MART-Lexical Complexity and Creativity

Key Features

Semantic Complexity Analysis

Baseline Generation:

Uses the BLIP-2 vision-language model (Salesforce/blip-image-captioning-large) to generate baseline descriptions of artwork Multilingual Processing: Employs the paraphrase-multilingual-MiniLM-L12-v2 sentence transformer for cross-lingual semantic analysis Complexity Measurement: Calculates semantic complexity using cosine distance from AI-generated baseline descriptions

Statistical Analysis

ANOVA Testing: Performs one-way ANOVA to compare semantic complexity across different experimental groups Post-hoc Analysis: Implements Tukey's HSD test for pairwise group comparisons Advanced Visualization: Creates comprehensive bar plots with significance indicators and statistical annotations

Dependencies

PIL (Python Imaging Library) PyTorch Transformers (Hugging Face) Sentence Transformers SciPy & StatsModels for statistical analysis Matplotlib & Seaborn for visualization Pandas & NumPy for data manipulation

Methodology

Image Processing: Loads and processes artwork images using PIL Baseline Creation: Generates standardized descriptions using BLIP-2 with token limits Semantic Embedding: Converts both baseline and participant descriptions into high-dimensional vectors Complexity Scoring: Measures semantic distance from baseline to quantify creativity/complexity Statistical Testing: Performs rigorous statistical analysis with multiple comparison corrections