The Grammaticalization Cycle
Languages can be roughly classed into three types: isolating, where each meaning element makes up a distinct word (e.g. Vietnamese and Chinese); agglutinative, where a word is likely to contain several meaning elements but these are clearly separable (e.g. Turkish and Swahili); and fusional,where a word will contain several meaning elements some of them being fusedtogether, so that a single vowel may simultaneously mark, say, tense,voice, and person and number of subject (e.g. Latin and Sanskrit).
Proof of genetic relationship rests heavily on grammatical elements, including affixes. The ideal situation is to have a proto-language that is agglutinative [Proto-Indo-European had a fairly fusional structure (with ablaut and stress shift). This structure can be reconstructed because these fusional features have been retained in many of the descendants (and we are fortunate in having old records)] and modern languages that vary from agglutinative to fusional. If a group of modern languages are all basically isolating, with few grammatical affixes, if may never be possible to prove genetic connection with the same degree of confidence as for an agglutinative group.
As languages change over time, they tend -- very roughly -- to move around a typological circle: isolating to agglutinating, to fusional, back to isolating, and so on. If we place the isolating type at the four o'clock position, agglutinative at eight o'clock and fusional at twelve o'clock,around a clock-face, it is possible to describe recent movements in various language families. Proto-Indo-European was at about twelve o'clock but modern branches of the family have moved, at different rates, towards amore isolating position (some to one or two o'clock, others toward three o'clock). Early Chinese is thought to have been at about three o'clock, Classical Chinese was a fairly pure isolating type at four o'clock, while Modern Chinese languages are acquiring a mildly agglutinative structure, towards five o'clock. Proto-Dravidian was on the isolating side of agglutinative, at about seven o'clock, and modern Dravidian languages have moved around the cycle towards nine o'clock. Proto Finno-Ugric may have been at around nine o'clock, with modern languages moving to ten or eleven o'clock...For Egyptian, which has a long recorded history, Hodge (1970)* describes how it moved right around the cycle from fusional back again to fusional over a period of about 3,000 years.
My graduate student noted that stress and syncopation seem to be the motor driving the grammaticalization cycle. Inflectional/fusional languages that have lots of grammatical meanings coded into single phonemes are vulnerable to losing them through syncopation or dropping of unstressed syllables. So they add on prepositions or postpositions to make the meanings clearer. Once all the inflections have been syncopated out, your grammatical meanings are expressed only by these separate pre/post-positional words. That's the isolating position. But over time, unstressed pre/post-positions tend to get attached to the neighboring words. (A good example can be seen with the infinitive marker "to" in English which is now in spoken speech attached to preceding auxilliary verbs: "D'you wanna go?" "No, I hafta clean my room" -- or more briefly "D'you wanna?" "I hafta," etc. The additional -a is now marking a particular type of auxilliary verb function.) So agglutinative endings are mostly former pre/post-positions. But as more and more such separate unstressed particles are "glued" onto words, they tend to combine and form complex endings coding for more than one grammatical meaning and voila! you have a fusional/inflecting language and the cycle starts all over again.
*Refers to Carleton Hodge, 1970. " The Linguistic Cycle," Language Sciences 13: 1-7