Maker's marks and hallmarks on historical gold- and silverwork are crucial evidence for provenance research, cataloging, and object dating. However, identifying these marks manually is highly labor-intensive. Even in high-quality museum photographs, marks are diminutive, frequently worn, and often obscured by reflections or patina. To address this bottleneck, this paper presents and evaluates a computer-vision pipeline designed to automatically retrieve visually similar marks from challenging real-world datasets. We evaluate the impact of different localization strategies, including fully automated bounding-box prediction, alongside off-the-shelf and fine-tuned feature extraction networks. Our experiments demonstrate that combining localized cropping with metric-learning fine-tuning significantly improves retrieval accuracy, allowing the model to prioritize subtle iconographic features over background noise. These performance gains represent a crucial step toward scalable, AI-assisted identification tools for domain experts. By enabling the accurate retrieval of impressions struck by the same physical stamp, this work lays a foundation for advanced analyses of historical metalwork, including forgery detection and stamp abrasion studies.

A stemma (pl. stemmata) is a genealogical trees of manuscripts used by philologists and editors to represent the history of textual traditions of manuscript texts since the early XIXth century. The emergence of the complex system approach to cultural dynamics as well as long-lasting methodological controversies in philology have driven the demand for a comprehensive, cross-corpus database of stemmata. The collection of such a database remains a tedious task due the fact that published stemmata are scattered among a vast body of scientific publications and critical editions, with widely heterogeneous representation conventions.

The object of this work is to present an automated detection and transcription pipeline of stemmata designed to extract relevant informations from large libraries of digitized publications, leveraging recent development in computer vision and vision-language models. The output of this pipeline is a a structured, XML-based representation of stemmata enriched with metadata on corresponding textual traditions and witnesses, following the guidelines of the OpenStemmata project.

We present a detailed outline of our protocol, with quantitative evaluations of the performances of the subtasks of this pipeline

This paper investigates how museum audio descriptions (ADs) linguistically reconstruct visual artworks and whether different artistic movements systematically shape the embodied qualities of their verbal mediation. While previous research has shown that AD tends to be lexically dense, metaphor-rich, and stylistically complex, little attention has been paid to the degree of cross-modal “semblance” between visual form and verbal expression. We propose that, as a form of intersemiotic translation, AD may employ iconicity (form–meaning resemblance) and sensorimotor grounding as strategies for evoking aesthetic experience.

Drawing on a corpus of approximately 400 English-language museum ADs describing paintings from six modern movements (Impressionism, Expressionism, Cubism, Postimpressionism, Surrealism, and Abstraction), we model lexical iconicity and multidimensional perceptual strength using two large-scale psycholinguistic resources: the Iconicity Norms (Winter et al. 2023) and the Lancaster Sensorimotor Norms (Lynott et al. 2019). For each description, we compute mean iconicity, iconicity selectivity, and modality-specific perceptual activation profiles. Sensorimotor dimensions are normalised both relative to the English lexicon and to the AD corpus itself, allowing identification of stylistically disproportionate sensory grounding.

Generalized linear mixed-effects models are used to estimate the influence of artistic movement while controlling for painter and commissioning institution. We examine overall iconicity, part-of-speech-specific effects, modality dominance, and the sensory grounding of highly iconic words.

The study contributes to Digital Humanities in three ways. First, it demonstrates how transparent, human-derived lexical norms can be operationalised for corpus-based modelling of cultural heritage discourse. Second, it provides a computational framework for linking aesthetic style with embodied linguistic structure. Third, it offers insight into how institutions construct perceptual experience for visually impaired audiences, framing audio description as a mode of embodied aesthetic mediation rather than mere informational transfer.