The exploration and retrieval of information from large, unstructured document collections remain challenging. Unsupervised techniques, such as clustering and topic modeling, provide only a coarse overview of thematic structure, while traditional keyword searches often require extensive manual effort. Recent advances in large language models and retrieval-augmented generation (RAG) introduce new opportunities by enabling focused retrieval of relevant documents or chunks tailored to a user’s query. This allows for dynamic, chat-like interactions that streamline exploration and improve access to pertinent information. This article introduces Topic-RAG, a chat engine that integrates topic modeling with RAG to support interactive and exploratory document retrieval. Topic-RAG uses BERTopic to identify the most relevant topics for a given query and restricts retrieval to documents or chunks within those topics. This targeted strategy enhances retrieval relevance by narrowing the search space to thematically aligned content. We utilize the pipeline on 4,711 articles related to nuclear energy from the Impresso historical Swiss newspaper corpus. Our experimental results demonstrate that Topic-RAG outperforms a baseline RAG architecture that does not incorporate topic modeling, as measured by widely recognized metrics, such as BERTScore (including Precision, Recall and F1), ROUGE and UniEval. Topic-RAG also achieves improvements in computational efficiency for both single and batch query processing. In addition, we performed a qualitative analysis in collaboration with domain experts, who assessed the system’s effectiveness in supporting historically grounded research. Although our evaluation is focused on historical newspaper articles, the proposed approach more generally integrates topic information to enhance retrieval performance within a transparent and user-configurable pipeline effectively. It supports the targeted retrieval of contextually rich and semantically relevant content while also allowing users to adjust key parameters such as the number of documents retrieved. This flexibility provides greater control and adaptability to meet diverse research needs in historical inquiry, literary analysis and cultural studies. Due to copyright restrictions, the raw data cannot be publicly shared. Data access instructions are provided in the repository, and the replication code is available on GitHub: https://github.com/KeerthanaMurugaraj/Topic-RAG-for-Historical-Newspapers.

Location mentions in local news are crucial for examining issues like spatial inequalities, news deserts and the impact of media ownership on news diversity. However, while geoparsing – extracting and resolving location mentions – has advanced through statistical and deep learning methods, its use in local media studies remains limited and fragmented due to technical challenges and a lack of practical frameworks. To address these challenges, we identify key considerations for successful geoparsing and review spatially oriented local media studies, finding over-reliance on limited geospatial vocabularies, limited toponym disambiguation and inadequate validation of methods. These findings underscore the need for adaptable and robust solutions, and recent advancements in fine-tuned large language models (LLMs) for geoparsing offer a promising direction by simplifying technical implementation and excelling at understanding contextual nuances. However, their application to U.K. local media – marked by fine-grained geographies and colloquial place names – remains underexplored due to the absence of benchmark datasets. This gap hinders researchers’ ability to evaluate and refine geoparsing methods for this domain. To address this, we introduce the Local Media UK Geoparsing (LMUK-Geo) dataset, a hand-annotated corpus of U.K. local news articles designed to support the development and evaluation of geoparsing pipelines. We also propose an LLM-driven approach for toponym disambiguation that replaces fine-tuning with accessible prompt engineering. Using LMUK-Geo, we benchmark our approach against a fine-tuned method. Both perform well on the novel dataset: the fine-tuned model excels in minimising coordinate-error distances, while the prompt-based method offers a scalable alternative for district-level classification, particularly when relying on predictions agreed upon by multiple models. Our contributions establish a foundation for geoparsing local media, advancing methodological frameworks and practical tools to enable systematic and comparative research.

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.

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.

Language models have the ability to identify the characteristics of much shorter literary passages than was thought feasible with traditional stylometry. We evaluate authorship and genre detection for a new corpus of literary novels. We find that a range of LLMs are able to distinguish authorship and genre, but that different models do so in different ways. Some models rely more on memorization, while others make greater use of author or genre characteristics learned during fine-tuning. We additionally use three methods – direct syntactic ablation of input text and two means of studying internal model values – to probe one high-performing LLM for features that characterize styles. We find that authorial style is easier to characterize than genre-level style and is more impacted by minor syntactic decisions and contextual word usage. However, some traits like pronoun usage and word order prove significant for defining both kinds of literary style.

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.