Imagine that you are going to dinner with a group of friends. Each one lives in a different part of town and has a different schedule. 5pm? Too early – John needs to pick up the kids. 9.30pm? Too late, Eva has to catch the last train home. How about 6pm then? Nope, still too early. Johanna needs to drive for an hour to get there, assuming traffic is good. Finally, you all agree that 7 would suite the group of friends. Next, you need to pick the restaurant. John is allergic to shellfish, Hannah has a gluten allergy and Fred is vegan. And so it begins again…. But by the end of it you have a plan. Johanna offers Eva a ride home if they end up being out late, you meet at 7 and you find a restaurant that suites everyones dietary constraints. In other words, you have a plan.
Planning is very complex task that has interested researchers working in Artificial Intelligence (AI) since the early days of computing science. Modern planning systems can take into account hundreds of constraints and possible actions. In fact, planning systems can create such large plans that people have problems understanding them. Understanding a plan is often necessary because even computers make mistakes (or the people who program the computers do) and anything important has to be approved by people. So one of the key questions is ``How do we present a plan to people so that they can understand it?”
A natural way of presenting a plan is to explain it in words. For example, “We will meet at 7 at the square. Then we will go to the restaurant YumYum.’‘ The process of producing a text from data is called Natural Language Generation (NLG) and the University of Aberdeen is one of the world leading universities in this field.
The SAsSy project aims to make computer generated plans more understandable by using natural language – English. As well as allowing people to understand a plan, it is beneficial to explain why certain decisions or trade-offs were made and this is also a part of our project.
It is a tough, but interesting set of problems, and we have got a great team. The project just started in October and will last another 3 years — we encourage you to get to know us follow our progress on this blog, and official webpage.