LLMs Underperform Graph-Based Parsers on Supervised Relation Extraction for Complex Graphs

Paolo Gajo, Domenic Rosati, Hassan Sajjad, Alberto Barrón-Cedeño · 2026

Read LLMs Underperform Graph-Based Parsers on Supervised Relation Extraction for Complex Graphs by Paolo Gajo, Domenic Rosati, Hassan Sajjad, Alberto Barrón-Cedeño (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-08752.txt.

Description

Gajo and three colleagues show language models still trail graph based parsers on supervised relation extraction when sentence graphs grow complex. Testing four language models against a graph parser on six datasets, they find the parser's lead grows with the number of relations in input documents. The much lighter graph parser thus wins on complex linguistic graphs, challenging the assumption that large language models dominate this knowledge graph building task.

Catalog record

CREATORPaolo Gajo, Domenic Rosati, Hassan Sajjad, Alberto Barrón-Cedeño
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35317
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.08752

Browse the full library · About the Tower of Babel