All learning paths Learn See Try Make Lead a collection with a team What this lesson is about Plan work you cannot finish alone: the sampling table, the people you hire, and the records that prove it.
The money for two hundred hours of Maithili speech arrives, and six people want the work. What comes first?
A Hire the six today, and draw the table up from what the first week brings in. B Write the sampling table before you hire: district, age band, gender, and target hours. C Start in the two nearest blocks, since age and gender even out over two hundred hours.
Check my recall Why this matters IndicVoices published the whole shape of this work. It reports 7,348 hours of speech from 16,237 speakers, across 145 districts and 22 languages. Of those hours, 1,639 were transcribed. Nobody records that alone. What holds it together is a table written before the hiring, and a record behind every speaker in it.
Plan and staff a collection too big for you to record it yourself.
IndicVoices published the whole shape of this work. It reports 7,348 hours of speech from 16,237 speakers, across 145 districts and 22 languages. Of those hours, 1,639 were transcribed. Nobody records that alone. What holds it together is a table written before the hiring, and a record behind every speaker in it.
Write the sampling table before you hire: district, age band, gender, and target hours. Every speaker leaves two records behind: her own consent copy and a signed payment line. Budget the checking hours beside the collecting hours, or the checking will not happen. Task Three months and money for two hundred hours of Maithili speech, and six people who want the work.
Weak approach Hiring the six you already know, and telling each to bring thirty hours from whoever agrees. The papers get sorted out when the files start arriving.
Stronger approach Splitting two hundred hours across four blocks, three age bands and both genders, and writing that table down first. Hiring six against it, training them on the same ten prompts, and paying per accepted hour. Every file carries a consent number and a payment line. Twenty of the two hundred hours are paid time for checking.
Why the stronger approach works The weak plan can still produce two hundred hours. What it cannot produce is a sentence saying who is missing from them, or a paper showing one person agreed.
Try a changed situation Not speech this time. Fifty thousand Bhojpuri sentences, typed by people you pay. What do you settle before the first batch?
Make something yourself Write the plan for a two-hundred-hour collection in a language you know. Give the sampling table with target hours in each cell. Add the team, the pay basis, and the two records every speaker leaves.
Lesson 10 of 12 on Bharat-language data craft next Measure the people collecting it Part 4 ends in a work sample you can send
Read the primary or official source: Javed et al., IndicVoices (arXiv)