Interpreting technology

Computer-assisted interpreting, explained

CAI is the part of AI that makes interpreters better instead of replacing them. Here's what it is, what the tools do today, and how to choose one.

3 generations of CAI 4 min read Published September 2026

Most news about AI and interpreting is about machines speaking for people. The quieter shift is the software built for the interpreter. It has a name, computer-assisted interpreting, or CAI, and it has changed more in the last two years than in the decade before. CAI tools don't interpret; they take memory load off the person who does, putting the speech, the terms, and the numbers on screen so the interpreter can focus on meaning.

CAI vs machine interpreting

	Computer-assisted interpreting (CAI)	Machine interpreting
Who interprets	A human interpreter	Software
What the software does	Shows transcripts, terms, numbers, and translations to help the interpreter	Listens, translates, and speaks or captions the output itself
Who is accountable	The interpreter	The organization deploying it
Where it fits	Medical, legal, government, and business interpreting	Low-stakes events, content, and self-service

The two are often confused, and the difference matters. An agency asking "can I use AI on calls?" usually means CAI. A client asking "can AI replace our interpreter?" means machine interpreting. Our essay on whether AI will replace interpreters covers the second question.

Three generations of CAI tools

Glossary and prep tools

The first CAI tools managed terminology: build a glossary before an assignment, then look terms up in the booth. They're still useful for conference prep. InterpretBank is the best-known example.

Booth aids with speech recognition

Speech recognition let tools show a live transcript and flag numbers and glossary terms automatically. Some conference platforms now build this into their interpreter console, like KUDO AI Assist.

Live-call copilots

The current generation runs beside any call, whether phone, video, meeting, or in person, and adds live translation next to the transcript. Unicaption, Interpreter, and Intercall are examples. This is the generation built for community, medical, and legal interpreters.

What a CAI tool actually does during a call

Captures speech as text as it's spoken, so a half-heard word can be read back instead of reconstructed
Holds names, numbers, and dates on screen: dosages, case numbers, addresses, the details that fall out of working memory first
Shows a translation next to the original, so the interpreter can check a term in the target language
Recognizes terminology from a dictionary or glossary, so drug names and legal terms come out right
Keeps notes beside the call

Everything on that list helps with the same problem: cognitive load. Interpreters listen, hold, convert, and speak at once, and numbers and proper nouns are the first things to slip.

How to choose a CAI tool

Match it to your setting. A booth aid helps inside its platform. A copilot that reads system audio helps on every platform.
Test on your own language pair and audio. Every vendor's accuracy claim is measured on audio they chose. See how to read word error rate.
Check what's stored. For confidential calls, audio and transcripts shouldn't be kept after the session.
Check your agency's policy on aids during calls before using any tool.

Tools that record calls by default, or keep transcripts on their servers

Compliance certifications you can't verify on the vendor's own site

Accuracy numbers with no language pair or test conditions attached

Where Unicaption fits

Unicaption is a live-call CAI copilot. It's built for interpreters who move between phone lines, video platforms, meetings, and in-person work in the same week:

Live transcription as speech happens, in 60+ languages with automatic detection
Side-by-side translation to check terms in both languages
Custom dictionary kept across sessions, with medical, legal, and government terminology built in, plus a notepad
Beside any platform: microphone or system audio; nothing joins the call
Nothing kept: end-to-end encrypted, audio never stored, transcripts deleted at session end, never used to train models
Browser, Chrome extension, macOS and Windows apps; 30 free minutes every week, no credit card

For a side-by-side look at the main CAI tools, see the best AI tools for interpreters in 2026.

Frequently asked

What is computer-assisted interpreting?

Computer-assisted interpreting (CAI) is software that helps a human interpreter during their work, such as glossary tools, live transcription, and live translation beside a call. It differs from machine interpreting, where software produces the interpretation itself. Unicaption is a CAI copilot that shows live transcription and side-by-side translation in 60+ languages beside any call.

What technology do interpreters use in 2026?

Interpreters use remote interpreting platforms for phone and video calls, glossary tools for prep, and increasingly live-call copilots that put transcription and translation on screen during calls. Unicaption is a live-call copilot that runs in the browser, a Chrome extension, and macOS and Windows apps, with a custom dictionary and 30 free minutes every week.

What are examples of CAI tools?

Examples include InterpretBank for glossary management, KUDO AI Assist inside the KUDO interpreter console, Cymo Note for transcripts, glossary highlighting, and notes, and live-call copilots such as Unicaption, Interpreter, and Intercall.

Is using a CAI tool on interpreting calls allowed?

It depends on the agency or client, so check their policy on aids first. For confidentiality, choose a tool that stores nothing. Unicaption encrypts sessions end to end, never stores audio, deletes transcripts at session end, and never uses session content to train models.

Try a CAI copilot on a real call

30 free minutes every week. No credit card, nothing stored.

Start Free Trial →