@inproceedings{fondekar-etal-2026-unigoa,
title = "Unigoa@{CH}i{PSAL} 2026: Early vs Late Fusion for Multimodal Hate and Sentiment Detection in {N}epali Memes",
author = "Fondekar, Ashweta and
Shivolkar, Milind and
Pawar, Jyoti",
editor = "Sarveswaran, Kengatharaiyer and
Vaidya, Ashwini",
booktitle = "Proceedings of the Second workshop on Challenges in Processing {S}outh {A}sian Languages ({CH}i{PSAL}2026)",
month = may,
year = "2026",
address = "Palma de Mallorca, Spain",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.chipsal-1.22/",
doi = "10.63317/37ooapnyzhbx",
pages = "229--236",
abstract = "Internet memes pose significant challenges for automatic content moderation due to the interaction of visual and textual cues, sarcasm, and cultural context. In this work, we participate in the CHiPSAL 2026 shared task on multimodal hate and sentiment understanding in Nepali memes. The task consists of two subtasks: binary hate speech detection and three-class sentiment classification. We investigate both early-fusion and late-fusion multimodal architectures. Our primary system employs a late-fusion dual-encoder architecture combining XLM-RoBERTa for multilingual text representation and CLIP for visual encoding. We further evaluate an early-fusion ViLT-based joint vision{--}language transformer using NepBERTa tokenization as a baseline. Experimental results show that late-fusion models consistently outperform early-fusion architectures, particularly for code-mixed memes containing Devanagari Nepali and Roman-script English text. Our best system achieves a Macro-F1 of 0.6564 for hate speech detection and 0.4859 for sentiment classification. We provide analysis highlighting the challenges of multilingual code-mixing, sarcasm, and implicit sentiment in low-resource multimodal settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="fondekar-etal-2026-unigoa">
<titleInfo>
<title>Unigoa@CHiPSAL 2026: Early vs Late Fusion for Multimodal Hate and Sentiment Detection in Nepali Memes</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ashweta</namePart>
<namePart type="family">Fondekar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Milind</namePart>
<namePart type="family">Shivolkar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jyoti</namePart>
<namePart type="family">Pawar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Second workshop on Challenges in Processing South Asian Languages (CHiPSAL2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kengatharaiyer</namePart>
<namePart type="family">Sarveswaran</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ashwini</namePart>
<namePart type="family">Vaidya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma de Mallorca, Spain</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Internet memes pose significant challenges for automatic content moderation due to the interaction of visual and textual cues, sarcasm, and cultural context. In this work, we participate in the CHiPSAL 2026 shared task on multimodal hate and sentiment understanding in Nepali memes. The task consists of two subtasks: binary hate speech detection and three-class sentiment classification. We investigate both early-fusion and late-fusion multimodal architectures. Our primary system employs a late-fusion dual-encoder architecture combining XLM-RoBERTa for multilingual text representation and CLIP for visual encoding. We further evaluate an early-fusion ViLT-based joint vision–language transformer using NepBERTa tokenization as a baseline. Experimental results show that late-fusion models consistently outperform early-fusion architectures, particularly for code-mixed memes containing Devanagari Nepali and Roman-script English text. Our best system achieves a Macro-F1 of 0.6564 for hate speech detection and 0.4859 for sentiment classification. We provide analysis highlighting the challenges of multilingual code-mixing, sarcasm, and implicit sentiment in low-resource multimodal settings.</abstract>
<identifier type="citekey">fondekar-etal-2026-unigoa</identifier>
<identifier type="doi">10.63317/37ooapnyzhbx</identifier>
<location>
<url>https://aclanthology.org/2026.chipsal-1.22/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>229</start>
<end>236</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Unigoa@CHiPSAL 2026: Early vs Late Fusion for Multimodal Hate and Sentiment Detection in Nepali Memes
%A Fondekar, Ashweta
%A Shivolkar, Milind
%A Pawar, Jyoti
%Y Sarveswaran, Kengatharaiyer
%Y Vaidya, Ashwini
%S Proceedings of the Second workshop on Challenges in Processing South Asian Languages (CHiPSAL2026)
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma de Mallorca, Spain
%F fondekar-etal-2026-unigoa
%X Internet memes pose significant challenges for automatic content moderation due to the interaction of visual and textual cues, sarcasm, and cultural context. In this work, we participate in the CHiPSAL 2026 shared task on multimodal hate and sentiment understanding in Nepali memes. The task consists of two subtasks: binary hate speech detection and three-class sentiment classification. We investigate both early-fusion and late-fusion multimodal architectures. Our primary system employs a late-fusion dual-encoder architecture combining XLM-RoBERTa for multilingual text representation and CLIP for visual encoding. We further evaluate an early-fusion ViLT-based joint vision–language transformer using NepBERTa tokenization as a baseline. Experimental results show that late-fusion models consistently outperform early-fusion architectures, particularly for code-mixed memes containing Devanagari Nepali and Roman-script English text. Our best system achieves a Macro-F1 of 0.6564 for hate speech detection and 0.4859 for sentiment classification. We provide analysis highlighting the challenges of multilingual code-mixing, sarcasm, and implicit sentiment in low-resource multimodal settings.
%R 10.63317/37ooapnyzhbx
%U https://aclanthology.org/2026.chipsal-1.22/
%U https://doi.org/10.63317/37ooapnyzhbx
%P 229-236
Markdown (Informal)
[Unigoa@CHiPSAL 2026: Early vs Late Fusion for Multimodal Hate and Sentiment Detection in Nepali Memes](https://aclanthology.org/2026.chipsal-1.22/) (Fondekar et al., CHiPSAL 2026)
ACL