The Web has been rapidly "deepened" by myriad searchable databases online, where data are hidden behind query forms. Helping users query alternative "deep Web" sources in the same domain (e.g., Books, Airfares) is an important task with broad applications. As a core component of those applications, dynamic query translation (i.e., translating a user's query across dynamically selected sources) has not been extensively explored. While existing works focus on isolated subproblems (e.g., schema matching, query rewriting) to study, we target at building a complete query translator and thus face new challenges: 1) To complete the translator, we need to solve the predicate mapping problem (i.e., map a source predicate to target predicates), which is largely unexplored by existing works; 2) To satisfy our application requirements, we need to design a customizable system architecture to assemble various components addressing respective subproblems (i.e., schema matching, predicate mapping, query rewriting). Tackling these challenges, we develop a light-weight domain-based form assistant, which can generally handle alternative sources in the same domain and is easily customizable to new domains. Our experiment shows the effectiveness of our form assistant in translating queries for real Web sources.