At the AIIC PRIMS meeting in Seville, Marty Zhu from Cymo.io presented a tool that embodies a simple but powerful principle: let machines do what they do best: capture terms, titles, numbers, and units. And let interpreters do what humans do best: tell stories and convey emotions.

That tool is Cymo Note, a Computer-Assisted Human Interpretation (CHI) system that's quietly changing how conference interpreters work. But more importantly, it's changing the conversation about what AI means for the profession.

The Problem: Cognitive Overload in the Booth

Conference interpretation is one of the most cognitively demanding tasks humans perform. Simultaneous interpreters must listen, understand, and produce speech in real time, often while managing an enormous working memory load. Terms, acronyms, numbers, proper names, units of measurement, all of these demand precious mental resources that could be better spent on what actually matters: understanding context, grasping nuance, and crafting a message that lands with the audience.

For decades, interpreters have relied on preparation, note-taking strategies, and sheer memorization to manage this load. But what if the machine could handle the memorization part?

That's the question a senior Mandarin interpreter asked Marty Zhu back in 2018: "Can interpreters use AI to lighten our memory burden and reduce note-taking work?"

Cymo Note was the answer.

How It Works: The Technical Side

Cymo Note operates on a straightforward premise. During a live interpretation—whether remote or in-person—the tool runs in the background, capturing the speaker's audio in real time. An automatic speech recognition (ASR) engine transcribes what's being said, and the tool highlights key terms, numbers, and proper nouns as they appear.

Interpreters can interact with this stream of information in several ways:

• Quick tagging: Double-click or drag to highlight a term you want to remember, and it gets added to your glossary instantly.
• Glossary management: Import pre-prepared glossaries as CSV or Excel files, or build them on the fly as you interpret.
• Force replace: If the engine misrecognizes a term, you can create rules to correct it automatically next time (though this isn't foolproof because AI is probabilistic, after all).
• Multi-engine flexibility: Different engines perform better on different languages. Chinese-made engines excel with Mandarin. Asian engines handle Middle Eastern Arabic better. Cymo Note lets you choose which engine to use based on your language pair.

The tool is available as a downloadable app for PC and iOS, as well as a web version in Chrome. There's also a consecutive mode with handwriting features, which is actually where the tool got its name, though simultaneous interpreters have increasingly adopted it too.

But here's the crucial part: the interpreter stays in control. You're not passively watching a transcript scroll by. You're actively deciding which information matters and what to do with it.

The Learning Curve: From Safety Net to Habit

Zhu was honest about one concern raised during the meeting: Will this become a crutch?

The answer is nuanced. Yes, initially. When interpreters first start using Cymo Note, they tend to rely on it heavily. There's a psychological comfort in knowing that the machine is capturing the terminology. But he describes a dependency curve that, over time, actually flattens out.

As interpreters become familiar with the tool's limitations, they develop a more sophisticated relationship with it. They stop staring at every word and instead glance at it strategically, when they sense trouble. They learn which engines perform best in their language pair. They know when to trust it and when to ignore the "jitter at the corner of the screen."

And importantly, they maintain their traditional skills. Because what happens when the internet drops? What happens in a venue with poor audio quality? The tool becomes a liability if you haven't kept your foundational interpretation abilities sharp.

ASR is probabilistic. It's affected by hardware, connectivity, accents, and background noise. Interpreters still need to train the old-fashioned way, to maintain context awareness, to hone their storytelling abilities, to have a fallback when technology fails.

What This Actually Means for the Booth

In practice, how do interpreters use Cymo Note during a real assignment?

It's not about reading the transcript. It's about having a safety net for the moments when memory fails. But there's a warning sign: if interpreters find themselves doing "sight translation" of the transcript, that's the moment to pull back. That defeats the purpose.

Preparation: A Different Kind of Tool

Say the client sends over briefing materials about an upcoming meeting. The tool processes it and highlights key terms. You can star the ones you wants to remember, and they automatically populate your glossary. Before the meeting even starts, you have already built a foundation of terminology.

This is where Cymo Note looks very well as a preparation tool. It's not doing the interpretation work—you are—but it's accelerating the research phase by automatically extracting and organizing the terms that matter.

The Bigger Question: Is This a Threat?

One of the underlying tensions at the meeting was the specter of fully autonomous interpretation. If AI can transcribe speech, can it also translate it, end-to-end, without human intervention?

This was addressed head-on. Yes, end-to-end AI interpretation exists. But here's the thing: if you define interpretation narrowly—as converting a message from one language to another—then autonomous AI might seem like a threat.

But that's not what interpretation actually is. "What we're doing here," Zhu said, "is to take emotion and context and package it in a new way that will leave an impression in the audience."

That's human work. That's storytelling. That's performance.

If you use the tool correctly you will find yourself moving away from memorization and toward narrative craft, paying more attention to how the message lands, less attention to getting every term exactly right, and focusing on the impression you want to leave.

This is why junior interpreters should take theater classes. There's a performance element to this work that machines can't replicate.

And if the profession can collectively shift its identity from "linguistic converter" to "narrative craftsperson," then the rise of AI tools becomes less a threat and more an opportunity. You're freed from the drudgery of memorization. You can focus on what you actually do better than any machine.

Outside the Booth: Privacy, Infrastructure, and the Real World

Of course, none of this happens in a vacuum. There are practical questions about data, privacy, and infrastructure. Cymo Note's approach is worth examining because it reveals the constraints of building AI tools for professional interpreters.

On privacy: Cymo Note uses multiple engines—Microsoft, Google, Chinese vendors, and Cymo Note 's own modified Google service. The tool doesn't store audio files on servers. The script stays local on your device. Only glossaries (up to five projects) are optionally backed up to the cloud. Cymo Note adheres to California privacy standards, which are considered on par with GDPR. For interpreters handling sensitive materials—diplomatic negotiations, legal proceedings, confidential business meetings—this matters enormously.

On servers and latency: Cymo Note operates servers in Silicon Valley (for the Western Hemisphere) and China (for Asia, Middle East, and North Africa). Latency can be an issue, which is why the tool lets you experiment with different engines to find which one performs best in your venue.

On carbon footprint: Here's where things get interesting. Large language models and ASR engines require enormous computational power. Data centers are carbon-intensive. Zhu acknowledged this, especially given that many of his clients discuss climate change.

Cymo is exploring quantized models—smaller, more efficient versions of AI systems that require far less computation—but the technology is still experimental. For now, the company is monitoring what its vendors do to reduce their carbon footprint. This is the unsexy, infrastructure-level reality of AI tools: they have environmental costs, privacy implications, and geographical constraints that don't always align with where interpreters need to work.

Consecutive Interpretation: Where It Started

With timestamps and handwriting support, Cymo Note becomes a digital note-taking system that ties your handwritten annotations directly to the speech transcript.

It's a different use case from simultaneous, but it shows how a single tool can adapt to different interpretation modes. Some users preferred an older interface with a split screen (notes on one side, transcript on the other).

What Actually Changes

So what does all of this mean for the profession?

Cymo Note isn't replacing interpreters. It's not even close. What it does is shift where cognitive load is distributed. The machine handles the rote memorization. The interpreter handles the meaning-making.

That's a meaningful change. It means preparation time might shift from drilling terminology to understanding context. It means the actual skill being tested in the booth might shift from memory to presence, how well you command the room, how effectively you convey not just words but tone and intention.

It also means the profession needs to be intentional about preserving the skills that aren't being automated. Because if every interpreter uses Cymo Note and forgets how to work without it, the first interpreter who encounters a dead battery, a lost Internet connection, or a venue with too much background noise will suddenly be in real trouble.

The tool is a genuine aid. But it's an aid, not a replacement. And maintaining that distinction might be the most important skill of all.