Examining linguistic shifts between preprints and publications

Area of Science:

• Biomedical research
• Scholarly communication
• Computational linguistics

Background:

• Preprints offer rapid dissemination of research findings before peer review.
• Previous studies on bioRxiv preprints focused on metadata, downloads, citations, and online discussion.
• The linguistic content of preprints and its evolution during peer review remains under-examined.

Purpose of the Study:

• To compare linguistic features of bioRxiv preprints with published biomedical literature.
• To investigate how peer review influences document language.
• To develop tools for analyzing preprint-to-publication linguistic relationships.

Main Methods:

• Comparative linguistic analysis of bioRxiv preprints and published biomedical articles.
• Development of document embeddings using a preprint-trained word2vec model.
• Creation of a web application for preprint-similarity searching.

Main Results:

• Significant linguistic differences were observed between preprints and published articles, particularly in typesetting and mentions of supplementary materials.
• Document embeddings successfully identified scientific concepts, linked preprints to peer-reviewed articles, and suggested journals publishing similar content.
• Preprints with multiple versions and substantial textual revisions exhibited longer publication times.

Conclusions:

• Linguistic analysis provides insights into the transformation of scientific documents during peer review.
• Computational methods, like document embeddings, can effectively bridge the gap between preprints and published literature.
• The developed web application aids researchers in navigating the scholarly landscape and understanding preprint-to-publication trajectories.