One of the distinctions between supervised and unsupervised learning in AI is the further use of such models in legal language machine development. Labeled data are used when training models based on input examples rather than on estimated values as performed by unsupervised approaches. This method is particularly important with respect to tasks encountered in the legal domain, such as – document classification where the legal texts are classified to types eg contracts or court opinions, and named entity recognition who identifies entities like names and dates within documents. As effective as it can be; supervised has its own setback since it requires a huge amount of labeled data which also may take much time and money in case of legal contexts.
Unsupervised learning differs from supervised learning in that it does not rely on labeled data. It instead focuses on finding patterns or representations in the data. In legal language modeling, unsupervised learning can be used for tasks like discovering “topics,” or underlying themes within a set of legal documents, as well as for “clustering,” or grouping together related documents based on similarities in their content. Unsupervised learning is particularly interesting within legal applications due to the relatively low availability of labeled instances (e.g., when compared to labeled reading materials available for training a comprehension test), but it can also be difficult to accurately interpret the results of unsupervised methods without a bit more explicit guidance from labeled examples.
Both of these approaches are important to the advancement of legal language models—supervised learning for tasks that need labeled data at the level of precision and recall we can currently achieve, and unsupervised learning to facilitate exploration and discovery in the vast space of unannotated legal text corpora.