The literal meaning of the term ‘lexicalization’ is ‘transfer into the lexicon’. This means that the expression undergoing lexicalization becomes a lexical item – representing a lexeme of, as we shall see, a form of a lexeme. It therefore presupposes that the expression in question is not in the lexicon at the beginning. It starts out as a complex expression formed by rules of the language system. In a first step, the combination becomes fixed and conventional. This may happen to compounds like those in .

At the moment of their formation, such compounds are completely transparent. The relationship between the significans and the significatum of the parts and their structure is completely regular. A construction displaying such a transparent relationship is compositional. As long as the construction is compositional, it makes no difference, from the systematic point of view, whether it is coined ad hoc by rules of the system, like a novel sentence, or has been stored as an item of the lexical stock. Speakers engaged in linguistic activity are free to recall it from their lexical memory, taking the holistic access, or to form it by the rules of their language system, taking the analytic access. It may be done one or the other way by different speakers or even by the same speaker on different occasions. The only symptom perceptible by the linguist describing the language system is that whenever the designatum in question comes up, speakers tend to use the same expression; i.e. conceivable and possible variation is reduced.

However, once the expression is a lexical item, it is no longer necessary for it to be compositional because speakers can take a holistic access to it. Then the relationship between significans and significatum may become irregular. Either the significans may undergo phonological change or the significatum may undergo semantic change or both may change. For instance, Mandarin fàn wǎn now designates a bowl no matter whether used for rice or four soup. On the other hand, it has undergone metonymic extension, being used for 'job' (Thomas Li p.c.). Such idiomatic semantic changes, presuppose the lexicalzation of the complex expression.

The following table contains some standard examples of lexicalization from Indo-European languages.

In the Romance example, only the significans is reduced, while the significatum remains intact. In the English examples, both significans and significatum change drastically.

Like grammaticalization, lexicalization is a reduction process. The difference consists in the access taken to the product: grammaticalization subjects the construction to rules, thus facilitating the analytic access to it; lexicalization obliterates the internal structure and compositionality of the association of significans and significatum, thus leaving the holistic access as the only possibility.

Two words which are adjacent in a syntactic construction can coalesce, both in grammaticalization and in lexicalization. The process has been called agglutination if one of the items involved is a grammatical operator on the other. It is called univerbation (occasionally and more recently, “chunking”) if this is not so (Lehmann 2020). Standard examples of univerbation include Spanish terremoto ‘earthquake’, univerbated from the Latin syntagma terrae motus (earth:GEN.SG movement:NOM.SG), and desde ‘since’, univerbated from Vulgar Latin de ex de (from out.of from).

Univerbation is to be distinguished from compounding. Some languages like Chinese, English and German have had productive compounding for millennia, and the schema can at any point serve as a model for new compounds. Compounds such as earthquake illustrate the point. The essential difference from univerbation is that the latter presupposes a syntactic construction at a diachronically earlier stage which is welded into one word. This is not the case with compounds. There is no syntactic construction earth quake which would be merged into one word in diachrony. Instead, it is formed by a rule of word-formation available at the relevant time. Thus, all instances of univerbation are ones of lexicalization. Compounds do not need to be lexicalized as long as their formation is compositional; but they are lexicalized once their formation is no longer compositional, as in the three examples of the above table.

Grammaticalization and lexicalization are both reductive processes, shifting their input down the hierarchy of grammatical levels. They differ in that lexicalization moves the output to the side of holistic access while grammaticalization moves it to the side of analytic access. This is visualized in the following diagram:

Lexicalization and grammaticalization in the significative system

Since lexicalization moves the input in the direction of the morphemicon, a corollary of this conception is that a combination like Span. desde is, at any rate, lexicalized. It may, in addition or at a later point in diachrony, be grammaticalized.

These examples illustrate, in their own way, the gradual transition between analytic and holistic access to a sign. Since there is no rule supporting the association of significans and significatum of a totally irregular form, this is of necessity stored in the inventory. The term ‘lexicalization’ is not commonly applied to such forms as went because they are not lexemes, but inflected forms. To keep the terminology consistent, we will have to assume that lexicalization may affect not only the direct representative of a lexeme, which is a stem, but also some of its inflected forms.