🥔 TATERS: Takes All Things, Extracts Relevant Stuff¶
Status: early and evolving. It already works for many common workflows, but expect the occasional rough edge and renaming as things mature. Pin a version if you need stability.
The backstory¶
Today, there are more ways to think about language as data than ever before in human history. The rapid expansion of text analytics and natural language processing into the social sciences has been great for us "word nerds" with programming backgrounds, but not so great for curious scholars and students who are lacking in technical expertise. For people who want a simple way to get up and running, the idea of learning a new programming language, figuring out entire ecosystems of package dependencies, and trying to get things to "just work" can be daunting. Even for those of us who do this for a living, it can be frustrating at times.
So, I spent years (and years, and years) building lots of little "point, click, voila!" applications to do different tasks, trying to lower the barrier to entry for colleagues and newcomers. But, even for me, it was exhausting to run my data through not one, not two, but many different applications to do more advanced analyses. I tried o solve that problem with a program called BUTTER — a more comprehensive application where you could build your own pipelines without writing a single line of code. The idea was simple: there are a lot of methods that social scientists commonly need, so why not put them all in one place? BUTTER was a good idea with a lot of shortcomings: it was written in native C#, which meant Windows-only, hard to extend, and forever cut off from the scientific ecosystem where all of the interesting breakthroughs were (and still are) happening.
So, here we go again. This time we're baking TATERS: written in Python, natively multi-platform, and it gets things done with gusto. I have to say, so far, I like it.
What it does¶
TATERS turns raw data — video, audio, or text — into analysis-ready features. And, depending on what your data looks like, hey, it'll even do some analyses for you. Let's say that you have a dataset of song lyrics and you want to know how the language changed over the years. Oh me, oh my, that sounds like you're going to have to do some natural language processing. Oh, wait, here we go, let's just tell TATERS where the data is, which columns we care about, and let it do the work for us. What if we have a folder of PDF files and want readability scores for each one? What if we have a CSV file of social media posts and want to run a topic model, measure lexical complexity, calculate some dictionary-based scores per user, and see which of those are best at predicting a person's age? Yep, just tell TATERS what you need and it'll do the heavy lifting for you.
TATERS asks what your data is and what you want out of it, works out the steps to get there, and runs them. What it leaves behind is an ordinary pipeline file you can re-run, edit, or hand to a colleague — see the app guide.
Importantly, the goal is to make it so that you do not have to write any code. If you do write code, every part of TATERS is also extremely modular, which means that you can take any part of it that you want, import it into your own Python codebase, and run any piece that you want, any way that you want.
The guides explain what each method is for, and the API reference documents every parameter.
What it isn't¶
It isn't edible. It also isn't a replacement for actually understanding the methods that you're using. I'll do my best to point users/readers to what I feel are some of the more valuable resources and citations for learning more about the rationale and nuances of different methods. There are a lot of great ideas out there, and it's a fast-moving field right now, so it's important to plug into the wider world of scientific discovery if you want to truly get a sense of what you're measuring, why you're measuring things a certain way, and how it all fits together.