Text Accuracy Evaluator with Normalization

In Natural Language Processing (NLP) and machine learning evaluation pipelines, measuring how accurately a model's predictions match expected outputs is fundamental. One widely used approach is Normalized String Matching, which compares predicted text against ground truth references after applying standardization transformations.

This technique is especially valuable when evaluating question-answering systems, text generation models, and information extraction pipelines, where superficial differences in formatting, capitalization, or punctuation should not penalize otherwise correct answers.

The Normalization Process: Before comparing strings, both predictions and references undergo a normalization pipeline:

1. Case Normalization: Convert all characters to lowercase to ensure case-insensitive comparison
2. Punctuation Removal: Strip all punctuation marks (periods, commas, apostrophes, quotes, etc.)
3. Whitespace Standardization: Collapse multiple consecutive spaces into a single space and remove leading/trailing whitespace

After normalization, strings are compared for exact equality. The accuracy score is computed as the fraction of predictions that perfectly match their corresponding normalized references.

Your Task: Implement a function that takes two parallel lists—predictions and references—and returns a float representing the accuracy score. The score should be the proportion of normalized predictions that exactly match their normalized counterparts, yielding a value between 0.0 (no matches) and 1.0 (all match).

Edge Case: If both input lists are empty, your function should return 0.0.