This site should be signed first and written second. If you can help us make that true, we want to hear from you.
What is LARRY?
LARRY is a format, not an app. It is a way to write signing down as plain text — nothing more, and that's the point. Think of what written text did for speech: once speech could be written, it could be sent, stored, searched, edited, and eventually spoken aloud by machines. Sign languages have never had a practical equivalent. LARRY (Linguistic Articulation Representation for Real-time sYnthesis) is an attempt to build one.
Unlike older systems, it captures what simple word-for-word glossing cannot: facial grammar, use of space, timing, and two hands doing different things at once — and a computer can check a LARRY file for errors the way a compiler checks code.
Discourse
Who is being talked about, and where they live in signing space.
Lexical
Which signs, in what order, with what grammar on the face and body.
Phonetic
Exact handshape, place, and movement — for new signs and classifiers.
# "My mother gave me a book yesterday."
YESTERDAY |
[topic] { poss(1p) MOTHER !est(mother @ ipsi-mid-near) } |
BOOK !est(book @ center-mid-near) GIVE :dir(mother -> 1p)One format in the middle
LARRY is an intermediary layer. Many different things can write it, and many different things can read it — and none of them need to know about each other. That is what a shared format buys: build one good animated signer, and every tool that writes LARRY can use it. Build one good camera recognizer, and everything downstream gets signing as text.
checkable
What can be built on it?
Interpreting is the obvious use — but the deeper comparison is text-to-speech and speech-to-text. Those aren't "interpreters for blind people"; they're direct interfaces anyone uses to talk to machines and to each other. LARRY aims to be that layer for signing:
Animated interpreting
Speech → translation AI → LARRY → animated signer, live. The announcement with no captions, the kiosk with no terp.
Sign in, like speech-to-text
Sign at a camera; recognition writes LARRY; the machine acts on it or renders it as text or voice. Talk to a computer — or through it — in your own language.
Sign out, like text-to-speech
Compose a message in LARRY (or from text) and an avatar signs it for you — no filming yourself, editable before sending, reusable.
Signed messages as text files
A signed message today is a video: heavy, uneditable, unsearchable, face required. As LARRY it's a few hundred bytes you can edit, quote, and search — and render as any signer you like.
Archives you can search
Signed content stored as text can be indexed, quoted, diffed, and studied — the way written language always could be.
Learning & research
Dictionaries, grammar checkers, corpora, and teaching tools that read and write the same open format.
Not a replacement for interpreters
A good interpreter reads the room, catches confusion as it forms, speaks up when the doctor needs to slow down, and carries real responsibility. Machines don't have that judgment. This project is not trying to build it.
The goal is the empty spaces: the announcement with no captions, the kiosk with no terp, the 2am emergency room before the interpreter arrives — and every place a hearing person would reach for text-to-speech or dictation, where a signer today gets nothing at all. Where the current option is nothing, a machine that signs is a win.
First bar to clear: sign better than a video-interpreter cart on three bars of hospital Wi-Fi. The bar is on the floor. We expect to trip over it a few times anyway.
Who is behind this — honestly
LARRY is being developed with AI assistance (Anthropic's Claude), directed by Dakota — a hearing CODA whose father is a late-deafened adult. The family signed PSE at home, not strict ASL. That is the motivation, not a credential: every linguistic claim in this project is a hypothesis until fluent Deaf signers and ASL linguists have reviewed it. Deaf collaborators hold real authority here — maintainer seats with binding review over all linguistic content. The next version does not ship without them.
The language is named for Dakota's father, Larry.
Get involved
ASL is a working language of this project. Feedback, reviews, and maintainer sign-offs are welcome as signed video — the translation work is on us, not you. Most wanted right now: fluent Deaf signers and sign linguists to challenge the spec's grammar; a rendered animation proof-of-concept; a core sign dictionary.
GitHub repository Read the spec