Comprehending technical texts: predicting and defining unfamiliar terms

We investigate how to improve access to medical literature for health consumers. Our focus is on medical terminology. We present a method to predict automatically in a given text which medical terms are unlikely to be understood by a lay reader. Our method, which is linguistically motivated and fully unsupervised, relies on how common a specific term is in texts that we already know are familiar to a lay reader. Once a term is identified as unfamiliar, an appropriate definition is mined from the Web to be provided to the reader. Our experiments show that the prediction and the addition of definitions significantly improve lay readers' comprehension of sentences containing technical medical terms.