Law, copyright, and machine learning.

ML & AI Concepts: Classifiers, Metrics, and Experiments

Experiments compared Logistic Regression, Decision Trees, Random Forest, Support Vector Machines, and K-Nearest Neighbors across several datasets.

The work reinforced three lessons: model performance depends on the dataset, tree-based systems can overfit, and tuning can improve metrics while increasing computational cost.

Predicting Copyright Case Outcomes with ML Models

This capstone explored whether machine-learning models could predict outcomes in civil copyright litigation using public court-opinion data.

Within the project dataset, Gradient Boosting performed best when legal text was included; tuned SVM improved substantially; Random Forest showed signs of overfitting; and KNN remained the weakest performer.

The larger lesson was methodological: legal language is not incidental metadata. Textual context materially affected model performance.

Observations on AI Translation of Legal Terms

An exploratory round-trip translation study examined how publicly accessible AI systems handled copyright and intellectual-property language between English and Spanish.

The translations were often broadly accurate, but some choices weakened precision. “Further purpose” shifted toward propósito adicional or propósito posterior; “supersedes” was sometimes reduced to reemplaza rather than the stronger suplanta.

This study became the hinge between legal-text research and later work on narrative flattening: a text may appear correct while still losing force, nuance, or specificity.