A framework for the emergence and analysis of language in social learning agents

Name: A framework for the emergence and analysis of language in social learning agents
Published: 2025-08-20
License: http://creativecommons.org/licenses/by/4.0

Open this dataset in the live repository

Important usage condition: Use and redistribution are governed by CC BY 4.0. Review and comply with the license before using the data.

Persistent identifier: doi:10.60507/FK2/Z7NCOP
Published version: 1.0
Publication date: 2025-08-20
License: CC BY 4.0

Description

Neural systems have evolved not only to solve environmental challenges through internal representations but also, under social constraints, to communicate these to conspecifics. In this work, we aim to understand the structure of these internal representations and how they may be optimized to transmit pertinent information from one individual to another. Thus, we build on previous teacher-student communication protocols to analyze the formation of individual and shared abstractions and their impact on task performance. We use reinforcement learning in grid-world mazes where a teacher network passes a message to a student to improve task performance. This framework allows us to relate environmental variables with individual and shared representations. We compress high-dimensional task information within a low-dimensional representational space to mimic natural language features. In coherence with previous results, we find that providing teacher information to the student leads to a higher task completion rate and an ability to generalize tasks it has not seen before. Further, optimizing message content to maximize student reward improves information encoding, suggesting that an accurate representation in the space of messages requires bi-directional input. These results highlight the role of language as a common representation among agents and its implications on generalization capabilities.

Creators

Wieczorek, Tobias J.
Tchumatchenko, Tatjana
Wert-Carvajal, Carlos
Eggl, Maximilian

Keywords

Computational Neuroscience, Social Learning, Agent-Based Modeling, Artificial Language

Files

Before downloading: Use and redistribution are governed by CC BY 4.0. Review and comply with the license before using the data.

File	Type	Bytes	Checksum
requirements.txt	text/plain	164	MD5 ac431b9da0854fe140232e67d96a18f7
README.md	text/markdown	7452	MD5 3041a7952943a03b5d8d4982dba00100
language_emergence_review.ipynb	application/x-ipynb+json	235616	MD5 7d6537faa0e1277b55e85a2913652c5b
data.zip	application/zip	65404724	MD5 9aaa729bb00833fed4a7334ba385cc0e
TobiasJWieczorek_DATASET_Readme.txt	text/plain	6383	MD5 f591f2c4ef71d18337874c706f4400f4

Citation

Wieczorek, Tobias J.; Tchumatchenko, Tatjana; Wert-Carvajal, Carlos; Eggl, Maximilian, 2025-08-20, A framework for the emergence and analysis of language in social learning agents, doi:10.60507/FK2/Z7NCOP, V1.0

Additional Dataverse fields

Id	463
Dataset Type	dataset
Internal Version Number	21
Latest Version Publishing State	RELEASED
Release Time	2025-08-20T13:09:41Z
Create Time	2025-07-30T13:01:44Z
Citation Date	2025-08-20
File Access Request	True

Export metadata

Static metadata exports available for this published dataset version:

Complete Dataverse metadata

Citation Metadata

Title title

A framework for the emergence and analysis of language in social learning agents

Alternative URL alternativeURL

https://github.com/meggl23/multi_agent_language

Author author

Author Name
Wieczorek, Tobias J.
Author Affiliation
University of Bonn
Author Name
Tchumatchenko, Tatjana
Author Affiliation
University of Bonn
Author Identifier Scheme
ORCID
Author Identifier
Type Name
authorIdentifier
Multiple
False
Type Class
primitive
Value
https://orcid.org/0000-0001-9137-809X
Expandedvalue
Person Name
Tchumatchenko, Tatjana
@id
https://orcid.org/0000-0001-9137-809X
Scheme
ORCID
@type
https://schema.org/Person
Author Name
Wert-Carvajal, Carlos
Author Affiliation
University of Bonn
Author Name
Eggl, Maximilian
Author Affiliation
University of Bonn
Author Identifier Scheme
ORCID
Author Identifier
Type Name
authorIdentifier
Multiple
False
Type Class
primitive
Value
https://orcid.org/0000-0001-5815-1045
Expandedvalue
Person Name
Eggl, Maximilian
@id
https://orcid.org/0000-0001-5815-1045
Scheme
ORCID
@type
https://schema.org/Person

Dataset Contact datasetContact

Dataset Contact Name
Tchumatchenko, Tatjana
Dataset Contact Affiliation
University of Bonn
Dataset Contact Email
ttchumat@uni-bonn.de

Ds Description dsDescription

Ds Description Value
Neural systems have evolved not only to solve environmental challenges through internal representations but also, under social constraints, to communicate these to conspecifics. In this work, we aim to understand the structure of these internal representations and how they may be optimized to transmit pertinent information from one individual to another. Thus, we build on previous teacher-student communication protocols to analyze the formation of individual and shared abstractions and their impact on task performance. We use reinforcement learning in grid-world mazes where a teacher network passes a message to a student to improve task performance. This framework allows us to relate environmental variables with individual and shared representations. We compress high-dimensional task information within a low-dimensional representational space to mimic natural language features. In coherence with previous results, we find that providing teacher information to the student leads to a higher task completion rate and an ability to generalize tasks it has not seen before. Further, optimizing message content to maximize student reward improves information encoding, suggesting that an accurate representation in the space of messages requires bi-directional input. These results highlight the role of language as a common representation among agents and its implications on generalization capabilities.

Subject subject

Medicine, Health and Life Sciences

Free Keyword freeKeyword

Free Keyword Value
Computational Neuroscience
Free Keyword Value
Social Learning
Free Keyword Value
Agent-Based Modeling
Free Keyword Value
Artificial Language

Publication publication

Publication Relation Type
IsCitedBy
Publication Citation
Wieczorek, T.J., Tchumatchenko, T., Wert-Carvajal, C. et al. A framework for the emergence and analysis of language in social learning agents. Nat Commun 15, 7590 (2024). https://doi.org/10.1038/s41467-024-51887-5
Publication IDType
doi
Publication IDNumber
10.1038/s41467-024-51887-5
Publication URL
https://www.nature.com/articles/s41467-024-51887-5#Sec17

Contributor contributor

Contributor Type
Supervisor
Contributor Name
Eggl, Maximilian
Contributor Type
Supervisor
Contributor Name
Wert-Carvajal, Carlos
Contributor Type
Project Leader
Contributor Name
Tchumatchenko, Tatjana
Contributor Type
Project Leader
Contributor Name
Wieczorek, Tobias J.

Grant Number grantNumber

Grant Number Agency
Deutsche Forschungsgemeinschaft
Grant Number Value
DFG

Depositor depositor

Walch, Sabrina

Date Of Deposit dateOfDeposit

2025-07-30

Expected crawler behaviour

Use a stable, truthful User-Agent with product/version and a working contact URL. Across all IP addresses and HTTP connections used by one crawler identity, allow no more than 5 requests in flight and wait at least 20 seconds between request starts. Crawl URLs listed in the catalog sitemap, including file pages and download URLs when they are published, use conditional requests, honor Retry-After, and apply exponential backoff after errors.

The welcome page may link to the interactive repository for human navigation. Automated clients must not treat that human link as a catalog crawl target.

Read the live machine-readable crawler policy before and during a crawl. Stop crawling when it reports CPU or memory utilization at or above 80% and 80% respectively.