This thesis compares eight agglomerative hierarchical clustering methods, two divisive hierarchical methods, and three k-means methods. The data used was a corpus of 689 texts written by school children in 1930s Ireland which have been transcribed by volunteers. The effects of stop word removal and stemming on each of these were investigated, as was the use of document embeddings as input instead of a document-term matrix. Overall, k-means methods produced the most desirable results, and document embeddings markedly improved output in most cases.