The purpose of this paper is to critically review so-called linguistic underdeterminacy in pragmatics. There has been a shared view that we don't usually say what we mean and we frequently mean much more than what we say. This view, however, has often overridden the radical contribution of morphosyntactic constructions to the import of utterances. In this paper it is argued that the role of linguistic manifestations (what-is-said) should not be underestimated as we consider how meanings are created in the communicator's mind. It is shown, with some typological examples, that even at the linguistically context-dependent layer inference is evidently at work in a systematic way in verbal interactions to make meanings. This type of meaning (what-is-entailed), for which the present author coined a new term in Japanese, is considered to lie between sentence-meaning (what-is-said) and speaker-meaning (what-is-implicated). Some pedagogical implications of this critical review of linguistic underdeterminacy are also discussed for foreign language teaching in relation to current views of second language development and the need of the reform in developing linguistic-pragmatics-oriented curricula for teacher education and in-service teacher training programs.