Nautilus recently published a great article from Chris Manning on the effect of deep learning in NLP. Only partway through reading it did I realize I had seen it before… it’s the same article that was published by Computational Linguistics in December 2015!
And I think the reason for this and the reason I actually read it all the way through this time, are at least partially due to the presentation. The big word cloud at the header, an added picture, larger text in a single column, and those highlights in a large font sprinkled throughout that are becoming more popular (I’m usually not too fond of those, but it could be a learned dislike based on the contexts I’ve seen it used in before).
I’ve always wondered, why do some conferences and journals prefer two-column articles, and others prefer one-column? Among other presentation decisions. I’ve never really seen this topic discussed much (point me in the direction of some thoughts on the matter if you’ve seen them!) Maybe it’s because, if you’re like me, it feels just a little amateurish to prefer the prettier version. It brings to mind certain online news outlets that try really hard to make their content go “viral”. But there’s no less merit to the content if it’s the same!
Maybe this is just a personal thing I can fix myself by e.g. zooming in more when I read research papers, but it’s not as easy to do that when printing PDFs. Also, there are often “prettier” formatted versions of the articles on the page itself in some journals, so maybe I could just opt to read those.
Recent developments like the distill project, and the way OpenAI is releasing their preliminary results could be interesting in this way. The PDF is tried, true, simple, and doesn’t require the most recent browser or HTML5. But even a little bit of makeup on articles could make them more accessible overall, and as a fan of open and honest sharing of science to the public, I hope it gets a bit more attention.