Following some of my favorite bloggers, I’ve written an overview of 2019. It was a very good year: I spent a lot of time traveling, meeting new people and learning new stuff. I’ve noticed I’m slowly getting better in the things that were new in the first year of my PhD, such as writing scientific papers and formulating research questions.
At the very beginning of 2019, I gave a talk at CLIN (a meeting of computational linguistics researchers that focus on the Dutch language) about an unfinished prototype of headline generator Churnalist. Here I also met Thomas Winters from KU Leuven, who ended up being my ICCC travel buddy later this year as we stayed at the same hotel.
I spend most of the winter months working on Churnalist and condensing everything I learned in my first full paper. The paper was accepted at ICCC 2019, so I got to travel to the US and meet the wonderful members of the computational creativity field. I wrote up my experiences in this blogpost.
To take a break from Churnalist, I took the month April off from my main research project so I could do a fun side project. I started analysing last year’s NaNoGenMo projects because I was curious to see what approaches and data people use if they only have 30 days to create a story generator. By looking at other people’s Github repositories (so many different languages!) I learned a lot about programming, procedural generation and low-cost story generation.
My notes about these projects turned into a second full paper, which was accepted at the StoryNLP workshop at ACL 2019 in Florence, Italy. Visiting ACL was an interesting experience because it was a HUGE conference, and only a tiny minority of the attendees were interested in creative text generation. I felt a bit lost in the beginning, but I had a great time in the end. I enjoyed the beautiful city of Florence (art! history! tiny cobblestone streets! icecream! truffle pasta!), met Dong Nguyen (who also did a PhD with my supervisor Mariët Theune) and found fellow story generation enthusiasts at the StoryNLP workshop.
In August, I visited COG 2019 to present a demo of my (now working!) prototype of Churnalist. People seemed to like my ICCC conference report, so I wrote another conference report about COG. A few weeks later, I published the code for Churnalist on Github.
I was able to meet some of my favorite Twitter people In Real Life at ICCC, ACL and COG. It was wonderful to meet so many of you!
It feels great to be part of a research community where research, tech and art are so intertwined. Especially in the procedural content generation (PCG) and computational creativity (CC) communities, people tend to wear many hats: game developers also publish about their projects in scientific journals, and some researchers also produce games. A large part of the community is active on Twitter, or writes blogposts, tutorials, or open source code, or is involved in organising creative events like game jams. All this sharing makes it easy to find resources and learn new things!
I had a lot of fun this year with all kinds of creative projects!
In March, I opened my t-shirt webshop with Spreadshirt, which means that people can now order my t-shirt designs online. I’ve created 5 new designs in 2019 and uploaded a bunch of my old designs (that were previously only for sale in the Netherlands). Spreadshirt prints t-shirts on demand and ships to all European countries, so it’s now much easier to buy one of my t-shirts! I also like that everything on Spreadshirt is customizable, so if you want a mug or a hat or a sweater with one of my designs, you can!
In September, I started streaming on Twitch. On Thursday evenings I play old and new games, drink tea and chat with the stream audience. Since my graduation I haven’t spent as much time playing video games as I would’ve liked, and thanks to my streaming schedule (and the contagious enthusiasm of my audience), I have finally played some games from my backlog & some nostalgic “oldies”. So far I’ve played: Old Man’s Journey, The Lion King (1994), House Flipper, Untitled Goose Game, Populous: The Beginning and Far From Noise. Besides these “Let’s play” streams, I also streamed the design process of my Death Jazz t-shirt.
To set up a stream with decent quality, I had to learn how to work with broadcasting software OBS Studio, fiddle around with my audio settings, find suitable public domain music, read up on copyright laws and build some streaming layouts. I also tried out different open source video editing tools and had to learn how Twitch works. I hadn’t realised how much time and effort is involved in running and producing and presenting a stream. It makes me appreciate well-produced live-shows like Critical Role and UnDeadWood even more.
I wrote 8 blogposts in 2019, which is a lot more than the 2 blogposts I wrote in 2018. Besides the conference reports, people seem to like the Twitter guide for researchers and the list of video game datasets.
I started an open source project in Python/Django called Josephine (named after Jo March from Little Women) for keeping track of the books I read. It was inspired by Hans de Zwart’s booklist which is a bit like GoodReads, but completely self-hosted. Unfortunately, Hans’ code is not open-source, so I decided to build my own version. If you like reading books and programming, feel free to contribute on Github.
I’m half-way through my PhD! I’ve updated the chart I made earlier this year, which lists all my PhD projects and their research topics:
I think I have a clear idea of the general direction of my research, so it’s time to start planning The Thesis (dun dun dunnnnn). I think I’m going to follow Austin Kleon’s example by making a paper prototype. Then I can just flick through it and see where the “plot holes” are – and start fixing them.
I’m currently working on evaluating Churnalist by asking people to judge Churnalist’s outputs. That means learning new research methodology (setting up human subject research was not part of my computer science curriculum), working with the ethical committee of my department and learning how crowdsourcing platforms work. It’s not going as fast as I had originally planned, but it’s all very interesting (I’ve had fun collecting materials for the questionaire) and my colleagues are helping me fill the gaps in my knowledge. I’m trying to the take the best practices from various ICCC papers, and Ehud Reiter’s blog also helps.
My next prototype will be a system that can generate fictional radio chatter. The Data2Game partners are working on a science fiction game and they were looking for procedurally generated messages for the comms station. Enter text generation! Currently there is a student working on this – she’s using NASA radio transcripts as source material – and I will continue this project in the rest of 2020.
Another student is working on NLP (sentiment analysis) on Skyrim dialogues. Since I’ve been working on sentiment analysis on the side for some time, I hope we can turn her research project into something that is useful for game developers or games researchers.
Judith van Stegeren is a Dutch computer scientist. She is working as PhD candidate at the University of Twente, where she researches natural language generation for the video games industry. She occassionaly works as a consultant in data engineering for textual data.