱蒼【うっそう】"thick; dense; luxuriant," and the kanji-info pane gives readings, stroke count, and SKIP for 鬱 directly.1714 The copy-paste fallback in Jisho returns the same kanji entry with readings ウツ / うっ.する・ふさ.ぐ・しげ.る, JLPT N1, SKIP 4a25.1, and stroke count 29.14
The hover path resolves the compound 鬱蒼 in one motion. The copy-paste path requires you to first identify 鬱 as the unfamiliar character, isolate it from the surrounding text, and paste it alone. Both arrive at the same dictionary entry, but the hover path stays inside the reading flow.
Path 2: Handwritten or printed text on paper
When the source is paper, a sign, a screenshot, or any image the dictionary cannot read directly, the workflow needs a capture step. The right capture tool depends on whether the text is general, such as signage, books, or menus, or manga-specific, such as speech bubbles in stylised fonts.
Camera OCR for printed text: Google Lens and Apple Live Text
Apple's iOS feature-availability page lists Japanese among the languages supported by Live Text. The same list includes Cantonese, Chinese (Simplified and Traditional), Korean, Thai, Vietnamese, and the European languages.3 Apple's official Live Text support article points to that feature-availability page as the canonical, up-to-date language list, since Japanese support was added after the initial iOS 15 / iPadOS 15.1 / macOS 12 Monterey release.16
Live Text reads text directly inside the Camera viewfinder and inside any photo in Photos. A tap-and-hold on detected text shows copy, look up, translate, and share actions, which makes sending a captured kanji into a dictionary app one tap.16 Google's Translate documentation describes an equivalent workflow on Android: point the camera at text and the app overlays the translation in place, or import a photo from the gallery.4
Google Lens is bundled into Google Translate, Google Photos, and the Chrome desktop sidepanel. It handles horizontal and vertical Japanese text, signage, and handwriting. The recognised text can be copied and pasted into Jisho or another dictionary.4 In practice, Live Text is the lowest-friction option on iPhone because no separate app launch is needed. Google Lens or Translate is the cross-platform default and the Android first choice.34
Manga-specific OCR: KanjiTomo and Mokuro
KanjiTomo "identifies Japanese characters from images" and performs dictionary lookup in one step by "pointing the mouse to any image on screen." It supports both horizontal and vertical Japanese text.9 It is a free Java desktop application. The developer states it has "been tested on Windows 10 operating system, other operating systems might also work but are not supported," and requires Japanese fonts installed. That places it firmly in the desktop manga and game-screenshot niche.9
Mokuro is "aimed towards Japanese learners, who want to read manga in Japanese with a pop-up dictionary like Yomitan."8 It pre-processes a manga volume offline and produces selectable text overlaid on the page images. The result loads in a browser, where Yomitan or 10ten can hover-lookup the text.178 The pipeline uses two purpose-built models: comic-text-detector to find speech bubbles and text regions, and manga-ocr to read the Japanese characters inside them. The output is a .mokuro file plus the original page images, opened in a web reader.8
General OCR, such as Google Lens and Apple Live Text, is trained on flat horizontal text. Manga speech bubbles are vertical, irregularly spaced, often in stylised fonts, and surrounded by sound-effect text in different orientations. Mokuro's text-detector targets that layout specifically. That is why a manga-specific tool can succeed where Lens drops characters.8
KanjiTomo and Mokuro are both free GPL-licensed community tools. They produce dictionary-lookup-ready text, but neither attempts translation. You still read the Japanese yourself, which is the point.89
Handwriting input on phone and desktop
The iOS Japanese keyboard includes a handwriting (手書き) input mode under Settings → General → Keyboard → Add New Keyboard → Japanese (Kana) → Romaji / Kana / Handwriting. On Android and desktop, Google Japanese Input provides a kanji drawing pad as an input source. That means Gboard on Android, or the standalone Google Japanese Input app on Windows, macOS, and Linux.
Jisho's drawing input is the browser-native equivalent: "Input kanji by handwriting. Just start drawing!" It includes clear and backspace controls, with no app install needed.2 That makes it the lowest-setup choice on an unfamiliar machine or a friend's computer.
Handwriting input is the right choice when you can see or imagine the shape but cannot type the reading. It is especially useful when the source character has been miscounted for strokes, is stylised beyond OCR, or otherwise blocks the camera path. The trade-off is that recognition quality depends on writing the strokes in something close to the conventional order, which is its own topic.
Worked example: a kanji you saw on a sign
Imagine the kanji 鬱 (29 strokes) painted on a pharmacy sign next to うつ病 in larger type.14 The iPhone path is: open Camera, frame the sign, tap the Live Text indicator (the bracketed box in the lower-right of the viewfinder), tap-and-hold on 鬱, choose "Look Up" or "Copy," then paste into Jisho or any kanji-dictionary app.316
The Android path is: open Google Translate, tap the camera, point at the sign, and tap any kanji in the live overlay to capture it. "Select all" then copies the recognised text to the clipboard for a Jisho paste.4 After either capture, pasting 鬱 into jisho.org returns 鬱【ウツ】"gloom; depression; melancholy; luxuriant," 29 strokes, SKIP 4a25.1, JLPT N1, junior-high jōyō.14
Both phone paths end at the same dictionary record. The choice between them depends on which phone you are holding, not on lookup quality.
Path 3: Partial recognition, no source to OCR
The fallback branch handles cases where the kanji exists only in your head, or on a paper page that resists capture. This is also the only branch that works without electricity or a network: a paper Halpern dictionary and a stroke count are enough.
Jisho's radical-selection grid: pick the components you can see
Jisho's by-radical interface is reachable from the search bar's radical icon and is documented as: "Find kanji by their parts. Click on to reset radicals." The grid groups radicals by stroke count from 1 through 17 strokes.2 The workflow is incremental: each radical you select narrows the result set. Keep adding radicals until the target kanji appears among the remaining candidates.2
Radical-grid lookup does not require you to know the kanji's primary radical, meaning the Kangxi-style head radical used to file the entry in a paper dictionary. Any visible component that is in the grid is usable as a query. That makes the grid forgiving when you cannot name what you are looking at.2
Multi-radical lookup: combine two or more components
The sci.lang.japan multi-radical tool at kanji.sljfaq.org/mr.html is one of the long-standing free implementations. Its help page recommends: "First identify a part of the kanji. It's usually best to start at the top or the left of the kanji if possible," then either add another part or scan the candidate list.6
Candidates are "sorted by the total number of strokes," which lets you narrow quickly if you can also estimate stroke count.6 Jim Breen's WWWJDIC offers the same pattern under "Multi-radical selection," operating on the same KANJIDIC2 dataset that Jisho also uses.6
Hover-extension users still need the multi-radical path in one specific case: the kanji is in their head, or visible on a paper page, but not yet captured by any OCR. Multi-radical lookup bridges "I can see it" and "I can type it."
SKIP code in Halpern dictionaries (paper fallback)
SKIP stands for "System of Kanji Indexing by Patterns." It was developed by Jack Halpern and used as the primary indexing system in The Kodansha Kanji Learner's Dictionary (Kodansha International, 1999) and in its predecessor New Japanese-English Character Dictionary (Kenkyūsha 1990; NTC 1993).101718 SKIP assigns each kanji a three-number code P-S1-S2, where P is the pattern category (1 through 4) and S1 and S2 are the stroke counts of the two parts of the kanji.175
The four pattern categories are summarised below.
| Pattern | Shape | Worked example | SKIP code |
|---|---|---|---|
| 1, Left-Right | Divides vertically into a left part and a right part | 明 (left 日 4 strokes, right 月 4 strokes) | 1-4-4135 |
| 2, Up-Down | Divides horizontally into a top part and a bottom part | 男 (top 田 5 strokes, bottom 力 2 strokes) | 2-5-2125 |
| 3, Enclosure | One element borders or surrounds another | 道, 国, 風, 区, 囲 | 3-S1-S25 |
| 4, Solid | Cannot be divided into the first three patterns | See sub-feature codes below | 4-S1-S25 |
Category 4 needs an extra detail. Its third digit is not a stroke count but a sub-feature code: 1 = horizontal line at top (e.g. 雨), 2 = horizontal line at bottom (e.g. 白), 3 = vertical dividing line (e.g. 虫), 4 = none of these (e.g. 丼).5
Jisho's #skip: operator accepts SKIP queries directly. For example, #skip:2-5-2 returns kanji matching that pattern, which is the fastest way to use SKIP without owning the paper dictionary.212 The system's appeal as a paper fallback is that the only skill it requires is counting strokes inside each visually obvious half of the character. It works without electricity, network, or knowledge of which Kangxi radical is the "official" one.1017
The SKIP system is protected by copyright; Halpern requires written permission for redistribution of the codes in bulk, though individual reference use is unrestricted.17 Personal lookup is fine; building a derivative work that republishes the codes is not.
Reading-based and stroke-count fallbacks
When neither components nor SKIP help, two coarser filters can still narrow the search. Jisho's #strokes:N operator returns all kanji of a given stroke count. You can combine it with another filter, which is useful when you can count strokes confidently but have no useful component handle.2
A reading-based guess is the second fallback. If you can guess a likely on'yomi from a known phonetic component, Jisho's plain-text search returns matching kanji directly. For example, pasting かん returns kanji and words containing that reading.2 The underlying heuristic is that many kanji inherit their on'yomi from a phonetic-series component, which the phonetic-semantic-compound category of kanji formalises.
Worked example: a kanji you can picture but cannot type
Take 薔 from the compound 薔薇 (bara, "rose"), JLPT non-listed, 16 to 17 strokes, SKIP 2-4-13, jōyō non-listed.19 This is a real-world case: you see 薔薇 on a flower-shop awning and remember what it means, but cannot type either character on a keyboard.
The radical-grid path on Jisho is to open the radical search, then pick 艹 (kusakanmuri, "grass," 3 strokes), since it is the visible top component. The result list narrows to grass-radical kanji. Adding a second visible component, such as 土 "earth" or 回 "return," both inside 薔, cuts the list further until 薔 appears.26
The SKIP-code path reaches the same character through shape. 薔 divides up-down (category 2) into 艹 on top (4 strokes) and 嗇 on the bottom (13 strokes), giving SKIP 2-4-13. The Jisho query #skip:2-4-13 returns 薔 in the candidate list, and both paths converge on the same character.2195
手前のガラス張りの建物は薔薇園なんです。20
"The glass-fronted building in front of you is the rose garden."
The convergence is the point. Once you can name two visible components, every Path 3 sub-method, whether radical grid, multi-radical, or SKIP, is just a different query language for the same underlying KANJIDIC2 dataset.
Good to know
Why hover lookup beats copy-paste even when both work
Yomitan and 10ten both resolve inflected verb forms and i-adjective forms back to their dictionary entry automatically. Copy-paste into Jisho requires you to identify and strip okurigana yourself, or risk a "no result" reply.172 The hover popup also shows auxiliary content that copy-paste does not, including audio, kanji components, frequency rank, and Anki export. A copy-paste workflow returns one screen of dictionary data, while hover lookup keeps the reading flow uninterrupted.17
Handwriting input fails on visually near-identical kanji
Handwriting recognition is shape-tolerant, but it cannot reliably distinguish visually near-identical kanji, such as 末 vs 未, 土 vs 士, and 戌 vs 戍 vs 戊, without correct stroke order. Learners report that wrong stroke order tends to surface the wrong candidate near the top of the list. If you have the shape right but the order wrong, you may still pick the wrong character. The systematic explanation of why stroke order matters belongs in the dedicated stroke-order article.
Recognising an OCR mis-read before trusting the gloss
Live Text and Google Lens are trained on horizontal, well-lit, machine-printed text. Predictable failure modes include low-contrast signs, calligraphic or stylised signage fonts, vertical text without clean column segmentation, handwriting, and small characters near complex backgrounds.3164
The visual traps for OCR are characters that differ by a single stroke or dot: 力 vs 刀 vs 刃, 王 vs 玉, 大 vs 太 vs 犬. When the returned dictionary gloss does not match the context, drop the OCR path and try the next branch, radical grid or SKIP, rather than trusting the misread.
Yomichan is not gone, it became Yomitan
Yomichan was sunset by its original owner on 26 Feb 2023. The codebase was forked, renamed, and is maintained by The Moe Way community as Yomitan.111 Older tutorials, screenshots, and Reddit threads referencing Yomichan still apply conceptually, but the install links in them are dead. Install Yomitan from the project's official browser-store listings rather than chasing those links.111
See also
- Writing Kanji by Hand: Is It Still Worth It?
- What Is Kanji? A Complete Beginner's Introduction
- How Many Kanji Do You Need? A Realistic Count
- How to Predict the Reading of an Unknown Kanji Compound: The On+On Default, Jūbako, Yutō, and the Look-It-Up Bucket