The Craft of Prose

Chapter 1 Episode 5

You may wonder: is there a gold template for academic papers, so that you can simply follow it to write a good paper? Well, there are “templates,” but they are closer to formatting rules for different venues. If you are talking about how you would narrate the fruit of your work, then no, there is no gold template; on the contrary, there are quite a lot of different genres.

Above all, the ultimate goal of academic writing is clear:

Academic paper writing requires that the paper be uniformly interpreted by diverse readers (i.e., readers with different backgrounds and years of experience) in the way the authors intend.

The goal is much more difficult to execute than it sounds. Achieving it requires you to know (1) what you want to convey, (2) what diverse readers can and cannot understand, (3) what questions readers may ask and what the answers are, and (4) how to articulate everything in a way readers can accurately understand and appreciate. Mastering these four requirements takes years of hard work to accumulate the necessary knowledge, which is not something a single episode can cover.

With that said, there are also many common practices we follow, and the purpose of this episode is to share such writing knowledge. I deliberately frame these as “advice” rather than “rules,” because I want to be clear that everything I share here is advice I use for my lab and myself, not submission policy at any venue. Specifically, I would like to offer five pieces of advice that help you approach the ultimate goal. The advice is designed to be task- and domain-agnostic, so that it generalizes across different contexts.

Advice 1: Make the Logical Connection Between Sentences Explicit

Let me use the motivation part of a paper, also known as the Introduction, as the example for this advice, because it is the section where logical connections matter most. Typically, each paragraph in the Introduction conveys one key statement, and the concatenation of these key statements forms the complete logical trajectory of your proposed work. Clearly, it is critical to ensure the logical trajectory is error-free, because a rejection is guaranteed if reviewers can identify logical flaws in your motivation.

The best way to make your motivation bulletproof is to make sure every consecutive pair of key statements has a clear and concrete logical relationship. Readers should never have to infer why the second statement follows the first, and they should never be able to challenge the logical relationship between them.

The key is to use explicit conjunctions — including those indicating cause and effect, progression, and contrast — to ensure consistent comprehension for all readers. Very simple.

Let us read a sample write-up:

“Current sepsis prediction models achieve high accuracy on retrospective datasets. Clinicians report low adoption rates in practice.”

What problem do you identify? Both sentences are true, but the relationship between them is unstated. Is the second sentence a contrast? A consequence? A separate observation? Readers have to guess, and if they guess wrong, they misunderstand the whole logic. If you cannot state the logical relationship between two consecutive sentences, either the sentences belong in a different order, or one of them does not belong in your rationale trajectory at all.

Advice 2: Every Sentence Must Contribute to the Rationale

A sentence can be factually correct, well-written, and properly cited, and still be noise. The test for whether a sentence should exist is whether it directly supports or advances the argument being built in the paragraph, as well as whether it directly feeds the next sentence. If a sentence does not contribute to the claim that follows it, it should be relocated to where it does contribute, or removed if it contributes nothing to the paper, regardless of how true or interesting it is on its own.

The rationale behind this advice is straightforward. If you think from a reader’s perspective, you do not know what the authors want to convey or what they propose to do before reading the paper, so you pay equal attention to every sentence. If a sentence does not contribute to the rationale, it will trigger an alert in the reader’s mind because of the gap, and may even lead to a misinterpretation of your work. Trust me, it is very frustrating to discover that a reviewer did not understand the motivation or the focus of your work, after they gave you quite a harsh review and a recommendation for rejection.

Here is an example:

“AI-based clinical decision support has attracted significant research interest in recent years. Many hospitals have adopted electronic health record systems. In particular, prediction models for sepsis have achieved AUC scores above 0.90.”

In this example, the second sentence is factually true and topically related, but it does not advance the argument from research interest (sentence 1) to prediction performance (sentence 3). It is noise that the reader has to step over, and such noise also costs your credibility, because it signals that you have not thought carefully about what the reader needs to know and in what order.

Advice 3: Every Pronoun Needs an Unambiguous Antecedent

When you write “this,” “it,” “they,” “these,” or “such,” the reader must be able to identify the exact referent without hesitation, confusion, or re-reading the previous sentence. A rule of thumb is that the referent of a pronoun is its nearest preceding entity. If the previous sentence is complicated enough that readers will struggle to identify the referent, replace the pronoun with the explicit entity you are referring to.

Consider this passage:

“We designed the system to support clinicians during triage. They found it helpful in identifying high-risk patients. This improved the overall workflow.”

What does “this” refer to? The system? The fact that clinicians found it helpful? The identification of high-risk patients? The reader has to guess, and different readers will interpret the pronoun differently, which means your sentence communicates something different to each person.

I single out this advice because ambiguous pronouns are the most frequent writing error (aside from logical errors) I encounter in student papers, and they are especially damaging in motivation sections, where every sentence is supposed to build on the previous one. If the reader cannot tell what “this” points to, the chain of reasoning breaks. And obviously, do not use a pronoun at the very beginning of a paragraph, since there is nothing preceding it for the pronoun to refer to.

Advice 4: Use Terms Consistently Throughout the Paper

Imagine you are reading a paper and encounter the term “cognitive load” in the introduction, “mental workload” in the methods, and “attentional demand” in the discussion. Are these three different constructs? Are they synonyms the authors are using interchangeably? As a reader, you have no way to know. You are forced to either guess or hold a mental note of uncertainty, both of which erode comprehension.

Terminological inconsistency is surprisingly common in student writing, often because the author has read broadly and absorbed multiple terminological traditions without realizing they are mixing them.

The rule is simple: once you choose a term for a concept, use that exact term every time you refer to that concept. If your construct is called “perceived ease of use,” do not call it “usability” in one paragraph and “user-friendliness” in another, even if those terms feel roughly synonymous to you. The reader cannot read your mind, only your text.

What if a concept genuinely has multiple widely used names in the literature? This happens often. In that case, acknowledge the variation early and then pick one. Add a footnote the first time the concept appears: “This construct is also referred to as X and Y in the literature. We use Z throughout this paper for consistency.” One sentence buys you clarity for the rest of the paper.

The underlying principle is the same as in Advice 1 and Advice 2: do not make the reader guess what you mean. Every time a reader has to pause and wonder whether two terms refer to the same thing, you have broken the chain of comprehension. A paper is a communication device where consistency enables understanding.

Advice 5: Details That Signal Care

The advice so far has addressed structural problems: pronouns, logical connections, relevance, and terminology. What follows is a set of smaller but collectively important practices. Individually, each might seem minor, but together, across a full manuscript, they signal to the reader (and to reviewers) whether the authors are careful or careless. Careful authors are trusted. Careless authors are scrutinized. You want to be in the first category.

1. Define domain-specific terms at first use. Readers come from different subfields, different disciplines, and different career stages. A term that feels basic to you, such as “confound,” “endogeneity,” or “ecological validity,” may be unfamiliar to a smart reader from an adjacent community. The first time a specialized term appears, define it. Doing so is hospitality toward your readers, not condescension.

2. Spell out abbreviations at first use. “The LLM was trained on EHR data from the ICU.” If the reader does not know what LLM, EHR, or ICU stand for, the sentence communicates nothing. Write “large language model (LLM)” and “electronic health record (EHR)” the first time each appears. After that, the abbreviation alone is fine.

3. One sentence, one piece of information. Long, multi-clause sentences force the reader to hold several ideas in working memory simultaneously while they wait for the sentence to resolve. When a sentence tries to do too many things at once (state a finding, qualify it, compare it to prior work, and note a limitation), the reader arrives at the period exhausted and uncertain. Break complex thoughts into multiple sentences. Let each sentence carry a single, clear piece of information. Your reader will absorb more, not less.

4. Do not use spoken English. I encounter this regularly in student drafts: “the model kinda struggles with noisy data,” “we ran a bunch of experiments,” “the results were pretty good.” In conversation, these phrases are perfectly functional. In academic writing, they signal a lack of seriousness, and the deeper issue is precision. “Kinda struggles” does not tell the reader whether the performance drop is 2% or 20%. Say what you mean with specific, formal language.

5. Spell out numbers below 10; use numerals for 10 and above. The convention is stylistic, but it aids visual parsing. “We recruited nine participants” and “We recruited 12 participants” scan differently, and the distinction helps the reader process quantities at a glance. Check your target venue’s style guide, since some have their own rules, but when in doubt, this convention is the safest default.

6. Avoid extreme language. Words like “never,” “always,” “all,” “none,” “completely,” and “definitively” are almost never true in research. More importantly, they invite the reader to search for counterexamples. If you write “no prior work has addressed this problem,” a reviewer who knows of even one tangentially related paper will doubt your thoroughness. Use hedges that match the actual state of knowledge: “to the best of our knowledge,” “prior work has largely focused on,” “our findings suggest.” Measured language is accurate language.

Advice Is Not Policy

I want to return to something I said at the beginning: these are pieces of advice, not rules enforced by any venue’s submission system. You will read published, well-cited papers at top venues that violate every single one of them. Published violations do not mean the advice is wrong, and they do not mean those papers are bad. Experienced writers know when a convention can be broken, because breaking it serves a specific rhetorical purpose.

As a PhD student, you are not in a position to know when that is. Follow the advice until you understand it well enough to know when to deviate. By the time you reach that point, you will no longer need this episode.