{
  "field_id": "F026",
  "field": "biomedical natural language processing",
  "directions": [
    {
      "direction_id": "G0",
      "name": "Reusable biomedical language representations and models",
      "scope": "Learn language models or semantic representations usable across multiple medical or biomedical tasks, including domain pretraining, subsequent medical specialization, and semantic representations of terms, sentences, or medical records."
    },
    {
      "direction_id": "G1",
      "name": "Biomedical language-model capability assessment",
      "scope": "Assess the capabilities and limitations of language models in biomedical and clinical NLP through cross-task benchmarks, systematic comparisons, or comprehensive analyses."
    },
    {
      "direction_id": "G2",
      "name": "Clinical NLP adaptation under limited or shifting evidence",
      "scope": "Study how clinical text models maintain or acquire usable capabilities when institutions, languages, data sources, or task-label conditions change, rather than addressing only one fixed downstream task."
    },
    {
      "direction_id": "G3",
      "name": "Acquiring labels for medical text extraction",
      "scope": "When training annotations for extraction are scarce, create or obtain usable labeled material; acquiring annotations is itself the research result."
    },
    {
      "direction_id": "G4",
      "name": "Patient condition, diagnosis and outcome inference",
      "scope": "Use patient language or clinical text evidence to make patient-level judgments, including identifying documented states, inferring diagnoses, and predicting subsequent risks or events; individual studies need not address all three."
    },
    {
      "direction_id": "G5",
      "name": "Structured facts and relations from biomedical text",
      "scope": "Turn local information in clinical or biomedical text into usable structures, covering entity mentions, attributes, and events, as well as relationships between entities."
    },
    {
      "direction_id": "G6",
      "name": "Mapping medical text to standardized concepts and codes",
      "scope": "Map medical expressions in text or document content to standardized terminology systems, and assess the accuracy and textual basis of those mappings."
    },
    {
      "direction_id": "G7",
      "name": "Interpretation of biomedical research findings",
      "scope": "Determine what interventions, outcomes, or findings research reports present, and whether research evidence supports or refutes a claim."
    },
    {
      "direction_id": "G8",
      "name": "Evidence-grounded medical text communication",
      "scope": "Transform existing clinical or biomedical material into coherent, trustworthy text suited to its intended readers, including care documentation, overviews of research evidence, and patient-facing explanations."
    },
    {
      "direction_id": "G9",
      "name": "Biomedical information seeking and answering",
      "scope": "Locate relevant material or formulate answers in response to medical or biomedical information needs; retrieval may be a standalone result or support answering."
    },
    {
      "direction_id": "G10",
      "name": "Privacy-preserving clinical text transformation",
      "scope": "Identify and remove or replace identifying information in clinical text while considering privacy protection and preservation of the text’s utility."
    }
  ]
}