Ann Sci. 2026 May 6:1-27. doi: 10.1080/00033790.2026.2638942. Online ahead of print.

ABSTRACT

This essay investigates the reception of William Gilbert's foundational work on magnetism, De Magnete, through a comprehensive analysis of extant copies of its early modern printed editions (1600, 1628, 1629, 1633). By employing a hybrid methodology combining quantitative and qualitative approaches to readers' annotations, this study charts patterns of engagement with Gilbert's text across diverse contexts and intellectual traditions. While celebrated for its experimental innovations and practical applications in navigation, it also elicited cosmological and humanist interests. Statistical analyses of readers' marks demonstrate a skewed distribution of engagement, with the majority of annotations concentrated in a small fraction of extant copies. This study moreover contributes to the historiography of early modern science by illustrating the methodological potential of combining large-scale digital datasets with close textual analysis, advocating for more systematic, collaborative approaches to the history of reading and book culture. In addition, a near-complete census of copies of De magnete is provided.

PMID:42091227 | DOI:10.1080/00033790.2026.2638942

Int J Archit Comput. 2024 Jan 13;22(3):277-294. doi: 10.1177/14780771231220903. eCollection 2024 Sep.

ABSTRACT

This article elaborates a computationally enabled approach to the study of design methods in 1960s North America. This entails the construction, visualization, and analysis of a digital database built from entries of the Design Methods Group Newsletter, a periodical published monthly between 1966-71. The article proposes a workflow that combines methods such as topic modeling and network visualization to activate the Newsletter as a source of anecdotal and informal knowledge, and to enable histories of connectivity and transaction that may elude archival investigations on singular actors or institutions. In doing so, the article contributes arguments and techniques for the study of design methods as a complex social, technical, and intellectual meshwork. The meshwork brings discursive themes, techniques, actors, and institutions at the same level of investigation and allows for layered cartographies of the field that advanced the systematic study of design and ushered in the development of early computer applications.

PMID:42038932 | PMC:PMC13104777 | DOI:10.1177/14780771231220903

PLoS One. 2026 Apr 21;21(4):e0347253. doi: 10.1371/journal.pone.0347253. eCollection 2026.

ABSTRACT

The overseas reception of classical literature through online platforms presents a critical lens for understanding cross-cultural dynamics in the digital age. This study investigates the overseas reception of English translations of Journey to the West by analyzing a corpus of 1,795 reviews from Amazon and Goodreads to examine temporal dynamics, cross-platform sentiment patterns, and topic modeling. The analysis covers four celebrated translators: Arthur Waley, Anthony C. Yu, Julia Lovell, and W.J.F. Jenner. Methodologically, we developed a hybrid sentiment lexicon by integrating a domain sentiment lexicon with AFINN, NRC, and VADER through weighted fusion, addressing the limited adaptability of general sentiment lexicons in translated literature analysis. LDA modeling was further applied to enable data-driven theme extraction. Key findings reveal a consistent year-on-year increase in review counts across all translations. Notably, despite an overall positive sentiment, significant cross-platform divergences emerge, reflecting the distinct evaluative mechanisms of digital platforms. Thematic analysis identifies three central reader concerns: translation quality, plot acceptance, and character portrayal, with plot acceptance exhibiting markedly higher negativity. Furthermore, translator-level analysis reveals performance variations across these themes. This study demonstrates how digital platforms reconfigure the valuation of literary translation, and pioneers a methodological framework for capturing the dynamic interplay between reader perception, media infrastructure, and textual mobility, offering new pathways for digital humanities research in translation studies.

PMID:42013105 | PMC:PMC13098970 | DOI:10.1371/journal.pone.0347253

Med Health Care Philos. 2026 Apr 16. doi: 10.1007/s11019-026-10347-1. Online ahead of print.

ABSTRACT

The uptake of social science methods by bioethics significantly expanded its methodological spectrum, raising new theoretical, methodological, and practical questions. Recently, we are witnessing another trend, adding advanced data science methods to bioethics' toolkit to aid, for example, in online data analysis, support scholarly writing, and inform clinical ethics. This article explores the emerging field of Digital Bioethics across its dimensions by analysing the tangled relationship between topics and methods, highlighting intersections between Digital Bioethics and Bioethics of the Digital, and advocating for a methods-based definition of the field. The use of advanced data science methods within bioethics must be interpreted in the context of the use of Artificial Intelligence (AI) in health care. At the same time, it presents unique opportunities and challenges. Defining, and thus demarcating, Digital Bioethics can create support for the new field but also requires navigating trade-offs. To do so, we take four kindred academic fields as points of comparison (Digital Humanities, Experimental Philosophical Bioethics, computational medicine and digitised biology) to analyse what each of them teaches for critically assessing and further developing Digital Bioethics. The article discusses potential pitfalls and concludes with recommendations on how the field can fully develop its potential to promote bioethical research and argument. Furthermore, the article discusses how a critical reflection of the use of AI methods within bioethics itself will also contribute to the ethical oversight of increasingly AI-driven branches of healthcare.

PMID:41989660 | DOI:10.1007/s11019-026-10347-1

Stud Hist Philos Sci. 2026 Mar 30;117:102151. doi: 10.1016/j.shpsa.2026.102151. Online ahead of print.

ABSTRACT

This paper examines large language models (LLMs) as research tools in the history, philosophy, and sociology of science (HPSS). Because LLMs can work directly with heterogeneous, unstructured texts and capture meaning-relevant associations from usage patterns, they offer new ways to bridge close reading and corpus-scale analysis, challenging the idea that computational scale and interpretive nuance must trade off. We provide a compact primer on LLMs, covering the main components of their neural network architecture, the differences between generative and full-context models, and adaptation strategies such as fine-tuning, prompt-based learning, and retrieval-augmented generation (RAG). Building on this foundation, we analyze how LLMs recast three classic methodological problems in HPSS: working with historically messy data, detecting and interpreting large-scale patterns, and modeling scientific change over time. Across these areas we synthesize recent work in HPSS and adjacent fields, and we clarify how LLM outputs can function as exploratory prompts, as inputs to more structured pipelines, or as evidence under stricter validation and documentation. We conclude with four lessons: 1) model choice embeds interpretive trade-offs, 2) responsible use requires LLM literacy, 3) HPSS should develop its own tasks and evaluation practices, and 4) LLMs should extend rather than replace established interpretive methods. We also situate these methodological questions within broader concerns about platform dependence, accountability, and the responsibilities attached to research infrastructures. Finally, we argue that HPSS is well positioned to both use LLMs and to interrogate what counts as explanation, evidence, and responsible use in interpretive research.

PMID:41916166 | DOI:10.1016/j.shpsa.2026.102151

Int J Transgend Health. 2025 Mar 15;27(2):902-916. doi: 10.1080/26895269.2025.2478092. eCollection 2026.

ABSTRACT

BACKGROUND: Research on the challenges, marginalization, and identity of trans women in India has sparked important discussions, contributing to progressive changes in society. While the visibility and recognition of trans women is steadily growing, beneficial schemes tailored to their unique challenges are often overlooked, underscoring the need for greater attention and action.

AIM: The article aims to identify the unique healthcare and migration challenges faced by South Indian trans women and the reach and utilization of government-provided facilities by their community.

METHOD: A survey of 53 and interviews with 4 South Indian trans women focused on the utilization of state and central government schemes. Data from public and private healthcare facilities in Madurai were collected and visualized using Airtable, with results disseminated in Tamil and English to ensure accessibility for the trans community.

RESULTS: A relationship was identified between the effectiveness of state government welfare schemes and the well-being of the trans women's community. However, central government schemes often fail to reach their entire population. Furthermore, state government transportation schemes do not sufficiently support their healthcare access and economic development.

DISCUSSION: To enhance the socio-economic development of trans women, policymakers can ensure that beneficial schemes comprehensively reach all segments of society. Increased promotion, awareness, and advancements in these schemes are necessary to meet the needs of the trans women community. Additionally, extending free bus fare facilities specifically to trans women is recommended to improve their healthcare access, mobility, economic opportunities, and integration into mainstream society.

PMID:41891076 | PMC:PMC13015021 | DOI:10.1080/26895269.2025.2478092

Med Humanit. 2026 Feb 12:medhum-2025-013580. doi: 10.1136/medhum-2025-013580. Online ahead of print.

ABSTRACT

Within the context of the eastward spread of Western learning during the late Ming and early Qing dynasties, the cross-cultural dissemination of medical knowledge exhibited complex characteristics of visual transformation. This study selects Taixi Renshen Shuogai (1623) and Zhenjiu Dacheng (1601) as research samples, employing a research method combining iconographic analysis and digital humanities to explore the different body concepts and their cultural connotations carried by Chinese and Western medical illustrations. The study constructs a three-dimensional analytical framework covering visual vocabulary, content expression and quantitative statistics to conduct an in-depth analysis of the cognitive differences in human body representation between the two medical traditions. The results show that the mechanistic view of the body advocated by Western anatomy and the organic view of the body upheld by traditional Chinese medicine form a sharp contrast at the image level; the former is characterised by precise analysis and structural reduction, while the latter centres on holistic grasp and functional correlation. This difference is not only reflected in visual elements such as composition patterns and expressive techniques, but more profoundly reflects the knowledge construction logic of different epistemological systems. The coexistence of two body cognition models during the Ming and Qing dynasties reveals the selective mechanism and creative transformation ability of Chinese culture in the process of knowledge acceptance, providing a new interpretive path for understanding the interaction model between traditional culture and foreign civilisations. This study expands the methodological boundaries of medical history research and provides a historical mirror for contemporary cross-cultural medical exchanges.

PMID:41679973 | DOI:10.1136/medhum-2025-013580

Data Brief. 2026 Jan 20;65:112495. doi: 10.1016/j.dib.2026.112495. eCollection 2026 Apr.

ABSTRACT

This dataset presents a valuable compilation of question-answer (QA) pairs derived from cultural texts and sources related to Durga mythology. A total of 21,395 QA pairs, encompassing textual materials such as scriptures, ritual narratives, temple inscriptions, and traditional storytelling records. Each entry includes the source reference, question, and corresponding answer, provided in a structured format compatible with Excel for seamless integration into downstream natural language processing (NLP) tasks. Data collection involved manual curation and annotation by domain experts, followed by preprocessing steps including text normalization, duplication removal, and verification of factual and contextual accuracy. The dataset is designed to support generative QA models, culturally aware chatbots, and digital preservation of heritage knowledge. It is particularly valuable for research in AI-driven cultural applications, educational tools, and digital humanities initiatives aiming to bridge traditional knowledge with computational methods. Researchers and practitioners may utilize the dataset for training generative models, creating interactive educational platforms, developing culturally sensitive AI agents, and supporting comparative studies in cross-cultural heritage. This openly accessible resource adheres to ethical standards, with proper attribution to source materials, and provides a foundational asset for both academic research and applied development in culturally informed artificial intelligence.

PMID:41657412 | PMC:PMC12874138 | DOI:10.1016/j.dib.2026.112495

IEEE Comput Graph Appl. 2026 Feb 2;PP. doi: 10.1109/MCG.2026.3660122. Online ahead of print.

ABSTRACT

This paper presents a case study focusing on the exploratory visual analysis of a unique historical dataset consisting of approximately 4000 visual sketches and associated captions from an encyclopedic book published in 1909-1910. The book, which offers insight into Vietnamese crafts and social practices, poses the challenge of extracting cultural meaning and narrative structure from thousands of drawings and multilingual captions. Our research aims to explore and evaluate the effectiveness of multiple visualization techniques in uncovering meaningful relationships within the dataset while working closely with professional historians. The main contributions of this study include refining historical research questions through task and data abstraction, combining and validating visualization techniques for historical data interpretation, and involving a focus group of historians for further evaluation. These contributions offer generalizable insights for the development of domain-specific visualization tools and support interdisciplinary engagement in historical data visualization and critical digital humanities research.

PMID:41628052 | DOI:10.1109/MCG.2026.3660122

Data Brief. 2026 Jan 7;64:112443. doi: 10.1016/j.dib.2025.112443. eCollection 2026 Feb.

ABSTRACT

Here we present a coherent, georeferenced and chronologically qualified corpus of fossil plant remains compiled from published archaeobotanical records from archaeological sites from Central Italy, focused on Olea europaea (olive) and Vitis vinifera (grape). The dataset is entirely based on secondary data and does not include newly generated primary archaeobotanical analyses. The dataset integrates site, context and all relevant archaeobotanical occurrences within a coherent relational and spatial model. The corpus was initiated through a structured bibliographic survey aided by the BRAIN database. Exclusively published literature was consulted, allowing to model archaeological sites and link them to excavation contexts and individual archaeobotanical occurrences (defined as the combination of a taxon and the specific plant part recovered, e.g., fruit, seed, rachis). The geodatabase was implemented using QGIS, with a local backend in GeoPackage, then migrated to PostgreSQL/PostGIS to support complex spatial/relational queries and future online outputs. All entities have a defined spatial placement accompanied by explicit quality-control parameters documenting positional uncertainty, source type and authority, as derived from the original published sources, ensuring transparent assessment of locational reliability. To enrich taxonomic information, an automated open thesaurus was built from CC BY/CC BY-SA resources (Floritaly, Acta Plantarum, and Wikimedia projects). The workflow employs REST-style access (or form-equivalent submissions), conservative rate-limiting, randomized waits, retries, and checkpoints; provenance and attribution (including noted transformations) are preserved. A standardized chronological table harmonizes relative cultural phases using ICCD nomenclature, with controlled fallbacks to Perio.do or peer-reviewed literature; a self-referential hierarchy (parent_id) ensures inheritance from sub-phase to broader period. Crucially, the use of open licenses, stable identifiers and cross-references makes the dataset interoperable and interlinked with the source ecosystems from which the secondary archaeobotanical data were extracted: records can resolve back to Floritaly and Acta Plantarum, and our forthcoming web portal can expose these connections for bidirectional navigation, automated updating and external reuse. The result is an interoperable, verifiable resource suitable for spatial and temporal analyses of plant remains based on aggregated and standardized published archaeobotanical data, while remaining legally reusable under the original licenses.

PMID:41624435 | PMC:PMC12855569 | DOI:10.1016/j.dib.2025.112443

Sci Rep. 2025 Dec 24;16(1):2505. doi: 10.1038/s41598-025-32302-5.

ABSTRACT

This study aims to achieve controllable generation of Chinese traditional cultural narrative content and enhance cultural identity. First, it constructs a tri-modal generation framework of text-image-style based on Stable Diffusion v2.1 and Contrastive Language-Image Pretraining (CLIP) models, realizing the joint modeling of traditional cultural semantics and visual imagery. Second, the study introduces the Low-Rank Adaptation (LoRA) mechanism to embed traditional cultural style features in a lightweight manner, improving the model's style adaptability under small sample conditions. Finally, a three-level evaluation system of "generation quality-semantic consistency-cultural identity" is built, covering both objective indicators and user feedback, to systematically verify the model's performance. Results show that the proposed model significantly outperforms existing methods in multiple dimensions: in terms of image quality, the Fréchet Inception Distance (FID) is 22.85, the Learned Perceptual Image Patch Similarity (LPIPS) is 0.298, and the style recognition accuracy reaches 86.4%. Regarding narrative consistency, the Bilingual Evaluation Understudy (BLEU) score is 0.325, the CLIP text-image similarity is 0.793, and the Narrative Style Match is 82.3%. On the cultural perception level, the average user narrative resonance is 4.32 points, the imagery accuracy score is 0.748, and the question-answer task pass rate is 82.6%. The comparative results indicate that the proposed method has significant advantages in expressive diversity and depth of cultural communication. When properly designed, Artificial Intelligence Generated Content (AIGC) technology can be effectively used for the generation and identity reconstruction of Chinese traditional cultural narrative content. This study provides a scalable technical path for the integration of AI and traditional culture, and expands the boundaries of digital humanities in content generation and reception research.

PMID:41444384 | PMC:PMC12820238 | DOI:10.1038/s41598-025-32302-5

Data Brief. 2025 Nov 8;63:112246. doi: 10.1016/j.dib.2025.112246. eCollection 2025 Dec.

ABSTRACT

Anime is an influential medium with global popularity, combining visual aesthetics with narrative depth and offering potential applications in content analysis, style transfer, and emotion recognition within computer vision research. Despite its widespread appeal, publicly available anime character datasets remain scarce. To address this gap, we propose the Attack on Titan: Anime Image Dataset, derived from the popular series Attack on Titan, to support anime-focused computer vision research. The dataset comprises 4041 high-quality images divided into 14 classes, each representing a prominent character from the series. These images are manually collected through high-resolution screenshots, capturing a wide range of character poses, expressions, costumes, and backgrounds. The dataset is suitable for various computer vision tasks, including character recognition, emotion detection, style classification, and domain adaptation.

PMID:41399437 | PMC:PMC12702017 | DOI:10.1016/j.dib.2025.112246

Sci Data. 2025 Dec 1;12(1):1804. doi: 10.1038/s41597-025-06241-9.

ABSTRACT

This dataset offers a comprehensive digital catalogue of 483 archaeological settlement sites in western Anatolia dating to the Middle and Late Bronze Age (c. 2000-1200 BCE). Compiled over a decade, it brings together evidence from excavation reports, systematic surveys, historical sources, and remote sensing. Each site is georeferenced and described through a standardized set of metadata, including chronological attribution, site function, material culture, bibliographic references, and associated ancient mineral resources. The dataset is published on Zenodo as a collection of openly accessible files, structured with consistent keys that ensure integration across records. To enhance semantic interoperability, settlement entries are linked to external reference datasets such as open knowledge bases, enabling opportunities for comparative, geospatial, and interdisciplinary research spanning archaeology, digital humanities, and historical geography. By combining standardized metadata with semantic linking, the resource facilitates reuse within broader digital infrastructures. It thereby provides a transparent, openly licensed foundation for analyzing regional settlement systems and encourages more comprehensive approaches to the study of Bronze Age Anatolia.

PMID:41326413 | PMC:PMC12669694 | DOI:10.1038/s41597-025-06241-9

Sci Rep. 2025 Nov 24;15(1):41531. doi: 10.1038/s41598-025-25466-7.

ABSTRACT

English literature and linguistics have long served as foundational disciplines in humanities education, cultivating critical analysis, linguistic proficiency, and cultural interpretation. Conventional teaching methods struggle to meet diverse learner needs, ensure consistent engagement, and provide personalized academic feedback. To improve learning with the help of modern techniques, this study proposes a comprehensive, multi-technique Artificial Intelligence (AI)-driven tools assessment framework aimed at enhancing English pedagogy through the integration of advanced artificial intelligence tools. The research work includes adaptation of a mixed-methods research design incorporating classroom case studies, in-depth interviews, and analysis of students' documents to evaluate their learnings. The framework employed statistical techniques to validate significant relationships among engagement, tool usage, and learning clarity. Key evaluation criteria is captured using the Fuzzy Delphi Technique which identifies high-importance attributes such as AI usage, usability, and analytical quality. Moreover, eXplainable AI (XAI) techniques including LIME and SHAP applied to enhance model transparency, offering both global and local interpretability of outcomes. To predict pedagogical effectiveness, a deep learning Bi-LSTM model was trained, achieving 90% accuracy, 92% precision, 93% recall, and 92% F1-score across key performance metrics for the usage analysis of AI-based tools.

PMID:41286088 | PMC:PMC12644666 | DOI:10.1038/s41598-025-25466-7

Front Sociol. 2025 Oct 31;10:1715327. doi: 10.3389/fsoc.2025.1715327. eCollection 2025.

NO ABSTRACT

PMID:41245508 | PMC:PMC12615465 | DOI:10.3389/fsoc.2025.1715327

Sci Rep. 2025 Oct 10;15(1):35498. doi: 10.1038/s41598-025-19436-2.

ABSTRACT

With the development of artificial intelligence education, the human-computer interaction and human-human interaction in virtual learning communities such as Zhihu and Quora have become research hotspots. This study has optimized the research dimensions of the virtual learning system in colleges and universities based on neural network algorithms and the value of digital intelligence in the humanities. This study aims to improve the efficiency and interactive quality of students' online learning by optimizing the interactive system of virtual learning communities in colleges. Constructed an algorithmic model for a long short-term memory (LSTM) network based on the concept of digital humanities integration. The model uses attention mechanism to improve its ability to comprehend and process question-and-answer (Q&A) content. In addition, student satisfaction with its use was investigated. The Siamese LSTM model with the attention mechanism outperforms other methods when using Word2Vec for embedding and Manhattan distance as a similarity function. The performance of the Siamese LSTM model with the introduction of the attention mechanism improves by 9%. In the evaluation of duplicate question detection on the Quora dataset, our model outperformed the previously established high-performing models, achieving an accuracy of 91.6%. Students expressed greater satisfaction with the updated interactive platform. The model in this study is more suitable than other published models for processing the SemEval Task 1 dataset. Our Q&A system, which implements simple information extraction and a natural language understanding method to answer questions, is highly rated by students.

PMID:41073596 | PMC:PMC12514068 | DOI:10.1038/s41598-025-19436-2

F1000Res. 2025 Sep 9;13:1305. doi: 10.12688/f1000research.156917.2. eCollection 2024.

ABSTRACT

BACKGROUND: Reproducibility is a basis of scientific integrity, yet it remains a significant challenge across disciplines in computational science. This reproducibility crisis is now being met with an Open Science movement, which has risen to prominence within the scientific community and academic libraries especially. At the Carnegie Mellon University Libraries, the Open Science and Data Collaborations (OSDC) Program promotes Open Science practices with resources, services, and events. Hosting hackathons in academic libraries may show promise for furthering such efforts.

METHODS: To address the need for reproducible computational research and promote Open Science within the community, members of the OSDC Program organized a single-day hackathon centered around reproducibility. Partnering with a faculty researcher in English and Digital Humanities, we invited community members to reuse Python code and data from a research publication deposited to Harvard Dataverse. We also published these materials as a compute capsule in Code Ocean that participants could also access. Additionally, we investigated ways to use ChatGPT to troubleshoot errors from rerunning this code.

RESULTS: Three students from the School of Computer Science participated in this hackathon. Accessing materials from Harvard Dataverse, these students found success reproducing most of the data visualizations, but they required some manual setup and modifications to address depreciated libraries used in the code. Alternatively, we found Code Ocean to be a highly accessible option, free from depreciation risk. Last, ChatGPT also aided in finding and addressing the same roadblocks to successfully reproduce the same figures as the participating students.

CONCLUSIONS: This hackathon allowed several students an opportunity to interact with and evaluate real research outputs, testing the reproducibility of computational data analyses. Partnering with faculty opened opportunities to improve open research materials. This case study outlines one approach for other academic libraries to highlight challenges that face reproducibility in an interactive setting.

PMID:41064702 | PMC:PMC12501581 | DOI:10.12688/f1000research.156917.2

J Med Ethics. 2025 Sep 4:jme-2025-111310. doi: 10.1136/jme-2025-111310. Online ahead of print.

ABSTRACT

We respond to Gillon's critique of our data-driven analysis of the history of Journal of Medical Ethics (JME), in which we used a topic model to trace intellectual trends in the journal's first 50 years. Gillon, drawing on his personal memories as JME's second (and longest serving) editor, challenges several of our findings, particularly those concerning the prominence and classification of topics such as Ethics education In this reply, we clarify misunderstandings that led to part of his criticisms of our method. At the same time, we also briefly discuss some nuances of topic modelling, in particular, its reliance on simplified representations of text, sensitivity to modeling choices and topic interpretations. Rather than viewing computational models and editorial memory as competing sources of insight, we propose that they are complementary: each illuminates different dimensions of the journal's evolution. Gillon's engagement with our work ultimately highlights the importance of methodological transparency and the value of combining digital humanities tools with lived experience in the historiography of academic disciplines.

PMID:40908135 | DOI:10.1136/jme-2025-111310

J Lesbian Stud. 2025 Sep 4:1-17. doi: 10.1080/10894160.2025.2555666. Online ahead of print.

ABSTRACT

Following the 2000s, India witnessed an emergence of cyber-queer spaces in the form of several online multimedia platforms, social media networking and dating sites which explored the intersections of gender and sexuality. Through an analysis of the queer 'I' in digital personal narratives by queer AFAB individuals on multimedia platforms like Agents of Ishq (AOI) and Point of View (POV), we argue that both the design of these websites and this written personal history contain a self-affirming erotic power. This power is enmeshed with and informed by an Indian lesbian political history which demands that we rethink our relationship with the challenges of lesbian desire, intersectionality, caste subalternity and the politics of inclusion both online and offline.

PMID:40905382 | DOI:10.1080/10894160.2025.2555666

Int J Digit Humanit. 2025;7(1):115-134. doi: 10.1007/s42803-025-00100-0. Epub 2025 May 12.

ABSTRACT

This study provides critical insights and evaluates the performance of state-of-the-art Handwritten Text Recognition (HTR) engines-PyLaia, HTR + , IDA, TrOCR-f, and Transkribus' proprietary Transformer-based "supermodel" Titan-to digitize historical documents. Using a diverse range of datasets that include different scripts, this research assesses each engine's accuracy and efficiency in handling multilingual content, complex styles, abbreviations, and historical orthography. Results indicate that, while all engines can be trained or fine-tuned to improve performance, Titan and TrOCR-f exhibit superior out-of-the-box capabilities for Latin-script documents. PyLaia, IDA, and HTR + excel in specific non-Latin scripts when specifically trained or fine-tuned. This study underscores the importance of training, fine-tuning, and integrating language models, providing critical insights for future advancements in HTR technology and its application in the digital humanities.

PMID:40584138 | PMC:PMC12202554 | DOI:10.1007/s42803-025-00100-0