• Research dataclose
• Other research productsclose
• Open Access close
• NARCIS close
• Digital Humanities and Cultural Heritage close

Past work in natural language processing interpretability focused mainly on popular classification tasks while largely overlooking generation settings, partly due to a lack of dedicated tools. In this work, we introduce Inseq, a Python library to democratize access to interpretability analyses of sequence generation models. Inseq enables intuitive and optimized extraction of models' internal information and feature importance scores for popular decoder-only and encoder-decoder Transformers architectures. We showcase its potential by adopting it to highlight gender biases in machine translation models and locate factual knowledge inside GPT-2. Thanks to its extensible interface supporting cutting-edge techniques such as contrastive feature attribution, Inseq can drive future advances in explainable natural language generation, centralizing good practices and enabling fair and reproducible model evaluations.

In this blog post, we reflect on the use of the Media Suite for multimodal storytelling practices in the international classroom. The CLARIAH Media Suite combines Dutch media data collections and makes audiovisual material from Dutch audiovisual collections more accessible for higher education (CLARIAH Media Suite - What is the CLARIAH Media Suite?, n.d.). From October 2021 until January 2022, we explored the use of the Media Suite together with a multicultural, interdisciplinary group of MA students with a background in media creation and innovation at the University of Groningen. The students used content from audiovisual collections in the Media Suite for creative storytelling practices. In this subject tutorial, we provide insights on how the Media Suite can be implemented for multimodal storytelling usage in the international classroom. Project: Making Multimodal Storytelling Experiences with Cultural Heritage Materials in the International Classroom (MAKE) - Media Suite Teaching Fellowship CLARIAH-PLUS (2019-2023) PI/Principal Investigator: dr. Berber Hagedoorn, Assistant Professor Media Studies & Audiovisual Culture (RUG), b.hagedoorn@rug.nl Project Period: October 1, 2021 to 1 July 2022

This is part 4 of the Rchon statistics course. It continues the basics of statistical testing in R. In this tutorial, we will treat the following statistical testing methods: Mann-Whitney test Kruskal-Wallis test Kolmogorov-Smirnov test Follow the instructions in Instructions Tutorial 4.pdf to start the tutorial. This course was originally created for Archon Research School of Archaeology by Philip Verhagen (Vrije Universiteit Amsterdam) and Bjørn P. Bartholdy (University of Leiden), and consists of an instruction, a tutorial, a test and two datafiles. All content is CC BY-NC-SA: it can be freely distributed and modified under the condition of proper attribution and non-commercial use. How to cite: Verhagen, P. & B.P. Bartholdy, 2022. "Rchon statistics course, part 3". Amsterdam, ARCHON Research School of Archaeology. https://doi.org/10.5281/zenodo.7458108

This is part 3 of the Rchon statistics course. It continues the basics of statistical testing in R. In this tutorial, we will treat the following statistical testing methods: chi square test Fisher's exact test Follow the instructions in Instructions Tutorial 3.pdf to start the tutorial. This course was originally created for Archon Research School of Archaeology by Philip Verhagen (Vrije Universiteit Amsterdam) and Bjørn P. Bartholdy (University of Leiden), and consists of an instruction, a tutorial, a test and two datafiles. All content is CC BY-NC-SA: it can be freely distributed and modified under the condition of proper attribution and non-commercial use. How to cite: Verhagen, P. & B.P. Bartholdy, 2022. "Rchon statistics course, part 3". Amsterdam, ARCHON Research School of Archaeology. https://doi.org/10.5281/zenodo.7457698

This final ARIADNEplus project report on Policies and Good Practices for FAIR Archaeological Data Management describes how focused and dedicated support has established structured policies and strategies for the creation of archaeological data of high quality. A standardised and online tool offers a Data Management Plan for archaeologists which helps researchers to conduct their research following standard quality criteria. Researchers can save a lot of time when preparing a Data Management Plan, since a motivation is only required when deviating from the standard reply, which the Domain Protocol for Archaeological Data Management offers as pre-formulated statements. A related Guide for Archaeological Data Management Planning helps users in their work to find good practices in archaeological data management.

Residue analysis is an established area of expertise focused on detecting traces of substances found on the surface of objects. It is routinely employed in forensic casework and increasingly incorporated into archaeological investigations.In archaeology, sampling and data interpretation sometimes lacked strict standards, resulting in incorrect residue classifications. In particular, molecular signals of salts of fatty acids identified by FTIR have been, at times, interpreted as evidence for adipocere, a substance formed as a consequence of adipose tissues' degradation.This article reviews and discusses the possibilities and limitations of the analytical protocols used in residue analysis in archaeology. The focus is on three main points: (1) reviewing the decomposition processes and the chemical components of adipocere; (2) highlighting potential misidentifications of adipocere while, at the same time, addressing issues related to residue preservation and contamination; and (3) proposing new research avenues to identify adipocere on archaeological objects.(c) 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

At the Vrije Universiteit Amsterdam, we have set out to explore the strengths and limitations of replication studies in the humanities in practice. We are doing so by replicating two original studies: one in the field of art history, the other in the field of history of science and religion. In this blog, we outline the design, purposes, and aims of these projects and explore some of the challenges.

Background and aim. Cancer and atrial fibrillation (AF) may be associated, and anticoagulation, either with vitamin K antagonists (VKAs) or direct oral anticoagulants (DOACs), is necessary to prevent thromboembolic events by reducing the risk of bleeding. The log incidence rate ratio (IRR) and 95% confidence interval were used as index statistics. Higgin's I-2 test was adopted to assess statistical inconsistencies by considering interstudy variations, defined by values ranging from 0 to 100%. I-2 values of less than 40% are associated with very low heterogeneity among the studies; values between 40% and 75% indicate moderate heterogeneity, and those greater than 75% suggest severe heterogeneity. The aim of this meta-analysis was to compare the safety and efficacy of VKAs and DOACs in oncologic patients with AF. Methods. A meta-analysis was conducted comparing VKAs to DOACs in terms of thromboembolic events and bleeding. A meta-regression was conducted to investigate the differences in efficacy and safety between four different DOACs. Moreover, a sub-analysis on active-cancer-only patients was conducted. Results. A total of eight papers were included. The log incidence rate ratio (IRR) for thromboembolic events between the two groups was -0.69 (p 0.9). The Log IRR was -0.38 (p = 0.008) for ischemic stroke, -0.43 (p = 0.02) for myocardial infarction, -0.39 (p = 0.45) for arterial embolism, and -1.04 (p = 0.003) for venous thromboembolism. The log IRR for bleeding events was -0.43 (p < 0.005), and the meta-regression revealed no statistical difference (p = 0.7). The log IRR of hemorrhagic stroke, major bleeding, and clinically relevant non-major bleeding between the VKA and DOAC groups was -0.51 (p < 0.0001), -0.45 (p = 0.03), and 0.0045 (p = 0.97), respectively. Similar results were found in active-cancer patients for all the endpoints except for clinically-relevant non-major bleedings. Conclusions. DOACs showed better efficacy and safety outcomes than VKAs. No difference was found between types of DOACs.

Increasingly code and algorithms are techniques also applied in textual scholarship, giving rise to new interactions between software engineers and textual scholars. This book argues that much of that process and its effects on textual scholarship are still poorly understood and go unchecked by otherwise normal processes of quality control in scholarship such as peer review. The text provides case studies in which some of these interactions become more apparent, as well as the academic challenges and problems that they introduce. The book demonstrates that the space between code creation and conventional scholarship is one that offers many affordances to textual scholarship that until now remain unexplored. The author argues that it is an intellectual obligation of programmers and textual scholars to examine the properties of digital text and how its existence changes and challenges textual scholarship.