Bin, Adriana; Pinto, Daniela Maciel; Pereira, Cesar A. Guimarães; Cristofoletti, Evandro Coggo; Mena-Chalco, Jesus Paschoal; Salles Filho, Sérgio; Sugimoto, Cassidy R.
Peer review plays a central role in research funding decisions, yet the evaluative language used in review reports remains insufficiently explored, particularly in funding agencies. This study examines how linguistic and sentiment characteristics of peer-review reports relate to funding outcomes at the São Paulo Research Foundation (FAPESP), Brazil, focusing on the Regular Research Grant (RRG) and Young Investigator Grant (YIG) schemes between 2000 and 2017. Drawing on 19,333 proposals and their corresponding review reports, we analyze three aggregated corpora — overall reports, proposal-related reports, and applicant track-record reports — using natural language processing and sentiment analysis. We compare approved and rejected proposals across fields of knowledge, reviewer gender, applicant gender, and peer-review report sections. Results show that reviewers use distinct textual and sentiment patterns when justifying approval or rejection. These patterns vary according to the section evaluated — applicant track record or technical proposal — and also differ by field of knowledge and, more locally, by reviewer–applicant gender configurations. The pattern is more regular and stable in RRG than in YIG. Overall, the findings show how evaluative language illuminates fairness, consistency, and transparency in grant peer review.
