How to write a readable paper?
Maybe many people will find this a well-understood problem. After all, there are numerous books and guidelines on this topic, and it isn't all that hard to follow their advice. However, my experience after reading so many papers is that a lot of papers, including those by experts, are terribly hard to read, even though they may be technically interesting and on the surface they may look quite professionally written: good grammar and spelling, correct equations, clear figures, and a structure that follows all those "best practices" precisely. Honestly, grammar, spelling or structure problems would never waste as much of the readers' time as being so hard to understand. Very often I spend hours going through the strange notation and complicated equations in a 20-page-long paper, and when I finally finish it (getting very sleepy in the process) the paper seems to offer no useful idea at all.
Some problems are inevitable, of course, simply due to the difficulty in catering to such a diverse audience: those new to the field, those who have some knowledge on the field in general but still unfamiliar with the topic, those who are familiar with the topic and want to extract the new idea out of the paper in the shortest time possible, and of course, picky reviewers that may (sometimes) know more about the field than the author. This reminds me of one of the papers I have been reading, about the difficulties in efficiently multicasting information to users with different channel qualities, and for which a joint source-channel coding scheme is proposed. For experts with little time, a few English sentences on the basic idea, possibly accompanied by a figure if it is really instructive, might be the best; yet for those new to the field, such acronym- and citation-filled English would rarely provide much insight, and equations with detailed steps of derivation, similar to those in most textbooks, are a must.
Yet, I don't believe the readers are the only ones to blame, just as it is suboptimal to optimize only the receiver in a communication system without looking at the transmitter. There are IMHO two objectives in writing a good paper: one is to be fun to read and easy to understand by newcomers to the field, and the other is to be quickly skimmable by an expert so that he/she can get the basic idea in a matter of minutes. Any paper should do one of these well--I believe it would also be possible to achieve both, given sufficient care--but a lot of papers achieve neither.
IMHO the most annoying problems are: long and hard-to-understand introductions, uninformative equations, and uninformative figures.
Most papers include in the introductory section a survey of the related work on the field, however I find such organization quite questionable. When I am not extremely familiar with the topic, such jargon-filled introductions, usually without any equations or other explanatory devices, are hardly understandable (though all these will often appear quite simple when I read more into the paper). It is often hard to skip such things, since this is, after all, an "Introduction" section and important ideas may be lurking anywhere. Maybe it would be better to put most of the related work after some detailed explanations, but I'm still not quite sure about the best organization.
Equations are another common problem. Of course, in my field, equations are practically a must, or otherwise with all the abstract English the paper would likely be only readable by the experts among experts. However, a problem with equations is that they are very verbose and take a long time to understand, by beginners or experts alike (experts may be able to save some time if the notation is familiar, by skipping familiar equations recognized by pattern-matching, but this is not always safe), although unlike English most equations can be understood given enough time. One particularly annoying thing about equations is that many of them have a simple form (e.g. the Bayes formula), yet it takes a lot of time to figure out what all the notation in them actually means, or whether they mean something important or just meaningless clutter. The authors are not totally to blame here--sometimes it is just the poor scoping rules in mathematics, in which you cannot introduce local variables very easily without introducing possible confusion. Still, trying to make sense of all the equations that may mean something or not (most do not) slows down paper-reading tremendously.
The same applies to figures. They should make things clearer, but very often the figures take more efforts to understand than reading the English!
Some problems are inevitable, of course, simply due to the difficulty in catering to such a diverse audience: those new to the field, those who have some knowledge on the field in general but still unfamiliar with the topic, those who are familiar with the topic and want to extract the new idea out of the paper in the shortest time possible, and of course, picky reviewers that may (sometimes) know more about the field than the author. This reminds me of one of the papers I have been reading, about the difficulties in efficiently multicasting information to users with different channel qualities, and for which a joint source-channel coding scheme is proposed. For experts with little time, a few English sentences on the basic idea, possibly accompanied by a figure if it is really instructive, might be the best; yet for those new to the field, such acronym- and citation-filled English would rarely provide much insight, and equations with detailed steps of derivation, similar to those in most textbooks, are a must.
Yet, I don't believe the readers are the only ones to blame, just as it is suboptimal to optimize only the receiver in a communication system without looking at the transmitter. There are IMHO two objectives in writing a good paper: one is to be fun to read and easy to understand by newcomers to the field, and the other is to be quickly skimmable by an expert so that he/she can get the basic idea in a matter of minutes. Any paper should do one of these well--I believe it would also be possible to achieve both, given sufficient care--but a lot of papers achieve neither.
IMHO the most annoying problems are: long and hard-to-understand introductions, uninformative equations, and uninformative figures.
Most papers include in the introductory section a survey of the related work on the field, however I find such organization quite questionable. When I am not extremely familiar with the topic, such jargon-filled introductions, usually without any equations or other explanatory devices, are hardly understandable (though all these will often appear quite simple when I read more into the paper). It is often hard to skip such things, since this is, after all, an "Introduction" section and important ideas may be lurking anywhere. Maybe it would be better to put most of the related work after some detailed explanations, but I'm still not quite sure about the best organization.
Equations are another common problem. Of course, in my field, equations are practically a must, or otherwise with all the abstract English the paper would likely be only readable by the experts among experts. However, a problem with equations is that they are very verbose and take a long time to understand, by beginners or experts alike (experts may be able to save some time if the notation is familiar, by skipping familiar equations recognized by pattern-matching, but this is not always safe), although unlike English most equations can be understood given enough time. One particularly annoying thing about equations is that many of them have a simple form (e.g. the Bayes formula), yet it takes a lot of time to figure out what all the notation in them actually means, or whether they mean something important or just meaningless clutter. The authors are not totally to blame here--sometimes it is just the poor scoping rules in mathematics, in which you cannot introduce local variables very easily without introducing possible confusion. Still, trying to make sense of all the equations that may mean something or not (most do not) slows down paper-reading tremendously.
The same applies to figures. They should make things clearer, but very often the figures take more efforts to understand than reading the English!