TY - CPAPER TI - Sounds Queer : Representation of LGBTQIA Identities in AI-generated Songs AU - Weber, Sabine AU - McLeod, Andrew PY - 2026 SP - 4375-4394 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/116541 LA - eng AB - The expansion of generative artificial intelligence into the domain of music production has introduced new possibilities and risks for cultural representation, particularly in its depiction of marginalized groups. In this paper we study how LGBTQIA identities are represented in text-prompt-based end-to-end generated songs by the black-box models of Suno and Mureka. To do so we use a template-based approach to generate prompts containing identity terms like “gay”, “lesbian” or “genderqueer” that we pass to the examined systems, generating a dataset of 9600 songs. Using qualitative and quantitative content analysis of the generated song lyrics and musicological analysis of the generated music, we find that while songs generally convey high regard and positive sentiment towards queer individuals, they often reproduce stereotypes. We uncover representational harms specifically in the representation of aromantic, asexual, lesbian and intersex people. Musically, the models produce songs that strongly align with the requested genre, but show only minor differences based on the identity terms used in the prompts. We release the created Sounds Queer dataset and all associated code under open Creative Commons licensing for further research. KW - Queer AI KW - NLP KW - Language Generation KW - Music Generation KW - Large Language Models KW - Stereotypes ER - TY - CPAPER TI - Sounds Queer : Representation of LGBTQIA Identities in AI-generated Songs AU - Weber, Sabine AU - McLeod, Andrew PY - 2026 T2 - Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency SP - 4375-4394 PB - ACM CY - New York SN - 979-8-4007-2596-8 DO - 10.1145/3805689.3806752 UR - https://fis.uni-bamberg.de/handle/uniba/116538 LA - eng AB - The expansion of generative artificial intelligence into the domain of music production has introduced new possibilities and risks for cultural representation, particularly in its depiction of marginalized groups. In this paper we study how LGBTQIA identities are represented in text-prompt-based end-to-end generated songs by the black-box models of Suno and Mureka. To do so we use a template-based approach to generate prompts containing identity terms like “gay”, “lesbian” or “genderqueer” that we pass to the examined systems, generating a dataset of 9600 songs. Using qualitative and quantitative content analysis of the generated song lyrics and musicological analysis of the generated music, we find that while songs generally convey high regard and positive sentiment towards queer individuals, they often reproduce stereotypes. We uncover representational harms specifically in the representation of aromantic, asexual, lesbian and intersex people. Musically, the models produce songs that strongly align with the requested genre, but show only minor differences based on the identity terms used in the prompts. We release the created Sounds Queer dataset and all associated code under open Creative Commons licensing for further research. KW - Queer AI KW - NLP KW - Language Generation KW - Music Generation KW - Large Language Models KW - Stereotypes ER - TY - CPAPER TI - Trust Me, I Can Convince You: The Contextualized Argument Appraisal Framework and the ContArgA Corpus AU - Greschner, Lynn AU - Weber, Sabine AU - Klinger, Roman PY - 2026 T2 - Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026) SP - 8327–8346 PB - European Languages Resources Association (ELRA) CY - Paris SN - 978-2-493814-49-4 DO - 10.63317/484rpnvebop5 UR - https://fis.uni-bamberg.de/handle/uniba/115018 LA - eng ER - TY - CPAPER TI - Categorical Emotions or Appraisals: Which Emotion Model Explains Argument Convincingness Better? AU - Greschner, Lynn AU - Bauer, Meike AU - Weber, Sabine AU - Klinger, Roman PY - 2026 T2 - Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026) SP - 8190–8203 PB - European Languages Resources Association (ELRA) CY - Paris SN - 978-2-493814-49-4 DO - 10.63317/3vrvrgvtnvhn UR - https://fis.uni-bamberg.de/handle/uniba/115020 LA - eng ER - TY - CPAPER TI - Less Is More?: The Role of Demographic Author Information in Emotion Classification of Ambiguous Text AU - Weber, Sabine AU - Greschner, Lynn AU - Klinger, Roman PY - 2026 T2 - Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026) SP - 8147–8161 PB - European Languages Resources Association (ELRA) CY - Paris SN - 978-2-493814-49-4 DO - 10.63317/2cw8tpo82h55 UR - https://fis.uni-bamberg.de/handle/uniba/115023 LA - eng ER - TY - CPAPER TI - Says Who? : Argument Convincingness and Reader Stance Are Correlated with Perceived Author Personality AU - Weber, Sabine AU - Greschner, Lynn AU - Klinger, Roman PY - 2026 T2 - The Proceedings for the 15th Workshop on Computational Approaches to Subjectivity, Sentiment Social Media Analysis (WASSA 2026) SP - 265–277 PB - Association for Computational Linguistics SN - 979-8-89176-378-4 UR - https://fis.uni-bamberg.de/handle/uniba/114414 LA - eng ER - TY - CPAPER TI - Which Demographics do LLMs Default to During Annotation? AU - Schäfer, Johannes AU - Combs, Aidan AU - Bagdon, Christopher Doyle AU - Li, Jiahui AU - Probol, Nadine AU - Greschner, Lynn AU - Papay, Sean AU - Menchaca Resendiz, Yarik AU - Velutharambath, Aswathy AU - Wuehrl, Amelie AU - Weber, Sabine AU - Klinger, Roman PY - 2026 SP - 17331–17348 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/114413 LA - eng AB - Demographics and cultural background of annotators influence the labels they assign in text annotation – for instance, an elderly woman might find it offensive to read a message addressed to a “bro”, but a male teenager might find it appropriate. It is therefore important to acknowledge label variations to not underrepresent members of a society. Two research directions developed out of this observation in the context of using large language models (LLM) for data annotations, namely (1) studying biases and inherent knowledge of LLMs and (2) injecting diversity in the output by manipulating the prompt with demographic information. We combine these two strands of research and ask the question to which demographics an LLM resorts when no demographics is given. To answer this question, we evaluate which attributes of human annotators LLMs inherently mimic. Furthermore, we compare non-demographic conditioned prompts and placebo-conditioned prompts (e.g., “you are an annotator who lives in house number 5”) to demographics-conditioned prompts (“You are a 45 year old man and an expert on politeness annotation. How do you rate {instance}”). We study these questions for politeness and offensiveness annotations on the POPQUORN data set, a corpus created in a controlled manner to investigate human label variations based on demographics which has not been used for LLM-based analyses so far. We observe notable influences related to gender, race, and age in demographic prompting, which contrasts with previous studies that found no such effects KW - - ER - TY - JOUR TI - Can We Build AI That Does Not Harm Queer People? AU - Weber, Sabine PY - 2025 JO - Communications of the ACM VL - 68 IS - 5 SP - 21-23 PB - Association for Computing Machinery (ACM) CY - New York, NY SN - 1557-7317 DO - 10.1145/3720537 UR - https://fis.uni-bamberg.de/handle/uniba/116845 LA - eng ER - TY - CPAPER TI - Language Models Are Poor Learners of Directional Inference AU - Li, Tianyi AU - Hosseini, Mohammad Javad AU - Weber, Sabine AU - Steedman, Mark PY - 2025 SP - 903–921 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/110866 LA - eng AB - We examine LMs’ competence of directional predicate entailments by supervised fine-tuning with prompts. Our analysis shows that contrary to their apparent success on standard NLI, LMs show limited ability to learn such directional inference; moreover, existing datasets fail to test directionality, and/or are infested by artefacts that can be learnt as proxy for entailments, yielding over-optimistic results. In response, we present BoOQA (Boolean Open QA), a robust multi-lingual evaluation benchmark for directional predicate entailments, extrinsic to existing training sets. On BoOQA, we establish baselines and show evidence of existing LM-prompting models being incompetent directional entailment learners, in contrast to entailment graphs, however limited by sparsity. KW - Language Models ER - TY - CPAPER TI - Cross-lingual Inference with A Chinese Entailment Graph AU - Li, Tianyi AU - Weber, Sabine AU - Hosseini, Mohammad Javad AU - Guillou, Liane AU - Steedman, Mark PY - 2025 SP - 1214–1233 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/110858 LA - eng AB - Predicate entailment detection is a crucial task for question-answering from text, where previous work has explored unsupervised learning of entailment graphs from typed open relation triples. In this paper, we present the first pipeline for building Chinese entailment graphs, which involves a novel high-recall open relation extraction (ORE) method and the first Chinese fine-grained entity typing dataset under the FIGER type ontology. Through experiments on the Levy-Holt dataset, we verify the strength of our Chinese entailment graph, and reveal the cross-lingual complementarity: on the parallel Levy-Holt dataset, an ensemble of Chinese and English entailment graphs outperforms both monolingual graphs, and raises unsupervised SOTA by 4.7 AUC points. KW - Chinese Entailment Graph ER - TY - CPAPER TI - Fine-grained General Entity Typing in German using GermaNet AU - Weber, Sabine AU - Steedman, Mark PY - 2025 SP - 138–143 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/110857 LA - eng AB - Fine-grained entity typing is important to tasks like relation extraction and knowledge base construction. We find however, that fine-grained entity typing systems perform poorly on general entities (e.g. “ex-president”) as compared to named entities (e.g. “Barack Obama”). This is due to a lack of general entities in existing training data sets. We show that this problem can be mitigated by automatically generating training data from WordNets. We use a German WordNet equivalent, GermaNet, to automatically generate training data for German general entity typing. We use this data to supplement named entity data to train a neural fine-grained entity typing system. This leads to a 10% improvement in accuracy of the prediction of level 1 FIGER types for German general entities, while decreasing named entity type prediction accuracy by only 1%. KW - GermaNet ER - TY - CPAPER TI - Zero-Shot Cross-Lingual Transfer is a Hard Baseline to Beat in German Fine-Grained Entity Typing AU - Weber, Sabine AU - Steedman, Mark PY - 2025 SP - 42-48 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/110856 LA - eng AB - The training of NLP models often requires large amounts of labelled training data, which makes it difficult to expand existing models to new languages. While zero-shot cross-lingual transfer relies on multilingual word embeddings to apply a model trained on one language to another, Yarowski and Ngai (2001) propose the method of annotation projection to generate training data without manual annotation. This method was successfully used for the tasks of named entity recognition and coarse-grained entity typing, but we show that it is outperformed by zero-shot cross-lingual transfer when applied to the similar task of fine-grained entity typing. In our study of fine-grained entity typing with the FIGER type ontology for German, we show that annotation projection amplifies the English model’s tendency to underpredict level 2 labels and is beaten by zero-shot cross-lingual transfer on three novel test sets. KW - Zero-Shot Cross-Lingual Transfer ER - TY - CPAPER TI - Different German and English Coreference Resolution Models for Multi-domain Content Curation Scenarios AU - Srivastava, Ankit AU - Weber, Sabine AU - Bourgonje, Peter AU - Rehm, Georg PY - 2025 SP - 48-61 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/110847 LA - eng AB - Coreference Resolution is the process of identifying all words and phrases in a text that refer to the same entity. It has proven to be a useful intermediary step for a number of natural language processing applications. In this paper, we describe three implementations for performing coreference resolution: rule-based, statistical, and projection-based (from English to German). After a comparative evaluation on benchmark datasets, we conclude with an application of these systems on German and English texts from different scenarios in digital curation such as an archive of personal letters, excerpts from a museum exhibition, and regional news articles. KW - Multi-domain Content Curation Scenarios ER - TY - CPAPER TI - Which Demographics do LLMs Default to During Annotation? AU - Schäfer, Johannes AU - Combs, Aidan AU - Bagdon, Christopher Doyle AU - Li, Jiahui AU - Probol, Nadine AU - Greschner, Lynn AU - Papay, Sean AU - Menchaca Resendiz, Yarik AU - Velutharambath, Aswathy AU - Wuehrl, Amelie AU - Weber, Sabine AU - Klinger, Roman PY - 2025 T2 - Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics SP - 17331–17348 PB - Association for Computational Linguistics SN - 979-8-89176-251-0 DO - 10.18653/v1/2025.acl-long.848 UR - https://fis.uni-bamberg.de/handle/uniba/109541 LA - eng AB - Demographics and cultural background of annotators influence the labels they assign in text annotation – for instance, an elderly woman might find it offensive to read a message addressed to a “bro”, but a male teenager might find it appropriate. It is therefore important to acknowledge label variations to not underrepresent members of a society. Two research directions developed out of this observation in the context of using large language models (LLM) for data annotations, namely (1) studying biases and inherent knowledge of LLMs and (2) injecting diversity in the output by manipulating the prompt with demographic information. We combine these two strands of research and ask the question to which demographics an LLM resorts when no demographics is given. To answer this question, we evaluate which attributes of human annotators LLMs inherently mimic. Furthermore, we compare non-demographic conditioned prompts and placebo-conditioned prompts (e.g., “you are an annotator who lives in house number 5”) to demographics-conditioned prompts (“You are a 45 year old man and an expert on politeness annotation. How do you rate {instance}”). We study these questions for politeness and offensiveness annotations on the POPQUORN data set, a corpus created in a controlled manner to investigate human label variations based on demographics which has not been used for LLM-based analyses so far. We observe notable influences related to gender, race, and age in demographic prompting, which contrasts with previous studies that found no such effects KW - - ER - TY - CPAPER TI - Language Models Are Poor Learners of Directional Inference AU - Li, Tianyi AU - Hosseini, Mohammad Javad AU - Weber, Sabine AU - Steedman, Mark PY - 2022 T2 - Findings of the Association for Computational Linguistics: EMNLP 2022 SP - 903–921 PB - Association for Computational Linguistics DO - 10.18653/v1/2022.findings-emnlp.64 UR - https://fis.uni-bamberg.de/handle/uniba/97222 LA - eng AB - We examine LMs’ competence of directional predicate entailments by supervised fine-tuning with prompts. Our analysis shows that contrary to their apparent success on standard NLI, LMs show limited ability to learn such directional inference; moreover, existing datasets fail to test directionality, and/or are infested by artefacts that can be learnt as proxy for entailments, yielding over-optimistic results. In response, we present BoOQA (Boolean Open QA), a robust multi-lingual evaluation benchmark for directional predicate entailments, extrinsic to existing training sets. On BoOQA, we establish baselines and show evidence of existing LM-prompting models being incompetent directional entailment learners, in contrast to entailment graphs, however limited by sparsity. KW - Language Models ER - TY - CPAPER TI - Cross-lingual Inference with A Chinese Entailment Graph AU - Li, Tianyi AU - Weber, Sabine AU - Hosseini, Mohammad Javad AU - Guillou, Liane AU - Steedman, Mark PY - 2022 T2 - Findings of the Association for Computational Linguistics: ACL 2022 SP - 1214–1233 PB - Association for Computational Linguistics DO - 10.18653/v1/2022.findings-acl.96 UR - https://fis.uni-bamberg.de/handle/uniba/97224 LA - eng AB - Predicate entailment detection is a crucial task for question-answering from text, where previous work has explored unsupervised learning of entailment graphs from typed open relation triples. In this paper, we present the first pipeline for building Chinese entailment graphs, which involves a novel high-recall open relation extraction (ORE) method and the first Chinese fine-grained entity typing dataset under the FIGER type ontology. Through experiments on the Levy-Holt dataset, we verify the strength of our Chinese entailment graph, and reveal the cross-lingual complementarity: on the parallel Levy-Holt dataset, an ensemble of Chinese and English entailment graphs outperforms both monolingual graphs, and raises unsupervised SOTA by 4.7 AUC points. KW - Chinese Entailment Graph ER - TY - CPAPER TI - Fine-grained General Entity Typing in German using GermaNet AU - Weber, Sabine AU - Steedman, Mark PY - 2021 T2 - Proceedings of the Fifteenth Workshop on Graph-Based Methods for Natural Language Processing (TextGraphs-15) SP - 138–143 PB - Association for Computational Linguistics DO - 10.18653/v1/2021.textgraphs-1.14 UR - https://fis.uni-bamberg.de/handle/uniba/97225 LA - eng AB - Fine-grained entity typing is important to tasks like relation extraction and knowledge base construction. We find however, that fine-grained entity typing systems perform poorly on general entities (e.g. “ex-president”) as compared to named entities (e.g. “Barack Obama”). This is due to a lack of general entities in existing training data sets. We show that this problem can be mitigated by automatically generating training data from WordNets. We use a German WordNet equivalent, GermaNet, to automatically generate training data for German general entity typing. We use this data to supplement named entity data to train a neural fine-grained entity typing system. This leads to a 10% improvement in accuracy of the prediction of level 1 FIGER types for German general entities, while decreasing named entity type prediction accuracy by only 1%. KW - GermaNet ER - TY - CPAPER TI - Zero-Shot Cross-Lingual Transfer is a Hard Baseline to Beat in German Fine-Grained Entity Typing AU - Weber, Sabine AU - Steedman, Mark PY - 2021 T2 - Proceedings of the Second Workshop on Insights from Negative Results in NLP SP - 42–48 PB - Association for Computational Linguistics DO - 10.18653/v1/2021.insights-1.7 UR - https://fis.uni-bamberg.de/handle/uniba/97228 LA - eng AB - The training of NLP models often requires large amounts of labelled training data, which makes it difficult to expand existing models to new languages. While zero-shot cross-lingual transfer relies on multilingual word embeddings to apply a model trained on one language to another, Yarowski and Ngai (2001) propose the method of annotation projection to generate training data without manual annotation. This method was successfully used for the tasks of named entity recognition and coarse-grained entity typing, but we show that it is outperformed by zero-shot cross-lingual transfer when applied to the similar task of fine-grained entity typing. In our study of fine-grained entity typing with the FIGER type ontology for German, we show that annotation projection amplifies the English model’s tendency to underpredict level 2 labels and is beaten by zero-shot cross-lingual transfer on three novel test sets. KW - Zero-Shot Cross-Lingual Transfer ER - TY - CPAPER TI - Construction and Alignment of Multilingual Entailment Graphs for Semantic Inference AU - Weber, Sabine AU - Steedman, Mark PY - 2019 T2 - Proceedings of the 2019 Workshop on Widening NLP SP - 77–79 PB - Association for Computational Linguistics UR - https://fis.uni-bamberg.de/handle/uniba/97229 LA - eng ER - TY - CPAPER TI - A Marker Passing Approach to Winograd Schemas AU - Fähndrich, Johannes AU - Weber, Sabine AU - Kanthak, Hannes PY - 2018 T2 - Semantic technology : 8th joint international conference, JIST 2018, Awaji, Japan, November 26–28, 2018, proceedings SP - 165-181 PB - Springer International Publishing CY - Cham, Switzerland SN - 9783030042837 DO - 10.1007/978-3-030-04284-4_12 UR - https://fis.uni-bamberg.de/handle/uniba/97230 LA - eng ER - TY - CPAPER TI - Different German and English Coreference Resolution Models for Multi-domain Content Curation Scenarios AU - Srivastava, Ankit AU - Weber, Sabine AU - Bourgonje, Peter AU - Rehm, Georg PY - 2018 T2 - Language Technologies for the Challenges of the Digital Age : 27th International Conference, GSCL 2017, Berlin, Germany, September 13-14, 2017, Proceedings SP - 48-61 PB - Springer International Publishing CY - Cham SN - 9783319737058 DO - 10.1007/978-3-319-73706-5_5 UR - https://fis.uni-bamberg.de/handle/uniba/97231 LA - eng AB - Coreference Resolution is the process of identifying all words and phrases in a text that refer to the same entity. It has proven to be a useful intermediary step for a number of natural language processing applications. In this paper, we describe three implementations for performing coreference resolution: rule-based, statistical, and projection-based (from English to German). After a comparative evaluation on benchmark datasets, we conclude with an application of these systems on German and English texts from different scenarios in digital curation such as an archive of personal letters, excerpts from a museum exhibition, and regional news articles. KW - Multi-domain Content Curation Scenarios ER -