3/4/2023 0 Comments Servicecenter uni hannover![]() Potential of encoding beliefs into argumentative texts, laying the ground for Of modeling users' beliefs based on their stances, but demonstrate the Uttered by someone with a respective belief. The generated claims in terms of informativeness and their likelihood to be In a manual study, we additionally evaluate Our automatic evaluation confirms the ability of our approach to adaptĬlaims to a set of given beliefs. State-of-the-art text generation models to generate claims conditioned on theīeliefs. To tackle this task, we model the people's priorīeliefs through their stances on controversial topics and extend This gap by studying the task of belief-based claim generation: Given aĬontroversial topic and a set of beliefs, generate an argumentative claim However, existing approachesĭo not perform any audience-specific adaptation. To address the automatic generation of arguments. Recently, the field of computational argumentation witnessed extensive effort When engaging in argumentative discourse, skilled human debaters tailorĬlaims to the beliefs of the audience, to construct effective arguments. We demonstrate the behavior of BSA on two standard embedding models for the three mentioned metrics with several word lists from existing research. The variance and rate of convergence of the bias values of each model then entail the robustness of the word lists, whereas the distance between the models' values gives indications of the general accuracy of the metric with the word lists. Given a biased and an unbiased reference embedding model, BSA applies the metric systematically for several subsets of the lists to the models. In particular, we present a generic method, Bias Silhouette Analysis (BSA), that quantifies the accuracy and robustness of such a metric and of the word lists used. In this paper, we study how to assess the quality of bias metrics for word embedding models. How suitable these lists actually are to reveal bias - let alone the bias metrics in general - remains unclear, though. Metrics such as ECT, RNSB, and WEAT quantify bias in these models based on predefined word lists representing social groups and bias-conveying concepts. Word embedding models reflect bias towards genders, ethnicities, and other social groups present in the underlying training data. In our evaluation, human annotators favored its counterfactuals in terms of both connotations, also deeming its meaning preservation better. According to automatic metrics, our model effectively controls for power while being competitive in agency to the state of the art. The verbs' probability is then boosted to encourage the model to rewrite both connotations jointly. In this paper, we thus develop a new rewriting model that identifies verbs with the desired agency and power in the context of the given sentence. We argue that this is insufficient, due to the interaction of agency and power and their dependence on context. Recent work targeted agency-specific verbs from a lexicon to this end. An effective way to mitigate bias is to generate counterfactual sentences with opposite agency and power to the training. When language models learn from respective texts, they may reproduce or even amplify the bias. Gender bias may emerge from an unequal representation of agency and power, for example, by portraying women frequently as passive and powerless ("She accepted her future'') and men as proactive and powerful ("He chose his future''). Vertretung der Belange studentischer Hilfskräfte.Ombudsperson für gute wissenschaftliche Praxis.Studium mit Beeinträchtigung / Behindertenbeauftragte.Allgemeiner Studierendenausschuss (AStA).Paderborn Center for Parallel Computing (PC²).Zentrum für Informations- und Medientechnologien (IMT).Betriebliche AnsprechPersonen - Prävention.Jenny Aloni Center for Early-Career Researchers.Welcome Service for international researchers and employees.EU-Projekte (Horizon Europe, Horizon 2020, FP 7).Außeruniversitäre Forschungskooperationen.Interdisziplinäre Forschungseinrichtungen.Forschungspreis der Universität Paderborn.Paderborner Wissenschaftskolleg „Data Society“.Netzwerk interdisziplinäre Forschung (NiFo).Nachhaltige Werkstoffe, Prozesse und Produkte.Forschungsorientierte Gleichstellungsstandards.Kommission für Forschung und wissenschaftlichen Nachwuchs.Akademische Auszeichnungen und Ehrungen.Kompetenzzentrum Hochschuldidaktik Mathematik (khdm). ![]() Stiftung Innovation in der Hochschullehre.Studiengangentwicklung & Akkreditierung.Studierenden- und Absolventenbefragungen.Qualitätsmanagement-System für Studium und Lehre.Förderpreis für Innovation und Qualität in der Lehre.Lehrpreis für den wissenschaftlichen Nachwuchs.E-Learning und Technische Unterstützung.Zentrum für Bildungsforschung und Lehrerbildung – PLAZ-Professional School.Studieren ohne Abitur und weitere Zugänge. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |