SYNUR

Dataset Card: SYNUR (Synthetic Nursing Observation Dataset) 1. Dataset Summary **Name**: SYNUR **Full name / acronym**: SYnthetic NURsing Observation Extraction **Purpose / use case**: SYNUR is intended to support research in structuring nurse dictation transcripts by extracting clinical observations that can feed into flowsheet-style EHR entries. It is designed to reduce documentation burden by enabling automated conversion from spoken nurse assessments to structured observations (Paper) **Version 1: Original**: Released with the EMNLP 2025 industry track paper **Version 2: MEDIQA-SYNUR**: Released with the MEDIQA-SYNUR 2026 shared task, including an additional test set: mediqa_synur_train and mediqa_synur_dev correspond to the original dataset mediqa_synur_test is the official test set for the MEDIQA-SYNUR 2026 challenge **License / usage terms**: cdla-permissive-2.0 2. Data Fields / Format `transcript`: string, the nurse dictation (raw spoken text) `observations`: JSON dumped of list of dictionaries with following format `id` (str): key of observation in schema. `value_type` (str): type of observation in {*SINGLE_SELECT*, *MULTI_SELECT*, *STRING*, *NUMERIC*}. `name` (str): observation concept name. `value` (any): value of observation. 3. Observation Schema The full schema (i.e., 193 observation concepts) is provided at the root of this dataset repo as `synur_schema.json`. It is a list of dictionaries with the following key-value pairs: `id` (str): key of observation concept. `name` (str): observation concept name. `value_type` (str): type of observation in {*SINGLE_SELECT*, *MULTI_SELECT*, *STRING*, *NUMERIC*}. `value_enum` (List[str], *optional*): set of possible string values for *SINGLE_SELECT* and *MULTI_SELECT* value types. 4. Contact **Maintainers**: {jcorbeil,georgemi,abenabacha}@microsoft.com 5. Citations If you use the original split and MEDIQA-SYNUR splits, please cite the papers below: @inproceedings{corbeil-etal-2025-empowering, title = "Empowering He