[…] since the rise in the fequency of a unit is nothing but a rise in the number of times that unit is used, any use of a linguistic unit contributes, even if only in an infinitesimal way, to an increase in its frequency. This automatically entails a lowering of its contribution to information and a tendency to reduce its cost, that is the output of energy it requires for its memorizing and production. Such a reduction is likely to determine changes in the phisical aspect of the unit. This could be summarized by saying that, strange as it may sound, a language changes because it is used. People as a rule are not conscious of any change taking place in their language.
- Martinet (1962), Pag. 135

But if the relationship between frequency and cost were of the same type as that between frequency and information, we should be able to state that, for a given frequency, a word should have 'n' phonemes, no more and no less, and that a word of 'n' phonemes should definitely belong to a given frequency range. This is, of course, not the case; 'conspicuous' with its four syllables and eleven phonemes is so much more frequent than 'dinosaur' with three syllables and seven phonemes.
- Martinet (1962), Pag. 144

The hypothesis that frequency could modify cost, i.e. exert an influence on the form of the units, phonemes as well as monemes, suggested that this was at least one of the channel through which communicative needs could act upon the economy of phonematic patterns.
- Martinet (1962), Pag. 159