@inproceedings{gampper-rutherford-2026-chulanlp,
title = "{C}hula{NLP} at {S}em{E}val-2026 Task 4: Neural Aspect Composition for Narrative Story Embeddings",
author = "Gampper, James and
Rutherford, Attapol",
editor = "Kochmar, Ekaterina and
Ghosh, Debanjan and
North, Kai and
Komachi, Mamoru",
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-1.341/",
pages = "2709--2714",
ISBN = "979-8-89176-414-9",
abstract = "Comparing stories and narratives has proven to be a difficult task to automate because traditional vector representations fail to capture the layered and multi-faceted aspects of stories such as theme, plot progression, and resolution. We address SemEval-2026 Task 4, which requires generating vector embeddings that preserve narrative similarity relationships. We propose Neural Aspect Composition, which functions by using a Large Language Model (LLM) to decompose stories into 13 semantic narrative aspects (theme, course of action, outcomes, etc.), encodes each aspect separately using an encoder model, and learns a global importance weight for each aspect through a trained weighting layer. Our approach achieves the official test scores of 0.64 on Track A and 0.61 on Track B. During validation, it outperformed vectors produced by inputting the raw story text directly into an encoder model and a sentence-averaging baseline. The analysis of the learned weights on the development set reveals that thematic elements and narrative resolutions were the primary drivers of perceived similarity, receiving significantly higher weights than intermediate plot events and other minor details such as character introductions."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gampper-rutherford-2026-chulanlp">
<titleInfo>
<title>ChulaNLP at SemEval-2026 Task 4: Neural Aspect Composition for Narrative Story Embeddings</title>
</titleInfo>
<name type="personal">
<namePart type="given">James</namePart>
<namePart type="family">Gampper</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Attapol</namePart>
<namePart type="family">Rutherford</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>
<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>Comparing stories and narratives has proven to be a difficult task to automate because traditional vector representations fail to capture the layered and multi-faceted aspects of stories such as theme, plot progression, and resolution. We address SemEval-2026 Task 4, which requires generating vector embeddings that preserve narrative similarity relationships. We propose Neural Aspect Composition, which functions by using a Large Language Model (LLM) to decompose stories into 13 semantic narrative aspects (theme, course of action, outcomes, etc.), encodes each aspect separately using an encoder model, and learns a global importance weight for each aspect through a trained weighting layer. Our approach achieves the official test scores of 0.64 on Track A and 0.61 on Track B. During validation, it outperformed vectors produced by inputting the raw story text directly into an encoder model and a sentence-averaging baseline. The analysis of the learned weights on the development set reveals that thematic elements and narrative resolutions were the primary drivers of perceived similarity, receiving significantly higher weights than intermediate plot events and other minor details such as character introductions.</abstract>
<identifier type="citekey">gampper-rutherford-2026-chulanlp</identifier>
<location>
<url>https://aclanthology.org/2026.semeval-1.341/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>2709</start>
<end>2714</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ChulaNLP at SemEval-2026 Task 4: Neural Aspect Composition for Narrative Story Embeddings
%A Gampper, James
%A Rutherford, Attapol
%Y Kochmar, Ekaterina
%Y Ghosh, Debanjan
%Y North, Kai
%Y Komachi, Mamoru
%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 gampper-rutherford-2026-chulanlp
%X Comparing stories and narratives has proven to be a difficult task to automate because traditional vector representations fail to capture the layered and multi-faceted aspects of stories such as theme, plot progression, and resolution. We address SemEval-2026 Task 4, which requires generating vector embeddings that preserve narrative similarity relationships. We propose Neural Aspect Composition, which functions by using a Large Language Model (LLM) to decompose stories into 13 semantic narrative aspects (theme, course of action, outcomes, etc.), encodes each aspect separately using an encoder model, and learns a global importance weight for each aspect through a trained weighting layer. Our approach achieves the official test scores of 0.64 on Track A and 0.61 on Track B. During validation, it outperformed vectors produced by inputting the raw story text directly into an encoder model and a sentence-averaging baseline. The analysis of the learned weights on the development set reveals that thematic elements and narrative resolutions were the primary drivers of perceived similarity, receiving significantly higher weights than intermediate plot events and other minor details such as character introductions.
%U https://aclanthology.org/2026.semeval-1.341/
%P 2709-2714
Markdown (Informal)
[ChulaNLP at SemEval-2026 Task 4: Neural Aspect Composition for Narrative Story Embeddings](https://aclanthology.org/2026.semeval-1.341/) (Gampper & Rutherford, SemEval 2026)
ACL