United States

Humans have the ability to rapidly understand rich combinatorial concepts from limited data. Here we investigate this ability in the context of auditory signals, which have been evolved in a cultural transmission experiment to study the emergence of combinatorial structure in language. We propose a neuro-symbolic generative model which combines the strengths of previous approaches to concept learning. Our model performs fast inference drawing on neural network methods, while still retaining the interpretability and generalization from limited data seen in structured generative approaches. This model outperforms a purely neural network-based approach on classification as evaluated against both ground truth and human experimental classification preferences, and produces superior reproductions of observed signals as well. Our results demonstrate the power of flexible combined neural-symbolic architectures for human-like generalization in raw perceptual domains and offers a step towards developing precise computational models of inductive biases in language evolution.

CogSci 2021

Learning Evolved Combinatorial Symbols with a Neuro-symbolic Generative Model

### Looking for Cogsci 2023? Please [click here](https://underline.io/events/405/reception) to switch to 2023 edition.


**Dear Colleagues,**

Welcome to the 43rd Annual Conference of the Cognitive Science Society! 

This year’s conference brings together the latest research from all Cognitive Science and also highlights this year’s theme **Comparative Cognition: Animal Minds.** 

The program features four plenary speakers: Friederike Range (University of Veterinary Medicine, Vienna), Onur Güntürkün (Ruhr University Bochum), Julia Fischer (University of Göttingen), and Erich Jarvis (The Rockefeller University). The keynotes and invited symposia showcase the broad spectrum of ideas that encompass the conference theme of comparative cognition. 

Further, the program includes the Rumelhart Prize keynote presentation by Susan Goldin-Meadow (University of Chicago), and symposia to celebrate the winner of the Elman Prize: Wei Ji Ma (New York University), the Heineken Award winner Robert Zatorre (McGill University), and the Robert J. Glushko Dissertation Prizes session.

The program committee for CogSci 2021 received 1,025 submissions, including 710 full papers, 275 member abstracts, 8 publication-based talks, as well as 15 proposals for symposia, and 17 for workshops and tutorials. After a rigorous review process, the committee selected 152 papers for oral presentation and inclusion in the conference proceedings (14.8%), 332 papers for poster presentation and inclusion in the proceedings (32.4%), and 175 papers for poster presentation with inclusion of abstracts in the proceedings (17.1%) We also selected 275 submitted member abstracts. In addition, we accepted 6 publication-based talks, 12 symposia, and 9 workshops and tutorials to make for an exciting and inclusive program.

We hope that you enjoy the conference this year!<br><br>

**Your Program Chairs,**
![](https://assets.underline.io/uploads/markdown_image/1/image/99984908caba5e6453e5fc5ba34f3507.jpg)

**Learn more about the Vienna Cognitive Science Hub at the University of Vienna, hosts of CogSci 2021**

<iframe width="560" height="315" src="https://www.youtube.com/embed/464h1m3iU68" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>

To review the proceedings for the conference or review the searchable program, please visit the [website](https://cognitivesciencesociety.org/program/).


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.

poster

Based on strategy-awareness (knowing which problem-solving strategy to use) and time-awareness (knowing when to use it), students are categorized into Rote (neither type of awareness), Dabbler (strategy-aware only) or Selective (both types of awareness). It was shown that Selective is often significantly more prepared for future learning than Rote and Dabbler (Abdelshiheed et al., 2020). In this work, we explore the impact of explicit strategy instruction on Rote and Dabbler students across two domains: logic and probability. During the logic instruction, our logic tutor handles both Forward- Chaining (FC) and Backward-Chaining (BC) strategies, with FC being the default; the Experimental condition is taught how to use BC via worked examples and when to use it via prompts. Six weeks later, all students are trained on a probability tutor that supports BC only. Our results show that Experimental significantly outperforms Control in both domains, and Experimental Rote catches up with Selective.

Preparing Unprepared Students For Future Learning

Frequency counters are computational mechanisms that track the frequency or probability of speech units. Such counters are idealizations which re-describe frequency effects in early word segmentation, not providing an underlying learning mechanism from which these effects arise. Previous work has shown that Implicit Chunking represents a plausible learning mechanism explaining infants’ sensitivity to statistical cues when segmenting small-scale artificial languages (French et al., 2011). However, no work has examined whether Implicit Chunking allows to segment naturalistic speech in a developmentally plausible way. Here, we show how a novel symbolic model of Implicit Chunking – CLASSIC-Utterance-Boundary - performs better or as well as previous frequency-based models (i.e., transitional probability, chunking) at predicting children’s word age of first production and a range of word-level characteristics of children’s vocabularies (word frequency, word length, neighborhood density, phonotactic probability). We suggest that explicit frequency counters are not necessary to explain infants’ speech segmentation in naturalistic settings.

Are Explicit Frequency Counters Necessary in Computational Models of Early Word Segmentation?

We test whether people are able to reason based on incidentally acquired probabilistic and context-specific magnitude information. We manipulated variance of values drawn from two normal distributions as participants perform an unrelated counting task. Our results show that people do learn category-specific information incidentally, and that the pattern of their judgments is broadly consistent with normative Bayesian reasoning at the cohort level, but with large individual-level variability. We find that this variability is explained well by a frugal memory sampling approximation; observer models making this assumption explain approximately 70% of the variation in participants' responses. We also find that behavior while judging easily discriminable categories is consistent with a model observer drawing fewer samples from memory, while behavior while judging less discriminable categories is better fit by models drawing more samples from memory. Thus, our model-based analysis additionally reveals resource-rationality in memory sampling.

One and known: Incidental probability judgments from very few samples

The `chattr` R package enables users to easily detect and describe temporal contingencies in pre-annotated interactional data. Temporal contingency analysis is ubiquitous across signal system research, including human and non-human animal communication. Current approaches require manual evaluation (i.e., do not scale up), are proprietary/over-specialized (i.e., have limited utility), or are constructed ad-hoc per study (i.e., are variable in construct). `Chattr`'s theoretically motivated, customizable, and open source code provides a set of core functions that allow users to quickly and automatically extract contingency information in data already annotated for interactant activity (via manual or automated annotation). We demonstrate the use of `chattr` by testing predictions about turn-taking behavior in three language development corpora. We find that the package effectively recovers documented variation in linguistic input given both manual and automatically created speech annotations and note future directions for package development key to its use across multiple research domains.

Analyzing contingent interactions in R with `chattr`

Categories such as 'animal' or 'furniture' play a pivotal role in processing, organizing, and communicating world knowledge. Many theories and computational models of categorization exist, but evaluation has disproportionately focused on artificially simplified learning problems (e.g., by assuming a given set of relevant features or small data sets); and on English native speakers. This paper presents a large-scale computational study of category and feature learning. We approximate the learning environment with natural language text, and scale previous work in three ways: We (1) model the full complexity of the learning process, acquiring learning categories and structured features jointly; (2) study the generalizability of categorization models to five diverse languages; and (3) learn categorizations comprising hundreds of concepts and thousands of features. Our experiments show that meaningful representations emerge across languages. We further demonstrate a joint model of category and feature acquisition produces more relevant and coherent features than simpler models, suggesting it as an exploratory tool to support cross-cultural categorization studies.

Categorization in the Wild: Category and Feature Learning across Languages

Cognates – words which share form and meaning across two languages – have been extensively studied to understand the bilingual mental lexicon. One consistent finding is that bilingual speakers process cognates faster than non-cognates, an effect known as cognate facilitation. Yet, there is no agreement on the underlying factors driving this effect. In this paper, we use computational modeling to test whether the effect can be explained by the cumulative frequency hypothesis. We train a computational language model on two language pairs (Dutch–English, Norwegian–English) under different conditions of input presentation and test it on sentence stimuli from two existing studies with bilingual speakers of those languages. We find that our model can exhibit a cognate effect, lending support to the cumulative frequency hypothesis. Further analyses reveal that the size of the effect in the model depends on its linguistic accuracy. We interpret our results within the literature on cognate processing.

Cumulative frequency can explain cognate facilitation in language models

The exercise of reopening optional social spaces following the COVID-19 lockdowns presents each individual with a complex problem in determining whether or not to attend these spaces given how the risks of virus transmission scale with crowding. In order to tackle this problem while recognizing individual cognitive capacity limits, this paper used a simulation model based on the El Farol Bar Problem and generated a population of agents relying on simple predictive strategies to determine their attendance to a retail location. It was determined that the more heterogeneous the threshold for crowding among agents was, the less variance there was in overall attendance numbers. This stability in daily attendance comes at the expense of the ability of the most cautious members to enjoy recreational spaces as these locations become the realm of only those least concerned with potential crowding.

Predicting Social Reopening Following COVID-19 Lockdown Using Bounded Rationality and Threshold Models

Here we examined whether eye movement measures in global-local perceptual processing tasks, where abnormalities were typically found in individuals with schizophrenia, could be used to predict schizotypy through Eye Movement analysis with Hidden Markov Models (EMHMM). Using both multiple regression analysis and Gaussian process classifier to predict schizotypy, we found that in addition to longer response times in contour integration, a less consistent eye fixation to locate a stimulus and a more consistent subsequent fixation to start engaging local processing in the embedded figures task predicted high schizotypy. These effects may be related to reduced top-down attention control due to deficient global processing and enhanced local processing bias respectively. In addition, performance in embedded figures could further enhance classification accuracy when being used in conjunction with the above predictors, suggesting the multifactorial nature of the identification problem. These predictors may be important endophenotype markers for schizotypal personality.

Eye movement consistency in global-local perceptual processing predicts schizotypy

A growing toolbox is emerging for linking neuroimaging data to computations supporting human cognition. In representational similarity analysis (RSA), for example, activity patterns over voxels are compared in response to similar vs. dissimilar stimuli. Resulting similarity matrices are compared to similarity matrices based on theoretical principles or computational models. Similarly, complex EEG or MEG time series can be compared to information-theoretic variables, such as stimulus entropy or surprisal, allowing inferences about the sensitivity of neural responses to aspects of signal information over time. A challenge is determining to what degree such analyses can identify hallmarks of specific computations rather than computationally non-specific resonance with inputs. Here, we apply RSA and information-theoretic analyses to one well-characterized model: TRACE. We consider whether these analyses identify known principles underlying TRACE, and whether TRACE exhibits sensitivity to information-theoretic variables similar to that observed in human brains.

Discovering computational principles in models and brains

What computations enable humans to leap from mere obser- vations to rich explanatory theories? Prior work has focused on stochastic algorithms that rely on random, local perturba- tions to model the search for satisfactory theories. Here we introduce a new approach inspired by the practice of ‘debug- ging’ from computer programming, whereby learners use past experience to constrain future proposals, and are thus able to consider large leaps in their current theory to fix specific de- ficiencies. We apply our ‘debugging’ algorithm to the mag- netism domain introduced by (Ullman, Goodman, & Tenen- baum, 2010) and compare its efficiency and accuracy to their stochastic-search algorithm. We find that our algorithm not only requires fewer iterations to find a solution, but that the solutions it finds more reliably recover the correct latent theo- ries, and are more robust to sparse data. Our findings suggest the promise of such constraint-based approaches to emulate the way humans efficiently navigate large, discrete hypothesis spaces.

Learning Evolved Combinatorial Symbols with a Neuro-symbolic Generative Model

Downloads

Next from CogSci 2021

Preparing Unprepared Students For Future Learning

Similar lecture

Visual representation of negation: Real world data analysis on comic image design