This paper presents recent advances toward an integrated framework that combines TEI-based digital publishing with embedding-based semantic search to support the preservation, exploration and analysis of Holocaust survivor testimonies. The corpus includes written and oral sources and preserves them within a XML-TEI model supported by an ODD customization that preserves provenance, structure and interpretability. A dedicated web application developed within the eXistdb platform provides guided access to the digital corpus and supports the management, visualization, and exploration of the encoded data. The project aims to investigate a specific research goal: to verify the presence of references to the Divine Comedy by Dante within Holocaust testimonies. To this end, we implement a semantic retrieval component based on SentenceTransformers’ embeddings and a vector database, enabling the discovery of both literal and non-literal Dantean passages within the testimonies. The paper presents the advances achieved toward this objective and the ethical constraints shaping access policies, resulting in a sustainable archive and a reproducible methodology for intertextual research in sensitive historical collections.

Digital humanities projects increasingly rely on heterogeneous collections of multimodal data, including video testimonies, scanned documents, and photographs. Despite the growing availability of such archives, researchers face challenges in efficiently locating relevant content due to the diversity of formats and the lack of unified retrieval methods. In this work, we present a general framework for semantic search over collections of multiple modalities. The framework integrates specific parsers and transforms all inputs into textual representations leveraging services like automatic speech recognition (ASR), optical character recognition (OCR), and generative-AI-based image captioning. Text is subsequently segmented into overlapping chunks, indexed in a vector database, and enriched through an automatic question generation (AQ) pipeline to create ground-truth queries for evaluation. We evaluate the framework on a constructed dataset derived from Holocaust-related archives, comparing two retrieval strategies (pure vector search vs. hybrid semantic-lexical search) under two chunking scenarios. Results demonstrate that hybrid search consistently outperforms vector-only retrieval, achieving high recall across modalities, and that semantic search is feasible even with diverse and noisy input sources. This framework provides a robust foundation for exploring complex multimodal archives, facilitating access to content that would otherwise remain difficult to discover.

The digitization and computational processing of Holocaust testimony interviews are essential for the long-term preservation and accessibility of survivors’ narratives. However, automatic speech recognition (ASR) for Yiddish—the primary language of most Holocaust victims and survivors—remains underdeveloped. This paper introduces the first ASR system for European Yiddish, focused on the Northeastern (“Lithuanian”) dialect and trained on Holocaust survivor testimonies from the Corpus of Spoken Yiddish in Europe (42 hours of speech segments from 60 survivors). A systematic comparison of CTC-based ASR models using transcripts with different orthographic representations reveals that a Hebrew-based phonemic system with precomposed Unicode is optimal, achieving a mean WER of 37.96% compared to 59.40% WER for romanized Yiddish and 99.67% WER (catastrophic failure) for standard Yiddish spelled with decomposed Unicode. Cross-domain testing on Yiddish audiobooks provides additional support for a phonemic representation (27.07% WER, 6.56% CER). Together, the results suggest that automatic transcription developed from oral Holocaust testimonies can support further technological innovation in service of Yiddish-speaking communities.

Polarity detection becomes substantially more challenging under domain shift, particularly in heterogeneous long-form narratives with complex discourse structure, such as Holocaust oral histories. This paper presents a corpus-scale diagnostic study of off-the-shelf sentiment classifiers on Holocaust oral histories, using three pretrained transformer-based polarity classifiers over a corpus comprising 107,304 utterances and 579,013 sentences. After assembling model outputs, we introduce an agreement-based stability taxonomy (ABC) to stratify inter-model output stability. We report pairwise percent agreement, Cohen’s κ, Fleiss’ κ, and row-normalized confusion matrices to localize systematic disagreement. As an external convergent descriptive signal, we apply a T5-based emotion classifier to stratified samples from each agreement stratum to compare emotion distributions across strata. The combination of multi-model label triangulation and the ABC taxonomy provides a cautious, interpretable framework for characterizing where and how sentiment models diverge in sensitive historical narratives. Inter-model agreement is low to moderate overall and is driven primarily by boundary decisions around neutrality.

This paper presents the EHRI Annotator, a web-based tool for multilingual named entity recognition (NER) and entity linking (EL) in Holocaust-related texts. The tool was developed to support services provided by the European Holocaust Research Infrastructure (EHRI), primarily the digital scholarly editions published by EHRI (EHRI Online Editions) by streamlining the process of detecting named entities in documents and linking them to their unique identifiers in EHRI and third-party controlled vocabularies and gazetteers. The EHRI Annotator builds upon previous work on domain-specific NER, taking it a step further to support multilingual EL. The tool adopts a dual entity linking architecture that uses a different matching approach depending on the type of the named entity. It performs semantic matching for entities to be linked to EHRI vocabularies and authority sets which are modestly sized, and string-matching-based retrieval for locations to be linked to the extensive GeoNames gazetteer using a domain-specific relevance weighting. A preliminary evaluation on 264 entities from a manually annotated dataset of Holocaust testimonies yields an Accuracy@5 of 77.7% when it comes to the linking component of the tool. User testing confirms the tool’s usability but also highlights areas for improvement.

Researchers in Holocaust studies have often distinguished between two styles of oral survivor testimony: the USC Shoah Foundation’s interviews tend to follow a structured, interviewer-guided format, whereas the Yale Fortunoff Video Archive generally favors a more free-form, open-ended style. This distinction has influenced both scholarly research and the development of later archives. In this study, we critically examine that claim by conducting a large-scale computational analysis of more than 1,600 testimonies from both collections. Leveraging discourse segmentation, topic modeling, and large language model (LLM) based analysis, we quantify the “structuredness” level of testimonies through topic coherence, interviewer–survivor dynamics, and the distribution of question types. Our results generally corroborate the structural differences identified in earlier research, while also revealing significant overlaps between the collections, both within individual interviews and across common narrative patterns. This complicates the simple “structured vs. free-form” dichotomy often applied to these oral histories. Beyond revisiting a foundational claim in Holocaust studies, our work provides a scalable, replicable framework for comparative corpus analysis. As a proof of concept, it suggests broader applications for digital oral history, narrative analysis, and the design of citizen-science annotation platforms.

We present MalachNER, a new multilingual dataset for Named Entity Recognition (NER) in testimonies of Holocaust survivors. MalachNER has been sourced from different archives and annotated based on comprehensive domain-specific guidelines refined by a collaboration of international experts. Covering 10 European languages, differs significantly from previously released datasets: It is primarily based on noisy, verbatim transcribed speech, rather than on digitized written documents. These transcripts are characterized, among other challenges, by fillers, dialectal speech, and in-line annotations indicating incomprehensible words, which are not commonly encountered in other datasets. However, large volumes of yet unprocessed oral history make such a dataset a necessity. In addition to the description of the dataset and its annotation guidelines, we show with baseline experiments that MalachNER is complementary with previously released data, and the key to training domain-specific language models that generalize well to written and oral testimony alike, achieving state-of-the-art performance on both types of documents.

Video interviews with Holocaust survivors and witnesses comprise, to date, the most globally distributed and comprehensive oral history documentation. As survivors among us disappear, these sources are increasingly important to understand the impact of the Holocaust and mechanisms to overcome the trauma experienced. While historians often rely on written transcripts, these omit emotional nuances conveyed through audiovisual cues such as facial expressions, pauses, and eye movements. This article outlines the resources, data-preparation steps, and analytical methods used during a 10-day Digital Humanities Hackathon project to examine emotions in Holocaust testimonies, incorporating video, audio, and text. The group aimed to determine whether audiovisual signals offer meaningful emotional or sentimental information beyond transcripts. To achieve this, the group worked with a sample of 10 interviews facilitated by the US Holocaust Memorial Museum (USHMM); which were separated into video, audio, and textual components for machine processing and realigned side-by-side for analysis. This resulting “cookbook” lays out a workflow, resources, and practical entry points for preparing oral history interviews for multimodal emotion and sentiment annotation, or to aid the detection of emotionally significant moments for deeper examination.

Documents related to the Holocaust increasingly move into the focus of Natural Language Processing research, including the digitization of written text, the automatic transcription of oral archives, and interpretive downstream tasks such as Named Entity Recognition. However, most modern language models are trained primarily on modern text, and thus struggle with historical language, historical entities, and domain-specific terminology. Furthermore, transcribed speech introduces challenges such as transcription errors, noise, filler words, and dialectal speech not often contained in textual datasets. We present XLM-RoBERTa-malach, a text encoder domain-adapted to oral testimonies of Holocaust survivors in seven languages. In addition to descriptions of the data acquisition via Automatic Speech Recognition, data augmentation via Machine Translation, and the continued pretraining of a state-of-the-art multilingual transformer, we evaluate the domain-adapted model on the Named Entity Recognition task. Experiments on this task show superior performance over the general-domain transformer in a multilingual domain-specific setting, including languages not seen during the domain adaptation.

Holocaust testimonies are key primary sources documenting survivors’ experiences, yet many remain inaccessible due to the labor-intensive nature of manual transcription. This paper presents a comprehensive evaluation of OpenAI’s Whisper automatic speech recognition (ASR) system on 1,847 testimonies from the Fortunoff Video Archive for Holocaust Testimonies at Yale University. We assess transcription quality across multiple languages including English, French, German, Hebrew, Yiddish, Ladino, Slovak, and American Sign Language (with English voice-over), using human-reviewed captions as ground truth. Our analysis reveals a mean Word Error Rate (WER) of 15.28%, with 90.9% of testimonies achieving “Fair” or better quality (WER ≤25%). We identify systematic error patterns including challenges with disfluencies, interrupted speech, and language-specific orthographic conventions, particularly in Ladino, where Whisper’s normalization to modern Spanish orthography creates systematic divergences from traditional Judeo-Spanish spelling. For Hebrew and Yiddish, we evaluate specialized models from ivrit-ai and find promising results for heritage language preservation. Our findings demonstrate that current ASR technology can substantially accelerate Holocaust testimony transcription while highlighting the need for domain-specific fine-tuning and post-processing for optimal results.

Large-scale corpora of Holocaust testimonies preserve vast amounts of historical, emotional, and narrative information, but their size and complexity can make accurate, systematic analysis challenging. This paper presents a cross-modal computational analysis of emotional and thematic trajectories in the CORHOH corpus, containing 500 Holocaust survivor testimonies as a language resource for computational analysis. We segment each testimony into ten segments and apply sentiment analysis, emotion recognition, and topic modeling to each of these segments to reveal how theme and emotion evolve over time in Holocaust testimonies. Results reveal a sharp decline from pre-war life in wartime and camp experiences, with sentiment and emotion remaining negative in post-war segments. Emotion analysis reveals decreasing joy and increasing sadness and fear during segments related to deportation and concentration camps, with limited emotional recovery. Topic modeling identifies coherent themes that align closely with sentiment and emotional patterns. We systematically examine correlations between sentiment, emotion, and topic trajectories, which demonstrate many strong associations between topic and emotion. This work demonstrates that combining sentiment analysis, emotion recognition, and topic modeling can reveal systematic patterns in large oral history corpora, and shows the value of computational approaches for studying historical narratives like the Holocaust.