AI · Interface design · UX

Language Analysis Interface

An AI-assisted workspace that helps a language analyst translate accurately, learn a new language on the job, and understand the people and events behind the words.

Role

UX research & interface design

Team

Kevin Ward & Ned Babbott

Partner

Laboratory for Analytic Sciences (LAS)

Timeline

Spring 2023

On this page Challenge Meet Cameron Research Design Walkthrough Features

01 · Challenge

Designing for software we weren’t allowed to see.

This project explored how AI, now and in the future, could support analysts translating foreign intelligence. The software analysts use today is classified, so we had only a rough idea of how their current interfaces look and work.

Our goal was to help analysts produce robust, reliable intelligence that conveys content, intent and context.

A note on the data

To present real data, the transcripts, names and events are adapted from the Nixon tapes, framed as foreign intelligence being translated into English.

02 · Persona

Meet Cameron.

LAS gave each team a persona. Cameron’s experience, strengths and weaknesses set the problems our interface had to solve: accurate translation, plus learning while working.

Persona

Cameron

Full Performance Language Analyst, just assigned to a new team

Strength

Three years as a language analyst

Challenge

Almost no experience with the new country’s language

Cameron’s day, before and after

As-is To-be

Today: where Cameron struggles through the day, and every tool they reach for along the way.

With our system: the same day, designed to remove the challenges that come from the tools themselves.

What analysts told us

There are too many tools analysts need to use, and they don’t work well together.

03 · Research

Learning a career we’d never heard of, as quickly as we can.

Interviews

We started with one-on-one interviews with LAS analysts. We couldn’t become experts in a semester, but we needed to get close enough to design accurate solutions.

Critiques

Formal critiques of our concepts followed. Expert access was invaluable, but we had to stay focused and communicate clearly so an influx of ideas didn’t pull us off course.

04 · Design

Three concepts, narrowed into one workspace.

Each early concept tackled a different pain point. Analyst feedback showed us which to pursue and how the tools could work together.

Concept A

Correcting errorful speech-to-text

Concept B

Visualizing contextual searches

Concept C

Mapping storylines

A visual language for focus

Dark edges draw the eye to the transcript in the center.

Bold color flags problems and new information.

Simple type keeps long transcripts readable.

Initial ideas as rough wireframes.

Narrowing down, sketching and developing the visual language.

05 · Walkthrough

See it in action.

Animation and voiceover by Ned Babbott

06 · Features

Four questions, six AI-assisted tools.

Pick a guiding question to see the tools we designed to answer it.

Q1Context Q2Language learning Q3Faster scanning Q4Speech-to-text

How can the analyst not only translate accurately, but also learn the context of the situation and the people involved?

Context InquiryInvestigate the context of an entity through AI-generated patterns and information.

Smart QueryAn AI-powered query tool that uses natural language to search for context in the document and in outside resources.

How can the system support continuous learning of a foreign language through grammar and word choice?

AI Grammar SuggestionTracks writing mechanics like grammar, spelling and word choice, and suggests alternatives.

How can information gathered by the analyst and the system be filtered for faster scanning?

AI Extracted EntitiesIdentifies and extracts entities from the translation: events, locations, people, organizations, indexes and data.

Context FilteringLets the analyst refine entity extraction by adding custom filters.

How can the system intelligently fix errorful speech-to-text, and make manual corrections easy?

STT Truth LabelingWhen the analyst accepts an AI recommendation for problematic audio, the fix is back-translated to correct the speech-to-text in the source language.

Where it went

Presented to the wider U.S. intelligence community as a case for involving designers in unexpected places.

This project was larger and more complex than I could have ever expected. I learned a lot not only about design but also about a whole field of career and study I had never been exposed to before. A huge thanks to everyone at LAS for the support throughout and being our biggest hype-people.

Next project Fading Frogs