Desemanticization has been described as the loss of semantic features. This can be made a little more precise. Let us assume that the meaning of a sign can be described by a set of elementary propositions which are linked by relations of propositional logic. For instance, let

P1: x holds y
P2: x has y.

Then P1 logically implies P2. The reverse does not hold, as x can have y without necessarily holding it. This is because hold has an additional semantic feature, representable by a proposition such as

P3: x has a physical contact with y which x controls.

Now P3 is responsible for the irreversibility of the implication. It is the most specific semantic feature of hold. This feature can get lost in semantic change. If it is eliminated, P2 remains. This is what happens when Latin tenere ‘hold’ emerges as Port. ter ‘have’.

The general principle here is that given two propositions characterizing the meaning of a sign which are in an implicational relationship, the implicans can be eliminated from the meaning, the implicatum cannot. Thus, hold can lose P3. By contrast, as long as P3 is there, there is no way of getting rid of P2, because logical implication (or entailment) is compulsory.

In many cases of desemanticization, the lost feature specifies the designatum as a concrete physical object or process of a certain kind. This is not only true of Port. ter, but also of the English immediate future construction be going to V, where the feature of physical movement associated with the basic meaning of go is lost. All in all, the grammaticalization of the be-going-to construction – displayed from top to bottom in the diagram – involves the following chain of implications between sets of propositions:

“Desemanticization” and even “loss of semantic features” are relatively loose ways of speaking. They can be made precise in the way shown. What comes out is, however, a rather narrow concept of desemanticization. It implies that when an item is desemanticized, the result is its own hyperonym,¹ so desemanticization is the same as semantic extension. This may not be compatible with semantic changes observable in certain cases of grammaticalization. Consider the case of prepositions. The relational noun top has the basic meaning ‘uppermost part [of something]’. Besides its many metaphorical uses which are lexical, there is also the expansion to a spatial region bordering on the uppermost part, as when a book is lying on top of another one. In this usage, the complex preposition on top [of X] is grammaticalized, and its meaning covers abstract regions somehow “superior” to the position occupied by the reference point, as in this politician is on top of the issues. Now this latter usage is not simply describable as loss of a semantic feature, since it also implies a control relation which is not inherent in the meaning of (on) top but is rather an implicature of typical relational conditions holding if a human being has a superior – spatial or abstract – position with respect to a reference point. In this respect, the meaning of the preposition is not bleached, but enriched.

Such phenomena are frequent in lexical semantic change, but not rare in histories of grammaticalization, either. We then have two methodological possibilities:

1. We rely on the prior recognition that grammaticalization is often accompanied by other types of linguistic change. Given that we do not exclude from grammaticalization such changes in which it is accompanied by analogical change or reanalysis, we could admit that a process of semantic enrichment may accompany a process of grammaticalization.
2. We eliminate desemanticization from the set of parameters which operationalize the concept of grammaticalization. Since the parameters concerning paradigmatic and syntagmatic relations are more constitutive of the concept than those parameters which only concern the individual item, viz. erosion and desemanticization (cf. the section on parameters), we are already prepared for such a step.

Solution #a does not seem to be available. We have to allow distinct types of change to cooccur in historical cases. However, desemanticization and semantic enrichment are not just distinct, they are opposite. Allowing opposite kinds of change to cooccur in a particular change amounts to not requiring anything about these kinds of change. Which leads us to solution #b. We will here abide by according loss of weight a subordinate status in the set of crucial parameters and leave better insight to further empirical and theoretical research.

¹ English dictionaries and anglophone linguists use the form hypernym instead. This is ill-formed both morphologically, since the base is onym, not nym, and phonetically, since in most pronunciations, the word is indistinguishible from its converse, hyponym.