The standard structural approach to word blends, such as noise1/sound2 → nound, has been to assume they involve a splicing together of the two words, where part of word2 is used to complete part of word1. The splice position has then been used as the source of mental lexicon generalizations. On the basis of 166 naturally occurring word blends, this article argues for a different approach—a “substitution” approach. The approach allows a comparison of the properties that word blends share with sublexical exchanges, anticipations, perseverations and substitutions; in addition, it accounts for the convergence of these properties. The substitution analysis allows a principled distinction between target and intruder; it predicts metrical structure output, and possible and impossible errors; moreover, the substitution analysis simplifies and constrains language production models.

Of 3,570 segmental malformations found in three corpora of English loanwords in French, h is the only segment that is never substituted. While interdentals are systematically adapted, h is systematically deleted. We attribute this to the fact that French — like Italian and Portuguese, where h is also deleted — does not phonologically employ the Pharyngeal node, the primitive essential in the adaptation of gutturals. The primitive cannot be dealt with (substituted or deleted), thus blocking phonetic interpretation of h. We encode this formally in the Non-availability Principle. Spanish, Greek and Russian, which employ the Pharyngeal node, replace h with the velar fricative x, a native sound, just as Fula replaces Arabic pharyngeals that it lacks with native laryngeals. The Non-availability Principle predicts that gutturals, whose sole articulatory specification is Pharyngeal (as is the case of Arabic pharyngeals and laryngeals) will be deleted in a language like French. This is confirmed. The Pharyngeal node, like tones and lexical accent, constitutes an unavailable primitive in French.