If one assumes that language does not exist separate from the speakers who use it then a major question arises for language change: how will a following generation know what changes are in progress in a current generation? The answer to this is that at any one time there co-exist two or more competing variants. Of these one is dominant and the other recessive.
Now one of the tenets of sociolinguistics is that children in the period of first language acquisition note not only what forms a language possesses but also what the variation among these forms is. For instance a child would note that both dived and dove are possible preterite forms of the verb dive. It would furthermore register other relevant aspects of the distribution, e.g. if one form is more common among older speakers, only used in more formal styles or conversely predominantly in colloquial usage, etc. By these means a child can observe the direction in which language change is moving and later contribute to this by unconsciously favouring forms which are inkeeping with his/her own linguistic preferences.
This conception of how language change is transmitted enables one to understand the notion of ‘drift’ better. If speakers at any point in time are aware of which variants are preferred and which are being increasingly neglected then the language can move in a definite direction as was the case with the drift from synthetic to analytic in the history of English.
The propagation of change would seem to follow a pattern which is found in other spheres apart from language. The pattern is termed an s-curve because of the approximate shape which it has. In essence an s-curve describes a change which starts slowly, picks up speed and proceeds rapidly but which stops — or at least slows down considerably — before it reaches completion.
One can think of an s-curve like a car which is started, accelerates to a given speed, travels the greatest distance and then slows down gradually when the driver takes his foot off the accelerator. Whether the car in this example will cross the finishing line depends on how much momentum it gains in the central phase.
A good linguistic example to illustrate an s-curve with is the case of shift from /ʊ/ to /ʌ/ in early modern English. This change would appear to have started in the mid 17th century and was active for at least 150 years after this in the south of England. Not all instances of /ʊ/ were shifted to /ʌ/. The small set of forms which did not undergo the shift from /ʊ/ to /ʌ/ are those which were phonetically resistant to the change, i.e. which were both preceded and followed by sound with inherent rounding. The elements which have such rounding are labial plosives, palato-alveolar fricatives and a velarised syllable-final [ɫ], hence pull, bull, push, bush, should still have /ʊ/ in standard English. These words occupy the gap between the top of the s-curve and the 100% mark on the vertical axis.
The stages of language change has been the subject of many sociolinguistic investigations and is given particularly detailed treatment in the following work (two volumes, the illustation is of the first).
Labov, William 1994. Principles of Linguistic Change. Vol. 1: Social Factors. Oxford: Blackwell.
There are also sections of the following volumes which are concerned with the progress of language change, especially the nature of S-curves.
Chambers, J. K., Peter Trudgill, Natalie Schilling-Estes (eds) 2002. The handbook of language variation and change. Oxford: Blackwell.
Hickey, Raymond (ed.) 2003. Motives for language change. Cambridge: University Press.