This research investigates the marginalizing effects of global digitalization and urbanization on local dialects through the metaphor of "Lost Frequencies." Focusing on the Maraş dialect, the study explores the intersection between the rupture in generational cultural transmission and the failure of digital systems (ASR) to recognize local linguistic codes. Theoretically, the paper argues that language is a primary vehicle for cultural identity, and its loss implies a weakening of social status and defense mechanisms (Arslan et al. 2024). Urbanization processes further demonstrate how language use shifts according to changing behavioral norms and rituals (Heinrich and Smakman 2018, 2), while younger generations increasingly perceive dialects as barriers to social mobility within urban spaces (Nortier and Svendsen 2015, 1). Additionally, the rise of social media fosters new linguistic forms that further disconnect traditional speech from formal digital recognition (Rianto and Juliarta 2024, 1).

Methodologically, the study proposes a framework leveraging the YouTube archives of Aksu TV (specifically the Hayat Ağacı and Maraş Konuşuyor series) as an unstructured data pool. The core output is a "Linguistic Error Taxonomy" that identifies phonetic, morphological, and semantic disconnections where algorithms fail to process non-standard variants. The study concludes that transforming these archives into a structured Digital Corpus is essential for the digital sustainability of local memory. By bridging the gap between traditional narratives and algorithmic recognition, this research provides a roadmap for preserving intangible cultural heritage in the digital era.

This paper introduces community-centred approaches to curating knowledge about cultural artefacts outside and beyond the Global North’s hegemony over knowledge ecologies and socio-technical infrastructures. I will argue that we, as scholars, have to “Do Cultural Her- itage” ourselves, while acknowledging the constrains and affordances we operate with. This is inspired by minimal computing, which requires us to address “what do we need?” with the answers to “what do we have at hand?”, while acknowledging the destruction of spaceship earth and thus keep things as simple as possible.The backdrop to this argument forms a concrete case study introducing the processes, conceptual framework, and experiences we made contributing bibliographic metadata for nearly 4,000 Arabic and Ottoman periodicals— including detailed information on more than 2,700 editors and approximately 350 holding institutions—to Wikidata. We thereby show that all the necessary components for implementing a functional, fully multilingual, discovery system serving five-star Linked Open Data are currently available and accessible for free to anyone possess- ing the means to connect to the internet

Rozwój dużych modeli generatywnych (LLM) uwypuklił problem deficytu danych dla niskozasobowych odmian językowych (low-resource languages). W wystąpieniu dokumentujemy proces budowy specjalistycznych zasobów dla lektu mazurskiego w ramach projektu infrastrukturalnego CLARIN-PL-BIZ-Bis (CLARIN-PL). Celem podjętych prac jest nie tylko ocalenie dziedzictwa językowego, ale przede wszystkim stworzenie rzetelnej platformy testowej i ewaluacyjnej dla nowoczesnych technologii językowych.

Istotą prezentowanego przez nas rozwiązania jest techniczny potok przetwarzania danych (data pipeline), obejmujący unikatową, czterostopniową transformację tekstów (zarówno drukowanych, jak i transkrypcji mowy):

- Zbiór surowy (zachowanie oryginalnej pisowni i artefaktów zapisu);

- Zbiór ujednolicony (normalizacja wewnątrz zbioru w celu eliminacji szumu ortograficznego);

- Zbiór normatywny (pełna korekta merytoryczna uwzględniająca systemowe cechy lektu mazurskiego);

- Zbiór translacyjny(równoległe tłumaczenie na język polski standardowy).

Efektem końcowym będą zbiory dwujęzyczne zrównoleglone w standardzie TMX, wzbogacone o pełną anotację morfologiczną i zaimplementowane w systemie KonText. Podczas wystąpienia omówimy również rolę wydzielonej części tych zasobów jako benchmarków w ewaluacji wydajności modeli LLM (w zakresie NLU). Przygotowane wielowariantywne dane mogą znaleźć zastosowanie do analizy progów jakościowych danych w uczeniu transferowym (transfer learning) na polskich danych tekstowych.

Named Entity Recognition (NER) is a core task in Natural Language Processing and Digital Humanities, yet its performance remains uneven for low-resourced languages and domain-specific genres. While English benefits from literary NER resources such as LitBank, Romanian lacks annotated datasets tailored to narrative fiction. This study addresses this gap by investigating domain-specific NER for Romanian fairytales, a challenging genre characterized by magical beings, symbolic entities, and fictional spaces that exceed standard annotation schemes.

We introduce FairyNERo, a manually curated gold-standard corpus of 101 open-access children’s fairytales (approximately 74,000 words). To capture genre-specific features, we developed a domain-adapted ontology extending the traditional PER, LOC, and TIME tags used in CoNLL. A hybrid human-in-the-loop workflow combined automatic pre-annotation with manual correction via a custom interface. Inter-annotator agreement reached 94.6%, indicating strong consistency.

We compared three approaches: (1) applying the general Romanian NER model RONEC, (2) fine-tuning a transformer model based on BERT-base-ro, and (3) prompting Microsoft Copilot in zero-shot and few-shot settings. The off-the-shelf model performed poorly due to domain mismatch. Fine-tuned BERT achieved the best results, with a strict span-level F1 score of 0.92 under CoNLL evaluation, demonstrating the effectiveness of domain adaptation. Large language model prompting outperformed the baseline but remained inferior to supervised fine-tuning.

Our findings highlight that ontology design and manual annotation are interpretive acts encoding genre knowledge essential for high-quality NER. By integrating ontology-driven annotation and transformer fine-tuning, this study provides the first Romanian fairytale NER resource and a replicable framework for domain-specific NER in low-resourced languages.

We report on the morphological tagging of Old Serbian in the Universal Dependencies framework. To facilitate the manual annotation, we pre-processed the data with the Old Church Slavonic 2.12 UDPipe model. The decision was based on the known similarity of these two languages as well as on the declared performance of this model compared to other models for historical varieties of Slavic languages. With over 3,000 manually annotated tokens, we evaluated the performance of the relevant pre-trained UDPipe2 models of historical Slavic languages. Besides, we also trained and evaluated custom models with UDPipe1 containing the annotated Old Serbian data. We have found that: (1) for this particular domain and amount of training data, the most suitable model is UD Old East Slavic – Birchbark 2.12, although its declared performance is much lower than that of Old Church Slavonic; (2) even 3,000 tokens of Old Serbian increase the performance of UDPipe1 models almost to the level of the Birchbark 2.12 model.

The Middle-Persian Corpus and Dictionary (MPCD) is a collaborative research infrastructure developed by the Institute of Iranian Studies (Freie Universität Berlin), the Center for Religious Studies (CERES, Ruhr-Universität Bochum), and the Cologne Center for eHumanities (CCeH) at the University of Cologne, funded as long-term project by the German research foundation (Deutsche Forschungsgemeinschaft, DFG). The project aims to create a corpus-driven dictionary of Zoroastrian Middle Persian that integrates lexicographic, morphosyntactic, and semantic analysis with full manuscript context. Central to the project is an agile, user-centered development process in which philologists act as primary stakeholders, ensuring that technical design decisions directly support concrete research questions.

The philological research conducted within MPCD addresses fundamental issues in Middle Persian studies, including the interpretation of heterograms in Book Pahlavi (Cantera 2023), verbal morphology and passive constructions (Cantera 2024), and the interaction between Avestan source texts and their Middle Persian Zand interpretations (Musavi 2024). A substantial part of the corpus consists of morphosyntactically and semantically annotated editions of Zoroastrian texts such as Dādestān ī dēnīg, Wizīdagīhā ī zādspram, Greater Bundahišn, and several Riwāyats, often represented by multiple manuscript witnesses (Jügel 2023, 2024, 2025; Musavi 2024; Colditz 2023, 2024). These annotated texts serve as the empirical basis for research on syntactic structure, lexical semantics, textual variation, and the transmission of religious and legal knowledge in Zoroastrian communities from the Sasanian to the early Islamic period (Colditz 2023; Rezania 2023, 2024).