As it was shown in §1, grammatical features are results of formal and meaningful oppositions of variants of units. But in actual speech we hear (or see) only one of the variants. It means that we have to discover Grammar so that we can describe it.
To do so we are to find those features that might give us lead to the underlying grammatical system. These features should be a generalised character that is they should be found with all units. Linguistic theory names two such features which represent two modes of Language, they are Syntagmatic properties of units. The syntagmatic properties reflect relationships of units to one another in strings of units of the same kind. The only question they can give answer to is "What follows and what precedes the selected unit?" Paradigmatic relationships reflect ability of a unit to substitute other units in the same position, and the only question they can give answer to is "What units can substitute the selected unit and what units the selected one can substitute?"
The syntagmatic relations depend on the speech, the paradigmatic relations depend upon the system of the language, but yet they interdependent as it what formulated by F. De Saussure, the paradigmatic relations are curtailed syntagmatic relations and the syntagmatic relations are curtailed paradigmatic relations.
We may start with collecting information about syntagmatic features, that is about the positions units may occupy in speech chains. This information then can be used to discover substitutional relations of units, paradigms of a language. This way of analysis is called Distributional and suggested as the main method of linguistic analysis by the American linguist L. Bloomfield in early 1930's and developed in detail by his followers. Though the distributional procedures were used as long as people analyse their languages, L. Bloomfield and his followers transformed it into one of the most powerful instruments of linguistic research.
We may start with collecting information about paradigmatic features of units, that is with their ability to substitute each other and establish as a result the system of language. Afterwards we have to collect information about the positions the variant of the units may occupy. This way of analysis was suggested as the main method of linguistic analysis by the Russian linguist A.I. Bauduin de Courtenay in late 1890's and developed in detail by his followers. Though these procedures were used as long as people analyse their languages, A.I. Bauduin de Courtney and his followers transformed it into even more powerful instruments of linguistic research. This method does not have any generally recognised mane. It is called often OPPOSITIONAL, or PARADIGMATIC, or CATEGORIAL. Because the method results in providing us with sets of categories of a language we shall refer to it as CATEGORIAL.
But there is one more constant feature of all units - they must be pronounced or written, that is they must be produced, created for speech. So we may base our grammatical analysis on the process of creation of linguistic units. This type of Grammar is known as Generative or Transformational. This way of analysis was developed as the main method of linguistic analysis by the American linguist N. Chomsky in late 1950's and developed in detail by his followers. And again, though the distributional procedures were used as long as people analyse their languages, N. Chomsky and his followers transformed it into one of the most powerful instruments of linguistic research.
These three types of analysis separately or in any combination are nowadays used to discover grammatical systems of languages. The methods of course have variants developed due to importance given to different additional features and extent to which other procedures are employed. But their purpose remains invariable - to produce as objective data of a language as possible for further interpretation.
Now we should consider what might be the aims a descriptive Grammar can be used for. We may describe Grammar to understand how it operates in a certain language at a certain period of its existence. As a result we receive a SYNCHRONIC GRAMMAR (distributional, categorial or transformational as a case may be). We may describe Grammar to understand how it changes with time. In this case we try to describe several stages of the same language and receive as a result a DIACHRONIC OR HISTORICAL GRAMMAR (distributional, categorial or transformational as a case may be).
If we want to compare languages then using the same criteria and analytic procedures we make descriptions of the Grammars of the languages we want to compare. Afterwards we may use the information either to register differences and similarities of the languages compared and receive CONFRONTATIONAL Grammar, or we may use it to understand the principles and possible rules of variability of grammatical systems and receive TYPOLOGICAL Grammar.
Thus we may conclude that descriptive Grammars can be:
A. due to analytic procedures:
I. DISTRIBUTIONAL, based on the analysis of positioning of units;
II. CATEGORIAL, based on the analysis of substitution of units;
III. TRANSFORMATIONAL, based on the analysis of procedures used to create units;
B. due to the purpose of description:
I. Synchronic;
II. Diachronic or Historical;
III. Confrontational;
IV. Typological.