It is desirable to represent a MUSE CSP as a directed acyclic graph (DAG) where the directed paths through the DAG correspond to instances of CSP problems. It is often easy to determine which variables should be shared and how to construct the DAG. The application presented in this section is one for which MUSE CSP is useful. A parsing problem is naturally represented as a DAG because of the presence of ambiguity. In many cases, the same word can have multiple parts of speech; it is convenient to represent those words as nodes in a MUSE CSP. In speech recognition systems, the identification of the correct words in a sentence can be improved by using syntactic constraints. However, a word recognition algorithm often produces a lattice of word candidates. Clearly, individually parsing each of the sentences in a lattice can be inefficient.