---
name: hebrish-lexicon-lookup
description: Look up how Hebrish spells a Hebrew word or Israeli name in Latin letters (for example makolet, arnona, teudat zehut), its domain, its meaning and how many dataset recordings contain it, using the public Hebrish lexicon at hebrish.app. Use when writing, checking or normalising English text with Hebrew words in it, building a vocabulary or hotword list for English-Hebrew speech recognition, or answering whether Hebrish covers a word.
metadata:
  publisher: hebrish.app
  updated: "2026-10-20"
  source: https://hebrish.app/data/lexicon.json
---

# Hebrish lexicon lookup

Hebrish writes every Hebrew word in **one canonical romanized spelling** and never in Hebrew
script. The lexicon is the list of those spellings: every Hebrew word and Israeli name the
Hebrish dataset covers.

## Data

| URL | What |
| --- | --- |
| `https://hebrish.app/data/lexicon.json` | Whole lexicon. Fetch once and search locally. |
| `https://hebrish.app/data/lexicon.csv` | Same as a spreadsheet: `term, domain_id, domain, meaning, recordings` (UTF-8 with BOM). |
| `https://hebrish.app/lexicon?d=<domain>&q=<text>` | Human view; give users this link. `&view=table` for the table. |

`lexicon.json` shape:

```json
{
  "built": "2026-10-19",
  "takes_revision": "5db41eef",
  "domains": { "food": "Food & shopping", "bureaucracy": "Bureaucracy & government", "...": "..." },
  "terms": [ { "t": "makolet", "d": "food", "n": 16 }, { "t": "mitzva", "d": "everyday", "m": "a commandment, or a good deed ('it's a big mitzva')", "n": 20 } ]
}
```

- `t`: the spelling to write. Keep its exact case: capitalised entries are names or brands.
- `d`: domain id. The 13 ids are: places, people, politics, religious, food, health,
  bureaucracy, money, home, army, transport, brands, everyday.
- `m`: English meaning. Optional; most entries have none, so search by spelling first.
- `n`: number of dataset recordings whose transcript contains the term (whole-word match;
  capitalised terms match case-sensitively). Refreshed daily. `n: 0` means the term is listed
  but not yet recorded, so the model has probably not learned it.

## How to look up a word

1. Fetch `lexicon.json` (no auth, CORS open).
2. Compare case-insensitively and ignoring diacritics; normalise curly apostrophes to `'`
   (spellings such as `tza'atzuim` contain one).
3. Prefer an exact match on `t`, then prefix matches, then matches in `m`.
4. Report the spelling exactly as in `t`, with its domain and recording count.

```bash
curl -s https://hebrish.app/data/lexicon.json \
  | jq -r '.terms[] | select(.t|ascii_downcase|startswith("teudat")) | [.t,.d,(.n|tostring)] | @tsv'
```

On hebrish.app itself, a browser agent can call the WebMCP tool `lookup_hebrew_term` instead.

## Rules

- Write the spelling from `t`. Do not invent a new romanization for a word that is listed,
  and do not output Hebrew script as if Hebrish produced it.
- Not listed does not mean wrong: the lexicon covers what the dataset covers. Say it is not in
  the Hebrish lexicon rather than guessing the spelling Hebrish would use.
- Known gap (2026-10-19): Torah, shul, daven(ing) and Shabbat appear in recordings but have no
  standalone entry (only compounds such as dvar Torah and Kabbalat Shabbat).
- The figure of 1,164 terms quoted on some pages is the upstream spelling lexicon
  (`Hebrew-Latin-Token-Classifier`, `data/terms.csv`) alone; `lexicon.json` is larger
  (3,562 terms in the 2026-10-19 build). Quote the count from the file you fetched.
- Spellings follow one convention. A user's preferred spelling may differ; that is a
  preference, not an error in either.

## Suggesting a missing word

Send people to `https://hebrish.app/words` (a list of words) or
`https://hebrish.app/contact?topic=word` (one word). Do not submit those forms on a user's
behalf without being asked.