One of the challenges of automating the tracking of patients
in the ICU is to incorporate the observed changes and actions. The
primary source of this information is the nursing notes. The nursing
notes present a special challenge to parsing and interpretation. They
are rapidly entered and highly telegraphic with many abbreviations,
some created on the spot. This talk will discuss our preliminary work
on this problem. So far we have a tokenizer to recognize the atomic
entities, a simple abbreviation recognizer that expands the common
abbreviations, a spelling corrector to fix simple errors and conjoined
words, and the UMLS to classify medical terminology. These are most
of the basic pieces, but the problem is far from solved.