Partner case

Seven code-switching languages
on short batch deadlines

A six-week transcription program for a global language services company, in seven languages mixed with English.

How we work with partners →
In short

How do you deliver multilingual transcription on short batch deadlines?

With a named owner for the account, a rule sheet that puts each language's exceptions first, and an internal deadline six hours before the client's, kept for review. That is how Translia delivered about 3,000 code-switching conversations in seven languages over about 50 batches. By the second month, time per task fell below half in most languages.

Where it was stuck
Seven languages,
seven sets of exceptions

A global language services company needed customer-service conversations transcribed in seven languages, each mixed with English: Mandarin, Cantonese, Taiwan Mandarin, Japanese, Korean, Turkish and Tagalog. The audio covered banking, finance, healthcare, IT, e-commerce and billing calls, some at 8 kHz telephone quality.

Batches arrived on short deadlines, and each language had its own rules for marking where speakers switch into English. Those language rules override the general ones, and that is where most errors happen. In Tagalog, a single English word inside a sentence is not tagged and numbers stay in digits; Chinese and Korean work the other way round. Japanese and Korean also need English rendered in their own script.

What we did
A named owner, and a rule sheet
for every language

A named owner, the same people on each language

More than 30 transcribers worked across the seven languages, and the same people stayed on a language from batch to batch.

One rule sheet per language

The client's specification and our own rules were merged side by side: how to tag foreign segments, spelled-out acronyms, overlapping speech, inaudible passages and cut-off words, with each segment kept under 15 seconds.

Our own deadline, six hours ahead

Every batch had an internal deadline six hours before the client's, reserved for internal review.

Checked twice before it left

Each task was self-checked item by item against a delivery checklist, then reviewed internally, then delivered.

What changed
About 3,000 conversations,
getting faster as the rules settled

Over roughly 50 client batches we delivered about 3,000 conversations (around 670 task links), totalling about 850 transcriber hours.

Most languages got faster as the rule sheet settled: in the second month, the average time per task fell below half of the first month's for most languages.

Tight batches and strict tagging rules?

We say yes or no first, then name an owner for the account and set up a rule sheet per language.

How we work with partners →