I prefer to use a more universal system like Generative Grammar instead of the classical Indian Grammar in order to develop a universal method, which may be applied not only to Sanskrit, but - theoretically - to all languages.
I assume that each actual sentence has evolved from a deeper, more logical and straightforward Deep Structure. For example, in this view the passive has an active counterpart in Deep Structure. I now analyse all verse lines, and re-write them in their Deep Structure form. In this way, the actual appearance plays no role, and the difference between the passive, middle and active is not visible anymore. This has as its advantage that the context, which plays an important role as to the choice of Mood and of emphasizing operations as inversion and extraction, is no longer a style determiner.
The analysis according to Generative Grammar yields a hierarchically organised tree-structure, as we will see in more detail in the first chapters. Each node is called a constituent, and consists of a head XP, and a specifier SPEC as its complement. Examples of heads are the nounphrase NP, the verbphrase VP and the prepositional phrase PP. There are three types of NP: the subject, the object and the NP within the PP. The object is internal to the VP, the subject external to it. In a simplified scheme,
(figure)
In addition, each nounphrase NP can be thought of as consisting of several specifiers, such as in the phrase sarve te manoharâ ashvâh "all those lovely horses'',
(figure)
Not all elements are necessarily present within each grammatical sentence. Some constituents may be empty, such as is often the case with the subject NP in Sanskrit if this subject NP consists of a personal pronoun. Personal pronouns are usually not expressed in subject position in this so-called PRO-drop language. The constituents, empty or not, are liable to movements and raisings, in order to yield the actual appearance. These movements may be understood as the evolutionary processes between the Deep Structure and the Surface Structure of a sentence or be thought of as processes between the various co-existing forms of one and the same sentence.
The most common example of a movement is the so-called wh-Movement. The whole relative clause is represented in Deep Structure as element of the NP to which it belongs, that is, as a Specifier to the Noun. These units can be moved as a whole: the wh-movement. No splitting up of constituents is allowed, only mutual interchangement of elements of one and the same subordinate clause is possible. wh- can be considered as a feature that appears in the surface form within a word but is abstractly associated with the NP of which this noun is the head (or the PP containing this NP). The characters wh come from the English relatives and interrogatives, both beginning with wh, as in what, whose, which.
In subordinate clauses, this element wh- (in English at least) moves to initial position. This rule can be called Front-wh (Chomsky 1986:69).
Similar movement in Sanskrit is not as clear as it is in English, still there must be a feature like Move-yad-, as Movement from the basic occupation of COMP to a more superficial place next to its coherent Noun often takes place. The yad-feature indeed often occupies the initial position of the subordinate clause.