ThinkASL: the grammar
of fingerspelling.

Fingerspelling isn't the alphabet. It's a grammar — whole words folded into one motion, single letters swapped into known signs — and it's the part most courses skip because it's the part you can only learn by producing, not by recognizing. ThinkASL is the course built around the producing. You watch a short clip, then fingerspell it back into your own laptop camera, and the recognizer reads what your hand actually did. It's coming soon. This is what we've built so far.

What ThinkASL is

Ask someone what fingerspelling is and they'll tell you it's the manual alphabet — twenty-six handshapes, one per letter, the chart on the classroom wall. That's where it starts and not where it lives. A fluent signer doesn't spell J-O-B one letter at a time. They sign #JOB: the spelling compressed into a single sign, the handshapes blurred together, the movement shortened, the whole word arriving as one gesture with its own rhythm. The letters are still in there, but you no longer read them as letters. A whole class of words lives behind that marker — #DO, #BANK, #SALE — lexicalized fingerspelled signs, and they are everywhere in adult conversation.

Then there's initialization: a single letter swapped into a known sign to make a new word — the same movement and palm orientation for Monday through Saturday, only the handshape changing. That, and the lexicalized signs, are the grammar of fingerspelling. ThinkASL teaches that grammar. It is not an A-to-Z alphabet trainer. The alphabet is the doorway; ThinkASL is the room behind it, the part that takes producing and not just recognizing — and the part almost no course measures, because measuring it means watching the hand.

What we built

The course content isn't the hard part — the structure for that already exists, twelve lessons deep, from common and proper nouns through the lexicalized and initialized signs. The hard part is the practice loop: the partner who sits across from you and watches your hand. We built that out of a camera you already own.

• The recognizer The same hand recognizer that runs across our other camera demos, loaded into the lesson page. It classifies a fixed, closed set of handshapes — sixteen letters, ten digits, and a few held gestures — and returns the letters you signed, nothing more. It reads a constrained token on demand, which is a tractable thing to read. It does not caption conversation, and it does not try to.
• The watch-then-spell loop You watch a short clip taught by a Deaf instructor, captioned by default and slowable to half speed. Then you fingerspell the answer back into the lens. Each letter holds for a beat to confirm before it counts, so a twitch doesn't register as a token. The recognizer reads your hand letter by letter and checks it against the word the clip was teaching.
• The honest fallback Some handshapes the recognizer can't tell apart cleanly — the J, the Z, the ones that move. Any answer containing one of those falls back to typed input automatically, one keystroke away, with no "you missed out" framing. A hard letter never costs you a question you actually knew. The camera path and the typed path earn the same credit.
• Frames stay on the device The recognizer runs entirely in the browser. The camera image is never sent anywhere — the only thing that leaves your machine is the parsed letter stream, the four letters F-R-E-D, not a video of your hand. That's the line that lets a camera sit in front of a learner without turning the course into a watch on them.

* Target. ThinkASL is coming soon — the twelve-lesson count is the launch target, not a live total. The recognizer's twenty-nine-token set, the on-device privacy line, and the two answer paths are built and working today.

Try the loop

Here's a small, working slice of the practice loop — no camera, no model, just the shape of it. Watch the clip, then spell the word back by tapping the letters, the way you'd fingerspell it into the lens. The dashed keys are the letters the recognizer can't read; in the real course those fall to typed input on their own. Press the lit key for the next letter, then check your answer. Nothing here leaves your browser.

The alphabet is where fingerspelling starts. The grammar is where it lives — and the grammar is the half you can only learn by producing it where something can watch. — a working note, written into the ThinkASL practice spec

Where it comes from

This one runs close to the bone. Five generations of us have been Deaf, and the structure under this course came from inside that line — a grammar of fingerspelling worked out by Deaf hands and taught hand to hand, across a kitchen table or a classroom, long before anyone put it on a screen. Our job wasn't to invent it. It was to carry it across, intact, to a place where the camera you already own becomes the partner who sits across from you. Bringing it online doesn't change what it is. It means the next person learning it doesn't have to find a partner first.

Why this fits 1891

The pitch on madeby1891.com is for systems built so ordinary people can run them, in Frederick, under one roof. ThinkASL is that pitch pointed at a skill that usually needs another person in the room. The camera is the practice partner, and the practice is real — you produce the sign, the system reads what you produced, and the feedback lands in the instant a teacher's would. None of that is an access feature bolted on at the end. Captions on by default, a typed path that earns the same credit, and frames that never leave the device are just what the course needs to work for the learner in front of it — the second-language adult, the new coworker, the interpreter sharpening for a living.

What's still unfair about this

The recognizer reads a constrained set on purpose, and that's a real limit, not a hidden one. A learner signing one deliberate letter on demand is a tractable thing to read; conversational signing is a research problem, and we don't pretend otherwise. ThinkASL does not caption ASL, and where a handshape sits outside what the recognizer can read cleanly, the honest move is the typed fallback, not a confident wrong guess. The course is built around that line, not in spite of it.

And it isn't open yet. The recognizer is running, the practice loop is built, and the work left is the careful part — the curriculum pass, the clips, the record at the end for the interpreters who earn it. We'd rather say that plainly than dress a half-built course as a finished one. When ThinkASL opens, the first lesson is free, the way the wall chart always was.