A server is configured to determine translations of queries from a first language into a second language, associated with a first specialized search engine, to obtain translated queries. The server is also configured to use a first model, associated with the first specialized search engine, to determine values for the translated queries. A value of the values, corresponding to a translated query of the translated queries, reflects a probability that the translated query is a type of query for which first specialized search results are responsive. The server is configured to create training data based on the queries and the values, and to create a second model based on the training data. The second model may be used to predict whether a particular query, received by the web search engine, is the type of query for which second specialized search results, from a second specialized search engine, are responsive.