where Latin is outwardly “formal.”  On the other hand, there are supposed to be languages[95] which have no true grasp of the fundamental relations but content themselves with the more or less minute expression of material ideas, sometimes with an exuberant display of “outer form,” leaving the pure relations to be merely inferred from the context.  I am strongly inclined to believe that this supposed “inner formlessness” of certain languages is an illusion.  It may well be that in these languages the relations are not expressed in as immaterial a way as in Chinese or even as in Latin,[96] or that the principle of order is subject to greater fluctuations than in Chinese, or that a tendency to complex derivations relieves the language of the necessity of expressing certain relations as explicitly as a more analytic language would have them expressed.[97] All this does not mean that the languages in question have not a true feeling for the fundamental relations.  We shall therefore not be able to use the notion of “inner formlessness,” except in the greatly modified sense that syntactic relations may be fused with notions of another order.  To this criterion of classification we shall have to return a little later.

[Footnote 95:  E.g., Malay, Polynesian.]

[Footnote 96:  Where, as we have seen, the syntactic relations are by no means free from an alloy of the concrete.]

[Footnote 97:  Very much as an English cod-liver oil dodges to some extent the task of explicitly defining the relations of the three nouns.  Contrast French huile de foie de morue “oil of liver of cod.”]

More justifiable would be a classification according to the formal processes[98] most typically developed in the language.  Those languages that always identify the word with the radical element would be set off as an “isolating” group against such as either affix modifying elements (affixing languages) or possess the power to change the significance of the radical element by internal changes (reduplication; vocalic and consonantal change; changes in quantity, stress, and pitch).  The latter type might be not inaptly termed “symbolic” languages.[99] The affixing languages would naturally subdivide themselves into such as are prevailingly prefixing, like Bantu or Tlingit, and such as are mainly or entirely suffixing, like Eskimo or Algonkin or Latin.  There are two serious difficulties with this fourfold classification (isolating, prefixing, suffixing, symbolic).  In the first place, most languages fall into more than one of these groups.  The Semitic languages, for instance, are prefixing, suffixing, and symbolic at one and the same time.  In the second place, the classification in its bare form is superficial.  It would throw together languages that differ utterly in spirit merely because of a certain external formal resemblance.  There is clearly a world of difference between a prefixing language like Cambodgian, which