← all posts

Two arks, one word

Open Første Mosebok 6:14 in NB2026 and Noah builds an ark. Open Andre Mosebok 2:3 and Moses' mother makes a kiste — a chest. In Hebrew they are the same word: תֵּבָה, tēbâ (H8392), which occurs twenty-five times in the Hebrew Bible and names those two objects and nothing else.

Then open Andre Mosebok 25:10, where Israel is told to build an ark for the testimony. Different word: אָרוֹן, ʾārôn (H727) — a box, used a hundred and seventy-four times for money-chests, for the coffin Joseph was laid in (Genesis 50:26), and for the ark in the tabernacle.

So Norwegian hides a link that is really there, and suggests one that is not. English does the opposite. It renders both words ark and loses the distinction altogether:

"Make thee an ark of gopher wood; rooms shalt thou make in the ark, and shalt pitch it within and without with pitch." (Genesis 6:14)

Neither language is making a mistake. Both are making a decision. That is the part I want to describe.

The same decision, over and over

We publish in thirty-six languages. A word like this comes up in a book about baptism, and again in a book about the tabernacle, and again in the Norwegian edition of the first book, and again in Farsi. Each time, somebody has to decide what tēbâ means and how to say it.

If that decision lives in the glossary at the back of one book, it lives in one book. The next book makes it again, slightly differently. Nobody notices for two years, and by then there is no version that is obviously the right one — only several that disagree.

Lexifye is where we moved those decisions.

The model is deliberately small. A dictionary holds entries; an entry holds definitions. A definition is authored there and nowhere else. Editing goes through about eighty small, single-purpose tools — add a definition, move it, look at its history, revert it — rather than by rewriting a document, because a dictionary edited as a document drifts like a document.

The part that made it worth building is provenance. When a Junifye book pulls a term into its glossary, the imported entry carries the id of the Lexifye dictionary it came from and the checksum of the exact definition it was built from. So a book does not contain an opinion about tēbâ. It contains a snapshot of one, and you can tell afterwards which one, and whether it has moved since.

There are nine dictionaries in the service today. The largest, The Consensus Bible Dictionary, has 13,668 entries. The one I use daily is Davar, our working Bible dictionary — 338 entries in English, 277 in Norwegian, 316 in Farsi, 254 in Russian, 196 in Ukrainian, each edition its own dictionary and its own decisions. Its lexical data comes through Darash, so a Hebrew or Greek word inside a definition carries its Strong's number attached to the word itself, rather than typed next to it and hoped for.

The honest part

You cannot read any of that.

Lexifye's public surface today is three pages: the front page, the tool contract at /tools, and a machine-readable summary at /llms.txt. There are no public dictionaries and no anonymous reads. Its robots.txt says so on the first line — the service is discoverable, the dictionaries are not — and that is not a stage we are passing through on the way to something open.

One membership set governs each dictionary for both reading and writing: the owner, the guest editors the owner adds, and members of a group attached to it. Every artifact — the PDF, the EPUB, the HTML reader, the JSON — takes the same gate. Knowing an id is not access.

So this is not an invitation to go and browse. It is a description of a press we run, written down because the press is now real. When a dictionary does become readable, it will be because its author opened that one — as a printed book, an EPUB, or a private reader for a study group.

What stays with me is the ark itself. One Hebrew word for the box that carried a household through a judgement and the box that carried a child down a river, and two languages that each hide half of it. Someone has to write that down once, carefully, and then mean the same thing every time afterwards.