Skip to content

Academic Writing Style Guide (Typst) ​

Table of Contents ​


Core Principles ​

1. Clarity Over Complexity ​

  • One idea per sentence
  • Avoid nested clauses when possible
  • Define terms on first use

2. Precision Over Vagueness ​

  • Use specific numbers instead of "several" or "many"
  • Quantify claims whenever possible
  • Avoid hedging without evidence

3. Active Voice (When Appropriate) ​

  • ✅ "We propose a novel method..."
  • ✅ "This paper presents..."
  • ❌ "A novel method is proposed by us..."

Sentence Length Guidelines ​

TypeWord CountUse Case
Short10-15Key findings, transitions
Medium15-25Most content
Long25-40Complex relationships
Very Long>40⚠️ Consider splitting

Paragraph Structure ​

Topic Sentence ​

First sentence states the main point.

Supporting Sentences ​

  • Evidence, examples, or elaboration
  • 3-5 sentences typical
  • Clear logical flow

Transition ​

Connect to next paragraph if needed.

Academic Vocabulary ​

Reporting Verbs (Neutral) ​

  • states, notes, observes, reports, describes

Reporting Verbs (Strong Agreement) ​

  • demonstrates, shows, proves, confirms, establishes

Reporting Verbs (Tentative) ​

  • suggests, implies, indicates, proposes, hypothesizes

Reporting Verbs (Critical) ​

  • claims, argues, asserts, alleges, maintains

Transition Words ​

Addition ​

  • Furthermore, Moreover, Additionally, In addition

Contrast ​

  • However, Nevertheless, Conversely, On the other hand

Cause/Effect ​

  • Therefore, Consequently, As a result, Thus

Example ​

  • For instance, For example, Specifically, In particular

Sequence ​

  • First, Second, Subsequently, Finally

Citation Integration (Typst Syntax) ​

Integral (Author as Subject) ​

typst
@smith2020 demonstrated that...
According to @jones2021, the method...

Non-Integral (Content Focus) ​

typst
Deep learning has shown remarkable success @smith2020 @jones2021 @wang2022.

Paraphrase (Preferred) ​

Restate the idea in your own words with citation.

Direct Quote (Sparingly) ​

Only for definitions or exceptionally well-phrased ideas.

Stacking 3+ references without individual discussion is a common AI writing pattern and is unacceptable in top-tier venues. These rules apply to both Introduction and Related Work sections.

Rules:

  1. Max 2 clustered citations without discussion per sentence

    • ❌ "Many methods have been proposed @smith2020 @jones2021 @wang2022 @li2023 @chen2024."
    • ✅ "@smith2020 proposed X for scenario A. Building on this, @jones2021 extended the approach to B, while @wang2022 addressed limitation C."
  2. Every cited work must earn its citation with at least one of:

    • A summary of its core contribution (1 clause minimum)
    • A comparison with another cited work
    • A specific limitation that motivates your work
  3. Narrative over parenthetical in Introduction and Related Work:

    • Use integral citations (author as subject) for key works: "@smith2020 demonstrated..."
    • Reserve non-integral citations for well-established facts only: "Gradient descent is widely used @smith2020 @jones2021."
  4. Funnel-appropriate density (Introduction):

    • Background paragraph (broad context): up to 2 clustered citations for established facts
    • Problem statement paragraph: each citation must be individually discussed
    • Gap/motivation paragraph: every cited limitation must reference a specific paper
  5. Categorical discussion (Related Work):

    • Group works by methodology/approach, not chronologically
    • Within each group, discuss each work's specific contribution and limitation
    • Use comparative language between works: "Unlike @smith2020, @jones2021 addresses..."

Positive patterns (Introduction):

typst
@smith2020 proposed method X, achieving Y% accuracy on dataset Z.
However, their approach assumes A, which limits applicability to B.
@jones2021 relaxed this assumption by introducing C, but at the cost of D.
In contrast, our method addresses both limitations by...

Positive patterns (Related Work):

typst
*Transformer-based methods.* @vaswani2017 introduced the self-attention
mechanism for sequence modeling. @li2019 adapted this architecture for
time series, but their method requires $O(n^2)$ memory. @zhou2021 proposed
ProbSparse attention to reduce complexity to $O(n log n)$, though at the
cost of approximation error.

Negative patterns (FORBIDDEN):

typst
Many researchers have studied this problem @a2020 @b2021 @c2022 @d2023 @e2024.
Several methods have been proposed @f2020 @g2021 @h2022 @i2023 @j2024 @k2024.

Common Section Patterns ​

Introduction ​

  1. General context → Specific problem
  2. Gap in existing work
  3. Contributions of this work
  4. Paper organization
  1. Categorize existing approaches
  2. Compare and contrast
  3. Position your work

Methodology ​

  1. Overview of approach
  2. Detailed steps
  3. Justification for choices

Results ​

  1. Experimental setup
  2. Quantitative results
  3. Qualitative analysis
  4. Comparison with baselines

Conclusion ​

  1. Summary of contributions
  2. Limitations
  3. Future work

Typst-Specific Formatting ​

Headings ​

typst
= Introduction          // Level 1
== Background          // Level 2
=== Related Work       // Level 3

Emphasis ​

typst
*bold text*            // Bold
_italic text_          // Italic
`code text`            // Code/monospace

Lists ​

typst
// Unordered list
- Item 1
- Item 2
  - Nested item

// Ordered list
+ First item
+ Second item
  + Nested item

Math ​

typst
// Inline math
The equation $x^2 + y^2 = z^2$ shows...

// Display math
$ x^2 + y^2 = z^2 $

// Numbered equation
$ x = (a + b) / 2 $ <eq:mean>

Figures and Tables ​

typst
// Figure
#figure(
  image("figure.png", width: 80%),
  caption: [Description of the figure.]
) <fig:example>

// Table
#figure(
  table(
    columns: 3,
    [Header 1], [Header 2], [Header 3],
    [Data 1], [Data 2], [Data 3],
  ),
  caption: [Description of the table.]
) <tab:example>

Cross-references ​

typst
As shown in @fig:example, the method...
Table @tab:example presents the results...
Equation @eq:mean defines the average...

Released under the MIT License.