A combining approach to find all taxon names (FAT)

Most of the literature on natural history is hidden in millions of pages stacked up in our libraries. Various initiatives aim now at making these publications digitally accessible and searchable, applying xml-mark up technologies. The unique biological names play a crucial role to link content related to a particular taxon. Thus discovering and marking them up is extremely important. Since their manual extraction and markup is cumbersome and time-intensive, it needs be automated. In this paper, we present computational linguistics techniques and evaluate how they can help to extract taxonomic names auto-matically. We build on an existing approach for extraction of such names (Koning et al. 2005) and combine it with several other learning techniques. We apply them to the texts sequentially so that each technique can use the results from the preceding ones. In particular, we use structural rules, dynamic lexica with fuzzy lookups, and word-level language recognition. We use legacy documents from different sources and times as test bed for our evaluation. The experimental results for our combining approach (FAT) show greater than 99% precision and recall. They reveal the potential of computational linguis-tics techniques towards an automated markup of biosystematics publications.