United States

CogSci 2025

Probing for experience-driven critical period effects in a large language model

statistical learning

learning

language acquisition

machine learning

### Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

Message from the General Chair:<br>
*It is my great pleasure and honor to welcome you to the 63rd Annual Meeting of the Association for  Computational Linguistics (ACL 2025), held in beautiful Vienna, Austria, from July 27 to August 1, 2025.  ACL2025continues our field’s tradition of excellence in scholarship, innovation, and inclusivity, and I am  deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.*<br>
[Read more](https://drive.google.com/file/d/1GI_hvOpjswAuYdUTromfeDiPpCcqidwg/view?usp=sharing)<br><br><br>



To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to ACL 2025". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you need to log in with the **email address you registered with**. 

ACL 2025

Welcome to The 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)

poster

### Welcome to CogSci Conference 2025!

The 47th Annual Meeting of the Cognitive Science Society was a hybrid meeting held in San Francisco. 

<div style="position:relative;padding-top:0;width:900px;height:500px;"><iframe style="position:absolute;border:none;width:100%;height:100%;left:0;top:0;" src="https://online.fliphtml5.com/ebtyf/amvr/"  seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe></div>

#### About

The Cognitive Science Society brings together researchers from around the world who hold a common goal: understanding the nature of the human mind. The mission of the Society is to promote Cognitive Science as a discipline, and to foster scientific interchange among researchers in various areas of study, including Artificial Intelligence, Linguistics, Anthropology, Psychology, Neuroscience, Philosophy, and Education.

The Society is a non-profit professional organization and its activities include sponsoring an annual conference and publishing the journals Cognitive Science and TopiCS.

#### Our History 

* **Society Creation**<br>
The Society was incorporated as a 501(c)(3) non-profit professional organization in Massachusetts in 1979. The organizing committee included Roger Schank, Allan Collins, Donald Norman, and a number of other scholars from psychology, linguistics, computer science, and philosophy. 
<br><br>
* **Conference Creation**<br>
The first conference on cognitive science was held at La Jolla, California in August, 1979, and has occurred annually since then. The proceedings of each conference are published, and those from most years are available through Lawrence Erlbaum Associates, Inc. The annual proceedings of the Cognitive Science Conference represent a major source of information on new work and new ideas in the scientific study of thinking. In 1990, the Society, with help from an anonymous donor, established the David Marr Prize for the best student paper at each annual meeting.
<br><br>
* **Journal Creation**<br>
The Journal, Cognitive Science, began publication in 1976, and is now published by Wiley-Blackwell. The Executive Editor is currently Richard P. Cooper of Birkbeck, University of London, and there are 18 Associate Editors and a 30-member editorial board. It serves as the premier outlet for research reports that intersect two or more disciplines. Copyrights for articles published in the journal are held by the Society. The Governing Board of the Cognitive Science Society voted in late 2006 to found a new journal, Topics in Cognitive Science (topiCS). The Editor in Chief is Wayne Gray, Cognitive Science Department, Rensselaer Polytechnic Institute. The journal seeks to fill a niche not occupied by Cognitive Science Journal or other cognitive science journals. Membership in the Society includes a subscription to Cognitive Science and TopiCS. Copyrights for articles published in the journal are held by the Society.
<br><br>

#### Code of Conduct

By attending the CogSci 2025 Conference, you are required to adhere to the society’s **[Code of Conduct](https://drive.google.com/file/d/1ChPuihLy6jE_BWqfO7J2KKgX35JW2zsM/view?usp=sharing)**.
<br><br>


You need to log in with the email address you registered with. 

Login credentials were sent to you from Underline -  subject line "Welcome to the CogSci 2025 Conference". Please be sure to check your spam/promotional inbox  if you do not see an email confirmation right away.





Please log in to join this event.

To access the site, please register [**here**](https://cognitivesciencesociety.org/registration/).

If you are registered and feel like you are seeing this message by mistake, please make sure you are logged in with the same email that you registered with. 

Please register!

The 47th Annual Meeting of the Cognitive Science Society presents the latest research across cognitive science and highlights the theme of Cognition in Context.

Storytelling, a cornerstone of human culture, thrives on creativity—a domain where modern AI systems often falter. While these systems excel at maintaining narrative coherence, they frequently produce technically sound yet predictable narratives. To address this gap, we introduce IMPROVISER, a collaborative system grounded in psychological theories of creativity, specifically the Blind Variation and Selective Retention (BVSR) framework. IMPROVISER leverages multiple AI personas, each embodying distinct narrative styles, to generate diverse story variations. Users iteratively refine these ideas through selective retention, balancing novelty, coherence, and personalization. In controlled experiments, IMPROVISER outperformed single-persona and chatbot-based systems on creativity and diversity, producing longer, more spatially rich stories without compromising readability. These results empirically validate BVSR’s role in computational creativity and establish a framework for human-AI co-creation, demonstrating how AI can amplify—rather than replace—human creative potential.

IMPROVISER: Multi-persona Co-creation System Enhances Story Creativity

Instructional designers face an overwhelming array of design choices, making it challenging to identify the most effective interventions. To address this issue, I propose the concept of a Model Human Learner, a unified computational model of learning that can aid designers in evaluating candidate interventions. This paper presents the first successful demonstration of this concept, showing that a computational model can accurately predict the outcomes of two human A/B experiments---one testing a problem sequencing intervention and the other testing an item design intervention. It also demonstrates that such a model can generate learning curves without requiring human data and provide theoretical insights into why an instructional intervention is effective. These findings lay the groundwork for future Model Human Learners that integrate cognitive and learning theories to support instructional design across diverse tasks and interventions.

Model Human Learners: Computational Models to Guide Instructional Design

This study investigated how the length of AI-generated text explanations in Japanese influences trust and acceptance in a route selection task, a subjective decision-making task. Experiment 1 demonstrated that AI explanations increased trust and acceptance compared with no explanation; however, there was no significant difference between the effects of 100- and 300-character explanations. Experiment 2 further explored the threshold of explanation length by including 50- and 1,000-character explanations. The results showed that 300- and 1,000-character explanations increased acceptance compared to 50-character explanations; however, no differences emerged among the 100-, 300-, and 1,000-character explanations. Additionally, trust ratings were unaffected by explanation length. These results suggest that AI explanations have a threshold; in this study, AI explanations exceeding 100 characters in Japanese (approximately 50 words in English) did not lead to further changes in users' decisions to accept or reject AI recommendations.

Effects of AI Explanation Length on User Trust and Acceptance

The SNARC effect would reflect a left-to-right mental number line (MNL). The aim of this study was to disentangle the roles of numbers’ magnitude and order in the SNARC effect by using playing cards as stimuli. While most people hold cards in ascending order (AO), consistent with the MNL, a subset of individuals stably hold cards in descending order (DO), opposite to the MNL, with low-value cards (e.g., 2) on the right and high-value cards (e.g., 6) on the left. In a magnitude classification task (Experiment 1) and in a parity judgement task (Experiment 2) involving both digits and playing cards as stimuli, AO participants showed regular SNARC effects for both digits and cards, whereas DO participants showed regular SNARC effects for digits but not for cards. These results suggest that the order of cards in DO participants prevents the SNARC effect from occurring, but does not reverse the effect.

A study on playing cards to disentangle order and magnitude in the SNARC effect

Predicting human decision-making under risk and uncertainty is a long-standing challenge in cognitive science, economics, and AI. While prior research has focused on numerically described lotteries, real-world decisions often rely on textual descriptions. This study conducts the first large-scale exploration of human decision-making in such tasks using a large dataset of one-shot binary choices between textually described lotteries. We evaluate multiple computational approaches, including fine-tuning Large Language Models (LLMs), leveraging embeddings, and integrating behavioral theories of choice under risk. Our results show that fine-tuned LLMs, specifically  GPT-4o, outperform hybrid models that incorporate behavioral theory, challenging established methods in numerical settings. These findings highlight fundamental differences in how textual and numerical information influence decision-making and underscore the need for new modeling strategies to bridge this gap.

Predicting Human Choice Between Textually Described Lotteries

The process by which people adjust, enrich, and revise their understanding of word meanings over time – so-called ‘Slow Mapping’ – has often been overlooked, particularly in terms of how a computationally bounded learner might approach such a task. To address this gap, we propose a process model of incremental word-meaning induction. This proposal is inspired by recent work on concept and theory change grounded in a probabilistic language of thought (pLOT). We focus on the problem of fixing the meanings of words from usage examples, taking kinship terms as our test domain. We frame word meaning induction at a computational level as a program induction problem, and hypothesize that individual learners search for possible meanings as evidence arrives via a mutative Markov-Chain Monte-Carlo search scheme. We show this idea provides a better description of how participants’ generalizations and tentative definitions of alien kinship words shift as evidence arrives, outperforming normative accounts and other baselines.

An Incremental Program Induction Model of Slow Mapping Words to Meanings

The approximately number system (ANS) helps people quickly estimate the numerosity of objects in their environment. In this study, we explore one proposed mechanism for visually perceiving numerosity: visual clustering. Participants completed a magnitude comparison task, magnitude estimation task, and a clustering task using the same set of numerosity stimuli. The stimuli varied in the spatial configuration of the points (cluster structure -- clustered or dispersed) and in the number of points present. Participants judged stimuli with dispersed cluster structure to be more numerous in the magnitude comparison task. However, there was a minimal effect of cluster structure in the magnitude estimation task and no effect of the number of clusters perceived in both tasks. We also found that the clusters people perceived in the third task did not explain the effects of cluster structure. These findings go against strong claims that people use visual clustering to judge numerosity and set the stage for further investigations into the mechanisms underlying the ANS.

Perceived clusters may not explain people's judgments of approximate numerosity

During reading, the mind continuously builds and updates a discourse model—a cumulative mental world representing the gleaned information. A key operator in this process is establishing novel discourse referents—entities in the model that can be picked out. For instance, ‘She bought wool, sponges and steel’ establishes three simple referents (italicized). By contrast, ‘She bought sponges of steel wool…’ establishes two simple referents forming a composite, itself referenceable (e.g., ‘…and glued them together.’). Here, in an online reading study, we target the cognitive basis of establishing simple and composite referents in the discourse model. Participants (n=43) read 72 five-sentence English stories sentence by sentence. The fourth, critical sentence featured: (i) three simple referents (‘wool, sponges and steel’; simple3); (ii) two simple referents (‘steel wool and sponges’; simple2), or (iii) two simple referents forming a composite referent (‘sponges of steel wool’; composite). Crucially, across conditions, critical sentences had identical lengths and lexical items, while remaining sentences were fully identical. A true/false comprehension task followed each story. We hypothesized that additional referents would increase reading times (RTs) beyond syntactic, semantic, and lexical factors. Multiple regression analyses showed significant effects of adding simple (RTsimple3>RTsimple2) and composite (RTcomposite>RTsimple2) referents. The effects appeared only on critical (but not on subsequent) sentences, possibly reflecting the cognitive operators of establishing, rather than maintaining, novel referents. Our findings pave the way for future work investigating hypotheses of hierarchical structure-building in mental discourse models.

Reading comprehension involves adding simple and composite discourse referents to a mental model

Derivation of pragmatic inferences typically assumes that both interlocutors behave rationally, as described by the cooperative principle. However, real-world communication often involves speakers who cannot behave fully rationally due to factors such as limited language proficiency, high cognitive load, or insufficient reasoning skills. In such cases, listeners may adjust their inferences to account for the speaker’s limitations. In this study, we investigate whether participants engage in ad-hoc reasoning about alternative messages available to the speaker when the speaker is explicitly literal. Our findings reveal that people overwhelmingly do not do so and instead behave as literal listeners, and that nudging participants to consider alternative messages does not improve performance. This suggests that while people readily consider the speaker's intent, they do not tend to engage in ad-hoc reasoning about probabilities of alternative messages in the absence of a rational speaker.

People do not engage in ad-hoc reasoning about alternative messages when interacting with a literal speaker

Contemporary transformer models have achieved human-like performance on many text-based tasks. However, real-world communication requires the integration of language with non-linguistic context (e.g., visual, social, etc.). Here, we study such information integration in three multimodal transformer models. We test these models’ pragmatic capabilities regarding referring expressions: when an object set contains two exemplars from the same category that differ in size, unambiguously referring to one of them requires a size adjective (e.g., the big hammer); the adjective is unnecessary if only one exemplar from the category is present. We evaluate these inferences when models process text-image inputs (via their surprisal for infelicitous vs. felicitous adjective use) and when they generate open-ended descriptions of images given text prompts. We find evidence for pragmatic integration of visual and linguistic context in all models. However, these inferences remain sensitive to the in-context statistics of visual inputs, unlike pragmatic inference in humans.

Downloads

Next from CogSci 2025

IMPROVISER: Multi-persona Co-creation System Enhances Story Creativity

Similar lecture

Developmentally-plausible Working Memory Shapes a Critical Period for Language Acquisition

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from CogSci 2025

IMPROVISER: Multi-persona Co-creation System Enhances Story Creativity

Similar lecture

Developmentally-plausible Working Memory Shapes a Critical Period for Language Acquisition

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads