A role-playing workshop on how to come up with an annotation schema for your NLP business problem. We will also touch on how to split a new problem into old well-established NLP tasks to minimise the amount of annotations needed.

You will play your part in a small team. You will be assigned one role in the team:

• NLP engineer - solving the problem with code, automating a manual process, creating an annotation guidance, splitting the problem into well-known NLP tasks ;
• Annotator - depending on the roll of the dice you might become a randomly clicking but speedy mechanical turker, an unreliable student, or a very detailed in-house expert asking lots of questions and working very slowly
• Subject Matter Expert - doing it all by hand and fearing automation despite finding lots of bugs in the model
• Project Manager/Business Stakeholder. You are defining the business value and answering the engineer’s stupid “why” questions. In return, you ask them every 5 minutes when the model is going to be finished with 99% accuracy. You are working hard on the slidedeck and the storyline for the Steering Committee Milestone Review meeting next month - who knows, they might decide to axe your project!

We will automate two text tasks - extracting crime reports from newspapers and determining whether a business is a baker or a butcher based on its web-page.

Even though we focus on the text tasks, the lessons learned apply to all kinds of data - automating a junior designer with image processing or a geospatial analyst with geodata algorithms.

You will get a 360-degree view of what is actually going on in an automation effort, beyond the narrow data scientist’s perspective. We believe that understanding the whole process will help you to become a more efficient link in the chain if not the driver.

No coding knowledge is needed, only familiarity with the ideas of automation and design.