Here is the first: is it possible to identify the language with the code and the speech with the message? This identification is impossible according to Hjelmslev's theory. P. Guiraud refuses it for, he says, the conventions of the code are explicit, and those of the language implicit; but it is certainly acceptable in the Saussurean framework, and A. Martinet takes it up.
We encounter an analogous problem if we reflect on the relations between speech and syntagm. Speech, as we have seen, can be defined (outside the variations of intensity in the phonation) as a (varied) combination of (recurrent) signs; but at the level of the language itself, however, there already exist some fixed syntagms (Saussure cites a compound word like magnanimus). The threshold which separates the language from speech may therefore be precarious, since it is here constituted by 'a certain degree of combination'. This leads to the question of an analysis of those fixed syntagms whose nature is nevertheless linguistic (glottic) since they are treated as one by paradigmatic variation (Hjelmslev calls this analysis morpho-syntax). Saussure had noticed this phenomenon of transition: 'there is probably also a whole series of sentences which belong to the language, and which the individual no longer has to combine himself.' If these stereotypes belong to the language and no longer to speech, and if it proves true that numerous semiological systems use them to a great extent, then it is a real linguistics of the syntagm that we must expect, which will be used for all strongly stereotyped 'modes of writing'.
Finally, the third problem we shall indicate concerns the relations of the language with relevance (that is to say, with the signifying element proper in the unit). The language and relevance have sometimes been identified (by Trubetzkoy himself), thus thrusting outside the language all the non-relevant elements, that is, the combinative variants. Yet this identification raises a problem, for there are combinative variants (which therefore at first sight are a speech phenomenon) which are nevertheless imposed, that is to say, arbitrary : in French, it is required by the language that the I should be voiceless after a voiceless consonant (oncle) and voiced after a voiced consonant (ongle) without these facts leaving the realm of phonetics to belong to that of phonology. We see the theoretical consequences: must we admit that, contrary to Saussure's affirmation ('in the language there are only differences'), elements which are not differentiating can all the same belong to the language (to the institution)? Martinet thinks so; Frei attempts to extricate Saussure from the contradiction by localising the differences in subphonemes, so that, for instance, p could not be differentiating in itself, but only, in it, the consonantic, occlusive voiceless labial features, etc. We shall not here take sides on this question; from a semiological point of view, we shall only remember the necessity of accepting the existence of syntagms and variations which are not signifying and are yet 'glottic', that is, belonging to the language. This linguistics, hardly foreseen by Saussure, can assume a great importance wherever fixed syntagms (or stereotypes) are found in abundance, which is probably the case in mass-languages, and every time non-signifying variations form a second-order corpus of signifiers, which is the case in strongly connated languages : the rolled r is a mere combinative variant at the denotative level, but in the speech of the theatre, for instance, it signals a country accent and therefore is a part of a code, without which the message of 'ruralness' could not be either emitted or perceived.
