ChiSCor: A Corpus of Freely-Told Fantasy Stories by Dutch Children for Computational Linguistics and Cognitive Science DOI Creative Commons
Bram van Dijk, Max van Duijn, Suzan Verberne

et al.

Published: Jan. 1, 2023

In this resource paper we release ChiSCor, a new corpus containing 619 fantasy stories, told freely by 442 Dutch children aged 4-12. ChiSCor was compiled for studying how render character perspectives, and unravelling language cognition in development, with computational tools. Unlike existing resources, ChiSCor’s stories were produced natural contexts, line recent calls more ecologically valid datasets. hosts text, audio, annotations complexity linguistic complexity. Additional metadata (e.g. education of caregivers) is available one third the children. also includes small set 62 English stories. This details shows its potential future work three brief case studies: i) show that syntactic strikingly stable across children’s ages; ii) extend on Zipfian distributions free speech obeys Zipf’s law closely, reflecting social context; iii) even though relatively small, rich enough to train informative lemma vectors allow us analyse use. We end reflection value narrative datasets linguistics.

Language: Английский

Strong and weak alignment of large language models with human values DOI Creative Commons
Mehdi Khamassi,

Marceau Nahon,

Raja Chatila

et al.

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Aug. 21, 2024

Minimizing negative impacts of Artificial Intelligent (AI) systems on human societies without supervision requires them to be able align with values. However, most current work only addresses this issue from a technical point view, e.g., improving methods relying reinforcement learning feedback, neglecting what it means and is required for alignment occur. Here, we propose distinguish strong weak value alignment. Strong cognitive abilities (either human-like or different humans) such as understanding reasoning about agents' intentions their ability causally produce desired effects. We argue that AI like large language models (LLMs) recognize situations presenting risk values may flouted. To illustrate distinction, present series prompts showing ChatGPT's, Gemini's Copilot's failures some these situations. moreover analyze word embeddings show the nearest neighbors in LLMs differ humans' semantic representations. then new thought experiment call "the Chinese room transition dictionary", extension John Searle's famous proposal. finally mention promising research directions towards alignment, which could statistically satisfying answers number common situations, however so far ensuring any truth value.

Language: Английский

Citations

6
Distributional Semantics: Meaning Through Culture and Interaction DOI
Pablo Contreras Kallens, Morten H. Christiansen

Topics in Cognitive Science, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 26, 2024

Mastering how to convey meanings using language is perhaps the main challenge facing any learner. However, satisfactory accounts of this achieved, and even what it for a linguistic item have meaning, are hard come by. Nick Chater was one pioneers involved in early development most successful methodologies within cognitive science discovering meaning: distributional semantics. In article, we review approach discuss its successes shortcomings capturing semantic phenomena. particular, dub "distributional paradox:" can models that do not implement essential dimensions human processing, such as sensorimotor grounding, capture so many meaning-related phenomena? We conclude by providing preliminary answer, arguing statistical scaffolding acquisition allows communication, which, line with Chater's more recent ideas, has been shaped features cognition on timescale cultural evolution.

Language: Английский

Citations

4
Więcej niż „trudny problem świadomości”: o możliwości powstania świadomej sztucznej inteligencji DOI Open Access
Piotr Przybysz

Człowiek i Społeczeństwo, Journal Year: 2025, Volume and Issue: 58, P. 55 - 87

Published: Jan. 27, 2025

The aim of this paper is to review the most important questions and problems concerning emergence conscious ai. In paper, I point out three such key problems: (1) how recognize that ai has acquired consciousness? (2) can emerge? (3) what properties have? argue these cannot currently be solved on basis purely experimental, computer science, engineering approaches, because path leads through areas marked by previous philosophical general theoretical reflection subject.

Language: Английский

Citations

0
Meta-learning contributes to cultivation of wisdom in moral domains: Implications of recent artificial intelligence research and educational considerations DOI
Hyemin Han

International Journal of Ethics Education, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 20, 2025

Language: Английский

Citations

0
Why do we need to employ exemplars in moral education? Insights from recent advances in research on artificial intelligence DOI
Hyemin Han

Ethics & Behavior, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 18

Published: April 29, 2024

In this paper, I examine why moral exemplars are useful and even necessary in education despite several critiques. To support my point, review recent AI research demonstrating that exemplar-based learning is superior to rule-based model performance training neural networks, such as large language models. particularly focus on aiming at promoting the development of multifaceted functioning can be done effectively by using exemplars, which like training. Furthermore, discuss potential limitations issues related exemplar-applied with findings from raising concerns about biases toxic outcomes. propose ways address regarding employing well. As remedies, suggest autonomy-supporting deliberative reflective should utilized. based discussion, how macroscopic socio-cultural aspects influence effectiveness education.

Language: Английский

Citations

3
Active Use of Latent Constituency Representation in both Humans and Large Language Models DOI
Nai Ding, Wei Liu, Ming Xiang

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: July 10, 2024

Abstract Understanding how sentences are internally represented in the human brain, as well large language models (LLMs) such ChatGPT, is a major challenge for cognitive science. Classic linguistic theories propose that brain represents sentence by parsing it into hierarchically organized constituents. In contrast, LLMs do not explicitly parse constituents and their latent representations remains poorly explained. Here, we demonstrate humans construct similar of hierarchical analyzing behaviors during novel one-shot learning task, which they infer words should be deleted from sentence. Both tend to delete constituent, instead nonconstituent word string. naive sequence processing model has access properties ordinal positions does show this property. Based on deletion behaviors, can reconstruct constituency tree representation both LLMs. These results tree-structured emerge

Language: Английский

Citations

0
Enhancing Pragmatic Nuance Decoding in Bidirectional Encoder Representation from Transformer DOI

Johnwendy Chinedu Nwaukwa,

Imianvan Anthony Agboizebeta

Published: April 2, 2024

Language: Английский

Citations

0
ChiSCor: A Corpus of Freely-Told Fantasy Stories by Dutch Children for Computational Linguistics and Cognitive Science DOI Creative Commons
Bram van Dijk, Max van Duijn, Suzan Verberne

et al.

Published: Jan. 1, 2023

In this resource paper we release ChiSCor, a new corpus containing 619 fantasy stories, told freely by 442 Dutch children aged 4-12. ChiSCor was compiled for studying how render character perspectives, and unravelling language cognition in development, with computational tools. Unlike existing resources, ChiSCor’s stories were produced natural contexts, line recent calls more ecologically valid datasets. hosts text, audio, annotations complexity linguistic complexity. Additional metadata (e.g. education of caregivers) is available one third the children. also includes small set 62 English stories. This details shows its potential future work three brief case studies: i) show that syntactic strikingly stable across children’s ages; ii) extend on Zipfian distributions free speech obeys Zipf’s law closely, reflecting social context; iii) even though relatively small, rich enough to train informative lemma vectors allow us analyse use. We end reflection value narrative datasets linguistics.

Language: Английский

Citations

1