How to Organize Your Writing

Write first, then do research

The primary objective of our work is to do great research. Great research is achieved by developing novel ideas and communicating these ideas to the research community. The best way to shape an idea and present it is to write a paper. Often people believe researchers spend most of the time planning experiments, implementing software, discussing ideas, and only write the actual paper the last two weeks before a conference deadline. Simon Peyton Jones (and me) believes this approach is broken. Doing unfocused research is often not very productive. Instead, one should start from a research idea written down as a paper draft -- title and abstract are sufficient -- and then focus the research work on whatever is needed to support the story of the paper. Experimental research (implementation, experiments, performance tuning) is hard work. The easiest way to save a lot of time is to not do experimental research at all. By having paper-writing guide us we can avoid any work that won't end up in the final paper. In addition, research is a highly collaborative task where communication is key. Meetings and whiteboard discussions are a great tool, but to become precise it is important to write. By having a paper draft early on, you have 6-9 months to get feedback from your collaborators on your ideas. Paper-writing is an important communication tool.

How to write a great research paper, Simon Peyton Jones, PDF

Present your idea, not your prototype

The most important objective of a research paper is to propose an idea. Hence, when writing a research paper, start by writing down your vision and deliberately move limitations of your research prototype to todo notes in the margin of your 'dreamed-up' research paper.

Having a written high-level idea:

• ensures a good and strong story for your paper
• identifies missing parts in your formalization or implementation
• identifies missing experiments

❗ Not every piece of your vision must be implemented. It is perfectly fine to state in your paper that your prototype does not implement 100% of your vision. In fact, it is never possible to implement a full vision as a product. However, this does not mean you need to keep your vision private. State it fully and then show by implementing and evaluating the critical parts that your vision is feasible.❗

❗ If you don't have an evaluation yet, draw the graphs you want to have on a piece of paper, take a picture with your phone, and inline the picture in your paper. This makes it easy to communicate what you expect your results to look like. ❗

⚖️ Academic honesty: Always state clearly the limitations of your prototype. However, don't make discussing prototype limitations the objective of your research paper. First bring forward your idea, then discuss to which degree your prototype supports this idea.

Don't write perfect text, write something and make it better

Good scientific publications are very well written and authors often struggle a lot (and loose a lot of time) with writing good scientific texts. The key secret here is: humans are really bad in coming up with something good, but humans are really good in identifying errors. Hence, come up with a list of bullet-points, translate them into a simple text, and start improving this text. Many good authors follow this very same style of writing and are are really good in it. But good means here not that they come up with the perfect first draft, rather good authors are very efficient in correcting errors. As a result they can correct their paper 20-50 times over the 6-9 months of writing it. Compared to the 2-3 times an average bachelor thesis is proof-read, no wonder they writing quality of scientific publications is a lot better. The 20-50 times a paper is proof-read is not wasted time. Instead, these reads come for free by using paper writing as a tool of scientific collaboration.

❗Make sure even your bachelor thesis has been proof read 20-30 times as well. ❗

Abstract

An abstract should consist of six main sentences:

1. Introduction. In one sentence, what’s the topic?
2. State the problem you tackle.
3. Summarize (in one sentence) why nobody else has adequately answered the research question yet.
4. Explain, in one sentence, how you tackled the research question.
5. In one sentence, how did you go about doing the research that follows from your big idea.
6. As a single sentence, what’s the key impact of your research?

Furthermore:

• Format an abstract as a single paragraph.
• Highlight your core experimental results (speedups) at the end.
• It is OK to write a longer abstract (using 2 instead of one sentence for the six points). While this makes the abstract less concise, it allows us to get more information into search engines, overview pages at ACM, and others. As for most papers only the abstracts are read this is very helpful in terms of science communication.

How to write a scientific abstract

Introduction

Provide an intuition -- don't try to be precise

The introduction (and abstract) in a scientific paper should provide a high-level intuition of the idea put forward in a paper. While scientific papers in general must be very precise (all terms must be defined) and detailed, when communicating an intuition we deliberately ignore details that distract from the high-level idea. This does not mean we can be incorrect, but it is OK to use terms that have not been well defined or to just forward-reference to later definitions as long as the high-level idea of a term remains clear to the reader.

Provide teaser data

People love data, hence the best way to motivate your work is to show actual data and use it to motivate your research work.

Explicitly list your contributions

Provide an explicit list of contributions at the end of your introduction. This list is what you claim to be novel in your paper, making it worth publishing. Being explicit about what you believe is new is very important. The remaining part of your paper is expected to provide evidence for your claimed contributions. Hence, it should be super clear which part of your paper makes which contribution (e.g., forward reference).

Don't describe 'the remainder of the paper'

Don't discuss the section structure of your paper. Many introductions end with a paragraph which gives an overview over the structure of the remaining paper. Don't do this. It is a lot better to reference the important sections from within the introduction -- there is no point in explaining that you have a background and / or conclusion section.

Content

Source: Kathryn S McKinley, Writing papers that get accepted to conferences

Paper = Recursively Nested Containers

A paper consists of nested containers:

• A paper contains sections.
• A section contains paragraphs.
• A paragraph contains sentences.

Intro → Discussion

Each container follows the same structure:

• Intro of new idea
• Discussion of idea

Paper

• The introduction section introduces the key idea.
• Subsequent sections provide a more detailed discussion of the key idea.

Sections

• First paragraph discusses the central topic of the section.
• Subsequent paragraphs provide a more detailed discussion of this topic.

Paragraphs

• First sentence states the single key message of the paragraph.
• Subsequent sentences support this message.

Sentences

• Each sentence states a single piece of information.

Consistent Storyline Across Levels

A well-written paper has consistent storylines across abstraction levels.

• Does only reading the section titles make a consistent story?

• Does only reading the first paragraph of each section make a consistent story?

• Does only reading the first sentence of each paragraph make a consistent story?

Why is a Recursive Structure important?

The human mind prefers a recursive structure

Even though the arguments that follow are only conjectures, thinking about how the human mind reacts to (un)structured arguments allows us to tell a nice story that may help our memory:

• The working memory of humans is very small

  Humans can keep less than 7 abstract objects in their immediate working memory. If understanding an argument requires the reader to remember a point that was made more than 7 "information pieces" ago, the reader needs to stop and re-read.

  A good recursive structure avoids long backward references. References are either local within a single paragraph or point to earlier paragraphs. Within a paragraph we typically do not have a large number of sentences. Earlier paragraphs can be remembered as a single object -- the key message of the earlier paragraph.

• The 'idea -> discussion' structure is easy to verify

  When a scientist reads a paper he continuously (and unconsciously) checks each sentence to understand if the sentence is correct and makes sense.

  Without a good paper structure, it is unclear which property should be verified. Hence, the reader scans his long-term memory for associations that either contradict the current sentence or place it into context. This process (a) slows down the reader, (b) makes him feel the paper is difficult to understand, and (c) increases the chance he finds almost random arguments to criticize the paper.

  With a good paper structure, each paragraph begins with a single argument. All subsequent sentences provide further details in support of this argument. The verification task is consequently very simple. The argument has been clearly stated at a predictable location that is in close temporal proximity. The reader now only has to verify that the 5-7 sentences that follow support the argument. If they do, the reader can forget about the 5-7 sentences and remember the argument as "valid". If the paper structure is consistent, a small positive validation event occurs at the end of each paragraph, always slightly increasing the satisfaction of the reader

Your audience does not have time!

• 90% only read the title
• 9% read only the abstract and the section titles
• 0.9% read only the first paragraphs and the first sentences (skim the paper)
• 0.09% actually read most of your paper

Even people who read your full paper typically start by skimming and only decide to read your paper fully after a positive first impression.
→ If you want that people read your work, make sure they understand it before they start reading in detail.

Copy Editing

Spell checked

Spell-checked using american-english. No mixed american/british english.

Agree on comma setting style.

Always use Oxford Comma.

Section titles use uppercase consistently?

• Section titles use uppercase consistently?

Literature

• On Writing Well, William Zinsser -- Probably the book to read on technical writing. Even the first chapters will be very helpful.

• Style, Joseph M. Williams