Evaluating Structured Output Robustness of Small Language Models for Open Attribute-Value Extraction from Clinical Notes

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Neveditsin, Nikita, Lingras, Pawan, Mago, Vijay
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908479111823360
author Neveditsin, Nikita
Lingras, Pawan
Mago, Vijay
author_facet Neveditsin, Nikita
Lingras, Pawan
Mago, Vijay
contents We present a comparative analysis of the parseability of structured outputs generated by small language models for open attribute-value extraction from clinical notes. We evaluate three widely used serialization formats: JSON, YAML, and XML, and find that JSON consistently yields the highest parseability. Structural robustness improves with targeted prompting and larger models, but declines for longer documents and certain note types. Our error analysis identifies recurring format-specific failure patterns. These findings offer practical guidance for selecting serialization formats and designing prompts when deploying language models in privacy-sensitive clinical settings.
format Preprint
id arxiv_https___arxiv_org_abs_2507_01810
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Evaluating Structured Output Robustness of Small Language Models for Open Attribute-Value Extraction from Clinical Notes
Neveditsin, Nikita
Lingras, Pawan
Mago, Vijay
Computation and Language
Information Retrieval
We present a comparative analysis of the parseability of structured outputs generated by small language models for open attribute-value extraction from clinical notes. We evaluate three widely used serialization formats: JSON, YAML, and XML, and find that JSON consistently yields the highest parseability. Structural robustness improves with targeted prompting and larger models, but declines for longer documents and certain note types. Our error analysis identifies recurring format-specific failure patterns. These findings offer practical guidance for selecting serialization formats and designing prompts when deploying language models in privacy-sensitive clinical settings.
title Evaluating Structured Output Robustness of Small Language Models for Open Attribute-Value Extraction from Clinical Notes
topic Computation and Language
Information Retrieval
url https://arxiv.org/abs/2507.01810