Transcribing Machines: HTR as a Question of Critical AI

Abstract

This talk examines Handwritten Text Recognition (HTR) not as a neutral tool of transcription, but as a site where salient debates of Critical AI emerge. HTR, it has been argued, is perhaps the most successful form of AI introduced into the archive (Neudecker). Yet, while it promises scaled access to the handwritten unread, it also unsettles traditional scholarly authority, raising fundamental questions about who produces and standardizes the creation of searchable text.

Drawing on multilingual, multi-community experiences with HTR, I argue that while these systems open up important critical perspectives in the computational study of texts, they also reshape authority and expertise under which documentary knowledge is produced. One of the central questions of this talk is whether the humanities community is treating automated transcription too simply as a technical problem, while overlooking forms of extraction hidden beneath promises of access.

Training corpora—often drawn from wealthy institutions of the Global North—privilege certain scripts, languages, and archival traditions, making the question of whose data and models are centered a critical one. As HTR transforms select handwritten materials into searchable corpora–readable by both humans and machines–it also runs the risk of deepening the gap between well-resourced textual traditions and those whose scripts or archival conditions remain difficult to computationally process. These inequalities are not merely technical, but they shape the conditions under which the past can be discovered and known.

Taken together, these tensions point toward a larger question about the future of HTR on which we can only speculate: will it remain shaped by scholarly commitments to transparency and context, or increasingly by AI infrastructures designed for scalability and generalized results? Automation may eliminate some older scholarly practices (Dubreuil) such as manual transcription, but it may also compel the humanities to focus more strongly on critical questions of systems, infrastructure, and methodological inclusivity.