Theories about language, language structure, and translation are based mainly on observations of individual languages or on the comparison of several languages. In the present article the author argues that observations of translation practice constitute an essential source of information about the nature of language which is not readily accessible from these other sources. The very fact that we can and do translate from one language into another in itself shows essential agreement in the expression of meaning in all languages. The fact that grammatical relationships expressed in one language can be expressed in all languages is evidence that these universals are not a function of the individual languages but of human cognitive processes.