We download the original newspaper articles automatically from the WWW2, and apply a number of processing stages sequentially. Lexical Tagger The tagger (Elworthy, 1994) assigns and ranks part-of-speech (PoS) tags for each word in a sentence using a rst-order HMM. The tagger includes an unknown word guesser with an accuracy of around 85%, and a large diskresident lexicon specialised to newspaper text. Morphological Analyser The morphological analyser (an enhanced version of the GATE project lemmatiser (Cunningham et al., 1996)) is based on nite state techniques, and performs an accurate and e cient in ectional analysis of the words in.
