In an objective sense the papers become less readable the older they are. For one, because at the beginning of the century many papers were still written in French, German or Russian, and also the notation and terminology has changed. It adds to this that back then people were discussing problems whose answer we know today, and it can be difficult to follow their trains of thought. And then, there's the physical readability that deteriorates. Printouts of scans, especially in small fonts with toner low, can give me a headache that is not conductive to my attention.
On one scanned paper that I read, an overactive software removed background noise, and in that process also erased all punctuation marks. In the text that was merely annoying, but unfortunately the authors had used dots and primes for derivatives.
However, in a subjective sense the papers seem to be getting less readable the newer they are, and that almost discontinuously. The style of writing has been changing.
Everything written before roughly 1990 is carefully motivated, edited, referenced and explained. One also finds very frequently errata, or constructive comments in the next issue of the journal, which seems to have fallen somewhat out of fashion after that. By the late 1990s, most papers are difficult to understand if one doesn't happen to work on a closely related topic or at least follows it; the motivation is often entirely missing or very narrow, common arguments are omitted and apparently just assumed to be known, variables are never introduced and believed to conform to some standard notation (that in 100 years nobody will recall), technical terms are neither explained nor referenced and yet hardly anybody ever seems to cite the textbooks that would explain them.
Needless to say, this is not the case for all papers, there are exceptions, but by and large that has been my impression. It's not so bad actually when you are familiar with the topic. In fact, I am often relieved if I don't have to read yet another introduction that says the same thing! But it is likely that to the reader not familiar with the topic, which in some decades might be pretty much all readers, the relevance and argumentation remains unclear.
So then I've been wondering why it seems that by the mid 1990 the style in which scientific papers were written changed. Here's some explanations that I came up with:
- That's just me. Everybody else thinks the newer a paper the better it is understandable, and people back then only wrote confusing garble.
- Selection bias. The old papers that are still cited today, or are at least at the root of citation trees, are the most readable ones.
- Specialization. There are many more physicists today than in 1920, and calculations have been getting more complicated. It would however take up too much space to explain all technical details or terminology from scratch, which makes papers increasingly opaque. There is certainly some truth to that, but that doesn't quite explain why this seemed to have happened so suddenly.
- Typesetting changes. Stefan pointed out that in the 1990s LaTeX became widely used. Before that, many papers were typed by secretaries on typewriters and the equations put in by hand, then the draft was send by mail. The ease and speed of the process today breads carelessness.
- Distribution changes. The pulse of academic exchange has been quickening. Today, researchers don't write to be understood in 100 years, they write to get a job next year. Errata or comments don't work towards that end. They don't add motivations because the people they want to reach share their opinion that the topic is relevant anyway.
Most likely it's a combination of the above. What do you think?
I've been wondering if not the future of the paper is an assembly of building blocks. Why does everybody have to write the motivation or explanation of techniques used all over again? I am thinking in ten years, when you download a paper you can choose an option for the level of detail that you want, and then get the paper customized for the knowledge you bring. That won't always work, but for research fields in stages 3 and 4, it might work quite well.