@inproceedings{rohan-etal-2025-generative,
title = "Generative Data Augmentation for Improving Semantic Classification",
author = "Rohan, Shadman and
Akhter, Mahmud Elahi and
Moosa, Ibraheem Muhammad and
Mohammed, Nabeel and
Ali, Amin Ahsan and
Rahman, Akmmahbubur",
editor = "Alam, Firoj and
Kar, Sudipta and
Chowdhury, Shammur Absar and
Hassan, Naeemul and
Prince, Enamul Hoque and
Tasnim, Mohiuddin and
Rony, Md Rashad Al Hasan and
Rahman, Md Tahmid Rahman",
booktitle = "Proceedings of the Second Workshop on Bangla Language Processing (BLP-2025)",
month = dec,
year = "2025",
address = "Mumbai, India",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.banglalp-1.28/",
pages = "347--356",
ISBN = "979-8-89176-314-2",
abstract = "We study sentence-level generative data augmentation for Bangla semantic classification across four public datasets and three pretrained model families (BanglaBERT, XLM-Indic, mBERT). We evaluate two widely used, reproducible techniques{---}paraphrasing (mT5-based) and round-trip backtranslation (Bn{--}En{--}Bn){---}and analyze their impact under realistic class imbalance. Overall, augmentation often helps, but gains are tightly coupled to label quality: paraphrasing typically outperforms backtranslation and yields the most consistent improvements for the monolingual model, whereas multilingual encoders benefit less and can be more sensitive to noisy minority-class expansions. A key empirical observation is that the neutral class appears to be a major source of annotation noise, which degrades decision boundaries and can cap the benefits of augmentation even when positive/negative classes are clean and polarized. We provide practical guidance for Bangla sentiment pipelines: (i) use simple sentence-level augmentation to rebalance classes when labels are reliable; (ii) allocate additional curation and higher inter-annotator agreement targets to the neutral class. Our results indicate when augmentation helps and suggest that data quality{---}not model choice alone{---}can become the limiting factor."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="rohan-etal-2025-generative">
<titleInfo>
<title>Generative Data Augmentation for Improving Semantic Classification</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shadman</namePart>
<namePart type="family">Rohan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mahmud</namePart>
<namePart type="given">Elahi</namePart>
<namePart type="family">Akhter</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ibraheem</namePart>
<namePart type="given">Muhammad</namePart>
<namePart type="family">Moosa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nabeel</namePart>
<namePart type="family">Mohammed</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amin</namePart>
<namePart type="given">Ahsan</namePart>
<namePart type="family">Ali</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Akmmahbubur</namePart>
<namePart type="family">Rahman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Second Workshop on Bangla Language Processing (BLP-2025)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Firoj</namePart>
<namePart type="family">Alam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sudipta</namePart>
<namePart type="family">Kar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shammur</namePart>
<namePart type="given">Absar</namePart>
<namePart type="family">Chowdhury</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naeemul</namePart>
<namePart type="family">Hassan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Enamul</namePart>
<namePart type="given">Hoque</namePart>
<namePart type="family">Prince</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohiuddin</namePart>
<namePart type="family">Tasnim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="given">Rashad</namePart>
<namePart type="given">Al</namePart>
<namePart type="given">Hasan</namePart>
<namePart type="family">Rony</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md</namePart>
<namePart type="given">Tahmid</namePart>
<namePart type="given">Rahman</namePart>
<namePart type="family">Rahman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Mumbai, India</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-314-2</identifier>
</relatedItem>
<abstract>We study sentence-level generative data augmentation for Bangla semantic classification across four public datasets and three pretrained model families (BanglaBERT, XLM-Indic, mBERT). We evaluate two widely used, reproducible techniques—paraphrasing (mT5-based) and round-trip backtranslation (Bn–En–Bn)—and analyze their impact under realistic class imbalance. Overall, augmentation often helps, but gains are tightly coupled to label quality: paraphrasing typically outperforms backtranslation and yields the most consistent improvements for the monolingual model, whereas multilingual encoders benefit less and can be more sensitive to noisy minority-class expansions. A key empirical observation is that the neutral class appears to be a major source of annotation noise, which degrades decision boundaries and can cap the benefits of augmentation even when positive/negative classes are clean and polarized. We provide practical guidance for Bangla sentiment pipelines: (i) use simple sentence-level augmentation to rebalance classes when labels are reliable; (ii) allocate additional curation and higher inter-annotator agreement targets to the neutral class. Our results indicate when augmentation helps and suggest that data quality—not model choice alone—can become the limiting factor.</abstract>
<identifier type="citekey">rohan-etal-2025-generative</identifier>
<location>
<url>https://aclanthology.org/2025.banglalp-1.28/</url>
</location>
<part>
<date>2025-12</date>
<extent unit="page">
<start>347</start>
<end>356</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Generative Data Augmentation for Improving Semantic Classification
%A Rohan, Shadman
%A Akhter, Mahmud Elahi
%A Moosa, Ibraheem Muhammad
%A Mohammed, Nabeel
%A Ali, Amin Ahsan
%A Rahman, Akmmahbubur
%Y Alam, Firoj
%Y Kar, Sudipta
%Y Chowdhury, Shammur Absar
%Y Hassan, Naeemul
%Y Prince, Enamul Hoque
%Y Tasnim, Mohiuddin
%Y Rony, Md Rashad Al Hasan
%Y Rahman, Md Tahmid Rahman
%S Proceedings of the Second Workshop on Bangla Language Processing (BLP-2025)
%D 2025
%8 December
%I Association for Computational Linguistics
%C Mumbai, India
%@ 979-8-89176-314-2
%F rohan-etal-2025-generative
%X We study sentence-level generative data augmentation for Bangla semantic classification across four public datasets and three pretrained model families (BanglaBERT, XLM-Indic, mBERT). We evaluate two widely used, reproducible techniques—paraphrasing (mT5-based) and round-trip backtranslation (Bn–En–Bn)—and analyze their impact under realistic class imbalance. Overall, augmentation often helps, but gains are tightly coupled to label quality: paraphrasing typically outperforms backtranslation and yields the most consistent improvements for the monolingual model, whereas multilingual encoders benefit less and can be more sensitive to noisy minority-class expansions. A key empirical observation is that the neutral class appears to be a major source of annotation noise, which degrades decision boundaries and can cap the benefits of augmentation even when positive/negative classes are clean and polarized. We provide practical guidance for Bangla sentiment pipelines: (i) use simple sentence-level augmentation to rebalance classes when labels are reliable; (ii) allocate additional curation and higher inter-annotator agreement targets to the neutral class. Our results indicate when augmentation helps and suggest that data quality—not model choice alone—can become the limiting factor.
%U https://aclanthology.org/2025.banglalp-1.28/
%P 347-356
Markdown (Informal)
[Generative Data Augmentation for Improving Semantic Classification](https://aclanthology.org/2025.banglalp-1.28/) (Rohan et al., BanglaLP 2025)
ACL
- Shadman Rohan, Mahmud Elahi Akhter, Ibraheem Muhammad Moosa, Nabeel Mohammed, Amin Ahsan Ali, and Akmmahbubur Rahman. 2025. Generative Data Augmentation for Improving Semantic Classification. In Proceedings of the Second Workshop on Bangla Language Processing (BLP-2025), pages 347–356, Mumbai, India. Association for Computational Linguistics.