In this paper, we investigate the retrieval, using statistical models, of strongly-textual cases, i.e. cases where both problem and solution components are described textually. We study two models borrowed from current research in natural language processing: Word co-occurrences and translation alignments. These approaches offer the potential to capture words associations between a problem description and its corresponding textual solutions. These lexical associations are injected in the retrieval process to guide the search for candidate solutions. We present the results of our experimentation and compare these with tf*idf vector-based similarity.