A Muse is a heavy resource in memoQ. It provides subsegment suggestions to users as they type in the translation grid. Muses collect single and multiword expressions in source and target content, identifies correlations between source and target, and suggests the target expressions as a suggestion in the predictive typing.
You can use Muses in local projects only. Muses also work in a local copy of an online project with offline copies of the TMs. You can retrain your Muse when for example your TM has grown.
Train a muse with a corpus of a few thousand segments. Let's assume that the corpus has the source expression "term base" and the target expression "Termdatenbank" and no explicit indication that both expressions are corresponding to each other. The Muse now detects correlations between such expressions. When you use the trained Muse now in a project, it compares the source segment with the extracted source expression from the training, and it suggests the target expression in the predictive typing during translation.