@inproceedings{purohit-etal-2026-thinkvision,
title = "{T}hink{V}ision at {S}em{E}val-2026 Task 6: A Transformer-Based Ensemble System for Clarity Detection",
author = "Purohit, Gourav Ghanshyam and
Swami, Praveen and
Sahoo, Shriyans S and
Bhati, Jenish and
Nadiger, Supriya and
Saumya, Sunil",
editor = "Kochmar, Ekaterina and
Ghosh, Debanjan and
North, Kai and
Komachi, Mamoru and
Zampieri, Marcos",
booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
month = jul,
year = "2026",
address = "San Diego, California, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.semeval-1.248/",
doi = "10.18653/v1/2026.semeval-1.248",
pages = "1973--1978",
ISBN = "979-8-89176-414-9",
abstract = "We study the problem of assessing the clarity of political question{--}answer pairs, where the goal is to determine whether a response directly addresses the question, avoids it, or remains ambiguous. This task is particularly challenging in political discourse, where evasiveness can be subtle and context-dependent.To tackle this problem, we propose an ensemble-based approach built on the transformer-based model DeBERTa-v3-base, fine-tuned on concatenated question{--}answer inputs. Special attention is given to class imbalance during training to ensure robust performance across all categories.To better capture uncertainty in borderline cases, we train multiple models with different random seeds and employ Monte Carlo Dropout at inference time. Final predictions are obtained by averaging logits across ensemble models and stochastic forward passes, yielding more stable and robust predictions.Our system achieves a Macro-F1 score of 0.76 on the evaluation dataset. Error analysis reveals that responses that partially engage with the question while failing to provide a direct answer remain the most challenging, highlighting the inherent difficulty of detecting nuanced evasiveness in political communication."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="purohit-etal-2026-thinkvision">
<titleInfo>
<title>ThinkVision at SemEval-2026 Task 6: A Transformer-Based Ensemble System for Clarity Detection</title>
</titleInfo>
<name type="personal">
<namePart type="given">Gourav</namePart>
<namePart type="given">Ghanshyam</namePart>
<namePart type="family">Purohit</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Praveen</namePart>
<namePart type="family">Swami</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shriyans</namePart>
<namePart type="given">S</namePart>
<namePart type="family">Sahoo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jenish</namePart>
<namePart type="family">Bhati</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Supriya</namePart>
<namePart type="family">Nadiger</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunil</namePart>
<namePart type="family">Saumya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 20th International Workshop on Semantic Evaluation (2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Kochmar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Debanjan</namePart>
<namePart type="family">Ghosh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kai</namePart>
<namePart type="family">North</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mamoru</namePart>
<namePart type="family">Komachi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marcos</namePart>
<namePart type="family">Zampieri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-414-9</identifier>
</relatedItem>
<abstract>We study the problem of assessing the clarity of political question–answer pairs, where the goal is to determine whether a response directly addresses the question, avoids it, or remains ambiguous. This task is particularly challenging in political discourse, where evasiveness can be subtle and context-dependent.To tackle this problem, we propose an ensemble-based approach built on the transformer-based model DeBERTa-v3-base, fine-tuned on concatenated question–answer inputs. Special attention is given to class imbalance during training to ensure robust performance across all categories.To better capture uncertainty in borderline cases, we train multiple models with different random seeds and employ Monte Carlo Dropout at inference time. Final predictions are obtained by averaging logits across ensemble models and stochastic forward passes, yielding more stable and robust predictions.Our system achieves a Macro-F1 score of 0.76 on the evaluation dataset. Error analysis reveals that responses that partially engage with the question while failing to provide a direct answer remain the most challenging, highlighting the inherent difficulty of detecting nuanced evasiveness in political communication.</abstract>
<identifier type="citekey">purohit-etal-2026-thinkvision</identifier>
<identifier type="doi">10.18653/v1/2026.semeval-1.248</identifier>
<location>
<url>https://aclanthology.org/2026.semeval-1.248/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>1973</start>
<end>1978</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ThinkVision at SemEval-2026 Task 6: A Transformer-Based Ensemble System for Clarity Detection
%A Purohit, Gourav Ghanshyam
%A Swami, Praveen
%A Sahoo, Shriyans S.
%A Bhati, Jenish
%A Nadiger, Supriya
%A Saumya, Sunil
%Y Kochmar, Ekaterina
%Y Ghosh, Debanjan
%Y North, Kai
%Y Komachi, Mamoru
%Y Zampieri, Marcos
%S Proceedings of the 20th International Workshop on Semantic Evaluation (2026)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, USA
%@ 979-8-89176-414-9
%F purohit-etal-2026-thinkvision
%X We study the problem of assessing the clarity of political question–answer pairs, where the goal is to determine whether a response directly addresses the question, avoids it, or remains ambiguous. This task is particularly challenging in political discourse, where evasiveness can be subtle and context-dependent.To tackle this problem, we propose an ensemble-based approach built on the transformer-based model DeBERTa-v3-base, fine-tuned on concatenated question–answer inputs. Special attention is given to class imbalance during training to ensure robust performance across all categories.To better capture uncertainty in borderline cases, we train multiple models with different random seeds and employ Monte Carlo Dropout at inference time. Final predictions are obtained by averaging logits across ensemble models and stochastic forward passes, yielding more stable and robust predictions.Our system achieves a Macro-F1 score of 0.76 on the evaluation dataset. Error analysis reveals that responses that partially engage with the question while failing to provide a direct answer remain the most challenging, highlighting the inherent difficulty of detecting nuanced evasiveness in political communication.
%R 10.18653/v1/2026.semeval-1.248
%U https://aclanthology.org/2026.semeval-1.248/
%U https://doi.org/10.18653/v1/2026.semeval-1.248
%P 1973-1978
Markdown (Informal)
[ThinkVision at SemEval-2026 Task 6: A Transformer-Based Ensemble System for Clarity Detection](https://aclanthology.org/2026.semeval-1.248/) (Purohit et al., SemEval 2026)
ACL