288 Words

I found an eight page PDF from TSMC. At the top it says Security C - TSMC Secret. Then it starts with ability, abnormal, abort, absence, acid, active, actual, add. A few lines later there is AI, air shower, alarm, alert, align.

There are 288 words.

TSMC makes some of the most advanced things humans know how to make. Inside its factories they use extreme ultraviolet light to manufacture structures almost impossibly small. Yet someone at TSMC also decided it was important that a technician knows that bottle means 瓶子 and bottom means 底部.

At first I thought this was funny. Then I kept reading.

Chemical, chip, clean chamber, contamination, cycle time, defect. Equipment, etching, EUV, fault, gloves, hold, inspect. Lithography, lot, machine, mask, material.

The factory slowly appears.

There are wafers, lots, cassettes, pods and masks. Things can be loaded and unloaded. They can be inspected, measured, held, released, reworked or scrapped. There are normal states and abnormal states. There is waiting. There is standby. There are alarms, warnings, faults and emergency stops.

I realized I wasn't really looking at an English vocabulary list anymore. I was looking at the factory through the words needed to operate it.

You could almost reconstruct the place from the list.

There are objects. There are actions. There are states. There are things that can go wrong. There are people who have authority over other people. There are materials moving from one stage to another. There are parameters and recipes and specifications. There is quality and yield. There is a route.

Even the stupid looking words start making sense. Top. Bottom. Wet. Dry. Same. Zero.

If you're standing in front of a machine worth millions of dollars and something is wet when it should be dry, or a lot needs to be held instead of released, these are not basic English words anymore. They are controls.

The vocabulary is part of the interface

This is where it becomes interaction design.

We usually talk about an interface as the thing on the screen. Buttons, colors, menus, icons. But before any of those things work, the human and the machine need to agree about what things are called.

What is a lot? What does hold mean? What does release mean? When do you reset something? When do you abort it? What can be reworked and what becomes scrap?

The vocabulary is part of the interface.

TSMC's technician job descriptions ask for basic English ability around machine interfaces. Suddenly the shape of this list makes sense. It is full of words like press, parameter, process, priority, reset, rework, run, SOP, save and scrap. You do not need literary English to operate the machine. You need to recognize the language of the system quickly enough that the system remains safe and predictable.

Maybe the interesting thing is that TSMC did not only make the machines understandable to people. It also made the people more understandable to the machines.

Thousands of people can enter the factory with very different backgrounds, but inside they learn the same words. The space of possible meanings becomes smaller. Hold means hold. Release means release. A warning is a warning. A wafer is a wafer.

It feels like a design system, except the components are words.

A normal design system standardizes buttons, colors, spacing and components. This standardizes the semantic layer underneath them. The things that exist. The actions that are possible. The states a thing can enter. The words used when something goes wrong.

Before you can design the button, you have to decide what the button means.

TSMC has effectively published a tiny piece of the ontology of its factory.

The old agreement

I started thinking about all the words computers made us learn. File. Folder. Save. Delete. Window. Username. Input. Output. Parameter. Execute. Some of these exact words are inside the TSMC vocabulary. The final page still has username, view, wafer, warning, window and zero.

For most of my life being good with computers meant learning how computers thought.

A desktop wasn't a desk. A folder didn't contain paper. A window didn't show you outside. You learned the model because the computer wasn't going to learn yours.

Factories pushed this even further. Here are the words of this world. Learn them before you enter.

There is something almost beautiful about how explicit this is. We normally hide the fact that every technical system asks humans to accept an ontology. The software presents a menu and we quietly learn what the designers decided exists. Account. Workspace. Project. File. Member. Permission. Archive. Delete.

After enough time these concepts stop feeling designed. They become reality.

But they are still somebody's decisions.

TSMC's list makes those decisions visible because the language has been pulled out of the interface and placed onto eight pages. You can see the boundary. On one side is ordinary life. On the other side is the fab. To cross it, learn these words.

That is interaction design too.

An interface is not necessarily a screen. Sometimes the interface is the vocabulary people are taught before they are allowed to touch the machine.

The direction is reversing

Now I can open an AI and just say what I want.

Remove the person behind me.

Find everyone I emailed last year who never replied.

Make this page feel quieter.

I don't necessarily know the command. Sometimes I don't even know what software function should do it. I describe the outcome and the machine figures out the translation.

This suddenly makes the TSMC PDF feel like an artifact from a very specific moment in computing.

For decades we made humans learn the language of machines. We learned the vocabulary, the commands, the menus, the file systems, the interfaces. Somewhere inside a semiconductor factory this relationship became so explicit that somebody could fit it into an eight page PDF.

288 words.

AI changes the direction of adaptation. The machine increasingly has to understand the human's language instead of requiring the human to memorize the machine's language.

That does not mean the ontology underneath disappears. A factory will still have wafers, lots, recipes, warnings, failures and things that need to be stopped. A database will still have tables. Software will still have permissions. The machine world still has structure.

What changes is who has to do the translation.

Maybe a future technician speaks Mandarin, Taiwanese Hokkien, Vietnamese or English to the same machine. Maybe the operator does not have to know the exact English label hidden underneath. The system understands the request and maps it to the factory's internal language.

There is a risk here too. Learning the machine's vocabulary gives you a clue about how the machine thinks. If natural language hides all of that structure, the interface may become easier while the system becomes harder to understand.

That might become the next interaction design problem.

How do we let people speak naturally without making the machinery underneath completely invisible?

I opened the PDF because I was curious why TSMC had made an English vocabulary list.

I ended up staring at 288 words describing the old agreement between humans and computers.

Sources

TSMC's official vocabulary PDF is publicly hosted on tsmc.com. TSMC technician job listings describe the required English ability specifically in relation to machine interfaces.

AI authorship note

This essay began with Kris Haamer finding TSMC's 288-word vocabulary PDF and asking why the company had created it. Kris identified the interaction design angle: the list is not only language training, but a semantic interface between technicians and the factory. AI supported online research, development of the argument, drafting, editing, and implementation. Kris selected the source material, framing, title, voice, direction, and decision to publish.