The literature on finite state automata generally assumes a relatively small alphabet, often 128 or fewer characters. Small alphabets allow one to implement an FSA efficiently and in a straight-forward manner.

Large alphabets (of which Unicode is a prime example) can make the efficient and compact implementation of automata difficult. This talk is a continuation of one given at IUC 21 in Washington DC, describing my latest work on space and time methods for building and dictionaries for large languages (100,000+ words) over large alphabets (5000+ symbols). I present an implementation of a recent incremental ADFSA algorithm due to Daciuk et al. that scales very well to large languages and is not impacted by alphabet size. I also present a method for representing the built automata in a compact way.