Regional vs. global robust spelling correction

We explore the practical viability of a regional architecture to deal with robust spelling correction, a process including both unknown sequences recognition and spelling correction. Our goal is to reconcile these techniques from both the topological and the operational point of view. In contrast to the global strategy of most spelling correction algorithms, and local ones associated with the completion of unknown sequences, our proposal seems to provide an unified framework allowing us to maintain the advantages in each case, and avoid the drawbacks.