All learning paths Learn See Try Make Write a guideline others can follow What this lesson is about Write rules with examples of the hard cases, so two people in two towns decide the same way.
Ten people must mark which Hindi sentences are polite enough for a bank message. They sit in four towns. What do you write?
A A list of polite words and rude words, taken from a school grammar book. B A guideline is rules plus worked examples of the hard cases. C Every rule you can think of, plus a promise to answer doubts on the phone.
Check my recall Why this matters Rules alone produce two people deciding two ways in two towns. The paper behind this lesson is a list of questions a stranger will ask later. A guideline has to answer them before they are asked.
Write a guideline that two people in two towns can follow to the same answer.
Rules alone produce two people deciding two ways in two towns. The paper behind this lesson is a list of questions a stranger will ask later. A guideline has to answer them before they are asked.
A guideline is rules plus worked examples of the hard cases. Write every disagreement you have already had into it as a decided case. Hand it to somebody who was not in the room; their questions are the missing parts. Task Ten people are marking which Hindi sentences are polite enough for a bank message.
Weak approach One line: mark the sentence polite if it sounds respectful.
Stronger approach The same line, plus six settled cases. Aap against tum. A missing greeting. A shortened word. Three sentences the team argued about.
Why the stronger approach works The weak guideline is not shorter, it is undecided. Every hard case it skips gets decided ten different ways.
Try a changed situation The team has moved to marking photos of shop signs, and two reviewers keep clashing over blurred signs. What belongs in the next version of the guideline?
Make something yourself Write a guideline of one page with at least five decided hard cases. Then give it to one person who was not in the room, and record every question they asked.
Lesson 4 of 12 on Bharat-language data craft next Settle the transliteration rules Part 2 ends in a work sample you can send
Read the primary or official source: Gebru et al., Datasheets for Datasets (arXiv)