An agglutinative language is a form of synthetic language where each affix typically represents one unit of meaning (such as "diminutive," "past tense," "plural," etc.), and bound morphemes are expressed by affixes (and not by internal changes of the root of the word, or changes in stress or tone). Additionally, and most importantly, in an agglutinative language affixes do not become fused with others, and do not change form conditioned by others.
Synthetic languages that are not agglutinative are called fusional languages; they sometimes combine affixes by "squeezing" them together, often changing them drastically in the process, and joining several meanings in one affix (for example, in the Spanish word comí [I ate], the suffix -í carries the meanings of indicative mood, active voice, past tense, first person singular subject and perfect aspect).
Agglutinative is sometimes used as a synonym for synthetic, although it technically is not. When used in this way, the word embraces fusional languages and inflected languages in general. The distinction between an agglutinative and a fusional language is often not sharp. Rather, one should think of these as two ends of a continuum, with various languages falling more toward one end or the other. In fact, a synthetic language may present agglutinative features in its open lexicon but not in its case system: for example, German, Dutch.
Agglutinative languages tend to have a high rate of affixes/morphemes per word, and to be very regular. For example, Japanese has only two irregular verbs (and not very irregular), Luganda has only one (or two, depending on how 'irregular' is defined) and Turkish has only one. Georgian is an exception; not only is it highly agglutinative (there can be simultaneously up to 8 morphemes per word), but there are also significant number of irregular verbs, varying in degrees of irregularity.
Examples of agglutinative languages include Basque, Blackfoot, Georgian, the Altaic languages (see Turkish), Japanese (sometimes grouped with Altaic), Korean (sometimes grouped with Altaic), many Tibeto-Burman languages, the Dravidian languages-particularly Tamil, many Uralic languages (the largest are Hungarian, Finnish and Estonian), Inuktitut, the Bantu languages (see Luganda), Indonesian & Malay, the Northeast, Northwest and South Caucasian languages, and some Mesoamerican and native North American languages including Nahuatl, Huastec, and Salish. Quechua and Aymara are Native American language of South America. Much of the ancient Near East also spoke such languages, such as Sumerian, Elamite, Hurrian, Urartian, Hattic, Gutian, Lullubi, and Kassite.
Agglutination is a typological feature and does not imply a linguistic relation, but there are some families of agglutinative languages. For example, the Proto-Uralic language, the ancestor of Uralic languages, was agglutinative, and most descended languages inherit this feature. But since agglutination can arise in languages that previously had a non-agglutinative typology and it can be lost in languages that previously were agglutinative, agglutination as a typological trait cannot be used as evidence of genetic relationship to other agglutinative languages.
Many separate languages developed this property through convergent evolution. There seems to exist a preferred evolutionary direction from agglutinative synthetic languages to fusional synthetic languages, and then to non-synthetic languages, which in their turn evolve into isolating languages and from there again into agglutinative synthetic languages. However, this is just a trend, and in itself a combination of the trend observable in Grammaticalization theory and that of general linguistic attrition, especially word-final apocope and elision. This phenomenon is known as language drift.