Systems translating keyword queries into SQL queries over relational databases are usually referred to in the literature as schema-based approaches. These techniques exploit the information contained in the database schema to build SQL queries that express the intended meaning of the user query. Besides, typically, they perform a preliminary step that associates keywords in the user query with database elements (names of tables, attributes and domain attributes). In this paper, we present a probabilistic approach based on a Hidden Markov Model to provide such mappings. In contrast to most existing techniques, our proposal does not require any a-priori knowledge of the database extension.