A research team is pre-training a language model. They integrate this model as a component within a larger system designed to classify news articles into categories like 'Sports', 'Technology', or 'Politics'. The language model's internal parameters are updated based on the entire system's success in correctly categorizing the articles. Which statement best analyzes the fundamental principle of this training methodology?

Optimizing a Sentiment Analysis Model

Explaining a Pre-Training Supervision Method

Sentence Comparison in Discriminative Training

Token Classification in Discriminative Training