Sydney Dolan

When I started graduate school, my understanding of scientific writing was largely based off the lab reports we had to write in my high school chemistry class. Unfortunately for my past self, academic writing is very different from high school scientific method. It is also different from the assigned textbook readings in college courses. The purpose of this post is to serve as an extensive guide for academic paper writing. I will cover conventionally expected elements, and more abstract parts like word choice, tone, and flow. This blogpost is for anyone new to academic writing and wanting to learn more about some of this stuff explicitly.

Parts of an Academic Paper

Academic papers typically have the following elements:  

• introduction 
• literature review
• methods
• results
• conclusion

In the sections below, I’ll provide a definition of each element, and what is conventionally expected.

Introduction

Your introduction serves to familiarize the reader with the topic you work on, the current gap in the field, and what you specifically have done.  Readers look for the following elements in an introduction: a broad problem description, a specific problem description, a clear description of the specific contributions of the paper, and a paragraph explaining the structure of the rest of the document.

The contributions of the paper should be written in a detailed, and should focus on the answer to the research question that you posed when you began the work. When describing your contributions, you should be writing in a factual tone, that realistically and accurately describes what you have achieved in your work. Optimistic or ‘sales’ type language, where you talk about how your work is the first of its kind and a massive step forward for the entire field, is not appropriate (more about appropriate tone will come later).

The introduction is the most accessible section of the document, meaning the first few paragraphs of the introduction can be written with a more broad audience in mind. An undergraduate in your field should be able to read your introduction and understand the broader problem that your work is addressing. In a well-written introduction, readers should be able to connect with the fundamental motivation of the research, even if they may not understand the technique.  

When writing introductions, one technique that I would recommend is the hourglass model. The hour glass, shown in the figure below is a tool used to structure introductions so that you craft your work into a compelling story. It makes your work more accessible by placing the emphasis on the motivation of the work and its implications.

The width of an hourglass represents the generality of the statements within. When you are establishing the significance of your work, this is meant to be parsable by a broad audience, whereas when you are discussing what you did in specific, the work should be much more technical, hence the pinch in the hourglass. Table 1 below contains an example of the difference in hourglass models depending on intended audience.

Hourglass stage	High School Talk	AI in Astrodynamics Journal
Establish significance	Craters on asteroids can give evidence from billions of years ago on how the solar system was formed. Due to their incredible speeds and distance locations, asteroids will require autonomous systems for approach and landing.	Due to their incredible speeds and distance locations, asteroids will require autonomous systems for approach and landing.
Describe the status quo	Current control techniques require an accurate model of the environment to control the spacecraft	Optimal control techniques provide guarantees surrounding satellite behavior, but require an apriori model of the dynamics.
Identify a gap	Reinforcement Learning (RL) has shown promise for this application due to its ability to learn control policies for a wide range of scenarios.	While model-free reinforcement learning (RL) allows us to develop control strategies for systems that are too complicated to model, the results are not guaranteed to follow basic laws of physics.
What did you do?	We apply reinforcement learning to the asteroid landing problem.	We apply a standard actor-critic reinforcement learning algorithm in conjunction with a physics informed neural network (PINN) to constrain the control strategies of our work. We use mean pooling and multi-head attention to prioritize strategies that maintain safety throughout.
Fill the gap	We found that RL is able to create more robust control policies than current methods.	We found that our technique is able to create more robust control policies than other model-free RL methods
Re-establish significance	Through our work, we have demonstrated the effectiveness of RL as a tool for fully autonomous asteroid landing.	Through our work, we have demonstrated the effectiveness of our joint PINN RL framework as a tool for fully autonomous asteroid landing.

For the more general high school audience, the hourglass model contains less details about the specific implementation (e.g. the ‘what did i do’ section). Additionally, the significance section is longer, as the audience may be unfamiliar with scientific value of visiting asteroids, and thus additional sentences should be written to provide the reader this context.

In contrast, the journal hourglass model contains more details about the specific implementation used in the paper. The gap that the work is addressing is described in much greater detail, and specific problem within the field of RL is mentioned rather than just mentioning RL as a tool.

The hourglass tool serves as an effective way of ensuring that you are fundamentally answering the motivations of your work, what you did, and the significance. While introductions are obviously longer than the sentences listed in the table above, and require additional details to link each element, the hourglass model can help you ensure that your introduction remains narratively strong.

Literature Review

The literature review adds more technical context to the specific research gap you are addressing in your work. The literature should contain field-specific information, and will not be accessible to general audiences.

In an effective literature review, you should be very deliberate and focused about the sources that you are including. A literature review is not a list of all papers you’ve read on a subject. It is only the most relevant papers needed to explain the necessary background information on your work, establish the current state-of-the-art, and imply a research gap exists.

When you do interdisciplinary research (ex. space and machine learning), you will need to think about the background of reader and tailor the level of detail in your literature review appropriately. For example, when I submitted to the Reinforcement Learning Conference, I change my literature review section to involve a lot more supplementary detail and explanation of more basic parts about space, as it is not expected the readers will know much about satellites.  In the sections about reinforcement learning, I kept those very technical and detailed, as that was the existing background of my audience.

Another factor to consider when writing your literature review is that the literature review and introduction are complimentary elements within a paper. In the introduction, you are setting up the problem, and in the literature review you get to explain the extent of the problem

Methods

The methodology section is where you explain your specific technique. There is a lot of variance in how methodology sections are written, and it is very field dependent. You should take note of how other papers in your field are structures and use that to determine the specific standard for your sub-discipline.

Typically, methods section start out with something that I would call the ‘gimme paragraph’. This is the paragraph that shows up in every single paper on the topic, where you described the notation or key assumptions that you will be using throughout. For example, I work in reinforcement learning, and oftentimes methods sections will start out with a description of a Markov Decision Process, and the key elements. This ‘gimme paragraph’ shows up all the time in papers, and it is fine if the way you describe the notation for your paper is very similar to others. The purpose of these paragraphs are to concretely layout any mathematical notation or key assumptions necessary to understand the rest of the method.

Following the ‘gimme paragraph,’ you will then need to explain your method in depth. In a well-written methods section, a person who is unaffiliated with your research project should be able to follow your instructions and reproduce your work. This means that you should be explaining both the theoretical approach and any simulation or experimental environment used to generate results.

If your project involves high power computing, you should mention what computing resources and platforms were used. If you are working in machine learning (LLMs, RL), you should include information about the weights you used, training datasets, and key training parameters.

In many cases, documenting your methodology thoroughly will exceed the space available in the main body of the paper. This is where appendices can be a super useful tool. For example, you can include more granular details such as hyperparameter settings, dataset preparation steps, or specific benchmarking protocols in a supplementary appendix. Similarly, if your results section includes comparisons with other methods, the setup and tuning of those comparison methods can also be documented in the appendices. Other common appendix materials include links to open-source code or full mathematical proofs that support your work. if it is not important to the reader’s understanding and intuition of your technique and the main contributions of the paper, then it is perfect for the appendix.

The methods section is not a literature review, although it may refer briefly to existing work if your method is building directly upon it. Literature reviews survey what others have done, the method section is focused on your specific approach. This section is meant to answer questions like, “what did you do” “how did you do it” and “why did you do it that way”. The level of detail you include depends heavily on the audience.

If you are writing for a specialized audience, like a conference where everyone else uses the same technique, you can spend more time writing about the specific implementation of your method and how your method deviates from existing techniques. But if you are writing for a more general audience, you should take more time to explain the theory behind the method, and help the reader build intuition.

So in general when writing a methods section, try to be as precise and transparent as possible. Remember that not everything has to be in the main text, and that appendices can offload less essential details to keep the main narrative of the text focused and digestible.

Results

In my opinion, results sections are the most fun to write, because that’s where you get to ‘flex on em’ [the reader]. A results section should clearly and concisely present key findings of your work. To do this, you should first be explaining the metrics that you used to evaluate your technique. Then, you should briefly introduce any comparisons to baseline models or control groups, and why they are relevant for your work.

Once you’ve established how your work is being evaluated, then you can explain your results. Figures and tables are the bread and butter of a good results section. When presenting tables and figures in a results section, think deeply on the clarity, relevance and overall integration with the text. Each table or figure should be numbered and include a descriptive caption. In academic writing, figures and tables should not just be dumped in without any explanation. They should be referred to explicitly in text, and the key takeaway of each figure should be described. Avoid saying very obvious things in the figure and in text, instead use your intext citation to highlight trends or comparisons. Most importantly, make sure each figure or table supports a specific claim relevant to the contributions that you mentioned in your introduction section. In this way, academic writing can be a very recursive process where you need to go back through different sections to ensure that you are narratively consistent throughout the paper.

Tips for Good Figure and Table Design

Obviously there are lots of figures that can exist. I’ll save figure design for a separate blogpost, because I also have a lot to say about effective figures. Below are some best practices. :

• Make the numbers and the lines on the figure large enough. Ensure that the linewidth, x-tick marks, y-tick marks, x-label and y-label, and title, are of sufficient font size to actually be readable. In general, fonts should be size 18 or larger. The default linewidth size in python is not thick enough to clearly read the line, you should increase it.
• Be intentional about color schemes. Use tools like Coloring for Colorblindness. which can check that colors for different elements in your figure have good contrast. Also consider that your figures could be printed in black and white, so secondary pattern elements (i.e. dashed lines) can be helpful for differentiating between colors. Your figures should work regardless of if the person is colorblind, or if the figure gets printed in black and white.
• Simplify as much you can. In a well-designed figure, a stranger should be able to review the figures in your document, without the text, and still understand your main takeaway.

Conclusion

Your conclusion serves to summarize your paper by reiterating your specific contributions, highlighting the main findings, and reinforcing the significance of your work. It is a nice neat little way to tell the readers about the problem you addressed, and core results.

However, one tendency I see among new writers is the tendency to use the conclusion to ‘sell’ the paper or introduce new ideas about the implications of the results. Do not introduce new ideas about your results in your conclusion. Do not save anything interesting or new for your conclusion. It should all be explicitly written already in your paper. The conclusion is a summary, it is not a eleventh hour plot twist to keep the reader invested.

If there is to be anything new in your conclusion at all, then it can be a short discussion on future work and weaknesses of the technique. For the weaknesses, ideally these should be touched upon already in the results section, but can be reiterated once again in the conclusion. 

Acknowledgements

It is often expected to have an acknowledgements section, where you should include information about the financial sources that contributed to the work, and any people you need to credit for key technical discussions. Depending on the document and formatting guidelines, this acknowledgement may be located at the very front of the thesis, or it may be at the end of the end. You should refer to the style guidelines for your specific document to determine the location.

The acknowledgements section in a technical paper serves to recognize individuals or organizations that contributed to the research but do not meet the criteria for authorship. This includes those who provided intellectual input, technical assistance, or financial support. The purpose of this section is to properly credit people who helped in the research process, and to maintain transparency regarding the contributions of the work. For an academic paper (talk, poster, conference, journal), it is not conventionally appropriate to mention friends or family in the acknowledgements. 

Below contains an example acknowledgements paragraph:

The authors would like to thank the MIT SuperCloud [36] and the Lincoln Laboratory Supercomputing Center for providing high-performance computing resources that have contributed to the research results reported in this paper.
This work was supported in part by NASA under grant #80NSSC23M0220 and the University Leadership Initiative (grant #80NSSC20M0163), but this article solely reflects the opinions and conclusions of its authors and not any NASA
entity. The research was sponsored by the Department of the Air Force Artificial Intelligence Accelerator and was accomplished under Cooperative Agreement Number FA8750-19-2-1000. The views and conclusions contained in this
document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the United States Air Force or the U.S. Government. The U.S. Government is authorized to reproduce and
distribute reprints for Government purposes notwithstanding any copyright notion herein. Sydney Dolan was supported in part by the National Science Foundation Graduate Research Fellowship under Grant No. 1650114. J. Aloor was also supported in part by a Mathworks Fellowship.

Key points to consider about this example

• The authors would like to thank the MIT SuperCloud [36] and the Lincoln Laboratory Supercomputing Center for providing high-performance computing resources that have contributed to the research results reported in this paper.
  • If you are using supercomputing resources to generate the results in your paper, the supercomputing facilities should be mentioned in the acknowledgements. Depending on the supercomputing facility, some facilities have a specific article or webpage that they would like you to formally cite in text when you are writing the acknowledgement.
• This work was supported in part by NASA…
  • You should cite specific grant numbers affiliated with the funding. Even if the funding only covers part of the work, you should mention partial funding sources, and highlight that they are partial sources. If you are on a paper and one author has funding, their funding source and specific grant number should be included if the paper is submitted as part of the credit for their grant source. 
• The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the United States Air Force or the U.S. Government.
  • Depending on your funding source (particularly if it is a government source), you may need to include text that mentions the copyright distribution rights of the work, as well as if the work is reflective of the opinions of the government. You should refer to your grant source to determine if these source of sentences are necessary. 

Even if you are not funded on an external grant, it is still important to include an acknowledgements paragraph. One example for this funding situation is below. 

The authors would like to thank Dr. Angela Crews for her insight and work to calibrate and validate microwave radiometer data, Dr. Benjamin Johnson for sharing his expertise in the CRTM, and to Victor Qin and Shashank A. Deshpande for technical discussions and feedback. The authors also would like to thank the reviewers for their time, attention, and feedback. Their constructive comments helped us to improve the quality of the article. The results contain modified Copernicus Climate Change Service information 2020. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains. Approved for public release. Distribution is unlimited. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Aeronautics and Space Administration. 

Key points to consider about this example

• The authors also would like to thank the reviewers for their time, attention, and feedback. Their constructive comments helped us to improve the quality of the article.
  • When you are at the end of a peer review (i.e. the peer is accepted) this sort of language is also common in the acknowledgements
• The results contain modified Copernicus Climate Change Service information 2020. Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains
  • Certain datasets ask you to state boilerplate phrases in the acknowledgements about how you used the data, and who has responsibility for the research findings. Review the webpages or sources where you have gotten data, to ensure that you are respecting their requests about how the data is represented in your research. 

If you are writing your PhD thesis or masters thesis, the acknowledgement page can have a more informal tone, and mention family members, friends, or other people in your life who have supported you emotionally. Unlike acknowledgements in journal papers/conferences, these sorts of acknowledgements can be pages long, and can even include inspirational quotes or psalms if that is of interest to you. This sort of tone (and length), is not appropriate for any other venue. While the tone in a thesis acknowledgements is noticeably lighter and more emotional,  it is not appropriate to discredit, make passive aggressive comments, or complain about any of your collaborators or institution.

Topic Sentences, Transition Sentences, Signposting, and Other Subtle Cues to Make your Writing Easier to Read

Topic Sentences

A topic sentence is a sentence within a paragraph that states the main thought. It is often placed at the very beginning. A well-organized paragraph supports or develops a single controlling idea, which is expressed in a sentence called the topic sentence. A topic sentence has several important functions, (1) it unifies the content of a paragraph and directs the order of the sentences (2) it advises the reader of the subject to be discussed and how the paragraph will discuss it. 

Topic sentences can help you create a roadmap for the reader, setting expectations for what exactly the rest of the paragraph will be about. Strong topic sentences make it easier for readers to follow an argument or narrative, especially in longer or more complex texts. Topic sentences are helpful for writers as well, as it can help you organize your thoughts logically and avoid getting off topic.

As you are writing and self-editing, review your topic sentences and see if the rest of the paragraph you wrote follows the same logical argument. By reviewing your topic sentences, you can check the overall flow and coherence of your writing.

Transition Sentences

Transition sentences are used to express transitions between thoughts that link them together. Unlike sign posts (read next section), which are key identifiable words readers use to recognize the structure of a sentence, transition sentences are entire sentences meant to link ideas. 

Transition sentences are important because they help the reader follow the flow of ideas in a clear way. Without effective transitions, writing can feel choppy and disjointed, which will make it hard for the reader to understand how one idea relates to the next. A strong transition sentence will connect what has already been discussed to what is coming next, which helps the reader follow the narrative more smoothly and enhances the overall readability.

As you are writing (introductions, literature reviews, any of the elements of a paper discussed above), review the final sentence of your paragraph. Does it neatly set-up the next topic sentence? Are you able to find transition sentences between different ideas throughout your writing? One useful exercise is to highlight each distinct idea in a different color, and then examine the transitions or signposts you’ve used to connect them. This editing technique can help you assess how clearly your ideas are flowing and how well you are guiding the reader through the argument.

Signposting

Transition sentences typically connect larger ideas between paragraphs, whereas signposts are shorter phrases or keywords that articulate the structure of a piece of writing. Signposting is a useful flag for readers to indicate the most important parts of an argument, signal transitions, and clarify stakes of an argument.

The idea behind using a signpost is that you’re telling the reader where they are going. In a well written paper, you want your reader to naturally have the intuition or thoughts about the next ideas you will explain. I would argue that signposting falls into these formats:

1. Overview signposting – Used to introduce the structure of a very long document (i.e. thesis or journal paper). This sort of signpost typically follows the form, “In this chapter, I do this“. It is an explicit way to tell your reader about the structure and overview of the document so that they don’t have to read or scan headings to track what the paper will be about.
2. Relational signposting – Used to show that you are constructing logical steps in your argument. When you’re articulating complex ideas, relational sign posts link complex ideas together and improve the readability of the individual sentences. 
   
   Examples of relational signposting
   To add onto a point/to build on a previous point – Additionally, Another
   To say a point is a further example of the previous – similarly
   To prepare the reader for an example – For example, for instance, to illustrate, in particular
   To tell the reader that a point is in opposition to the previous – in contrast, in comparison, rather
   To tell the reader that this point is a result of the previous – Therefore, consequently, accordingly, thus, as a result, subsequently
   Prepares the reader for a summary of previous points – In summary, Overall, altogether, in brief, in short

You might have noticed that a lot of the words that I am listing as signposts are also transition words that you were taught in school when you were learning how to write multi-paragraph essays. Transition words, by definition, are classical examples of signposts. However, the significance of using them properly cannot be overstated. Excess signposting creates unnecessary wordiness and can give the impression that you don’t trust the readers ability to follow your argument or that you’re using signposts to compensate for a poorly articulated argument. Signposting should be used sparingly, and as a way to enhance the sentence structure of an argument. Not every sentence in your paragraph needs to start with a signpost or a transition phrase. Transition words should be used to match the logic of the relationship you are emphasizing or connection you are making. Graduate-level academic writing requires you to write at a more advanced level that requires more careful and intentional selection of transition phrases. Simply relying on transition words is not sufficient to show how a paragraph builds to form one coherent argument. Oftentimes, it is much more effective to actually write out the idea that links two sentences than to have transition word.

Miscellaneous Points on Readability

Topic sentences and transitions are the most significant way you can improve readability and flow in your writing. Below are a few additional writing and self-editing techniques that I use to improve the readability of my own writing.

Parallel Structures – Parallel structures are created by constructing two or more phrases or sentences that have the same grammatical structure and use the same parts of speech. By creating parallel structures, you make your sentences clearer and easier to read. In addition, repeating a pattern in a series of consecutive sentences helps your reader see the connections between ideas. 

Overcooked Word Check – Overreliance on certain adjectives can decrease the effectiveness of your writing. If every single result is described as significant or essential, then it gives the reader the sense that none of the results are significant. You should use control-f throughout your document to sanity check how frequently you are using certain adjectives to determine if each are really justifiable.

Using the Appropriate Tone in Academic Writing

Tone is the general character or attitude of a work, and it is dependent on word choice and structure. In academic writing, particularly for researchers, maintaining objectivity is crucial. As researchers, our goal is to advance knowledge through unbiased analysis. In academic contexts, readers are expecting logical, evidence-based arguments, with critical but respectful discussion of previous research. Therefore, word choice and objective phrasing in sentences is very important.

Words that reveal a subjective attitude can impose the writer’s perspective on the reader, preventing them from forming their own conclusion on the evidence presented. Usually, these words can be omitted without taking away from the substance of the sentence. For example, phrases like “ground breaking” or “revolutionary” sound hyperbolic and inflated, which is inappropriate for academic writing. Instead of using such words, it is better to frame the sentence more neutrally, such as: “notable” or “relevant”. This approach keeps the focus on the content rather than a personal perspective. Below I’ve added some common words used in academic writing cases. This is not mean to be a comprehensive list, nor is it meant to be absolute rules, its purely meant to be a reference.

Case	Good Word Choice Examples	Unadvisable Word Choice Examples
When you need to describe something as good	notable, relevant, paradigm shift	revolutionary, groundbreaking
When you need to describe something as bad	poor, inefficient	horrific, terrible, incomprehensible
When you want to discuss implications of work but not something you’ve formally analyzed or have results for	suggest, offer, provide	demonstrate, prove

Another common issue students have with writing in an academic tone is the tendency to use awkward or overly formal phrasing in an attempt to seem more objective. This often leads to the use of passive voice and unnatural sentence structures with unclear or abstract subjects.

Passive voice occurs when the subject of the sentence is acted upon, rather than performing the action. For example, instead of saying “The researchers conducted the action”, a passive variant would be “The experiment was conducted by the researchers.” While passive can be appropriate in certain cases, overusing it can make the writing feel vague or unnecessarily complex.

A common example of this issue is the use of phrases like ” One must understand the brightness of the object as function of its luminosity.” While the use of first person (“I vs We”) is an active area of debate, using “one” as an alternative is generally considered unnatural and unsuitable. A clearer and more natural revision in active voice would be, “To understand the brightness of the object, we must consider its luminosity”. Writing in active voice makes the sentences more direct and readable. It allows you to maintain a professional tone, without sounding distant or artificial

The next issue I’ve seen with writing in an academic tone is the tendency to slip into overly informal, list-like writing. Avoid phrases like “First, I did XX. Then I did YY.” Academic writing is not a personal narrative, so it is not necessary to recount every step you took. As mentioned earlier in the topic sentences/transition sentence section, each paragraph should present a clear claim and develop a logical argument around it. Merely listing things chronologically, particularly in first person, weakens the structure of your writing.

Summary/Final Thoughts

I talked about a LOT in this guide, and it’s probably a lot to process. Academic writing is a broad and nuanced field, and this guide was meant to serve as a helpful introduction and reference point for anyone working to improve their writing. Hopefully, you found some practical tips, and came away with a more clear understanding of what makes academic writing effective.

Additional Resources

• Parts of a Scientific Paper – A poster I made  explaining parts of scientific papers
• Structuring a Paper and the Hourglass Model – A poster I made, contains more about the hourglass model as a writing tool
• Creating an Outline for your Master’s Thesis – Instructional Article I wrote about how to outline a masters thesis and what is conventionally expected in these documents (most relevant to MIT students, but has general purpose advice)
• Indiana University Bloomington’s Paragraphs and Topic Sentences – A really nice article about paragraph structure and how to improve flow from the writing center at IU. 
• Tone and Audience – Explanation of how to identify audience and expectations for academic tone