@inproceedings{saban-etal-2026-lakefront,
title = "Lakefront {AI} Ramblers at {MEDIQA}-{SYNUR} 2026: Hybrid Retrieval and {LLM} Verification for Open-Source Schema-Guided Clinical Information Extraction",
author = "Saban, Michael T. and
Yaghoubi, Arsalan and
Eslami, Behnaz and
Tootooni, Samie and
Dligach, Dmitriy",
editor = "Ben Abacha, Asma and
Bethard, Steven and
Bitterman, Danielle and
Naumann, Tristan and
Roberts, Kirk",
booktitle = "Proceedings of the 8th Workshop on Clinical Natural Language Processing (Clinical {NLP}) @ {LREC} 2026",
month = may,
year = "2026",
address = "Palma, Mallorca (Spain)",
publisher = "ELRA Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.clinicalnlp-1.23/",
doi = "10.63317/2jfca3zi8p4o",
pages = "212--221",
abstract = "Schema-constrained clinical information extraction requires identifying text-supported observations and outputting exact schema identifiers and values. In the MEDIQA-SYNUR 2026 shared task, synthetic nursing dictations were mapped to structured JSON outputs aligned with a 193-concept clinical schema under strict exact-match evaluation. We extended the baseline pipeline, which consists of transcript segmentation, schema retrieval, and LLM-based extraction, with hybrid schema retrieval, supervised fine-tuning (SFT) of open-source LLMs, and LLM-based verification. Our hybrid retrieval approach combined dense embeddings with sparse BM25 representations using a convex combination strategy, improving schema coverage to 0.994 recall@60 on the development set. We evaluated GPT-4o, GPT-4o-mini, Llama-3-8B-Instruct, and Llama-3.3-70B-Instruct, applying LoRA-based SFT to open-source models. On the official test set, our best submitted configuration (Llama-3.3-70B-Instruct-SFT with union voting and GPT-4o-mini verification) achieved 0.711 F1. Post-competition experiments showed that Llama-3-8B-Instruct-SFT (train + dev) reached 0.723 F1 under the same post-processing pipeline. For reference, GPT-4o achieved 0.791 F1 and did not benefit from post-processing. Performance differences across development and test splits further highlight the sensitivity of post-processing strategies to variation across split distribution. Overall, integrating high-recall retrieval, SFT, and LLM verification substantially narrows the performance gap between open- and closed-source models for schema guided clinical extraction."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="saban-etal-2026-lakefront">
<titleInfo>
<title>Lakefront AI Ramblers at MEDIQA-SYNUR 2026: Hybrid Retrieval and LLM Verification for Open-Source Schema-Guided Clinical Information Extraction</title>
</titleInfo>
<name type="personal">
<namePart type="given">Michael</namePart>
<namePart type="given">T</namePart>
<namePart type="family">Saban</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Arsalan</namePart>
<namePart type="family">Yaghoubi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Behnaz</namePart>
<namePart type="family">Eslami</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Samie</namePart>
<namePart type="family">Tootooni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dmitriy</namePart>
<namePart type="family">Dligach</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 8th Workshop on Clinical Natural Language Processing (Clinical NLP) @ LREC 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Asma</namePart>
<namePart type="family">Ben Abacha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Steven</namePart>
<namePart type="family">Bethard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Danielle</namePart>
<namePart type="family">Bitterman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tristan</namePart>
<namePart type="family">Naumann</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kirk</namePart>
<namePart type="family">Roberts</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>Schema-constrained clinical information extraction requires identifying text-supported observations and outputting exact schema identifiers and values. In the MEDIQA-SYNUR 2026 shared task, synthetic nursing dictations were mapped to structured JSON outputs aligned with a 193-concept clinical schema under strict exact-match evaluation. We extended the baseline pipeline, which consists of transcript segmentation, schema retrieval, and LLM-based extraction, with hybrid schema retrieval, supervised fine-tuning (SFT) of open-source LLMs, and LLM-based verification. Our hybrid retrieval approach combined dense embeddings with sparse BM25 representations using a convex combination strategy, improving schema coverage to 0.994 recall@60 on the development set. We evaluated GPT-4o, GPT-4o-mini, Llama-3-8B-Instruct, and Llama-3.3-70B-Instruct, applying LoRA-based SFT to open-source models. On the official test set, our best submitted configuration (Llama-3.3-70B-Instruct-SFT with union voting and GPT-4o-mini verification) achieved 0.711 F1. Post-competition experiments showed that Llama-3-8B-Instruct-SFT (train + dev) reached 0.723 F1 under the same post-processing pipeline. For reference, GPT-4o achieved 0.791 F1 and did not benefit from post-processing. Performance differences across development and test splits further highlight the sensitivity of post-processing strategies to variation across split distribution. Overall, integrating high-recall retrieval, SFT, and LLM verification substantially narrows the performance gap between open- and closed-source models for schema guided clinical extraction.</abstract>
<identifier type="citekey">saban-etal-2026-lakefront</identifier>
<identifier type="doi">10.63317/2jfca3zi8p4o</identifier>
<location>
<url>https://aclanthology.org/2026.clinicalnlp-1.23/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>212</start>
<end>221</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Lakefront AI Ramblers at MEDIQA-SYNUR 2026: Hybrid Retrieval and LLM Verification for Open-Source Schema-Guided Clinical Information Extraction
%A Saban, Michael T.
%A Yaghoubi, Arsalan
%A Eslami, Behnaz
%A Tootooni, Samie
%A Dligach, Dmitriy
%Y Ben Abacha, Asma
%Y Bethard, Steven
%Y Bitterman, Danielle
%Y Naumann, Tristan
%Y Roberts, Kirk
%S Proceedings of the 8th Workshop on Clinical Natural Language Processing (Clinical NLP) @ LREC 2026
%D 2026
%8 May
%I ELRA Language Resources Association (ELRA)
%C Palma, Mallorca (Spain)
%F saban-etal-2026-lakefront
%X Schema-constrained clinical information extraction requires identifying text-supported observations and outputting exact schema identifiers and values. In the MEDIQA-SYNUR 2026 shared task, synthetic nursing dictations were mapped to structured JSON outputs aligned with a 193-concept clinical schema under strict exact-match evaluation. We extended the baseline pipeline, which consists of transcript segmentation, schema retrieval, and LLM-based extraction, with hybrid schema retrieval, supervised fine-tuning (SFT) of open-source LLMs, and LLM-based verification. Our hybrid retrieval approach combined dense embeddings with sparse BM25 representations using a convex combination strategy, improving schema coverage to 0.994 recall@60 on the development set. We evaluated GPT-4o, GPT-4o-mini, Llama-3-8B-Instruct, and Llama-3.3-70B-Instruct, applying LoRA-based SFT to open-source models. On the official test set, our best submitted configuration (Llama-3.3-70B-Instruct-SFT with union voting and GPT-4o-mini verification) achieved 0.711 F1. Post-competition experiments showed that Llama-3-8B-Instruct-SFT (train + dev) reached 0.723 F1 under the same post-processing pipeline. For reference, GPT-4o achieved 0.791 F1 and did not benefit from post-processing. Performance differences across development and test splits further highlight the sensitivity of post-processing strategies to variation across split distribution. Overall, integrating high-recall retrieval, SFT, and LLM verification substantially narrows the performance gap between open- and closed-source models for schema guided clinical extraction.
%R 10.63317/2jfca3zi8p4o
%U https://aclanthology.org/2026.clinicalnlp-1.23/
%U https://doi.org/10.63317/2jfca3zi8p4o
%P 212-221
Markdown (Informal)
[Lakefront AI Ramblers at MEDIQA-SYNUR 2026: Hybrid Retrieval and LLM Verification for Open-Source Schema-Guided Clinical Information Extraction](https://aclanthology.org/2026.clinicalnlp-1.23/) (Saban et al., ClinicalNLP 2026)
ACL