@inproceedings{kuzman-pungersek-etal-2026-state,
title = "State of the Art in Text Classification for {S}outh {S}lavic Languages: Fine-Tuning or Prompting?",
author = "Kuzman Punger{\v{s}}ek, Taja and
Rupnik, Peter and
Porupski, Ivan and
Dini{\'c}, Vuk and
Ljube{\v{s}}i{\'c}, Nikola",
editor = "Montejo-Raez, Arturo and
Grisot, Cristina and
Blochowiak, Joanna and
Ljube{\v{s}}i{\'c}, Nikola and
Battaner, Elena and
Rigau, German",
booktitle = "Proceedings of Shaping Multilingual, Multimodal {AI} for the Social Sciences and Humanities ({LLM}s4{SSH}) @ {LREC} 2026",
month = may,
year = "2026",
address = "Palma de Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.llms4ssh-1.1/",
doi = "10.63317/2fudy99w2taz",
pages = "1--17",
abstract = "Until recently, fine-tuned BERT-like models provided state-of-the-art performance on text classification tasks. With the rise of instruction-tuned decoder-only models, commonly known as large language models (LLMs), the field has increasingly moved toward zero-shot and few-shot prompting. However, the performance of LLMs on text classification, particularly on less-resourced languages, remains under-explored. In this paper, we evaluate the performance of current language models on text classification tasks across several South Slavic languages. We compare openly available fine-tuned BERT-like models with a selection of open-weight and closed-source LLMs across three tasks in three domains: sentiment classification in parliamentary speeches, topic classification in news articles and parliamentary speeches, and genre identification in web texts. Our results show that LLMs demonstrate strong zero-shot performance, often matching or surpassing fine-tuned BERT-like models. Moreover, when used in a zero-shot setup, LLMs perform comparably in South Slavic languages and English. However, we also point out key drawbacks of LLMs, including less predictable outputs, significantly slower inference, and higher computational costs. Due to these limitations, fine-tuned BERT-like models remain a more practical choice for large-scale automatic text annotation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kuzman-pungersek-etal-2026-state">
<titleInfo>
<title>State of the Art in Text Classification for South Slavic Languages: Fine-Tuning or Prompting?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Taja</namePart>
<namePart type="family">Kuzman Pungeršek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peter</namePart>
<namePart type="family">Rupnik</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ivan</namePart>
<namePart type="family">Porupski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vuk</namePart>
<namePart type="family">Dinić</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nikola</namePart>
<namePart type="family">Ljubešić</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 Shaping Multilingual, Multimodal AI for the Social Sciences and Humanities (LLMs4SSH) @ LREC 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Arturo</namePart>
<namePart type="family">Montejo-Raez</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Cristina</namePart>
<namePart type="family">Grisot</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joanna</namePart>
<namePart type="family">Blochowiak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nikola</namePart>
<namePart type="family">Ljubešić</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Elena</namePart>
<namePart type="family">Battaner</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">German</namePart>
<namePart type="family">Rigau</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>Until recently, fine-tuned BERT-like models provided state-of-the-art performance on text classification tasks. With the rise of instruction-tuned decoder-only models, commonly known as large language models (LLMs), the field has increasingly moved toward zero-shot and few-shot prompting. However, the performance of LLMs on text classification, particularly on less-resourced languages, remains under-explored. In this paper, we evaluate the performance of current language models on text classification tasks across several South Slavic languages. We compare openly available fine-tuned BERT-like models with a selection of open-weight and closed-source LLMs across three tasks in three domains: sentiment classification in parliamentary speeches, topic classification in news articles and parliamentary speeches, and genre identification in web texts. Our results show that LLMs demonstrate strong zero-shot performance, often matching or surpassing fine-tuned BERT-like models. Moreover, when used in a zero-shot setup, LLMs perform comparably in South Slavic languages and English. However, we also point out key drawbacks of LLMs, including less predictable outputs, significantly slower inference, and higher computational costs. Due to these limitations, fine-tuned BERT-like models remain a more practical choice for large-scale automatic text annotation.</abstract>
<identifier type="citekey">kuzman-pungersek-etal-2026-state</identifier>
<identifier type="doi">10.63317/2fudy99w2taz</identifier>
<location>
<url>https://aclanthology.org/2026.llms4ssh-1.1/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>1</start>
<end>17</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T State of the Art in Text Classification for South Slavic Languages: Fine-Tuning or Prompting?
%A Kuzman Pungeršek, Taja
%A Rupnik, Peter
%A Porupski, Ivan
%A Dinić, Vuk
%A Ljubešić, Nikola
%Y Montejo-Raez, Arturo
%Y Grisot, Cristina
%Y Blochowiak, Joanna
%Y Ljubešić, Nikola
%Y Battaner, Elena
%Y Rigau, German
%S Proceedings of Shaping Multilingual, Multimodal AI for the Social Sciences and Humanities (LLMs4SSH) @ LREC 2026
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma de Mallorca (Spain)
%F kuzman-pungersek-etal-2026-state
%X Until recently, fine-tuned BERT-like models provided state-of-the-art performance on text classification tasks. With the rise of instruction-tuned decoder-only models, commonly known as large language models (LLMs), the field has increasingly moved toward zero-shot and few-shot prompting. However, the performance of LLMs on text classification, particularly on less-resourced languages, remains under-explored. In this paper, we evaluate the performance of current language models on text classification tasks across several South Slavic languages. We compare openly available fine-tuned BERT-like models with a selection of open-weight and closed-source LLMs across three tasks in three domains: sentiment classification in parliamentary speeches, topic classification in news articles and parliamentary speeches, and genre identification in web texts. Our results show that LLMs demonstrate strong zero-shot performance, often matching or surpassing fine-tuned BERT-like models. Moreover, when used in a zero-shot setup, LLMs perform comparably in South Slavic languages and English. However, we also point out key drawbacks of LLMs, including less predictable outputs, significantly slower inference, and higher computational costs. Due to these limitations, fine-tuned BERT-like models remain a more practical choice for large-scale automatic text annotation.
%R 10.63317/2fudy99w2taz
%U https://aclanthology.org/2026.llms4ssh-1.1/
%U https://doi.org/10.63317/2fudy99w2taz
%P 1-17
Markdown (Informal)
[State of the Art in Text Classification for South Slavic Languages: Fine-Tuning or Prompting?](https://aclanthology.org/2026.llms4ssh-1.1/) (Kuzman Pungeršek et al., LLMs4SSH 2026)
ACL