This question may not suit to here, but I couldn't find a better place to ask (it was closed in SO).

I find research papers on computer science hard to understand. Of course the subjects are complicated. But after I understand a paper usually I can tell it to someone in simpler terms, and make them understand. If somebody else tells me what is done in that research I understand too.

I think the best example that I can tell here is: I have tried to understand SIFT paper for a long time, and I found a tutorial while googling, in a couple of hours I was ready to implement the algorithm. If I was to understand the algorithm from the paper itself it might have taken a couple of days I think.

My question is: is it only me who finds research papers this hard to understand? If not how do you deal with it? What are your techniques? Can you give tips?

  • 4
    When you crosspost next time, please include links, preferably in both directions. stackoverflow.com/questions/4484932/… – Tsuyoshi Ito Dec 19 '10 at 21:07
  • 1
    There is no royal road to learning. That said, people may have something to share on this subject. I will wait for a while before voting to close. – Tsuyoshi Ito Dec 19 '10 at 21:11
  • 4
    When reading a paper from the ACM Digital Library I often think that the authors never use short simple statements when long complicated ones will do. I blame this on school papers having required word/page counts. :) – Zan Lynx Dec 20 '10 at 18:09
  • 2
    It helps when you're a native English too. When you're not, it's more pain. – toto Feb 26 '11 at 17:06
  • 2
    @TsuyoshiIto ask the question here is definitely one royal road to learning. – zinking Jun 28 '12 at 9:08
up vote 148 down vote accepted

Unfortunately, research conferences generally do not place a premium on writing for readability. In fact, sometimes it seems the opposite is true: papers that explain their results carefully and readably, in a way that makes them easy to understand, are downgraded in the conference reviewing process because they are "too easy" while papers that could be simplified but haven't been are thought to be deep and rated highly because of it. So, if you rephrase your question to add another word, is it not just you who finds some research papers unnecessarily hard to read, then no, it is not. If you can find a survey paper on the same subject, that may be better, both because the point of a survey is to be readable and because the process of re-developing ideas while writing a survey often leads to simplifications.

As for strategies to read papers that you find hard, one of them that I sometimes use is the following: read the introduction to find out what problem they're trying to solve and some of the basic ideas of the solution, then stop reading and think about how you might try to use those ideas to solve the problem, and then go back and compare what you thought they might be doing to what they're actually doing. That way it may become clearer which parts of the paper are just technical but not difficult detail, and which other parts contain the key ideas needed to get through the difficult parts.

  • 3
    I second turkistany, in particular I liked the idea of reading the introduction and trying to complete the solution or coming up with a new one. (I know some people also use this for reviewing papers, i.e. to make sure the paper is not made unnecessarily hard to read by the author.) – Kaveh Dec 20 '10 at 0:07
  • 21
    As someone who recently served on a committee, I can tell you that we appreciated well written papers, and were annoyed with sloppy, obscure, or otherwise badly written papers. We actually ended up rejecting papers that were so badly written we couldn't verify correctness. Of course, if a paper seems correct, and the result is great, it will be accepted, even though the authors didn't figure out the best way to write it. – Dana Moshkovitz Dec 20 '10 at 0:17
  • 4
    It helps to look for published journal versions, but these do not appear for at least a year after the original conference version. Some authors post a "full" version to their website or the ECCC so the full details are known. – Derrick Stolee Dec 20 '10 at 3:13
  • 1
    "....while papers that could be simplified but haven't been are thought to be deep and rated highly because of it".. I am curious if there are nice examples inside TCS of this case? – Brout Oct 12 '14 at 10:01
  • @J.A perhaps post as a question? – usul Oct 13 '14 at 13:25

There is a very large gap between understanding deeply a result (history, motivation, what it implies, etc.) and just applying it (implementation is one of the way of applying a research results) !

This is why it can be hard to understand a research paper, and why an intuitive explanation can give enough to implement...

My only tip is the following. When I was a master student, I start reading research papers, it took me weeks to understand "easy" (in fact old papers, so papers with well known results) papers in details. I spent basically my first year of PhD reading hundreds of papers. And reading papers is still by far the task where I spend most of my time. Now I can more easily understand what a paper is about, and if the paper is about incremental results in a familiar area I have quickly a good comprehension, but It is still a hard task to understand new results. So, my tip is then: read a LOT of papers, and spend a LOT of time on a paper if necessary.

  • 6
    Thank you for your answer, I am in a similar position, the remedy of not being able to understand what I am reading seems like reading even more :) – nimcap Dec 19 '10 at 23:50

I like to give my students Keshav's "How to Read a Paper" (ACM DL)(PDF). He outlines some pretty effective strategies. In general, I'd say practice makes perfect and you just have be very patient with the process. Try different strategies and just keep reading and re-reading the paper until it makes sense. If you have to read and re-read one paragraph for 30 minutes, so be it. Treat it as a non-linear process and don't be afraid to stop and start so you can check some formulas or skip around the paper if necessary. As you get more practice reading the style, reading research papers begins to feel less hard. When people are new to reading research papers I think they confuse hard for different. We assume that previous experience reading translates, but research-style writing is, in my opinion, totally different than any of the styles you've previously encountered. You have to adjust to the style and that takes time and practice.

  • 1
    "How to Read a Paper" is a great suggestion. I'll be sure to read it as soon as I figure ou.. Comment truncated -- Stack overflow. – Mike Samuel Jun 20 '12 at 23:48
  • It's worth mentioning that "How to Read a Paper" shows up on the big-list question about Papers everyone should read. In the context of this question and my response, that list is probably an excellent place to find Research papers with which you can hone your reading skills. Papers Everyone should Read – Logan Mayfield Jun 21 '12 at 12:18
  • Re-reading a paper you're currently capable of understanding to really pin down the details is sensible, but sometimes you just don't have the prerequisite information in either the paper itself or your head. Knowing when to back off and find additional information elsewhere is an important skill. It's a fallacy that if you stare at something incomprehensible long enough you'll get it - that's rarely true. – Stuart Golodetz Jan 12 '14 at 20:02
  • Well. I didn't mean re-read as in "acknowledge a sequence of words over and over". Reading is an active and critical thought process. Hence the non-linearity. If you have to step away to learn some prerequisite material, then do that. So yes, re-reading is useless if all you're doing is acknowledging words, which is not what I was recommending. It is, unfortunately, what I see a fair number of students do when they encounter hard material. – Logan Mayfield Jan 12 '14 at 23:04

I found the following helpful for understanding papers, especially when they are highly specialized or from a different subfield:

Read the introduction very carefully. That is where the main idea is explained, while the rest of the paper proves it. One word can change the meaning of a whole sentence, especially when special vocabulary is used or if, unfortunately, a word has a completely different meaning in another subfield. Therefore it is a good idea to stop whenever you feel that you do not fully understand a sentence and look up the words , as you would do if you were learning a new (human) language. Making notes next to words or even better making a small "dictionary" helps a lot.

Also, it's better to spend some time on the preliminaries, building on the foundations used, especially if you want to work on this paper rather than just understand the result. Make sure you have every necessary tool before reading the proof.

As for the proof, I do not have much advice other than reading it very carefully and make sure you understand every step in the way. It is also a good idea to make a small list of the things that have been proven before the main proof, so that you can see what result is used where, and to have a quick index to look it up if you don't remember it exactly.

The same goes for additional lemmas... if you feel you cannot follow the explanation of a lemma , it's best to read the main proof again.

I think this is a type of "educational snowball effect".

The academics writing the papers have "snowballed" so much information about their subject, that when they write their paper, they speak entirely in terms of their subject. Any given sentence in that paper, however, might make mention of several other research topics that you have no idea about. For example:

In particular, we discuss how the harmonic basis, which is derived analytically from a model of an object, can be used in a linear subspace-based object recognition algorithm, in place of a basis derived by performing SVD on large collections of rendered images.

So every bolded set of words there actually requires a substantial amount of information for you to even know what they're talking about. If you don't have the prereqs to understand the paper, the paper will be opaque to you.

The author assumes you just know, because he doesn't have the space to explain absolutely everything, and advanced readers (his target audience) would find the paper "too basic", or filled with stuff they already know. People in the field want the author to "get to the good bits".

"Tutorials" are all about filling in those gaps that papers tend to leave. This is just the entry barrier research topics have. Once you know a bit about your topic, you'll find it easier to read the papers. Just keep at it.

  • 1
    +1 This is the key point - that the problem in understanding is often due to a lack of background. Writing quality does vary, and people sometimes don't explain things as clearly as they should, but the main issue is that authors are writing for their colleagues in the field, and you have to read your way into a field before you can understand some of the papers in it - they're often building (bit by bit) on 10+ years of research that you don't know about. The trick is to find a shortish route from the bottom up. – Stuart Golodetz Jan 12 '14 at 20:10

The excellent answers so far have left out one reason that I think is critical to both understanding and publishing papers: the reviewers are not really the eventual audience.

A paper's reviewers are generally experts in the sub-field, and experts on the topic if possible. A paper's readership are generally newcomers, or are only passingly familiar with the topic. The reviewers are generally professors. Most readers are graduate students.

This contributes to the unpleasant effects on the clarity of the writing that you've been struggling with.

It also makes publishing a readable paper very difficult. For example, I would love to include a short, targeted tutorial in every paper I publish because very few readers in my field are familiar with my application area. But my reviewers are usually familiar enough with it that they don't need a tutorial, and suggest removing it in favor of expanding the results, regardless of how efficient the paper's presentation is. They just feel like something must be missing if there's room for a tutorial.

That's just the most obvious effect, though.

Normal research papers are written primarily for other researchers, typically people in the same field, who are also at the cutting edge of research themselves. It is not surprising that they are hard to understand for normal professionals.

Some top researchers do know that a wide readership is a prize to be sought and develop skills to write in such a way that they can reach it as well as impress their colleagues. But that is hard to do. A typical research author would write whatever he/she can get accepted by the referees. So they are our prime audience, perhaps not even all the conference attendees. But conference talks are generally more accessible and, these days, we put the slides on the web so that everybody can benefit from them.

Perhaps there is room for a journal (or journals) that invite papers that are understandable to professionals, even if they are rewrites of earlier research papers. It is worth a thought.

  • 3
    or just if we promoted a culture of blogging about your own papers. Then after it is published, your blog can serve as a publicly accessible and interactive (in the comments) tutorial for the paper. – Artem Kaznatcheev Nov 17 '13 at 17:51

Usually the main idea of a paper is simple. However it is often the case that the translation from the simple idea to mathematically rigorous proofs adds quite a bit of minor technical details, which contribute to the difficulty of understanding the paper. If you ask an author to explain their result in person, they are often able to do so pretty efficiently on a black board or even over a quick lunch.

Ideally a paper should explain every mathematical proof in simple and plain English in addition to the formal terms. Alas few researchers put that into practice for several reasons, most notably time constraints and space limitations for the conference version and the fact that our journal papers, if they exist, are typically minor revisions of the conference submission (basically, with the omitted proofs back in the text). There are other factors mentioned by others, for example the unfortunate fact that technically difficult papers look more appealing to some reviewers.

  • 5
    I'm not sure I agree with the premise that "usually the main idea of a paper is simple". – Yuval Filmus Jun 15 '12 at 21:27

Your Answer

 
discard

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.