Investigating writing style as a contributor to gender gaps in science and technology

This paper investigates whether gender differences in writing styles contribute to observed gender gaps in the evaluation of scientific and technical work. The authors ground their investigation in a framework by Biber (1988) that contrasts informational linguistic features (which emphasize facts) with involved linguistic features (which emphasize relationships). Using a large sample of single-authored academic papers from Web of Science and patents from PatentsView, the authors find significant differences in writing style by gender.

Specifically, the paper states: we find significant differences in writing style by gender, with women using more involved features in their writing. The authors document that "even in such restricted genres as academic writing and invention abstracts, female authors tend to use more 'involved' features than male authors, a pattern that holds across all scientific fields and patent subcategories. In the patent data, they find that the gender of the lawyer appears to have more impact on the writing than the gender of the inventor, suggesting both the importance of attorneys in crafting patent text and the universal nature of the patterns we observe, even across individuals with very different professional training."

The paper also examines citation patterns and finds that Papers and patents with more involved features also tend to be cited more by women. More specifically, papers with higher rates of utilization of involved features in their abstracts tend to be cited at a higher rate by papers with a female first author and papers with a female last author. The authors note that papers with higher rates of utilization of involved features are cited at a lower rate by papers with a male first author.

The authors conclude that scientific text is not devoid of personal character, which could contribute to bias in evaluation, thereby compromising the norm of universalism as a foundational principle of science. They suggest that in order to mitigate bias in evaluation processes, the diversity among the contributors should be matched by the same diversity among the evaluators, such as increasing the number of female reviewers and encouraging citation diversity.

Based on the paper, here are specific improvements for AI systems:

• Improvement: Add linguistic style features (involved/informational rates) as inputs to citation recommendation algorithms, alongside content similarity.

• What it can do: Predict citation behavior more accurately by accounting for the finding that papers with higher involved features are cited more by female authors, while informational features attract male citers. This reduces gender bias in academic recommendation systems.

• Improvement: Train classifiers to flag when a manuscript's linguistic style (e.g., high involved rate) may trigger unconscious bias from reviewers of a particular gender.

• What it can do: Alert editors when a submission's style is likely to be systematically undervalued by male reviewers, enabling interventions (e.g., assigning more diverse reviewers) to uphold universalism in evaluation.

• Improvement: Implement a post-processing module that adjusts involved and informational feature ratios in draft text toward a neutral baseline, without altering content.

• What it can do: Help female applicants reduce the stylistic gap that contributes to lower funding rates, by automatically suggesting rephrasing that lowers involved features (e.g., reducing pronouns, questions, and connectors) while preserving meaning.

• Improvement: Build a system that distinguishes between inventor and lawyer contributions, using the finding that lawyer gender (not inventor gender) more strongly predicts style.

• What it can do: For patent attorneys, provide real-time feedback on abstract style to minimize gender-linked variation, potentially reducing examiner bias and improving grant rates for female inventors.

• Improvement: Integrate the Involved-Informational Ratio as a feature in bibliometric models that track citation flows by gender.

• What it can do: Enable research teams to map how linguistic style propagates through citation networks, identifying whether certain fields or journals systematically favor one style, and thus where gender gaps are likely to emerge.

• Improvement: Use the framework to generate two versions of an abstract—one optimized for involved style (to attract female citers) and one for informational style (to attract male citers).

• What it can do: Allow authors to strategically tailor their abstract for target audiences, or to create a balanced version that maximizes cross-gender citation potential.

• Improvement: Implement a field-aware adjustment layer that accounts for the paper's finding that gender differences are larger in Social Sciences/Arts & Humanities and smaller in Physical Sciences/Technology.

• What it can do: Automatically adjust the sensitivity of bias-detection thresholds by field, preventing over- or under-flagging of style differences in contexts where they are more or less normative.

A growing stream of research finds that scientific contributions are evaluated differently depending on the gender of the author. In this article, we consider whether gender differences in writing styles - how men and women communicate their work - may contribute to these observed gender gaps. We ground our investigation in a framework for characterizing the linguistic style of written text, with two sets of features - informational (i.e., features that emphasize facts) and involved (i.e., features that emphasize relationships). Using a large sample of academic papers and patents, we find significant differences in writing style by gender, with women using more involved features in their writing. Papers and patents with more involved features also tend to be cited more by women. Our findings suggest that scientific text is not devoid of personal character, which could contribute to bias in evaluation, thereby compromising the norm of universalism as a foundational principle of science.