Category: Fréttir

The recording is available here.

Alexander Peter Pfaff, University of Iceland
How Many Ways Can You Say It? Measuring Language Diversity with Digital Methods

Understanding how languages vary in the way they build expressions is essential for studying texts across periods and languages, yet traditional descriptions capture only a fraction of this diversity. Patternization is an approach that combines traditional text analysis and corpus linguistics with mathematical methods into a Python tool that allows us to explore such syntactic variation: it treats phrases as sequences of category labels and compares the patterns that actually appear with the full range of patterns that could appear. This task is hardly feasible if performed manually, illustrating how digital methods can reveal nuances otherwise invisible to the human eye.

The recording is available here.

Eiríkur Smári Sigurðarson, University of Iceland, MSHL
Tengjum allt við allt!

Hvernig getum við breytt einangruðum gagnasöfnum í lifandi og samtengdan þekkingarvef? Í þessu erindi er fjallað um tækifæri sem verða til þegar menningarleg gögn eru stafræn og tengd með merkingarfræðilegri tækni. Með því að nýta aðferðafræði tengdra gagna (e. Linked Data) má rjúfa múra á milli ólíkra stofnana og verkefna, þannig að upplýsingar flæði á milli kerfa.

Með innblæstri frá hinu finnska Sampo-módeli, skoðum við hvernig hægt er að smíða stafrænt vistkerfi þar sem handrit, myndlist, örnefni og sögulegar persónur mætast í einu vistkerfi. Markmiðið er að hætta að líta á gögn sem stakar eyjar og byrja að tengja allt við allt. Við ætlum ekki bara að varðveita menninguna, heldur gera hana aðgengilega og gagnvirka.

The recording is available here.

Trausti Dagsson, Árni Magnússon Institute for Icelandic Studies
Ísmús: Víðfeðmur vefur þjóðsagna, tónlistar og radda fortíðarinnar

Vefurinn Ísmús er samstarfsverkefni Árnastofnunar og Tónlistarsafns Íslands sem nú er hluti af Landsbókasafni. Á vefnum má nálgast upptökur úr þjóðfræðisafni Árnastofnunar ásamt gríðarmiklum upplýsingum um tónlistarsögu Íslands. Í fyrirlestrinum verður fjallað um hvernig önnur stór gagnasöfn hafa runnið saman við Ísmús og þannig skapað víðfeðmt yfirlit yfir þjóðfræði, sagnir, ævintýri, alþýðukveðskap, þjóðlög og hljómsveitir. Einnig verður sýnt hvernig sjálfvirkri talgreiningu var beitt á upptökur í þjóðfræðisafninu og hvernig önnur rafræn söfn tengjast Ísmús–eins og Nafnið.is, Bækur.is og Handrit.is.

The lecture will be postponed to autumn 2026.

Ondřej Tichý, Charles University, Prague
Benchmarking Large Language Models for DH Research

Recent advances in Large Language Models (LLMs) have shown that they can increasingly replace traditional NLP techniques across a range of linguistic tasks, such as part-of-speech tagging or orthographic normalization, and can even approach or surpass human annotators in tasks such as speech-act classification or genre annotation. However, the results of previous studies are often difficult to replicate, given the wide range of factors influencing LLM output, its inherent stochasticity, and the breakneck lifecycle of the individual models. Consequently, state-of-the-art results, their systematic comparison and generalization remain challenging.

In this paper, I propose a set of guidelines and Python scripts designed to make benchmarking LLMs on linguistic tasks more accessible, reproducible, and comparable. I also conduct several benchmarking experiments using this methodology to validate it and to identify best practices for applying LLMs to linguistic research, as well as to determine which current models perform best in specific tasks.

The proposed guidelines define input and output data structures for a variety of linguistic tasks, recommend parameter settings (e.g. temperature, top_p, chain-of-thought reasoning, retrieval-augmented generation), and outline how to interpret outputs such as self-reported confidence scores and token-level probabilities. The accompanying scripts enable researchers to (re)run tests on new tasks or new models and to generate comparable reports. While model fine-tuning is outside the scope of our framework, we support both zero- and few-shot prompting, allowing users to provide ground-truth data for evaluation and, optionally, as few-shot examples.
In my own tests, I will focus on tasks that have not been largely solved by NLP (avoiding e.g. PoS tagging in English) and that are commonly performed by empirical and more specifically corpus linguists. While most tasks target Present-Day English, I also investigate how LLMs handle low-resource languages and non-standardized varieties by including Czech and earlier stages of English. The selected tasks range from morphological and syntactic classification (e.g. identifying nominal number in Old English or the syntactic role of non-finite verbs in Present-Day English) to pragmatic annotation (e.g. contextual functions of like), semantic disambiguation, and historical spelling normalization.

I benchmark both major commercial models (e.g. ChatGPT, Gemini, Claude) and leading open-source or smaller models (e.g. gpt-oss, LLaMA, DeepSeek, Mistral), including different quantizations and configurations (leveraging the resources of the e-infra.cz research infrastructure). This enables evaluation not only of their performance but also of factors such as cost, accessibility, and data security, as well as testing claims such as smaller models outperforming larger ones on simple binary classifications.

My preliminary findings suggest that some linguistic tasks, such as text normalization and basic morphological classification, are already well-suited to LLM applications. In contrast, more complex tasks requiring extensive context and elaborate hierarchical categorization, such as discourse and pragmatic annotation, still fall significantly short of human performance.

The recording is available here.

Eydís Huld Magnúsdóttir, Tiro ehf.
Fjársjóðsleit með máltækni

Hvernig er það, er ekki hægt að nýta máltækni og gervigreind til að yfirfæra tal sem finnst á hljóðupptökum fortíðarinnar yfir í leitarbæran texta? Jú enimitt, máltæknifyrirtækið Tiro hefur tekið þátt í verkefnum af því tagi með Árnastofnun og RÚV. Ávinningurinn fyrir þessa aðila hefur verið mikill og jafnvel meiri en búist var við í upphafi. Svo er aldrei að vita nema að það finnist áður óþekktir fjársjóðir inná milli?

The recording is available here.

Timothy Liam Waters, Ohio State University
Virtual Vikings: A Case of Identity Formation and Techno-Skepticism in the in the Digital Vernacular

Debates about generative artificial intelligence often frame it as a threat to human creativity. This paper argues instead that AI extends a longer history of hybrid vernacular expression in digital culture. Focusing on AI-generated and AI-suspect images of the Viking Age, it examines how online communities negotiate the epistemic status of visual media. These disputes center on legitimacy rather than accuracy: what counts as authentic, what counts as AI, and who has the authority to decide. I identify three claimant groups, each advancing distinct epistemologies. Generative AI emerges as an epistemic flashpoint that destabilizes existing criteria of authenticity and reshapes how vernacular authority is produced online.

The recording is available here.

Sigríður Regína Sigurþórsdóttir, University of Iceland & National Gallery of Iceland
Variability, Iterability, Re-presentation: Time Based Media Art and Archival Research at the National Gallery of Iceland

Since the turn of the last century, conservators specialising in time‑based media (video, audio, digital, performance, etc) have grappled with the many practical, ideological, and ethical questions raised by the nature of technological development. The challenge is not new, but artists, museums, and conservators continually face new issues as artworks are created with ever‑changing technologies. What meaning‑making is embedded in a curator’s decision‑making regarding the technical aspects of installing these works and what influences these decisions? What lessons can be drawn from the challenges posed by obsolete media? Can we apply those lessons to works created with today’s newest technologies?

The recording is available here.

Hannes Högni Vilhjálmsson, Reykjavík University
Experiencing another time and place in 3D: The Historical Postal Route project

I will present a project bringing to life a historic postal route from Vopnafjörður in the northeast of Iceland to Kelduhverfi in the north. Over a century ago, postal carriers on horses would travel this route of over 200 km along coasts, over rugged highlands, across rivers and through narrow mountain passes. The sheer scale of the journey has been hard to envision, which is why the founders of the postal history museum at Ásbrandsstaðir in Vopnafjörður, approached us with the idea of recreating it in interactive 3D. We had been working on technology for turning readily available geological and natural data into interactive experiences and decided to accept this challenge. The project was delivered by the end of April, and this will be the first public glimpse of the results.

The recording is available here.

Paola Peratello, École nationale des chartes – PSL, Paris
From Script to Data: Automating the Reading of Old Swedish Texts

As part of the ANR-funded project „Digital Approaches to the Survival and Loss of Old Norse Romances“ (PI: CPJ Katarzyna Anna Kapitan), this presentation aims to introduce the results in the development of the first ATR model for Old Swedish handwritten texts, dating from the early 14th to the mid-16th century, within the eScriptorium environment. Building on Bifrost, a CC BY-licensed HTR model for Old Norse, the work involves the production of ground truth data using public domain digital surrogates from manuscripta.se and Handrit.is. The presentation addresses the methodological challenges encountered, ranging from script variation and transcription conventions to the construction of evaluation benchmarks from scratch, and reflects on best practices for ATR in the broader landscape of medieval Scandinavian studies.