The author raises the question of how to relate modern transformational grammars to a body of empirical fact, and suggests why paraphrasing might legitimately be considered a feature of behavior relevant to linguistic competence. This study is introduced by a discussion of the empirical basis of descriptive linguistics, followed by sections discussing empirical evidence for grammatical organization and linguistic performance and competence. Section III presents a description of the author's experimental study of the use of compound nouns, involving the free generation of paraphrases, forced-choice, and some replication. The final section discusses the perception of paraphrastic relations in the light of individual differences, grammaticalness and paraphrasability, and semigrammaticalness and error-type. General comments on linguistic method and innovation and creativity in language use conclude the study. Appended are a list of stimuli, tables indicating distribution of errors in the generating task and a new scoring technique, and a bibliography. (AMM)