Abstract The article offers empirical foundations for machine text classification within the immediacy-distance continuum. By introducing an innovatively stratified collection of German texts from diverse media and communicative situations, we compute prototypical characteristics based on text-internal language properties. The application of a linguistically motivated feature set to supervised machine learning not only enables automatic classification but also produces interpretable explanatory models. Findings, and particularly methodological insights, should be transferable to other languages.
Abstract If the same in-game events are arrayed in different sequences, players may find some sequences to be more attractive, exciting, engaging, or intriguing than others. But how can such sequences be explained, understood, planned, compared, or rated? Do models exist that analytically describe various sequences? To address this question, this study seeks to identify conceptual models for the strategic ordering of game play events . Such models can describe, for instance, a pattern of increasing difficulty, a mix of positive and negative events, or phases of intense action broken up by pauses. The study identifies relevant exemplary contributions in the discourses of game design, design patterns, Dynamic Difficulty Adjustment, tabletop roleplay, literary drama, Flow theory, and Operant Conditioning theory. It is found that surprisingly few models exist. The predominant model appears to be repeated variations of rather simple experiential sequences of tension and relaxation on an overall increasing level, similar to the dramatic arc . The study’s main contribution to the narratology discourse is to show how the dramatic arc as a powerful conceptual model can inform the understanding of the strategic ordering of game play events.
Abstract Manchu, a low-resource language essential for understanding early modern Eastern Eurasian history, lacks effective optical character recognition (OCR) systems that can handle real-world historical documents. This study develops high-performing OCR systems by fine-tuning three open-source vision–language models (Llama-3.2-11B-Vision, Qwen2.5-VL-7B, and Qwen2.5-VL-3B) on 60,000 synthetic Manchu word images using parameter-efficient training. Llama-3.2-11B-Vision achieved exceptional performance with 98.6 percent word accuracy and a character error rate of 0.002 on synthetic data, while maintaining 87.4 percent word-level accuracy (95% CI: 85.1%–89.7%) on scanned word images from formal Qing administrative manuscripts and printed documents. Comparative evaluation reveals substantial advantages over traditional approaches: while a convolutional recurrent neural network baseline achieved 99.9 percent synthetic accuracy, it suffered severe degradation to 56.8 percent on scanned documents. Our approach demonstrates effective synthetic-to-real domain transfer, providing a cost-effective solution deployable on accessible infrastructure. This work establishes a transferable framework for endangered language OCR that reduces technical and financial barriers in digital humanities, enabling historians and linguists to process historical archives without specialized computing resources.
Abstract This study integrates multilingual text analytics with time-series modeling to examine how geopolitical narratives emerging around the 2022 Russia–Ukraine conflict align across Russian and Colombian media, drawing on a corpus spanning 2013–2023 that also encompasses the 2014 conflict period. Drawing on a corpus of more than 38,000 Spanish- and Russian-language news articles (2013–2023), we employ a six-stage pipeline combining multilingual NER, document-level sentiment scoring, e5-large-instruct embeddings, BERTopic clustering, and a semi-automated human-LLM labeling workflow. Narrative salience and tone were aggregated to weekly, log-normalized series and analyzed using linear regression and Granger causality tests (lags 1–4 weeks), providing a descriptive view of temporal coupling rather than evidence of direct causal influence. The period surrounding the 2022 invasion produced the clearest and most coherent patterns. Across four macro-narratives ( Security & Conflict , Diplomacy , Economy , and Politics & Society ), Russian and Colombian coverage exhibited sequential alignment in which shifts in narrative volume were often followed by shifts in evaluative tone. Thirty-two of forty-four target–topic pairs displayed significant lag structures prior to false-discovery correction ( p < 0.05 $p<0.05$ p less than 0.05 ), with 25 surviving Benjamini–Hochberg adjustment. Categories with lower geopolitical salience ( Technology , Health , and Off-topic ) displayed less consistent temporal coupling, though this contrast is offered as descriptive context rather than a formal control. While these results do not imply coordinated messaging, they indicate recurring forms of narrative mimicry , understood here as patterned convergence in framing and sentiment across media ecosystems. By demonstrating how transformer-based semantic representations can be integrated with human-in-the-loop interpretation and classical time-series analysis, this study contributes a reproducible workflow for tracing narrative trajectories at scale. The approach provides a methodological foundation for examining how geopolitical frames circulate, mutate, and align across linguistic and regional boundaries, offering new possibilities for computational humanities research on transnational discourse.
Abstract Colonial Korean printshops (1910–1945) operated within severe material constraints that produced apparent typographic standardization. Manual bibliography has nevertheless identified subtle shop-specific signatures in letterforms. This article tests whether computational methods can detect these signatures systematically across large corpora. We apply multiple instance learning to 57,583 page images from four major printshops in Kyŏngsŏng (Seoul). To avoid same-book leakage, we evaluate page classification with all pages from each tested book held out from training. Under this stricter protocol, the model assigns pages to printshops with 84.5 percent accuracy while producing interpretable attention maps. Clustering high-attention patches reveals that the four presses occupy the same regions of the embedding space, reflecting their shared typographic infrastructure. Differentiation emerges through density patterns within this shared space rather than through distinct typographic vocabularies. Each press concentrates disproportionately in clusters capturing specific sub-character features, such as terminal angles, serif weight, and stroke thickness. These computationally identified patterns align with features that De Fremery and Ryu documented through intensive manual analysis of individual volumes. The correspondence supports the approach while extending it to scales that manual bibliography cannot achieve. This framework shows how computational analysis can test and extend humanities scholarship on historical print culture by surfacing the specific visual features that distinguish institutional outputs within materially constrained production systems.
Abstract This study addresses the challenges of performing narratological analysis on low-resource languages, with a focus on Old Church Slavonic. Understanding the roles, interactions, and networks of persons is central to narrative analysis, yet such investigation is hindered by the scarcity of experts and the limited availability of annotated resources. We explore both established natural language processing (NLP) methods and large language models (LLMs) for analyzing pre-modern Slavic Lives of Saints, including several Slavic versions, the Greek original, and an English translation. Pre-modern Slavic texts pose particular difficulties due to rich morphology, orthographic variation, and limited standardization, which complicate the application of both traditional NLP tools and off-the-shelf LLMs. Through experiments using annotated and non-annotated ground truth data, we demonstrate that while conventional NLP methods often reach their limits on such low-resource, highly variable texts, LLMs provide complementary capabilities that can support narratological insights, especially in tracking persons and their interactions, albeit with important caveats regarding accuracy and coverage.
Abstract RUCAI is an AI-based open-source software platform designed for humanities and social sciences educational workflows that require traceability to assigned course materials. RUCAI has a teacher-only workspace in which teachers can curate class exercises grounded in curriculum documents (articles, book chapters, and notes). From here, the teacher can set up workspaces grounded explicitly in curriculum materials, learning outcomes, and course requirements for students to use throughout the course. The system is set up to run on institution-controlled infrastructure using FastAPI, PostgreSQL, and a local LLM model serving via Ollama, and is intended to be deployable on a GPU-enabled virtual machine. RUCAI’s primary contributions are practical: (1) it presents a software architecture for local-first, source-grounded educational AI that treats the course as the primary software object; (2) it documents concrete implementation choices for reliable ingestion and retrieval across heterogeneous course materials, including a three-tier extraction pipeline, word-based chunking, and intent-aware retrieval; and (3) it describes a publish-to-student mechanism that lets teachers freeze course context and prompt constraints into student-facing instances. The paper also reports pilot observations, current limitations, and future development directions.
Abstract Large-language models (LLMs) have transformed natural language processing and opened new possibilities for the computational social sciences and digital humanities. Yet translating historical sources remains difficult because early modern varieties are scarcely represented in contemporary training corpora and because standard tokenizers fragment their non-standard orthography. This article tackles these gaps by adapting open LLMs to early modern Dutch-to-English translation and advances two concrete contributions: (i) a memory-efficient fine-tuning workflow that runs on a single consumer GPU, comparing order-reward policy optimization with the Unsloth supervised fine-tuning approach and (ii) a verifiable evaluation protocol that combines embedding-based metrics with systematic expert review. Experiments on testimonial texts (1680–1792) show that fine-tuning choice decisively shapes quality: the Unsloth-tuned Mistral model attains the highest BERTScore and METEOR values and most faithfully preserves historical nuance. The framework supports a collaborative workflow where machine-generated drafts accelerate expert translation, making archival texts more accessible while maintaining scholarly oversight through domain-expert validation.
Abstract In recent years, digital humanities (DH) research has evolved from its textual origins to encompass film and video studies as critical areas of inquiry as well. Nevertheless, much of this research has remained tied to the formal levels of description most readily revealed by automatic processing. This maintains a gap between treatments in terms of formal technical features and the concerns of many researchers involved in film analysis of a more qualitative, interpretative nature, thereby reiterating the classic tension within DH as such: that is, how to relate levels of description that are “computable” and those more responsive to broader humanities-oriented interests. In this article, we set out an approach to this challenge that incorporates a multi-layered analytic framework capable of specifying increasingly abstract descriptions in terms of patterns at lower levels. This enables us to start bringing concerns of narrative organization and interpretation into analysis at scale. We set out the overall approach and show several examples of its use.
The ending speech in Aesopic fables, where stories conclude with direct utterances from characters, is not merely a didactic tool but a crucial narrative device constructing hermeneutic complexity. This study systematically examines the narrative function of ending speech through computational analysis of 600 Aesopic fables from Laura Gibbs’ edition. We quantitatively analyzed the complex relationships between ending speech, story content, explicit morals and speaker identity using natural language processing techniques. The analysis reveals three key findings. First, the average similarity of ending speeches (0.1820) is significantly lower than that of stories (0.3578), confirming that ending speech forms a unique semantic domain rather than serving as a simple summary of the narrative. Latent Dirichlet allocation analysis also shows that ending speeches are differentiated into 13 topics, displaying a more complex structure than stories (seven topics). Second, we found that ending speech constitutes a distinct narrative domain from epimythium, with an overwhelming ratio of their relationships being either independent (76.8%) or tensional (21.4%). This indicates that the ending speech is a narrative device that amplifies interpretive complexity, often clashing with the epimythium rather than reinforcing it. Third, 249 different ending speech speakers each represent unique voices and perspectives, with the frequency of utterances – fox (34 times), lion (19 times) and wolf (18 times) – demonstrating a value system in Aesopic fables where wisdom is prioritized over physical strength. These findings indicate that the ending speech establishes complex and sometimes tensional relationships with both story and epimythium, thereby transforming fables into “open work” that can be newly interpreted. This study provides empirical evidence for understanding Aesopic fables not as simple didactic tales but as complex narratives with structural features supporting polyphonic interpretation, demonstrating the potential of computational narratology.
This article implements a critical method for assessing bias in large historical datasets that we term the “Environmental Scan.” The Environmental Scan sheds new light on newspaper collections by linking newly available “reference metadata” gathered from historical sources to existing full-text and catalogue metadata. The rise of computational methods in history and the social sciences, in tandem with newly “datafied” source materials, creates a challenge for researchers to adapt their existing critical practices to the increasing scale and complexity of computational research. To help address this challenge, the Environmental Scan situates big historical datasets in much greater context, including estimating what materials are missing, thereby revealing the ways digital collections can be “oligoptic” in nature. Using the British Newspaper Archive (BNA) as a case study, we diagnose the biases and imbalances in the digitised Victorian press. We determine which voices are under- or over-represented in relation to the political composition of the collection as well as its content and we trace the origins of these biases in the digitisation process. This article informs future interdisciplinary discussions about data bias and offers a conceptual model adaptable to diverse historical datasets. The Environmental Scan provides a more nuanced and accurate understanding of how newspaper data reflects past societies, making it a valuable tool for researchers.
In this study, we perform a comprehensive evaluation of sentiment classification for German language data using three different approaches: (1) dictionary-based methods, (2) fine-tuned transformer models such as BERT and XLM-T and (3) various large language models (LLMs) with zero-shot capabilities, including natural language inference models, Siamese models and dialog-based models. The evaluation considers a variety of German language datasets, including contemporary social media texts, product reviews and humanities datasets. Our results confirm that dictionary-based methods, while computationally efficient and interpretable, fall short in classification accuracy. Fine-tuned models offer strong performance, but require significant training data and computational resources. LLMs with zero-shot capabilities, particularly dialog-based models, demonstrate competitive performance, often rivaling fine-tuned models, while eliminating the need for task-specific training. However, challenges remain regarding non-determinism, prompt sensitivity and the high resource requirements of large LLMs. The results suggest that for sentiment analysis in the computational humanities, where non-English and historical language data are common, LLM-based zero-shot classification is a viable alternative to fine-tuned models and dictionaries. Nevertheless, model selection remains highly context-dependent, requiring careful consideration of trade-offs between accuracy, resource efficiency and transparency.
In this article, we evaluate several large language models (LLMs) on a word-level translation alignment task between Ancient Greek and English. Comparing model performance to a human gold standard, we examine the performance of four different LLMs, two open-weight and two proprietary. We then take the best-performing model and generate examples of word-level alignments for further finetuning of the open-weight models. We observe significant improvement of open-weight models due to finetuning on synthetic data. These findings suggest that open-weight models, though not able to perform a certain task themselves, can be bolstered through finetuning to achieve impressive results. We believe that this work can help inform the development of more such tools in the digital classics and the computational humanities at large.
Texts, whether literary or historical, exhibit structural and stylistic patterns shaped by their purpose, authorship and cultural context. Formulaic texts, which are characterized by repetition and constrained expression, tend to differ in their information content (as defined by Shannon) compared to more dynamic compositions. Identifying such patterns in historical documents, particularly multi-author texts like the Hebrew Bible, provides insights into their origins, purpose and transmission. This study aims to identify formulaic clusters: sections exhibiting systematic repetition and structural constraints, by analyzing recurring phrases, syntactic structures and stylistic markers. However, distinguishing formulaic from non-formulaic elements in an unsupervised manner presents a computational challenge, especially in high-dimensional and sample-poor data sets where patterns must be inferred without predefined labels. To address this, we develop an information-theoretic algorithm leveraging weighted self-information distributions to detect structured patterns in text. Our approach directly models variations in sample-wise self-information to identify formulaicity. By extending classical discrete self-information measures with a continuous formulation based on differential self-information in multivariate Gaussian distributions, our method remains applicable across different types of textual representations, including neural embeddings under Gaussian priors. Applied to hypothesized authorial divisions in the Hebrew Bible, our approach successfully isolates stylistic layers, providing a quantitative framework for textual stratification. This method enhances our ability to analyze compositional patterns, offering deeper insights into the literary and cultural evolution of texts shaped by complex authorship and editorial processes.
This article explores the potential of large language models (LLMs), particularly through the use of contextualized word embeddings, to trace the evolution of scientific concepts. It thus aims to extend the potential of LLMs, currently transforming much of humanities research, to the specialized field of history and philosophy of science. Using the concept of the virtual particle – a fundamental idea in understanding elementary particle interactions – as a case study, we domain-adapted a pretrained Bidirectional Encoder Representations from Transformers model on nearly a century of Physical Review publications. By employing semantic change detection techniques, we examined shifts in the meaning and usage of the term “virtual.” Our analysis reveals that the dominant meaning of “virtual” stabilized after the 1950s, aligning with the formalization of the virtual particle concept, while the polysemy of “virtual” continued to grow. Augmenting these findings with dependency parsing and qualitative analysis, we identify pivotal historical transitions in the term’s usage. In a broader methodological discussion, we address challenges such as the complex relationship between words and concepts, the influence of historical and linguistic biases in datasets, and the exclusion of mathematical formulas from text-based approaches.
This article introduces a strategy for the large-scale corpus analysis of music audio recordings, aimed at identifying long-term trends and testing hypotheses regarding the repertoire represented in a given corpus. Our approach centers on computing evolution curves (ECs), which map style-relevant features, such as musical complexity, onto historical timelines. Unlike traditional approaches that rely on sheet music, we use audio recordings, leveraging their widespread availability and the performance nuances they capture. We also emphasize the benefits of pitch-class features based on deep learning, which improve the robustness and accuracy of tonal complexity measures compared to traditional signal processing methods. Addressing the frequent lack of exact work dates (year of composition) in historical corpora, we propose a heuristic method that aligns works with timelines using composers’ life dates. This method effectively preserves historical trends with minimal deviation compared to using actual work dates, as validated against available metadata from the Carus Audio Corpus, which spans 450 years of choral and sacred music and contains 5,729 tracks with detailed metadata. We demonstrate the utility of our strategy through case studies of this corpus, showing how ECs provide insights into stylistic developments that confirm expectations from musicology, thus highlighting the potential of computational studies in this field. For example, we observe a steady increase in tonal complexity from the Renaissance through the Baroque period, stable complexity levels in the 19th and 20th centuries, and consistently higher complexity in minor-key works compared to major-key works. Our visualizations also reveal that vocal music was more complex than instrumental music in the 18th century, but less complex in the 20th century. Finally, we conduct comparative analyses of individual composers, exploring how historical and biographical contexts may have influenced their works. Our findings highlight the potential of this strategy for computational corpus studies in musicological research.
