I recently submitted my first paper as a PhD student (arxiv). It is now under review and will probably stay on the journal editor’s desk for a few months. I did learn a lot about the paper writing process. Later on, I finished a writing course and learned even more about how to write. After that, I can now see the mistakes I made in my first paper. I wish I knew some of these information before. I will get to apply them in my next paper. Here is what I learned from experience and the Coursera course.
We divide the paper writing process into three steps. Prewriting, writing the first draft, and Editing. To give an overview of the time and efforts share, prewriting takes about 70% of time and efforts, writing the first draft takes about 10% and the editing takes about 20%.
In the prewriting step, we simply do two tasks: Gather data and plan the organization. We collect all the main results from our experiments and bits from the references we want to talk about. We decide how we would like to present the data. That is, we decide what sections our paper will have. In each section, we also plan how we will break down the information into paragraphs. A good step we can use to figure out the structure is explaining the paper to a friend or a colleague. This is significantly better than a conference presentation.
Do not merge this with the writing step. This is the common mistake. It becomes too hectic to later on change the structure. The writing process takes longer if we don’t have all the data ready. Besides, we can write better if the entire picture is clear in our head.
Writing the first draft
After we gather all the data and plan the structure of our paper, we start the writing process. This is the easiest part. While writing, we do not worry about the “good writing practices”. Instead, we just produce the first draft as quickly as possible. The idea is to produce a complete (possibly not so good) version of the paper, and then improve it by rigorous editing.
We should follow the following order for writing our first draft.
Tables and figures
The readers often look at the tables and the figures first. They tell the main story. We find the main idea we want to present and try to convey it through a table or a figure (because readers like tables and figures more than the text). Use diagrams to explain the processes. The readers should be able to read the tables and figures without having to read the surrounding text. Only put the relevant information in tables that are absolutely needed to tell the main story (cut the clutter). For example, we should use only one or two decimal places unless we have a good reason to use more.
Summarize at a higher level what’s on the tables and figures. They should not repeat what’s exactly on the tables and figures, but rather complement the information in them. Highlight only the most important numbers. Leave the interpretation of the finding for the discussion section.
The main experiment setup and procedures. Break it down into multiple parts to make it easier to understand. Again, try to use figures to explain the complex parts.
Use about two to five paragraphs for the introduction. The information in the introduction is usually in this order: Background (known information), Knowledge gap (unknown information), Hypothesis or question, Approach or proposed solution, Why our approach is better or new or different. The idea is to start with a broad image and then narrow down to the specific part.
Here we do exactly the reverse of introduction. We start with the specific and then discuss the bigger picture. The order is as follows: Answer the question asked in the introduction. Support the conclusion using justification from the data presented in the results section. Describe the limitations of our methods. Finally, give the big picture take home message.
This is perhaps the easiest part after writing all the above parts. Think of it as a trailer. All we need to do is to pick the highlights from each section and dump it into abstract. It should still have a clear flow for reading.
The last step is editing. Here we go through the paper multiple times and improve it. I discussed some “good writing practices” in an earlier post. A common check-list of editing steps include the following:
Organization review: Check if the sections and paragraphs follow a logical ordering that is easy to read. This is the major part of editing.
Cut the clutter: Remove the extra words, or any other information, without which we can still convey the main idea.
Tense check: Check if we have used the correct tense in each sentence. Use the present tense for things that are still valid. For example, “the data shows…”, or “We observe from the data…”. Use the past tense for the things done in the past. For example, “We computed the average…”.
Voice check: Check if we are using the active voice in our sentences.
Verb check: Underline the main verb of each sentence. See if we can use a better verb. See if we can convert the nouns into verbs.
The editing process can take multiple iterations. I used to think that it is a bad thing. But, it is needed to write a good paper. All writers (even the experienced ones) have to go through multiple editing rounds to produce a good paper.
I hope this was helpful. Here is my other post on writing science effectively.
This trimester, I will be working as a teaching assistant (TA) for the machine learning course at poly. I will be taking refresher tutorials for probability and linear algebra. There are many other things I will be doing as part of my TA job. I will write about this experience somewhere around the end of the course.
Video: When chess is played perfectly, it is a draw! (Don't get offended, this is for fun and learning only)
Quote: “There is no wealth like knowledge, no poverty like ignorance.” — Ali Ibn Abi Talib. Retrieved from the book 'The Unfair Advantage'.