Knights of Academia discussion

Over on the KOA server I promised cs42 I would write something about how to make sense of a paper he had read, and similar papers. I'm going to write here, and then post on KOA if cs42 doesn't want to visit here.

I'm going to write this before I read the article, as a way of testing my advice.

No surprise, my basic moves are going to be those in How to Read a Book.

1. First figure out what kind of paper it is. My idea is that a journal article is probably one of two types: a survey or a research paper.

A survey doesn't claim to include new research -- it can be a literature survey or something like "recent developments in X". But it lays out an overview of a body of existing work.

A research papers does claim to have something new to say.

cs42's sample paper looks to me like a research paper so I'm not going to talk about surveys any more.

2. A research paper is in some broad ways similar to a murder mystery (and to other types of writing). In a murder mystery we know there is going to be -- wait for it -- a murder. Most murder mysteries involve figuring out over time who done it. In addition to the murder itself, there are certain necessary parts to a murder mystery, including some kind of scene in which we learn who done it.

In a similar way, a research paper will have (I claim) a problem and a solution to the problem. The solution is the new content that came from the research. Once you have identified the problem and solution (even if you don't understand them), the rest of the paper should fall into line.

Try to identify the problem and solution before you read the article.

Unlike a murder mystery, the authors are not trying to be mysterious. They will state the solution and problem prominently. The first place to look is the paper's abstract if there is one. This should tell you the problem, the solution, and something about how the authors came up with or tested their solution. All you care about now is identifying the problem and the solution, even if you don't understand them.

In addition to or in place of the abstract, look at the first several paragraphs and the last several paragraphs of the article. The beginning of the article should describe the problem. The end of the article, or the end of the next-to-the-last section of the article should summarize the solution. (The very last section may point to further research that needs to be done on the problem.)

Overall, you should expect to see:

a. Statement of the problem
b. Information necessary to understand the problem: survey of previous research or experience, definition of key terms, etc.
c. Then possibly a more detailed review of previous work or experience.
d. Then likely a detailed description of the researchers' approach and why it is, or might be, a solution.
e. Then a detailed description of the approach including if appropriate a narrative. In a math paper you get the deep math here. If the authors claim to have made something more efficient, you'll get details of what they did and how they tested it.
f. Then a summary statement of the solution and the degree to which it solved the problem that was originally posed.
g. Then possibly words about future research that needs to be done, acknowledgements, etc.

It's quite likely that there will be headings indicating the major sections.

So first you identify the murder. Then you identify other necessary sections, like the background, the circumstances that make the murder hard to solve, etc.

3. In the research article before you begin to read in detail you should be able to identify the problem and solution even if you don't understand them. Highlight those sentences, if possible state them in your own words.

You should be able to identify key terms used in the statement of the problem, the description of the approach, and the summary of the solution. Quite often a research paper will have a list of keywords at the front. Maybe a fifth of those keywords are crucial. Grab that list and keep it as a list of "terms I might need to look up if they do turn out to be important."

4. Now and only now you read in detail. You keep aware of each section and its purpose. Usually the first and last paragraphs of a section state and summarize the material that is in the section. But you are not just reading a long string of words -- you are reading sections, each of which has a necessary purpose. It's like a GPS. "Now I'm the background section." "Now I'm in the narrative of the work they did." Etc.

You may actually read the entire paper and not understand the problem and solution. But you know what they are, as the authors stated them. You've identified key terms. You have precise things to look up or work on, rather than just "I don't know what they're talking about." Now you look up stuff, selectively. You revisit parts of the article selectively. You follow references (footnotes) selectively.

Your overall goal is to be able to state in your own words what the problem was, how these authors approached solving it, what their solution was, and to what degree it solved the problem.

One addition that has occurred to me: Research papers describe a problem and a solution. They could be theoretical (entirely math, or information theory) or empirical.

Support for a theoretical paper will be math or logic or suchlike.

Support for an empirical paper will include actual testing, but is also likely to have a theoretical or architectural component that may be supported by non-empirical methods.

In the paper that cs42 provided as an example, the section labeled "Introduction" was very valuable and had the best statement of the innovation that the authors had done. The method as stated above didn't mention the introduction separately. The example paper had a very useful one.

JFYI: Over on the KOA server the discussion moved over to the #computer-science channel.