At present the analytical level represents the main level for structural annotation. Here linear annotation is abandoned, where each word was taken separately irrespective of context, and it is the structure of the sentence that is introduced into the annotation of the text (however, not the structure of the text). All the original words of the text are preserved and obtain their functions in the resulting structure.
By "structure" a tree is understood (a directed acyclic graph having one root) whose nodes are marked with attributes (Note: the word "attribute" is used here in its mathematical, rather than linguistic meaning - see List of analytical functions.)
Every node has 12 attributes at present. The edges (arcs) of the tree represent relations of dependency between lexical units, or, as the case may be, relations of some other type (according to the values of the attributes). The basic objective of the annotation is to render correctly the structure of the sentence and to denote the type of dependency. A type of dependency represents one of the attributes of each node and is, understandably, oriented "upwards", i.e., towards its "governing" node.
A detailed description of the analytical level is contained in Chapter 2., Principles of Annotation and Chapter 3., Rules of annotation. The volume of data to be attained is the same as at the morphological level, i.e., 1 million words, using the same texts.