@inproceedings{marongiu-sassolini-2026-lemmatization,
title = "From Lemmatization to Legal Terminology: Assessing an Hybrid Pipeline on Justinian{'}s Digest",
author = "Marongiu, Paola and
Sassolini, Eva",
editor = "Sprugnoli, Rachele and
Passarotti, Marco",
booktitle = "Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages ({LT}4{HALA} 2026) @ {LREC} 2026",
month = may,
year = "2026",
address = "Palma, Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.lt4hala-1.43/",
doi = "10.63317/4x2mcetz9f5r",
pages = "418--428",
abstract = "This paper evaluates a hybrid NLP pipeline for supporting the extraction of Roman legal terminology from Jus- tinian{'}s Digest. Our goal is not to optimize lemmatization in isolation, but to assess whether integrating a Large Language Model (GPT-4o-mini) as a post-processing component improves lemma quality in ways that are critical for downstream glossary construction. Using LatinPipe as a baseline (F1 = 95.05), we test the integration of GPT-4o-mini under three experimental settings (zero-shot with and without prior lemma information, and few-shot prompting) against a manually annotated gold standard of 3,703 sentences and an expert-validated list of legal Latin technical terms. Results show improvement across all settings, with the best performance achieved in the few-shot configuration. Our analysis shows that the hybrid configuration produces selective improvements, significantly more likely for frequent lemmas and verbs forms, suggesting that the LLM layer primarily assists in resolving morphologically ambiguous inflected forms. Although our experimental conditions may not hold in real-world scenarios, we argue that the main contribution of this work is methodological: demonstrating how evaluation can be aligned with downstream terminological goals, rather than proposing a general-purpose solution to domain-specific lemmatization."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="marongiu-sassolini-2026-lemmatization">
<titleInfo>
<title>From Lemmatization to Legal Terminology: Assessing an Hybrid Pipeline on Justinian’s Digest</title>
</titleInfo>
<name type="personal">
<namePart type="given">Paola</namePart>
<namePart type="family">Marongiu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Eva</namePart>
<namePart type="family">Sassolini</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 Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Rachele</namePart>
<namePart type="family">Sprugnoli</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marco</namePart>
<namePart type="family">Passarotti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca (Spain)</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>This paper evaluates a hybrid NLP pipeline for supporting the extraction of Roman legal terminology from Jus- tinian’s Digest. Our goal is not to optimize lemmatization in isolation, but to assess whether integrating a Large Language Model (GPT-4o-mini) as a post-processing component improves lemma quality in ways that are critical for downstream glossary construction. Using LatinPipe as a baseline (F1 = 95.05), we test the integration of GPT-4o-mini under three experimental settings (zero-shot with and without prior lemma information, and few-shot prompting) against a manually annotated gold standard of 3,703 sentences and an expert-validated list of legal Latin technical terms. Results show improvement across all settings, with the best performance achieved in the few-shot configuration. Our analysis shows that the hybrid configuration produces selective improvements, significantly more likely for frequent lemmas and verbs forms, suggesting that the LLM layer primarily assists in resolving morphologically ambiguous inflected forms. Although our experimental conditions may not hold in real-world scenarios, we argue that the main contribution of this work is methodological: demonstrating how evaluation can be aligned with downstream terminological goals, rather than proposing a general-purpose solution to domain-specific lemmatization.</abstract>
<identifier type="citekey">marongiu-sassolini-2026-lemmatization</identifier>
<identifier type="doi">10.63317/4x2mcetz9f5r</identifier>
<location>
<url>https://aclanthology.org/2026.lt4hala-1.43/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>418</start>
<end>428</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T From Lemmatization to Legal Terminology: Assessing an Hybrid Pipeline on Justinian’s Digest
%A Marongiu, Paola
%A Sassolini, Eva
%Y Sprugnoli, Rachele
%Y Passarotti, Marco
%S Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F marongiu-sassolini-2026-lemmatization
%X This paper evaluates a hybrid NLP pipeline for supporting the extraction of Roman legal terminology from Jus- tinian’s Digest. Our goal is not to optimize lemmatization in isolation, but to assess whether integrating a Large Language Model (GPT-4o-mini) as a post-processing component improves lemma quality in ways that are critical for downstream glossary construction. Using LatinPipe as a baseline (F1 = 95.05), we test the integration of GPT-4o-mini under three experimental settings (zero-shot with and without prior lemma information, and few-shot prompting) against a manually annotated gold standard of 3,703 sentences and an expert-validated list of legal Latin technical terms. Results show improvement across all settings, with the best performance achieved in the few-shot configuration. Our analysis shows that the hybrid configuration produces selective improvements, significantly more likely for frequent lemmas and verbs forms, suggesting that the LLM layer primarily assists in resolving morphologically ambiguous inflected forms. Although our experimental conditions may not hold in real-world scenarios, we argue that the main contribution of this work is methodological: demonstrating how evaluation can be aligned with downstream terminological goals, rather than proposing a general-purpose solution to domain-specific lemmatization.
%R 10.63317/4x2mcetz9f5r
%U https://aclanthology.org/2026.lt4hala-1.43/
%U https://doi.org/10.63317/4x2mcetz9f5r
%P 418-428
Markdown (Informal)
[From Lemmatization to Legal Terminology: Assessing an Hybrid Pipeline on Justinian’s Digest](https://aclanthology.org/2026.lt4hala-1.43/) (Marongiu & Sassolini, LT4HALA 2026)
ACL