United States

Features of the physical world may be acquired from the statistical properties of language. Here we investigate how the Transformer Language Model T5 is able to gain knowledge of the visual world without being able to see or feel. In a series of four studies, we show that T5 possesses an implicit understanding of the relative sizes of animals, their weights, and their shapes, but not their colors, that aligns well with that of humans. As the size of the models was increased from 60M to 11B parameters, we found that the fit to human judgments improved dramatically, suggesting that the difference between humans and these learning systems might ultimately disappear as the parameter sizes grow even larger. The results imply that knowledge of the perceptual world—and much of semantic memory—might be acquired in dis-embodied learning systems using real-time inferential processes

CogSci 2021

What Transformers Might Know About the Physical World: T5 and the Origins of Knowledge

### 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! 

**Your Program Chairs,**
![](https://s3.eu-west-1.amazonaws.com/underline.prod/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.

**31st IEEE International Conference on Robot & Human Interactive Communication**
 This conference is a leading forum where state-of-the-art innovative results, the latest developments as well as future perspectives relating to robot and human interactive communication are presented and discussed.
 
 The conference covers a wide range of topics related to Robot and Human Interactive Communication, involving theories, methodologies, technologies, empirical and experimental studies. Papers related to the study of robotic technology, psychology, cognitive science, artificial intelligence, human factors, ethics and policies, interaction-based robot design and other topics related to human-robot interaction are welcome.
 
 **CONFERENCE THEME:**
 Social, asocial, and antisocial robots
 
 Robots are starting to populate our lives: at work, in our homes, hotels, shops. However, not everybody is comfortable with robots finding a place in society. For example, people may be afraid that robots will replace humans in their jobs. This belief is exacerbated by the media that, on one side, raise expectations of robots’ capabilities, while amplifying fears and worries on the other side.
 
 In the next years, due to the increasing request for interaction with people in social environments, academic research and market analyses predict that robots will become more and more “social”. But, at the same time, robots may incur the risk of being perceived as “asocial” as their pervasiveness in everyday environments may become an obstacle to social interaction between humans. Or they may even be rejected as “antisocial” if researchers do not take care to develop robots that behave according to the customs and habits of the communities where they are expected to operate.
 
 To prevent this from happening, robotic researchers can no longer take the luxury of being mere technicians. They must be capable of addressing all the challenges raised by future robot-populated communities in a multidisciplinary way. They need to share their knowledge with the public to make people aware of what robots really are, what expectations we can have from them, the risks we may incur, and the possible mitigation plans.
 
 While being open to any kind of contribution related to Human-Robot Interactive Communication, the RO-MAN 2022 conference will emphasize these aspects of robotic research. How can we create robots capable of being welcomed as our assistants and companions in society while, at the same time, not being perceived as “asocial” or “antisocial?” How can we create robots perceived as facilitators of the natural development of human beings and communities, rather than obstacles?
 
 Contributions from different disciplines, not exclusively related to technological development, will be welcomed as crucial to pave the way to global acceptance of future robots with an informed attitude.
 
 [![](https://s3.eu-west-1.amazonaws.com/underline.prod/uploads/markdown_image/1/image/af259abe105b3791c498ae1552d75710.png)](https://drive.google.com/file/d/18w73BZPB_JVOgcrcdZDqYpEupwVt4CnU/view?usp=sharing)

A registration is required to access this event page. If you have not already done so, please use the below link to register for the event. 

[Register Here](http://www.smile.unina.it/ro-man2022/welcome-to-ieee-ro-man-2022/)

Registration Is Required

IEEE RO-MAN 2022

31st IEEE International Conference on Robot & Human Interactive Communication
This conference is a leading forum where state-of-the-art innovative results, the latest developments as well as future perspectives relating to robot and human interactive communication are presented and discussed.

poster

Evidence shows comparing events helps children learn verbs (e.g., Childers et al., 2016), but studies are needed to understand whether event type is important. Study 1 examines two types of experiences during learning: seeing similar events than varied, or all varied events. Two½-(n=22), 3½-(n=20) and 4½-year-olds(n=14) learned 4 novel verbs. BS:similar events first or all varied events; pointed at test. A 3(Age: 2,3,4 years) x 2(Condition: similar,varied) univariate ANOVA,dv=proportion correct, showed main effect of Age, F(2, 55)= 7.6, p=.001 only. All children succeeded (one sample t-tests,ps< .03), whereas in a prior study, younger children failed. A second study confirms these results with a different set of video events, and events separated by 1-minute distractors. Here, only 2½- and 3½-year-olds who saw similar events extended verbs,perhaps because events were separated in time. These studies show that verb learners can benefit from comparing events, and comparing similar events can be especially useful.

Verb learning in young children: Are types of comparisons important?

As time series analysis continues to capture the interest of cognitive and behavioral researchers, it is increasingly important to evaluate these methods and compare their respective insights. Here, we evaluate three popular analyses: vector autoregression, cross-correlation, and cross-recurrence quantification analysis. Using social cohesion data derived from Twitter and daily counts of real-world events during the Arab Spring, we present a case study using these methods and evaluate their benefits, limitations, and differences in results. We propose that researchers interested in time series analysis consider these differences and use multiple methods to assure reliability of their results.

Vector Autoregression, Cross-Correlation, and Cross-Recurrence Quantification Analysis: A Case Study in Social Cohesion and Collective Action

The purpose of this study is to capture the structure of the interactive role-making process and introduce an integrated perspective to view actors’ creativity in the process. In particular, this study focuses on the characteristics of utterances in an acting training, which emphasizes paying attention to a partner and communicating. Statistical analyses are conducted on temporal changes, individual differences, and the granularity of utterances. The influence of attention toward partners on utterances is discussed, as well as the possible internal changes of actors, which reflects a perspective from which to view creative action as is referred to in the five A’s framework. In addition, by applying the theory of constructed emotion, the present study attempts to provide a possible demonstration of how truthful emotion is born under imaginary settings in acting.

Emotion Expression Captured by Utterances in Acting and Underpinning Internal Changes in Actors

Zipf’s Law of Abbreviation (ZLA) states that the more frequently a word is used, the shorter its length tends to be. This arises due to the optimal trade-off between competing pressures for accuracy and efficiency in communication, known as the Principle of Least Effort. Existing research has not focused on how individuals adapt their language use to remain optimal despite language change and whether social factors like common ground affect this. To investigate this, we replicated and extended the artificial language learning paradigm and communication game of Kanwal et al. (2017). We found participants were able to re-optimize their language use according to ZLA after a language change event, but this ability was hampered by common ground. This research identifies common ground as one potential cause for observed sub-optimalities in human languages and may have implications for understanding the dynamics of language change across communities where common ground varies.

Zipf's law of abbreviation and common ground: Past communicative success hampers the re-optimization of language

Previous studies have found that bilingual children receive different Give-Number task knower-levels across their languages as subset-knowers. However, recent work reveals that the reliability of Give-Number task is moderate for subset-knower levels. This raises the possibility that the differences found in knower-levels across bilinguals’ languages are explained by a lack of reliability of Give-Number. To address this possibility, we presented bilingual children with Give-Number and a subitizing task that they had to perform in both languages. If differences in knower-levels are due to true differences in children’s understanding of small numbers and not random noise, then differences in subitizing abilities should also be non-random. Data collection is still in progress, but preliminary planned analyses (N=13/64) revealed no differences in subitizing abilities across languages, based on knower-levels classification. We discuss implications of these findings on knowledge transfer across bilinguals’ languages.

Subitizing Abilities of Bilingual Subset-Knowers

Children typically do not spontaneously recognize accented productions of known words until approximately 19 months. In 15-month-olds, however, this ability is correlated with vocabulary size. Vocabulary size may support cross-accent accommodation by decreasing the likelihood that a variant production is considered to be an unknown word. We simulated a cross-accent word identification experiment, with word tokens generated from a two-dimensional Gaussian space, and accented productions simulated via linear transforms. Simulated participants were Bayesian classifiers with large or small vocabularies. Our large vocabulary group accurately classified more accented tokens and were less likely to classify an accented token as an unknown word. Thus, one way a growing vocabulary size may foster cross-accent accommodation is through increasing infants’ propensity to fit accented variants to known words, rather than treating them as unknown words.

Modeling a direct role of vocabulary size in driving cross-accent word identification

Language acquisition is supported by an ability called phonological awareness, which allows children to become intentionally aware of phonological units. It is known that erroneous pronunciation appears during the formation of phonological awareness. To clarify children's internal processes during this formation, this research aims to examine the factors that correct an erroneous phonological generation process. To do this, we utilized the innate and experiential factors of the memory retrieval mechanism in the cognitive architecture ACT-R. Specifically, we performed simulations to examine the interaction tasks that contribute to the acquisition of phonological awareness. As a result, it was shown that repeating a single task causes incorrect convergence and that this convergence can be prevented by performing other kinds of learning between tasks. In the future, it will be necessary to examine learning between tasks that can be associated with real situations and to confirm the overall process of phonological awareness formation.

Simulating the factors that correct the erroneous process of phonological generation in Japanese

The speed-accuracy tradeoff (SAT) method has produced several prominent findings in sentence processing. While a substantial number of SAT studies has yielded statistical null results regarding the degree to which certain factors influence the speed of sentence processing operations, the statistical power of the SAT paradigm is not known. As a result, it is not entirely clear how to interpret these findings. We addressed this problem by means of a simulation study in which we simulated SAT experiments for a range of known effect sizes in order to determine the statistical power in typical SAT experiments. We found that while SAT experiments appear to have quite satisfactory power to detect differences in asymptotic accuracy, that is not the case for speed-related parameters. We conclude that the failure to find an effect in speed-related parameters in SAT experiments may be less meaningful than previously thought.

Statistical Power in Response Signal Paradigm Experiments

How many rhymes are possible in English? How much alliteration or assonance? This paper explores the space of common phonologically driven poetic devices. I investigate instances of these poetic devices across unique words in the dictionary, assuming perfect matching in relevant corresponding elements of word pairs. Documenting the frequencies of poetic device matches across words provides a baseline for understanding the diversity and use of these forms in the wild. It also allows us to describe the distribution of words across these patterns within each poetic device. We show that certain devices, such as alliteration and stress, support a relatively small number of unique phonological patterns, providing more consolidated and predictable resources than other devices. Forms like masculine and feminine rhyme, assonance, and consonance display relatively more sound patterns respectively, and unique words are distributed across them differently. These results are discussed in terms of their poetic and cognitive implications.

Of Pieces and Patterns: Modeling Poetic Devices

Bilinguals are reported to have language learning advantages. One possible pathway is a language-specific transfer effect, whereby sensitivity to structural regularities in known languages can be brought to novel languages that share features. To test for specific linguistic feature transfer, we designed a task for bilinguals with homogeneous language exposure (bilingual in same languages) and heterogeneous feature representation (differing proficiencies). As Hindi and Mandarin have retroflexion in phoneme contrasts, we conducted a pre-registered study with a statistical learning task of a Hindi dental-retroflex contrast on parallel English-Mandarin bilinguals with varied Mandarin proficiency. Unlike the pilot study (N = 15), the main study (N = 50) found no evidence for a learning effect, and language-experience did not explain learning variance. As these stimuli have shown learning effects in children, learning effects for this feature may be fragile in this adult population, and language-specific neural commitments may prevent learning of the contrast.

What Transformers Might Know About the Physical World: T5 and the Origins of Knowledge

Downloads

Next from CogSci 2021

Verb learning in young children: Are types of comparisons important?

Similar lecture

The Impact of an In-Home Co-Located Robotic Coach in Helping People Make Fewer Exercise Mistakes

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES