Some linguists like to talk about disparagingly about prescriptive linguists, or prescriptive grammarians, as distinct from descriptive and generative linguists. When ideology is cast aside, one finds that the three complement each other, and the disparagement is misguided.
Generative linguists strive to create a system of rules from which a novice can potentially generate an infinite number of new sentences that a native speaker would recognize as “well-formed.” They often delight in taking evidence of use from casual registers of the language, but a complete survey of any language must begin with a descriptive stance and take into account all registers, from the most casual to the most formal.
Casual speech and formal speech both have their rules, which may be described descriptively or generatively, and which become prescriptive when taught. One may argue that a native speaker cannot make an error, but native speakers may misuse words that they haven’t heard much, and may make faux pas by following the rules of one register in a situation that requires another. A native speaker of a dialectical form of English, such as Cockney, usually needs to learn the standard dialect in order to get a job, and a native speaker comfortable only with casual registers of English cannot get a job that requires a more formal register. Instruction in standard dialects and formal registers is effective only insofar as it is prescriptive.
Except for certain written dialogues, written language generally follows the same rules as the formal language. When I work as an editor, I am therefore a prescriptivist, helping the writer to follow the rules of the written language with full respect for what is considered correct.
When I work as a lexicographer, I am primarily descriptive, but when I write usage notes, I am also prescriptive, because the dictionary user is best served by a dictionary that explains what is acceptable in the written and formal spoken language as well as what is common in casual speech. Many of the articles posted here are prescriptive, because writers and speakers need to know the rules of the formal language in order to be effective. When I teach language, I teach multiple registers, so I am at once descriptive and prescriptive, but teaching rules that allow students to generate more sentences than they have heard or read is also generative.
Similarly, when I work as a computational linguist, my work is primarily generative, but if I don’t take adequate account of the actual data, my rules won’t work; a theory that breaks when confronted by data is useless. Generative theory must pass muster descriptively as well, or it is mere sophistry.
In short, the distinction among the three approaches is largely exaggerated; any adequate account of the language must take all three into account.