No Cover Image

Journal article 15 views 2 downloads

BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning

Arash Ashrafzadeh, Julian Hough Orcid Logo, Arash Eshghi Orcid Logo

Languages, Volume: 11, Issue: 5, Start page: 99

Swansea University Author: Julian Hough Orcid Logo

  • languages-11-00099.pdf

    PDF | Version of Record

    © 2026 by the authors. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.

    Download (5.72MB)

Check full text

DOI (Published version): 10.3390/languages11050099

Abstract

Recent research in grounded language learning has seen remarkable success due to advances in large vision and language models (VLMs). However, these models (i) are extremely costly to train and update; (ii) struggle with generalisation; and (iii) do not support continual learning. In this paper, we...

Full description

Published in: Languages
ISSN: 2226-471X
Published: MDPI 2026
Online Access: Check full text

URI: https://cronfa.swan.ac.uk/Record/cronfa72028
first_indexed 2026-06-09T09:08:06Z
last_indexed 2026-06-09T09:08:06Z
id cronfa72028
recordtype SURis
fullrecord <?xml version="1.0" encoding="utf-8"?><rfc1807 xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema"><bib-version>v2</bib-version><id>72028</id><entry>2026-06-09</entry><title>BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning</title><swanseaauthors><author><sid>082d773ae261d2bbf49434dd2608ab40</sid><ORCID>0000-0002-4345-6759</ORCID><firstname>Julian</firstname><surname>Hough</surname><name>Julian Hough</name><active>true</active><ethesisStudent>false</ethesisStudent></author></swanseaauthors><date>2026-06-09</date><deptcode>MACS</deptcode><abstract>Recent research in grounded language learning has seen remarkable success due to advances in large vision and language models (VLMs). However, these models (i) are extremely costly to train and update; (ii) struggle with generalisation; and (iii) do not support continual learning. In this paper, we introduce baby-ds integrating the Dynamic Syntax (DS) framework with automated planning within the multimodal BabyAI platform as a testbed. We provide methods whereby DS lexicons are induced continually from teacher demonstrations within BabyAI. We study (i–iii) by experimenting with the compositional complexity of natural language instructions in the data to compare data efficiency, generalisation, and continual learning properties of baby-ds with a simple neural model. The results show that the baby-ds model: (i) needs much less data than the neural model to reach threshold performance; (ii) generalises much faster to more complex instructions; and (iii) is a more effective continual learner. We argue that it is the attendant linguistic bias within DS and the rich inferential power of TTR that enables (i–iii), highlighting the importance of further research on hybrid grammar–neural approaches. Finally, we discuss several important limitations of baby-ds and sketch a path forward for further DS research.</abstract><type>Journal Article</type><journal>Languages</journal><volume>11</volume><journalNumber>5</journalNumber><paginationStart>99</paginationStart><paginationEnd/><publisher>MDPI</publisher><placeOfPublication/><isbnPrint/><isbnElectronic/><issnPrint/><issnElectronic>2226-471X</issnElectronic><keywords>grammar induction; neural semantic parsing; computational semantics; grounded language learning</keywords><publishedDay>12</publishedDay><publishedMonth>5</publishedMonth><publishedYear>2026</publishedYear><publishedDate>2026-05-12</publishedDate><doi>10.3390/languages11050099</doi><url/><notes/><college>COLLEGE NANME</college><department>Mathematics and Computer Science School</department><CollegeCode>COLLEGE CODE</CollegeCode><DepartmentCode>MACS</DepartmentCode><institution>Swansea University</institution><apcterm>Another institution paid the OA fee</apcterm><funders>Hough’s work is supported by the EPSRC grant EP/X009343/1 ‘FLUIDITY’.</funders><projectreference/><lastEdited>2026-06-09T10:09:02.4739884</lastEdited><Created>2026-06-09T10:03:52.0569202</Created><path><level id="1">Faculty of Science and Engineering</level><level id="2">School of Mathematics and Computer Science - Computer Science</level></path><authors><author><firstname>Arash</firstname><surname>Ashrafzadeh</surname><order>1</order></author><author><firstname>Julian</firstname><surname>Hough</surname><orcid>0000-0002-4345-6759</orcid><order>2</order></author><author><firstname>Arash</firstname><surname>Eshghi</surname><orcid>0000-0003-4711-4091</orcid><order>3</order></author></authors><documents><document><filename>72028__36891__ae4099d0226b471d95e7a204935e5963.pdf</filename><originalFilename>languages-11-00099.pdf</originalFilename><uploaded>2026-06-09T10:03:52.0377508</uploaded><type>Output</type><contentLength>6001338</contentLength><contentType>application/pdf</contentType><version>Version of Record</version><cronfaStatus>true</cronfaStatus><documentNotes>© 2026 by the authors. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.</documentNotes><copyrightCorrect>true</copyrightCorrect><language>eng</language><licence>https://creativecommons.org/licenses/by/4.0/</licence></document></documents><OutputDurs/></rfc1807>
spelling v2 72028 2026-06-09 BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning 082d773ae261d2bbf49434dd2608ab40 0000-0002-4345-6759 Julian Hough Julian Hough true false 2026-06-09 MACS Recent research in grounded language learning has seen remarkable success due to advances in large vision and language models (VLMs). However, these models (i) are extremely costly to train and update; (ii) struggle with generalisation; and (iii) do not support continual learning. In this paper, we introduce baby-ds integrating the Dynamic Syntax (DS) framework with automated planning within the multimodal BabyAI platform as a testbed. We provide methods whereby DS lexicons are induced continually from teacher demonstrations within BabyAI. We study (i–iii) by experimenting with the compositional complexity of natural language instructions in the data to compare data efficiency, generalisation, and continual learning properties of baby-ds with a simple neural model. The results show that the baby-ds model: (i) needs much less data than the neural model to reach threshold performance; (ii) generalises much faster to more complex instructions; and (iii) is a more effective continual learner. We argue that it is the attendant linguistic bias within DS and the rich inferential power of TTR that enables (i–iii), highlighting the importance of further research on hybrid grammar–neural approaches. Finally, we discuss several important limitations of baby-ds and sketch a path forward for further DS research. Journal Article Languages 11 5 99 MDPI 2226-471X grammar induction; neural semantic parsing; computational semantics; grounded language learning 12 5 2026 2026-05-12 10.3390/languages11050099 COLLEGE NANME Mathematics and Computer Science School COLLEGE CODE MACS Swansea University Another institution paid the OA fee Hough’s work is supported by the EPSRC grant EP/X009343/1 ‘FLUIDITY’. 2026-06-09T10:09:02.4739884 2026-06-09T10:03:52.0569202 Faculty of Science and Engineering School of Mathematics and Computer Science - Computer Science Arash Ashrafzadeh 1 Julian Hough 0000-0002-4345-6759 2 Arash Eshghi 0000-0003-4711-4091 3 72028__36891__ae4099d0226b471d95e7a204935e5963.pdf languages-11-00099.pdf 2026-06-09T10:03:52.0377508 Output 6001338 application/pdf Version of Record true © 2026 by the authors. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. true eng https://creativecommons.org/licenses/by/4.0/
title BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning
spellingShingle BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning
Julian Hough
title_short BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning
title_full BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning
title_fullStr BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning
title_full_unstemmed BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning
title_sort BabyDS: Visually Grounded Grammar Induction with Online Curriculum Learning
author_id_str_mv 082d773ae261d2bbf49434dd2608ab40
author_id_fullname_str_mv 082d773ae261d2bbf49434dd2608ab40_***_Julian Hough
author Julian Hough
author2 Arash Ashrafzadeh
Julian Hough
Arash Eshghi
format Journal article
container_title Languages
container_volume 11
container_issue 5
container_start_page 99
publishDate 2026
institution Swansea University
issn 2226-471X
doi_str_mv 10.3390/languages11050099
publisher MDPI
college_str Faculty of Science and Engineering
hierarchytype
hierarchy_top_id facultyofscienceandengineering
hierarchy_top_title Faculty of Science and Engineering
hierarchy_parent_id facultyofscienceandengineering
hierarchy_parent_title Faculty of Science and Engineering
department_str School of Mathematics and Computer Science - Computer Science{{{_:::_}}}Faculty of Science and Engineering{{{_:::_}}}School of Mathematics and Computer Science - Computer Science
document_store_str 1
active_str 0
description Recent research in grounded language learning has seen remarkable success due to advances in large vision and language models (VLMs). However, these models (i) are extremely costly to train and update; (ii) struggle with generalisation; and (iii) do not support continual learning. In this paper, we introduce baby-ds integrating the Dynamic Syntax (DS) framework with automated planning within the multimodal BabyAI platform as a testbed. We provide methods whereby DS lexicons are induced continually from teacher demonstrations within BabyAI. We study (i–iii) by experimenting with the compositional complexity of natural language instructions in the data to compare data efficiency, generalisation, and continual learning properties of baby-ds with a simple neural model. The results show that the baby-ds model: (i) needs much less data than the neural model to reach threshold performance; (ii) generalises much faster to more complex instructions; and (iii) is a more effective continual learner. We argue that it is the attendant linguistic bias within DS and the rich inferential power of TTR that enables (i–iii), highlighting the importance of further research on hybrid grammar–neural approaches. Finally, we discuss several important limitations of baby-ds and sketch a path forward for further DS research.
published_date 2026-05-12T10:09:04Z
_version_ 1867509813209989120
score 11.108223

Similar Items