How to Read OJAD: The Online Japanese Accent Dictionary
Reading OJAD, the Online Japanese Accent Dictionary, comes down to four features, one overline convention, and a few important limitations.1 Learners who already know basic pitch-accent notation can use OJAD as a single page that answers every "where does the drop fall?" question about a word, a conjugated form, or a whole sentence.
Overview
What OJAD is (and what it is not)
OJAD stands for the Online Japanese Accent Dictionary.1 It is a free academic web resource maintained by the Minematsu Laboratory and the Saito Laboratory in the Graduate School of Engineering at the University of Tokyo.1
OJAD is a pitch and intonation reference, not a meaning dictionary. Word Search returns accent type and audio, never an English gloss.12
The Interspeech 2013 paper introduces OJAD as "the first online and free framework for teaching and learning Japanese prosody including word accent and phrase intonation."3 Prosody means the rhythm, pitch, and intonation of speech. The four official features are Word Search, Verb Suffix Search, Text Search, and Suzuki-kun: Prosody Tutor.14
OJAD does not gloss words. The reliable workflow is to confirm meaning and reading in a meaning dictionary first, then carry the kana or kanji into OJAD for the accent.5
When to reach for OJAD
Use OJAD when the question is pitch: citation form, conjugated form, or sentence-level contour. First settle the meaning elsewhere.15
Each of the four features answers a different version of that question. Word Search returns "the accent type of the conjugated forms (apps. 10 kinds) of declinable words." Here, declinable means words that change form, such as verbs and adjectives. Verb Suffix Search returns "the accent type of verb-suffixes." Text Search lets you paste a Japanese sentence to "determine the accent types of all the declinable words." Suzuki-kun predicts "the accent kernel position" and "the pitch contour when read aloud."4
What the dataset covers
OJAD's home page gives the scale directly: "Over 9000 nouns and 3500 declinable words including verbs, i-adjectives, and na-adjectives, making it possible to search approximately 42,300 conjugations of words."1
The accent-change patterns are organised around "twelve kinds of fundamental conjugation," meaning twelve core forms for verbs and adjectives.3 Tofugu summarises the audio side of the dataset as "approximately 42,300 conjugation patterns, with audio recordings by both male and female voice actors."5
"Each conjugated form given in the dictionary is accompanied by both male and female audio." OJAD does not provide noun audio samples.1
The four features of OJAD
The site's top navigation shows four tools that share one accent dataset. Each tool answers a different question: a single word, a suffix family, a full sentence, or a full sentence with playback audio.
1. Accent Dictionary (word search)
Official description: "You can look up the accent type of the conjugated forms (apps. 10 kinds) of declinable words."4
Inputs can include kanji. The help page demonstrates 楽 as a search term, and the search box accepts comma- or space-separated multi-word lookups such as 遊ぶ 食べる 起きる 寝る 歩く 旅行する.2
The live filter row lets you narrow results by textbook chapter (Minna no Nihongo, GENKI, Marugoto, できる日本語, and others), part of speech, accent type (All, Accentless / Type 0, Initial-high / Type 1, Mid-high, Final-high, Accented Type 2 through 6+), word difficulty, and Old JLPT (Japanese-Language Proficiency Test) level (Level 4, 3, 2, 1).26
Accent representation: "The accent type is depicted with a straight line or the ￢ mark."2
For audio, the help page says, "Click either 'FEMALE' or 'MALE' and either female or male speech will be played." You can download samples for offline use by shift-clicking the sample.2
2. Verb Handbook (suffix and conjugation search)
Official description: "you can look up the accent type of verb-suffixes."4 In other words, this view shows how suffixes affect pitch.
The conjugation columns on the OJAD search results page cover twelve forms: dictionary form, ～ます, ～て, ～た, ～ない, ～なかった, ～ば, causative, passive, imperative, potential, volitional.6
The pedagogical aim, per the SLaTE (Speech and Language Technology in Education) 2013 paper, is "visual, auditory, systematic, and comprehensive illustration of patterns of accent change (accent sandhi) of verbs and adjectives, focusing on changes resulting from twelve kinds of fundamental conjugation."37 Tofugu summarises the same feature as "suffix search for grouped words like 〜ます, て form, ない."5 These are common endings learners meet early.
3. Text Search
Official description: Text Search lets you "enter a Japanese sentence of your own or one you grabbed from another source" to "determine the accent types of all the declinable words" in that sentence.4
Tofugu notes a per-query length of "up to 1500 words" for Text Search.5
4. Suzuki-kun (sentence prosody tutor)
Official function: "With Suzuki-kun: Prosody Tutor, (1) the accent kernel position will be predicted and shown and (2) the pitch contour for when the sentence is read aloud will be displayed for the sentence entered into the text box."8 The accent kernel is the place where the pitch drop is predicted.
The Interspeech 2013 paper specifies that Suzuki-kun relies on "an accent change prediction module that we developed for Japanese text-to-speech (TTS) synthesizers."3 The Suzuki-kun help page references the Fujisaki model for pitch-contour parameters, a technical model for describing intonation.8 Tofugu describes Suzuki-kun more plainly as "an automatic speech processing system. It generates the audio and the pitch pattern of the sentence you put into the text box."5
How to read an OJAD entry
The accent number (and what the small digit next to a verb means)
OJAD's accent-type filter exposes the standard numeric categories: Accentless (Type 0), Initial-high (Type 1), Mid-high, Final-high, Accented Type 2 through 6+.6
The numbering follows the established Japanese accent-dictionary tradition. The integer marks the mora after which the H→L drop falls, and 0 means no drop (heiban). OJAD adopts this convention in its UI labels.6
Full coverage of the convention is out of scope here. For the deeper tour, use the dedicated pitch-accent-notation article in this pillar.
The overline diagram
Official wording: "The accent type is depicted with a straight line or the ￢ mark."2
The straight horizontal line marks the stretch of high pitch. The small downward ￢ tick at the right edge of the line marks where the pitch drops.2
Particle behaviour is drawn into the same diagram. This lets you see whether the high contour continues onto a following particle (the heiban case) or ends at a drop inside the word.2
Suzuki-kun lets the learner switch the contour visualisation between "Mountain-hill Contour," "Pitch Contour with Accent (Beginner)," and "Pitch Contour with Accent (Advanced)": three levels of visual detail over the same prosody.9
Multiple accent patterns on one entry
Some entries list more than one accepted accent. OJAD's home page states the ordering rule directly: "when more than one accent type is displayed, the accent type at the top is the standard accent type."1
That is the only formal rule the site publishes about alternates. For drilling, start from the top-listed accent and treat the others as legitimate variants the dictionary acknowledges.
Audio samples: male and female
"Each conjugated form given in the dictionary is accompanied by both male and female audio."1 OJAD does not provide noun audio samples.1
The male voice samples for certain conjugated forms "are read without voiceless devoicing."1 A vowel that would often be weakened or devoiced is still pronounced. The Word Search help page describes the UI control as: "Click either 'FEMALE' or 'MALE' and either female or male speech will be played," and samples are downloadable by shift-clicking the sample.2
A worked lookup: from a Jisho hit to an OJAD entry
The full loop runs from meaning lookup to sentence-level contour in five steps. Each step uses the tool that fits that part of the job.
Step 1: confirm meaning and reading in Jisho
Jisho returns the reading and gloss; OJAD does not. The hand-off between a meaning dictionary and an accent dictionary is the core of the workflow.5
Step 2: paste the kana (or kanji) into OJAD word search
The OJAD help page demonstrates kanji input (楽) and multi-word kanji input (遊ぶ 食べる 起きる 寝る 歩く 旅行する).2
Before submitting, narrow the result set with the filter row. It includes part of speech, accent type, textbook, and Old JLPT level.26
Step 3: read the accent number and the overline together
The result table shows the accent type "with a straight line or the ￢ mark" alongside the numeric type label. The same information appears twice on the same row.26 Cross-check the two notations against the cached audio sample before moving on.
A simple verb query:
食べる2
"to eat."
An i-adjective query, for comparison across parts of speech:
高い2
"high; tall; expensive."
Step 4: for a verb, switch to the Verb Handbook view
The learner sees these columns: dictionary form, ～ます, ～て, ～た, ～ない, ～なかった, ～ば, causative, passive, imperative, potential, volitional.6
You can read accent class membership (heiban versus accented) across the row. A heiban verb stays flat across all twelve cells; an accented verb has a drop locus that migrates predictably.3
Step 5: paste the sentence into Suzuki-kun
The input box accepts plain Japanese sentences. The forward slash / is the explicit phrase-break delimiter. The OJAD example is 今日は、いい天気だから/出かけよう。.9
今日は、いい天気だから／出かけよう。9
"It's nice weather today, so let's go out."
OJAD plus Yomitan: a reading-and-pitch loop
Yomitan, a browser-based pop-up dictionary, is the natural front end for reading practice. Its overlap with OJAD is partial. Yomitan answers the citation-form question on hover; OJAD answers everything beyond that. Yomitan setup is covered in its own guide, so this section stays at the workflow level.
What each tool covers in the reading loop
The clean division of labour is simple: use Yomitan for instant meaning and citation accent during reading. Use OJAD for the conjugation matrix and sentence-level contour the popup cannot show.
| Question while reading | Tool that answers it best |
|---|---|
| What does this word mean? | Yomitan (pop-up) |
| What is its citation accent? | Yomitan with a pitch-accent dictionary installed |
| How does the accent shift across the twelve conjugated forms? | OJAD Verb Handbook6 |
| What is the pitch contour of the whole sentence? | OJAD Suzuki-kun48 |
When Suzuki-kun is the only answer
Sentence-level prosody, or pitch and rhythm across a whole sentence, is the part of the loop a popup dictionary cannot do. Suzuki-kun "predicts the accent kernel position and the pitch contour when read aloud" for whole sentences,4 which is out of scope for any hover-and-gloss tool, including Yomitan.
The practical move during reading is to keep an OJAD tab open. When a word's conjugated accent is interesting, switch to the Verb Handbook view; when the whole sentence is interesting, paste it into Suzuki-kun.
OJAD for verb and adjective conjugation accent
The 12 conjugation patterns OJAD tracks
The conjugation matrix on OJAD's Word Search results page has twelve columns: dictionary form, ～ます, ～て, ～た, ～ない, ～なかった, ～ば, causative, passive, imperative, potential, volitional.6 The Interspeech 2013 paper calls these "twelve kinds of fundamental conjugation" and treats the visual-auditory display of accent change across them as a core OJAD function.3
| # | Column | # | Column |
|---|---|---|---|
| 1 | dictionary form | 7 | ～ば |
| 2 | ～ます | 8 | causative |
| 3 | ～て | 9 | passive |
| 4 | ～た | 10 | imperative |
| 5 | ～ない | 11 | potential |
| 6 | ～なかった | 12 | volitional |
The Word Search help page's worked example uses the input row 遊ぶ 食べる 起きる 寝る 歩く 旅行する. It displays these twelve cells side by side across verbs of different conjugation classes.2
Reading the matrix: where the drop moves
The pedagogical rationale, from the SLaTE 2013 paper, is "visual, auditory, systematic, and comprehensive illustration of patterns of accent change (accent sandhi) of verbs and adjectives, focusing on changes resulting from twelve kinds of fundamental conjugation."37
The matrix makes the accent-sandhi pattern, the way accent changes when forms combine, legible across forms. In one row, you can see whether a verb's drop locus stays put, moves by one mora, or disappears under a given suffix.26 A dedicated article in this pillar covers the underlying pitch-shift rules across conjugations.
Spotting accent-class membership in one glance
Heiban verbs appear with the high contour unbroken across every column. Accented verbs appear with a visible ￢ tick whose position shifts predictably across the twelve forms.26 The visual difference is the fastest way to read accent class from the page.
Suzuki-kun: sentence-level prosody
Inputs Suzuki-kun accepts
Suzuki-kun accepts plain Japanese sentences with mixed kana and kanji.89
The forward slash / is the explicit phrase-break delimiter. The official example is 今日は、いい天気だから/出かけよう。.9 The Suzuki-kun help page does not document an explicit input length cap.89
The analyse step (kernel prediction and pitch pattern)
The help page labels the button "Prosody Prediction." Pressing it triggers the analysis that produces "H (high) and L (low) accent patterns and pitch contour."8
The home-page note on devoicing is worth carrying over here. Male voice samples for certain forms "are read without voiceless devoicing,"1 meaning Suzuki-kun audio may not always render predicted devoiced morae as devoiced. Use the visual prediction as the authoritative cue.
The display step (contour styles, playback, print)
Suzuki-kun offers three visualisation choices for the same prediction: "Mountain-hill Contour," "Pitch Contour with Accent (Beginner)," and "Pitch Contour with Accent (Advanced)."9 A "Print-view" option is available via the "Display Type" listbox for printable materials.9
Tofugu summarises the output as "the audio and the pitch pattern of the sentence you put into the text box."5
Using Suzuki-kun for shadowing
A simple drill loop follows from what Suzuki-kun documents: paste the target sentence, click Prosody Prediction, read the contour, play the audio, then shadow it. Repeat at the sentence level until your own contour matches the predicted one.
OJAD does not publish a shadowing protocol of its own. The loop above is an editorial workflow built on the documented capabilities: sentence input, prosody prediction, and audio playback. Treat the contour as a strong first draft, not a recording. See the limitations section below.
OJAD's limitations
Tokyo / standard accent tradition
OJAD's accent data follows the Tokyo / standard accent tradition. The home page states that, when alternates are listed, "the accent type at the top is the standard accent type,"1 which presupposes a single accent standard rather than regional coverage.
The Interspeech 2013 abstract describes OJAD as a framework "for teaching and learning Japanese prosody including word accent and phrase intonation." It does not claim regional or dialectal coverage.3 Across 1, 4, 3, and 7, the published descriptions make no reference to Kansai, Tohoku, accentless dialects, or other regional accent systems.
A dedicated treatment of regional pitch-accent variation lives elsewhere in this pillar.
Low-frequency vocabulary gaps
The dataset is bounded at "over 9000 nouns and 3500 declinable words."1 Words outside that set are absent from the index. The search returns no entry rather than a guess. Tofugu's review does not contradict this scope.5
Suzuki-kun is a prediction, not a recording
OJAD says so directly: "The performance is not 100% because morphological analysis, accent estimation, and accent boundary estimation are used."8 The help page goes further: "Suzuki-kun cannot deal with word-focus, question sentences, emotional expressions, and so on."8
The Interspeech 2013 paper describes the prediction module as one OJAD adapted from TTS synthesisers.3 That is the source of the prediction-versus-recording distinction. Treat Suzuki-kun's contour as a strong first draft, not an oracle.
Ambiguous syntactic environments, such as compound boundaries, proper nouns, and novel coinages, are where the parser is most likely to mis-segment or mis-accent. When Suzuki-kun's contour surprises you on a compound, cross-check the head noun in Word Search before changing how you say it.
Multiple acceptable accents
"When more than one accent type is displayed, the accent type at the top is the standard accent type."1 The ordering encodes editorial preference: the top entry is the recommended one for drill purposes. It does not declare other listed accents incorrect. A learner who treats the first listed accent as the only correct one misreads the convention.
Compound accent variation is a frequent source of multi-accent entries; a dedicated treatment of compound pitch accent lives elsewhere in this pillar.
Good to know
Why the URL lives under gavo.t.u-tokyo.ac.jp
OJAD is maintained by the Minematsu Laboratory and the Saito Laboratory in the Graduate School of Engineering, The University of Tokyo.1 That academic provenance supports the data scale: over 9,000 nouns and 3,500 declinables with approximately 42,300 conjugated forms, prepared and voiced for pedagogical use.1 A learner blog or community wiki cannot match that scope.
Suzuki-kun's contour is not a recorded reading
Do not treat a Suzuki-kun playback as authoritative on a contested word. If Suzuki-kun plays こうえん with a drop on こ, do not conclude that "こうえん is therefore atamadaka." Read it as "Suzuki-kun predicted atamadaka here, and the prediction may be wrong."
The documentation is explicit: "The performance is not 100% because morphological analysis, accent estimation, and accent boundary estimation are used."8 The engine is the same accent-change prediction module OJAD's authors developed for Japanese TTS synthesisers.3 It inherits the failure modes of a parser, not the certainty of a recorded dictionary entry.
The correct interpretation is a Word Search lookup you can verify directly:
公園2
"park."
Romaji input is not the reliable mode
The Word Search help page demonstrates kanji input and multi-word kanji input.25 Romaji input is not documented as a primary option on the official help pages. If you rely on it, expect inconsistent behaviour and fall back to kana or kanji as the reliable input mode.
"OJAD is what; Jisho is why; NHK is whose."
This is a short mnemonic for the three-tool dictionary stack. OJAD answers "what is the pitch, and how does it shift under conjugation?" Jisho answers "what does the word mean?" The NHK accent dictionary answers "whose authority backs this accent?"15 OJAD's accent-type filter explicitly uses categories from the standard accent-dictionary tradition: Accentless, Initial-high, Mid-high, Final-high, Accented Type N.6
Use the slash to scope Suzuki-kun
The forward-slash phrase-break delimiter is Suzuki-kun's key input convention. The official example 今日は、いい天気だから/出かけよう。9 shows the parser where to place a phrase boundary instead of guessing. In a long sentence, a single well-placed / can clean up the predicted contour more than rewording the Japanese.
See also
- Japanese Pitch Accent: A Complete Beginner's Guide
- Should You Learn Pitch Accent? An Honest Cost-Benefit Analysis
- Japanese Pitch-Accent Minimal Pairs: The Drill List You Must Hear
- Nakadaka (中高): The Middle-High Japanese Pitch-Accent Pattern
- Odaka (尾高): The Tail-High Japanese Pitch-Accent Pattern
- Japanese Pronunciation Drills: A Daily 5-Minute Protocol with Minimal Pairs, Shadowing, and Record-and-Compare