This paper conducts a comparative study on the performance of various machine learning ("ML") approaches for classifying judgments into legal areas. Using a novel dataset of 6,227 Singapore Supreme Court judgments, we investigate how state-of-the-art NLP methods compare against traditional statistical models when applied to a legal corpus that comprised few but lengthy documents. All approaches tested, including topic model, word embedding, and language model-based classifiers, performed well with as little as a few hundred judgments. However, more work needs to be done to optimize state-of-the-art methods for the legal domain.

No abstract

No abstract

Legal academics were once thought to be parasitic on the work of judges, so much so that citing academic work was said to weaken a judgment's authority. Recent times have however seen prominent academics appointed to the highest courts, and judicial engagement with academic materials appears to have increased. In this light, this article empirically studies academic citation practices in the Singapore High Court. Using a dataset of 2,772 first-instance High Court judgments, we show that citation counts have indeed increased over time. This increase was distributed across most legal areas, and was not limited to, though more pronounced in, judgments authored by judges with post-graduate law degrees. Books, not journal articles, have consistently accounted for the bulk of the court's citations. The study sheds new statistical light on the evolving relationship between judges and academics, particularly in the context of an Asian, first-instance court.

We propose and evaluate generative models for case law citation networks that account for legal authority, subject relevance, and time decay. Since Common Law systems rely heavily on citations to precedent, case law citation networks present a special type of citation graph which existing models do not adequately reproduce. We describe a general framework for simulating node and edge generation processes in such networks, including a procedure for simulating case subjects, and experiment with four methods of modelling subject relevance: using subject similarity as linear features, as fitness coefficients, constraining the citable graph by subject, and computing subject-sensitive PageRank scores. Model properties are studied by simulation and compared against existing baselines. Promising approaches are then benchmarked against empirical networks from the United States and Singapore Supreme Courts. Our models better approximate the structural properties of both benchmarks, particularly in terms of subject structure. We show that differences in the approach for modelling subject relevance, as well as for normalizing attachment probabilities, produce significantly different network structures. Overall, using subject similarities as fitness coefficients in a sum-normalized attachment model provides the best approximation to both benchmarks. Our results shed light on the mechanics of legal citations as well as the community structure of case law citation networks. Researchers may use our models to simulate case law networks for other inquiries in legal network science.