Lingo

Streamlining the BuzzFeed video translation workflow.

  • User experience
  • user research
  • prototyping
  • user testing

Background

In 2017, increasing videos views in international editions was a company priority. The higher view count would mean more potential ad inventory.

We needed more video content but didn’t have the resources to produce new content for every edition. However, we had a large library of English language content that we could translate. The translation workflow at that time didn’t have the throughput to meet the desired demand. My team was asked to retool the process.

This was a tricky problem given the variety of formats we produced. Some videos use only dialogue, others use on-screen text, and some use a combination of both. We didn’t have the resources to tackle all of these formats at once. We needed a point of focus.

Finding a Focus

Tasty was our most popular video brand at the time, and thus the most lucrative opportunity for international expansion.

Tasty recipe videos follow a strict format: they are shot from overhead, feature the recipe as on-screen text, and have no dialogue. This would be the format constraint of our MVP.


Research

We spent the first month learning as much as we could about the translation workflow.

Stakeholders

Social Media Editors (SMEs)

Request video translations for their editions and are based in Brazil, France, Germany, Japan, and Mexico.

Video Adaptation Editors (VAEs)

Create the translated videos and are based in Los Angeles.

Translation Workflow Overview

  1. Request video translation
  2. Create Zendesk ticket for Working files
  3. Add translation to VAE workflow table
  4. Transcribe video
  5. Create transcript
  6. Translate transcript
  7. Approve transcript translation
  8. Check for Premier working files in archive
  9. Close Zendesk ticket
  10. Produce translated video
  11. Request video review
  12. Review and approve translated video

3 roles, 8 tools, and 6 timezones.

Workflow Assessment

  • The transcript quality from the Turks was inconsistent because it depended on how they interpreted our instructions.
  • SMEs often translated transcripts without referencing the original videos because switching between the translation spreadsheet and video player was too cumbersome. Meaning they didn’t see their translations in context until the review stage. Due to the time zone differences between the SMEs and VAEs, copyedits could take a day or up to a week.
  • The process involved too many tools making it difficult to track progress of translation job.

Product Objectives

  • Increase translation throughput
  • Improve transcription quality
  • Improve translation accuracy
  • Provide visibility and accountability

Product Focus

Increasing the translation candidate pool

The requests from SMEs weren’t enough, we needed another source, ideally an automated one.

Data Science to the Rescue

Our data scientist created a data model that analyzes published videos and recommends the two markets the video will most likely perform the best in when translated.

Automating transcription

To increase transcript quality we replaced Amazon Turks with a computer vision transcription service.

Automation Advantages

  • Significantly faster turn around.
  • Capturing start and end time for every occurrence of on-screen text.
  • Capturing text formatting.
  • Ability to optimize the OCR model for our lexicon.

Design Exploration and Refinement

Our goal was to consolidate transcript request, transcript translation, and translation request fulfillment into a single tool.

Queue Management

Here are some early workflow wireframes showing a number of concepts to show workflow status like lanes, tabs, and filters…

Final Design

To reduce complexity we decided to have separate optimized queue views for the SMEs and VAEs. The SME view focuses on requesting transcripts, translating the received transcripts, and then sending the approved translations to the VAEs for production. The VAE view focuses on reviewing and claiming video requests.

SME Queue
VAE Queue

Transcript Translation

Our primary goal was to make it easy to reference the original video while translating. We explored a few options for placing the video and ultimately located it to the right of the transcript, following along as a user scrolled. When a transcript translation input field is in focus, the video jumps to the point when that text appears.

Other Notable Functionality

  • Adding video editing notes.
  • Flagging transcriptions that should be excluded from the translated video.
  • Basic text formatting.

Early Transcript View Wireframes

We explored various permutations of the video player and transcript UI…

Final Designs

Here too, the SMEs and VAEs had transcript views optimized for their tasks.

SME Translation View

Video jumps to the start time for the in focus transcription.
Transcription actions menu.

VAE Translation Reference View

Optimized for quickly copying text and reviewing and editing notes, as well as downloading assets for the original video.


Launch

As we predicted, having the original video as reference led to more accurate translations and far fewer edits during the review process. Average translation time was dramatically reduced from two weeks to a couple of business days.

Stakeholder Feedback

I just translated a video during a commercial break cos i didn’t have to change tabs!!!!

—Social Media Editor

…i just did a video for you in 10 minutes (excluding export time which takes as long as it takes.) ITS SO FAST

—Video Adaptation Editor

Lingo suggested a video to me yesterday afternoon, i had the transcript and translated it within about 20 minutes, sent it off and today it’s in frame.io for me to check

—Social Media Editor