Recent advances in large language models (LLMs), such as GPT-4, have spurred interest in their potential applications across various fields, including actuarial work. This paper introduces the use of LLMs in actuarial and insurance-related tasks, both as direct contributors to actuarial modelling and as workflow assistants. It provides an overview of LLM concepts and their potential applications in actuarial science and insurance, examining specific areas where LLMs can be beneficial, including a detailed assessment of the claims process. Additionally, a decision framework for determining the suitability of LLMs for specific tasks is presented. Case studies with accompanying code showcase the potential of LLMs to enhance actuarial work. Overall, the results suggest that LLMs can be valuable tools for actuarial tasks involving natural language processing or structuring unstructured data and as workflow and coding assistants. However, their use in actuarial work also presents challenges, particularly regarding professionalism and ethics, for which high-level guidance is provided.

The rise of populism concerns many political scientists and practitioners, yet the detection of its underlying language remains fragmentary. This paper aims to provide a reliable, valid, and scalable approach to measure populist rhetoric. For that purpose, we created an annotated dataset based on parliamentary speeches of the German Bundestag (2013–2021). Following the ideational definition of populism, we label moralizing references to “the virtuous people” or “the corrupt elite” as core dimensions of populist language. To identify, in addition, how the thin ideology of populism is “thickened,” we annotate how populist statements are attached to left-wing or right-wing host ideologies. We then train a transformer-based model (PopBERT) as a multilabel classifier to detect and quantify each dimension. A battery of validation checks reveals that the model has a strong predictive accuracy, provides high qualitative face validity, matches party rankings of expert surveys, and detects out-of-sample text snippets correctly. PopBERT enables dynamic analyses of how German-speaking politicians and parties use populist language as a strategic device. Furthermore, the annotator-level data may also be applied in cross-domain applications or to develop related classifiers.

Stance detection is identifying expressed beliefs in a document. While researchers widely use sentiment analysis for this, recent research demonstrates that sentiment and stance are distinct. This paper advances text analysis methods by precisely defining stance detection and outlining three approaches: supervised classification, natural language inference, and in-context learning. I discuss how document context and trade-offs between resources and workload should inform your methods. For all three approaches I provide guidance on application and validation techniques, as well as coding tutorials for implementation. Finally, I demonstrate how newer classification approaches can replicate supervised classifiers.

Several disciplines, such as economics, law, and political science, emphasize the importance of legislative quality, namely well-written legislation. Low-quality legislation cannot be easily implemented because the texts create interpretation problems. To measure the quality of legal texts, we use information from the syntactic and lexical features of their language and apply these measures to a dataset of European Union legislation that contains detailed information on its transposition and decision-making process. We find that syntactic complexity and vagueness are negatively related to member states’ compliance with legislation. The finding on vagueness is robust to controlling for member states’ preferences, administrative resources, length of texts, and discretion. However, the results for syntactic complexity are less robust.

Multilingual question answering (MQA) is an effective access to multilingual data to provide accurate and precise answers, irrespective of language. Although a wide range of datasets is available for monolingual QA systems in natural language processing, benchmark datasets specifically designed for MQA are considerably limited. The absence of comprehensive and benchmark datasets hinders the development and evaluation of MQA systems. To overcome this issue, the proposed work attempts to develop the EHMQuAD dataset, an MQA dataset for low-resource languages such as Hindi and Marathi accompanying the English language. The EHMQuAD dataset is developed using a synthetic corpora generation approach, and an alignment is performed after translation to make the dataset more accurate. Further, the EHMMQA model is proposed to create an abstract framework that uses a deep neural network that accepts pairs of questions and context and returns an accurate answer based on those questions. The shared question and shared context representation have been designed separately to develop this system. The experiments of the proposed model are conducted on the MMQA, Translated SQuAD, XQuAD, MLQA, and EHMQuAD datasets, and EM and F1-score are used as performance measures. The proposed model (EHMMQA) is collated with state-of-the-art MQA baseline models for all possible monolingual and multilingual settings. The results signify that EHMMQA is a considerable step toward the MQA system utilizing Hindi and Marathi languages. Hence, it becomes a new state-of-the-art model for Hindi and Marathi languages.

Housing is an environmental social determinant of health that is linked to mortality and clinical outcomes. We developed a lexicon of housing-related concepts and rule-based natural language processing methods for identifying these housing-related concepts within clinical text. We piloted our methods on several test cohorts: a synthetic cohort generated by ChatGPT for initial infrastructure testing, a cohort with substance use disorders (SUD), and a cohort diagnosed with problems related to housing and economic circumstances (HEC). Our methods successfully identified housing concepts in our ChatGPT notes (recall = 1.0, precision = 1.0), our SUD population (recall = 0.9798, precision = 0.9898), and our HEC population (recall = N/A, precision = 0.9160).

Innovation, typically spurred by reusing, recombining and synthesizing existing concepts, is expected to result in an exponential growth of the concept space over time. However, our statistical analysis of TechNet, which is a comprehensive technology semantic network encompassing over 4 million concepts derived from patent texts, reveals a linear rather than exponential expansion of the overall technological concept space. Moreover, there is a notable decline in the originality of newly created concepts. These trends can be attributed to the constraints of human cognitive abilities to innovate beyond an ever-growing space of prior art, among other factors. Integrating creative artificial intelligence into the innovation process holds the potential to overcome these limitations and alter the observed trends in the future.

This paper demonstrates workflows to incorporate text data into actuarial classification and regression tasks. The main focus is on methods employing transformer-based models. A dataset of car accident descriptions with an average length of 400 words, available in English and German, and a dataset with short property insurance claims descriptions, are used to demonstrate these techniques. The case studies tackle challenges related to a multilingual setting and long input sequences. They also show ways to interpret model output and to assess and improve model performance, by fine-tuning the models to the domain of application or to a specific prediction task. Finally, the paper provides practical approaches to handle classification tasks in situations with no or only few labelled data. The results achieved by using the language-understanding skills of off-the-shelf natural language processing (NLP) models with only minimal pre-processing and fine-tuning clearly demonstrate the power of transfer learning for practical applications.

This article sheds light on the significant yet nuanced roles of shame and guilt in influencing moral behaviour, a phenomenon that became particularly prominent during the COVID-19 pandemic with the community’s heightened desire to be seen as moral. These emotions are central to human interactions, and the question of how they are conveyed linguistically is a vast and important one. Our study contributes to this area by analysing the discourses around shame and guilt in English and Japanese online forums, focusing on the terms shame, guilt, haji (‘shame’) and zaiakukan (‘guilt’). We utilise a mix of corpus-based methods and natural language processing tools, including word embeddings, to examine the contexts of these emotion terms and identify semantically similar expressions. Our findings indicate both overlaps and distinct differences in the semantic landscapes of shame and guilt within and across the two languages, highlighting nuanced ways in which these emotions are expressed and distinguished. This investigation provides insights into the complex dynamics between emotion words and the internal states they denote, suggesting avenues for further research in this linguistically rich area.

Handling nominal covariates with a large number of categories is challenging for both statistical and machine learning techniques. This problem is further exacerbated when the nominal variable has a hierarchical structure. We commonly rely on methods such as the random effects approach to incorporate these covariates in a predictive model. Nonetheless, in certain situations, even the random effects approach may encounter estimation problems. We propose the data-driven Partitioning Hierarchical Risk-factors Adaptive Top-down algorithm to reduce the hierarchically structured risk factor to its essence, by grouping similar categories at each level of the hierarchy. We work top-down and engineer several features to characterize the profile of the categories at a specific level in the hierarchy. In our workers’ compensation case study, we characterize the risk profile of an industry via its observed damage rates and claim frequencies. In addition, we use embeddings to encode the textual description of the economic activity of the insured company. These features are then used as input in a clustering algorithm to group similar categories. Our method substantially reduces the number of categories and results in a grouping that is generalizable to out-of-sample data. Moreover, we obtain a better differentiation between high-risk and low-risk companies.