The assimilation and deletion of consonants at word boundaries poses a problem for the continuous recognition of English speech. This is because the location of word boundaries in the input is not known, Two methods for accounting for these phonological processes are discussed. The offline approach increases the size and complexity of the lexicon, while the online approach generates possible but non-occurring words. A third type of algorithm, Finite State Transducers (FST's), is shown to combine the advantages of these two approaches while avoiding their drawbacks. The results of implementing and testing FST's are given. Consideration is given to the use of FST's in accounting for similar consonantal phonological processes within words.