What is the difference between a glossary, pitch, and frequency dictionary in the Yomitan format?
YuzuLingo · Nov 1, 2026
A glossary dictionary holds readings and definitions in term_bank files. Pitch and frequency dictionaries hold accent graphs and corpus ranks in term_meta_bank files. All three use the same Yomitan format 3 zip layout with an index.json at the root.
Glossary dictionaries
Glossaries are what most people mean by “a dictionary.” term_bank_*.json rows carry expression, reading, tags, and sense text. Structured-content glossaries such as Jitendex still live in those banks. Without at least one glossary, a popup has nothing to show as a definition.
Pitch dictionaries
Pitch packages skip glossary banks. They put accent data in term_meta_bank_*.json with mode pitch: headword, reading, and mora downstep positions. Readers attach the first matching pitch entry after a glossary hit so the card can draw a downstep graph. Kanjium-derived zips are the usual free source. Commercial NHK data is a different product and license.
Frequency dictionaries
Frequency packages also use term_meta banks, with mode freq. Each row ties an expression (and often a reading) to a rank or score from a corpus such as Wikipedia or novels. The popup shows that rank so learners can judge how common a word is. Frequency is language-agnostic in the schema: the same bank shape works for Japanese, Chinese, or other packs when the data exists.
How readers stack them
Install order usually means glossary first for the gloss and reading, then separate pitch and frequency zips. One zip can legally combine kinds when the files are present together, but immersion setups keep them split so each source can update on its own schedule. index.json still declares format 3 for every package.
How to tell kinds apart in a zip
Open the archive and look at the filenames and meta modes. term_bank files mean glossary content. term_meta_bank files with pitch or freq modes mean pitch or frequency. Some catalogs also stamp a kind label when they host the file; the zip contents remain the source of truth for a custom import.
