United States

There are at least three ways of learning how the world works: learning from observations, from interventions, and from explanations. Prior work on causal inference focused on how people learn the underlying causal structure of a model through observation and intervention. Our study is the first to look at what people can learn from explanations in the context of a causal structure learning paradigm. We develop a normative inference model that learns from observations and explanations, and compare the model&#39;s predictions to participants&#39; performance. The task is to infer the causal connections in 3-node graphs, based on information about their co-activation, and explanations of the kind &quot;B activated because A activated&quot;. We find that participants learn better from explanations than from observations. However, while the normative model benefits from having observations in addition to explanations, participants did not.

CogSci 2023

Show and tell: Learning causal structures from observations and explanations

## Welcome to EMNLP 2022!
I am delighted to welcome you to EMNLP 2022! I believe this conference has been complicated beyond any precedent. Over the past year, it’s been thrilling to see the organization team approach each new puzzle with creativity and enthusiasm. We hope that those participating in Abu Dhabi as well as those joining remotely will leave the conference feeling newly inspired by the program and newly connected to our ever-growing community. Following EMNLP 2021 and major NLP conferences since, EMNLP 2022 is “hybrid,” serving both virtual and in-person participants.

Our key innovations for EMNLP 2022 include:

* EMNLP 2022 is “hybrid” in a second sense, as well: we allowed both direct and rolling review paper submissions, building on the pilot experiment of EMNLP 2021, which considered a small number of ARR submissions. 
* Familiar from NAACL but new to EMNLP, we’ve added an industry track.
* During the conference, “portals” will link virtual poster sessions to in-person conference participants during poster sessions each day.
* The first ACL-family conference in the United Arab Emirates.

 *Message from Noah A. Smith, University of Washington and Allen Institute for AI, Seattle, Washington, USA* 
***EMNLP 2022 General Chair***
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/9eec7d4a287ee18c278b08229290aa83.png)](https://drive.google.com/file/d/1OlPv6QBeo62VVTughj2jkiLeyHd1WnUt/view)
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/a3db7a768409f05192210d98601edb25.png)](https://emnlp2022.rocket.chat/)

To access this site you need to register. Please register [here](https://2022.emnlp.org/registration/).

Register here

EMNLP 2022

Welcome!
EMNLP 2022 will take place in Abu Dhabi from December 7th to December 11th, 2022. And it will be held in hybrid mode, both online and offline.

There are at least three ways of learning how the world works: learning from observations, from interventions, and from explanations. Prior work on causal inference focused on how people learn the underlying causal structure of a model through observation and intervention. Our study is the first to look at what people can learn from explanations in the context of a causal structure learning paradigm. We develop a normative inference model that learns from observations and explanations, and compare the model's predictions to participants' performance. The task is to infer the causal connections in 3-node graphs, based on information about their co-activation, and explanations of the kind "B activated because A activated". We find that participants learn better from explanations than from observations. However, while the normative model benefits from having observations in addition to explanations, participants did not.

poster

The 45th Annual Meeting of the Cognitive Science Society will take place in Sydney, Australia as a modified hybrid conference. Presentations from the main-session room will be live-streamed every day for all events. This will include all invited talks and award sessions. Recordings of all CogSci 2023 sessions will be made available for on-demand viewing after the conference. 
The in-person component will be held at the International Convention Centre Sydney at Darling Harbour. The Conference will take place Wednesday July 26th – Saturday July 29th, 2023 (Wednesday, July 26th is the Workshops Day).

Join us!

 [**The Program at a Glance can be found here**](https://cognitivesciencesociety.org/wp-content/uploads/2023/07/COgSci-PAAG-V2.pdf)
 [**The Detailed Program can be found here**](https://online.fliphtml5.com/ebtyf/mswz/)
 

#### **CogSci 2023 Program Book**
<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/tzgk/" seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>

The 45th Annual Meeting of the Cognitive Science Society presents the latest research across cognitive science and highlights the theme of Cognition in Context.

Children exhibit a sophisticated explanatory judgment: they expect, generate, and judge explanations. Do children also exhibit a metaphysical explanatory judgment, such that everything must have an explanation? If they do, this would conform to what philosophers have called the Principle of Sufficient Reason (PSR). In this study, 6-9-year-old children (N = 55, Mage = 8.07, SDage = 1.32) were shown a number of statements across domains (Psychology, Biology, Nature, Physics, Religion, and Supernatural) and were asked if they agree or disagree that each statement has to have an explanation. As a comparison, children were also asked about coincidences, which, metaphysically, do not have an explanation. We found that children conform to the PSR: of the statements that children thought were true, children of all ages believed that the statements must have an explanation, except for coincidences. This is the first step at uncovering the development of our metaphysical explanatory judgment.

Must there be an explanation? Children and the Principle of Sufficient Reason

In science, we use common graphical representations to indicate changes in events over time, independent of domain. Are children also sensitive to abstract patterns in the ways events change over time? In a series of four experiments, we show that young children (range: 48-84 months) distinguish different function families (Exp 1). Children can also distinguish specific function types within function families (e.g., between linear and sigmoid monotonic functions; Exp 2); nonetheless, they will group different function types within a family together, rather than with functions in a different family (Exp 3). Finally, we show that children’s sensitivity to functions is abstract, allowing them to match observable causes to verbal descriptions of their effects (Exp 4). These results suggest that although some aspects of function understanding, like learning how to interpret graphs, requires formal education, the ability to identify abstract functional relationships is intuitive and early emerging.

Representations of Abstract Relations in Early Childhood

Previous work has shown that even young infants can “add” or “subtract” quantities of objects using the non-symbolic object tracking system. We asked how these non-symbolic addition skills may change as children learn to count. Thirty 3-year-old children completed a non-symbolic exact addition task, in which children saw an animal collect two sets of objects inside a container, and children were asked to choose the summed total from two alternatives. We also measured children’s knowledge of the count list and their working memory capacity. We found that children with higher count knowledge performed better in solving non-symbolic addition problems with small sets of objects (<3). However, even strong counters struggled when one of the operands exceeded working memory capacity (>=3). Our results suggested that non-symbolic addition skills in 3-year-old’s may be influenced by both their emerging number knowledge and the object tracking system.

How count-list knowledge relates to 3-year-olds’ exact non-symbolic addition abilities

Motion quality can differentiate experts from novices in fields like surgery (Ershad et al., 2018). We extend approaches used by researchers in that field to examine the relationship between motion and skill in a children’s block-building task. We ask whether the relationship between these two variables is detected equally well by humans and machines—in this case, motion sensors embedded in the blocks. We investigate whether adults’ judgments about motion quality and children’s overall building skill reflect children’s actual construction ability, and whether data from embedded motion sensors predict children’s skill as well as adult judgments do. We find that human raters outperform the motion sensor data. Our findings raise questions about how people form such intuitive judgments of expertise, and how automated judgments of skill can be enhanced to more accurately predict expertise in block building and other similar cognitive tasks.

Naive human judges can accurately predict expertise in children's block building. Can embedded motion sensors do just as well?

Brain function connectivity, derived from functional magnetic resonance imaging (fMRI), has enjoyed high popularity in studies of Autism Spectrum Disorder (ASD) diagnosis. Albeit rapid progress has been made, most studies still suffer from several knotty issues: (1) the hardship of effectively modeling the sophisticated brain neuronal connectivity; 
(2) the dimensionality explosion resulted from excessive voxels in each fMRI sample; (3) the poor interpretability giving rise to unpersuasive diagnosis. To ameliorate these issues, we propose a GCN-based model, namely ConResNet. Specifically, a convolutional restructuring attention is designed for efficiently ROI-interaction feature extraction. Besides, we embed a salient graph pooling method to sift salient nodes. Extensive experiments conducted on ABIDE dataset demonstrate ConResNet achieves state-of-the-art performance. Moreover, the most salient brain regions predicted by ConResNet are closely consistent with theoretical knowledge in the medical domain, providing potential biomarkers for ASD clinical diagnosis. Codes have been uploaded to Github.

GCN-based Autism Spectrum Disorder Diagnosis via Convolutional Restructuring Attention

Most models in cognitive and computational neuroscience trained on one subject don’t generalize to other subjects due to individual differences. An ideal individual-to-individual neural converter is expected to generate real neural signals of one subject from those of another one, which can overcome the problem of individual differences for cognitive and computational models. In this study, we propose a novel individual-to-individual EEG converter, called EEG2EEG, inspired by generative models in computer vision. We applied THINGS EEG2 dataset to train and test 72 independent EEG2EEG models corresponding to 72 pairs across 9 subjects. Our results demonstrate that EEG2EEG is able to effectively learn the mapping of neural representations in EEG signals from one subject to another and achieve high conversion performance. Additionally, the generated EEG signals contain clearer representations of visual information than that can be obtained from real data. This method establishes a novel and state-of-the-art framework for neural conversion of EEG signals, which can realize a flexible and high-performance mapping from individual to individual and provide insight for both neural engineering and cognitive neuroscience.

Generate your neural signals from mine: individual-to-individual EEG converters

Recent advances in Natural Language Processing (NLP) make it possible to quantify relationships among different words extracted from large-scale human text corpora. Using a word embeddings model, we quantified the semantic distance between pairs of adjectives that could describe people or objects (e.g., smart, friendly; round, wooden) and scanned participants using fMRI while they had the opportunity to generalize from one attribute to another across parametrically varying degrees of semantic distance (e.g., given that this person is smart, how likely are they to be friendly?; given that this furniture is round, how likely is it to be wooden?). Across categories, we observed a positive parametric effect of semantic distance on activation in the dorsomedial prefrontal cortex (DMPFC). Results connect to this region’s role in abstraction and inference under reducible uncertainty, with implications for understanding how people generalize beyond what they know to make inferences about novel individuals, items, or experiences.

Measures of Semantic Distance from Word Embeddings Predict Neural Responses During Inferences about People and Objects

Action perception plays a crucial role in human life and is significant both evolutionarily and socially. It is supported by the well-established action observation network (AON), which consists of three core nodes in the occipitotemporal, parietal, and pre-motor cortex of the human brain. In the present study, we investigated how top-down attention affects the AON using fMRI and representational similarity analysis. In the first session, human participants viewed short videos while they performed tasks in which they attended different features of actions in separate blocks, including the actor, the target, or the effector. In a second fMRI session, participants viewed the videos in a passive manner to be able to compare the neural representations in the AON across passive and active attention tasks. Our results show that attention to different features of observed actions modulates the nodes of the AON differently indicating their distinct roles in supporting action perception.

Top-down effects of attention on the action observation network

The accurate reconstruction of deep cortical regional source activities from EEG recordings is crucial for understanding spatiotemporal cognitive processes and the origins of impairments in brain disorders. Recently, several state-of-the-art source localization algorithms have been developed that claim to outperform standard methods. However, a rigorous comparison of the strengths and weaknesses of these methods in EEG source localization has yet to be conducted. This study aims to evaluate the algorithms, namely Full Dugh, Thin Dugh, and Mackay, against standard methods, namely minimum norm and eLORETA, using a realistic head model and simulated spatially-propagated source signals. Results indicate that Thin Dugh performed superiorly in both spatial and temporal source reconstruction, followed by Mackay. Both Thin Dugh and Mackay effectively identified contributing sources and diminished the amplitudes of non-contributing sources. Additionally, we have also proven that there are serious theoretical flaws within Full Dugh algorithm through both mathematical analysis and simulation results.

Evaluating State-of-the-Art EEG Source Localization Algorithms Using Spatially Propagated Sources and Realistic Head Model

Brain networks have evolved to be highly efficient and robust while consuming minimal energy. They possess unique properties such as sparsity, complexity, and small-world nature. Studies suggest that a trade-off between minimal wiring cost and efficient communication shapes brain network development. However, it's not entirely clear what factors are influential and to what extent each factor contributes to this development. This research examines several potential underlying factors and finds that with connectivity guaranteed by fixed degree distribution, minimizing wiring cost has the largest impact on network structure among all factors, such as maximizing the clustering coefficient or coefficient of variation for wiring length distribution. Without degree constraints, the cost-efficiency balance reproduces optimally brain networks of five species, but minimizing wiring cost remains the primary determinant.

Premium content

Downloads

Next from CogSci 2023

Must there be an explanation? Children and the Principle of Sufficient Reason

Similar lecture

Faithful, Interpretable Model Explanations via Causal Abstraction