Abstract. In the document-centric modeling of information systems, the flow of data between individual documents and the collections stored in the database plays an important role. The types of data that can be stored in the collections and their constraints are given by the data models supported by the database management software. We show the semantically fundamental usage difference between the documents and the database. In addition to managing relational rows, the relational data model that is the most important starting point also contains a basic document type, namely the tabular document type, as a representation of a relation consisting of relatively few rows. With this dual task in mind, we introduce the extended relational model. In the standard relational model, each relation name identifies a collection of occurrences of a single row type. Therefore, in schema-level instructions and operations, each relation name/table name has a fixed set of attributes (ordered or unordered). In our paper, we introduce a collection containing tables of different but similar schemas. Actions can be specified at this level, which applies to several tables at the same time. The similarity of the schemes is given by belonging to a formal language over a finite set of attribute names. Every sentence of a given formal language can be a potential schema. An attribute name can therefore occur more than once in a specific schema, and the order is fixed. An instance of a schema collection consists of a finite number of table types. This is called an extended relation. Among the formal languages, we first discuss the use of regular languages. In the case of context-free languages, we use the language's terminal and non-terminal symbols as attribute names. With this, embedded relational schemas can also be handled. We use graph representations of formal languages as schema graphs. With the help of this, we specify the operations at the schema and instance level. The definition and implication problems of functional dependencies are investigated too. We show the close connection with the XML ELEMENT declaration. We also specify an implementation solution. After introduction, we shortly examine its use in the document centric modelling of information systems. A novel X-merge solution is specified between free documents and X-relations.