This paper addresses the critical need for detecting bias in government documents, an underexplored area with significant implications for governance. Existing methodologies often overlook the unique context and far-reaching impacts of governmental documents, potentially obscuring embedded biases that shape public policy and citizen-government interactions. To bridge this gap, we introduce the Dutch Government Data for Bias Detection (DGDB), a dataset sourced from the Dutch House of Representatives and annotated for bias by experts. We fine-tune several BERT-based models on this dataset and compare their performance with that of generative language models. Additionally, we conduct a comprehensive error analysis that includes explanations of the models' predictions. Our findings demonstrate that fine-tuned models achieve strong performance and significantly outperform generative language models, indicating the effectiveness of DGDB for bias detection. This work underscores the importance of labeled datasets for bias detection in various languages and contributes to more equitable governance practices.

We are interested in the computation of ontology extracts based on forgetting from large ontologies in real-world scenarios. Such scenarios require nearly all of the terms in the ontology to be forgotten, which poses a significant challenge to forgetting tools. In this paper we show that modularization and forgetting can be combined beneficially in order to compute ontology extracts. While a module is a subset of axioms of a given ontology, the solution of forgetting (also known as a uniform interpolant) is a compact representation of the ontology limited to a subset of the signature. The approach introduced in this paper uses an iterative workflow of four stages: (i)~extension of the given signature and, if needed partitioning, (ii)~modularization, (iii)~forgetting, and (iv)~evaluation by domain expert. For modularization we use three kinds of modules: locality-based, semantic and minimal subsumption modules. For forgetting three tools are used: NUI, LETHE and FAME. An evaluation on the SNOMED CT and NCIt ontologies for standard concept name lists showed that precomputing ontology modules reduces the number of terms that need to be forgotten. An advantage of the presented approach is high precision of the computed ontology extracts.

In the presence of an ever growing amount of information, organizations and human users need to be able to focus on certain key pieces of information and to intentionally ignore all other possibly relevant parts. Knowledge about complex systems that is represented in ontologies yields collections of axioms that are too large for human users to browse, let alone to comprehend or reason about it. We introduce the notion of an ontology excerpt as being a fixed-size subset of an ontology, consisting of the most relevant axioms for a given set of terms. These axioms preserve as much as possible the knowledge about the considered terms described in the ontology. We consider different extraction techniques for ontology excerpts based on methods from the area of information retrieval. To evaluate these techniques, we propose to measure the degree of incompleteness of the resulting excerpts using the notion of logical difference.