Publications 📗

@inproceedings{Muraji_LACATODA2024,

title = {Evaluation of active generation of interlocutor profiling sentences from utterances and their implicit context},

author = {Shinji Muraji and Rafal Rzepka and Toshihiko Itoh},

editor = {Rafal Rzepka and Michal Ptaszynski and Pawel Dybala and Siaw-Fong Chung and Jordi Vallverdu},

year  = {2024},

date = {2024-11-19},

urldate = {2024-11-19},

booktitle = {Proceedings of the 9th Linguistic and Cognitive Approaches To Dialog Agents Workshop - LaCATODA 2024 co-located with the Pacific Rim International Conference on Artificial Intelligence (PRICAI 2024)},

pages = {45-55},

publisher = {WS-CEUR},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Ji2024pricai,

title = {Fame Bias -- Large Language Models Change Their Judgement Depending on Personal Name},

author = {Huizhong Ji and Rafal Rzepka},

editor = {Rafik Hadfi, Patricia Anthony, Alok Sharma, Takayuki Ito, Quan Bai},

doi = {https://doi.org/10.1007/978-981-96-0122-6},

year  = {2024},

date = {2024-11-18},

urldate = {2024-11-18},

booktitle = {PRICAI 2024: Trends in Artificial Intelligence, Lecture Notes LNAI 15283, 21st Pacific Rim International Conference on Artificial Intelligence, PRICAI 2024, Kyoto, Japan, November 18–24, 2024, Proceedings, Part III},

pages = {15--20},

series = {Lecture Notes in Computer Science},

abstract = {This paper focuses on bias caused by changing personal names in the input of proprietary LLMs, fine-tuned language models, and lexicon-based sentiment analysis tool. It extracts examples from NLI and story cloze task datasets with personal names, replaces them with names of celebrities, ordinary people or mixtures. Results show that altered names significantly influence model assessments, especially with more than two choices, indicating potential real-world application hazards. A mitigation method for “fame bias” is proposed but the problem requires further research.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Rzepka2024AGI,

title = {Semantic Primes-Inspired Tacit Knowledge Dataset for Simulating Basic Perception Capabilities of Cognitive Architectures},

author = {Rafal Rzepka and Ryoma Shinto and Kenji Araki},

url = {https://doi.org/10.1007/978-3-031-65572-2_16},

doi = {10.1007/978-3-031-65572-2_16},

isbn = {978-3-031-65571-5},

year  = {2024},

date = {2024-08-15},

urldate = {2024-01-01},

booktitle = {Artificial General Intelligence: 17th International Conference, AGI 2024, Seattle, WA, USA, August 13–16, 2024, Proceedings},

pages = {145–154},

publisher = {Springer-Verlag},

address = {SEATTLE, WA, USA},

abstract = {In this paper we present a novel dataset of tacit knowledge represented in natural language (Japanese) inspired by semantic primes categories. The main goals of this data is to a) allow investigations regarding influence of perception data in various cognitive tasks, b) mimic signals for cognitive processes of an artificial agent to extend the understanding of the world and c) testing cognitive capabilities of intelligent instances like foundation models. We describe the dataset and share results of preliminary experiments showing that the tacit knowledge recognition is still hard for language models. We also discuss how such redirecting neural approaches to cognition only and then perform reasoning in a symbolic realms could become beneficial for new type of simulations before AGIs are equipped with more sophisticated sensory apparatus.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{10487735,

title = {Expert Evaluation of Export Control-Related Question Answering Capabilities of LLMs},

author = {Rafal Rzepka and Shinji Muraji and Akihiko Obayashi },

url = {https://doi.ieeecomputersociety.org/10.1109/CSDE59766.2023.10487735},

doi = {10.1109/CSDE59766.2023.10487735},

year  = {2023},

date = {2023-12-01},

urldate = {2023-12-01},

booktitle = {2023 IEEE Asia-Pacific Conference on Computer Science and Data Engineering (CSDE)},

pages = {1-6},

publisher = {IEEE Computer Society},

address = {Los Alamitos, CA, USA},

abstract = {In this paper we introduce evaluation experiments performed by an expert to assess Large Language Models ability to answer questions related to export control in Japanese language. We compare outputs of two popular models, namely ChatGPT and GPT-4, and measure their accuracy in predicting intention of the question and label of the answer. Prediction results indicate that both models were better in guessing question intention but failed to recognize what type of an answer would be most probable. Furthermore, expert evaluation of outputs generated by both models show that most of the answers contain mistakes which can mislead users asking a dialog system for an expertise. We analyze these outputs and enumerate problems to be tackled in the future.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{10487682,

title = {Can LLMs Determine Political Leaning of Polish News Articles?},

author = {Joanna Szwoch and Mateusz Staszkow and Rafal Rzepka and Kenji Araki},

url = {https://doi.ieeecomputersociety.org/10.1109/CSDE59766.2023.10487682},

doi = {10.1109/CSDE59766.2023.10487682},

year  = {2023},

date = {2023-12-01},

urldate = {2023-12-01},

booktitle = {2023 IEEE Asia-Pacific Conference on Computer Science and Data Engineering (CSDE)},

pages = {1-6},

publisher = {IEEE Computer Society},

address = {Los Alamitos, CA, USA},

abstract = {This study explores how Large Language Models (LLMs) can be used to analyze the political and establishment leanings of opening paragraphs in Polish online news articles that discuss controversial topics. This task, often time-consuming and costly when performed by humans, holds promise for enhancing media fairness and balance cost-effectively. It is one of the first attempts to analyze this problem in an under-resourced language like Polish. Findings reveal the challenge of task difficulty for LLMs and human annotators. Accuracy of LLMs varies, with better performance on questions related to political leaning than establishment stance. Notably, the text-davinci-002 model achieves the highest scores in wing orientation detection, whereas gpt-3.5-turbo excels in recognizing pro-/anti-government bias. This study highlights LLMs' limitations for this demanding task in an under-resourced language like Polish, emphasizing the need for alternative detection approaches.},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}

@inproceedings{Takeshita2023Moral,

title = {Towards Theory-based Moral AI: Moral AI with Aggregating Models Based on Normative Ethical Theory},

author = {Masashi Takeshita and Rafal Rzepka and Kenji Araki},

year  = {2023},

date = {2023-08-01},

booktitle = {Proceedings of Ethics and Trust in Human-AI Collaboration: Socio-Technical Approaches, IJCAI 2023 Workshop},

address = {Macao, China},

keywords = {},

pubstate = {published},

tppubtype = {inproceedings}

}