cassielee04/automated-main-concept-analysis: Automated Main-Concept Analysis v0.1.0

Abstract

Initial research release This is the first archived release of Automated Main-Concept Analysis, a research pipeline for measuring main-concept production and discourse coherence in spoken Cinderella narratives. The pipeline processes AWS Transcribe JSON files, segments transcripts into utterances, and compares those utterances with 34 reference Cinderella story concepts using sentence embeddings. It produces participant-level features for control, aphasia, and dementia cohorts. Included Transcript preprocessing and filler normalization Stanza-based utterance segmentation Semantic main-concept matching using sentence-transformers/all-mpnet-base-v2 Main-concept counts, match ratios, and story-sequence measures Rule-based local and global coherence classifications Cutoff-independent utterance caching Participant-grouped cross-validation feature extraction Classification and sensitivity-analysis notebooks Precomputed Cinderella concept embeddings Sun Grid Engine job-submission scripts Existing derived feature tables, figures, and analysis outputs Research status This release is an initial archival snapshot of an active research codebase. It preserves the code and analysis materials associated with the current study, but additional packaging, documentation, dependency pinning, path cleanup, and reproducibility improvements are planned for later releases. Several scripts retain paths and settings from the original research-computing environment. Raw participant transcripts are not included. The software is provided for research and reproducibility purposes. It is not a clinical diagnostic tool, and its outputs require expert interpretation and independent validation. Known limitations Dependencies and model versions are not yet pinned. Some scripts contain environment-specific paths and filenames. The direct cohort entry point references an external ciu helper that is not included in this snapshot. The coherence classifications are heuristic and should not be treated as clinical ground truth. The repository does not yet include a formal automated test suite. See the repository README for setup instructions, input requirements, pipeline details, and output definitions.