The paradigmaticity of a linguistic sign is the extent to which it is a member of a closed and homogeneous paradigm. This condition entails that members of such a paradigm are in opposition in the same syntagmatic slot. Paradigmaticity increases with grammaticalization in a process called paradigmaticization.¹ A grammaticalized item may either generate, together with other words or constructions grammaticalized at the same time, a new paradigm in the language system, or else it may join a pre-existent paradigm and integrate itself formally and functionally into it.

The Chinese aspect markers reviewed in the section on erosion become formally and functionally alike when they are grammaticalized. To the set may be added guò ‘pass’, grammaticalized to a marker of an experiential perfect (Sun & Bisang 2020: 543). All of these are in opposition in a postverbal position.

Another compelling example of paradigmaticization is offered by the grammaticalization of the articles in Germanic and Romance languages (ch. 1.2.4). These stem from distinct sources. The indefinite article was a member of the set of numerals, which have a rather free distribution in Latin. The definite article was a demonstrative, thus, a member of a rather small and homogeneous paradigm of grammatical words. The following table shows the demonstrative pronouns in the middle, flanked by the personal pronouns and the demonstrative adverbs. It shows how tightly integrated this paradigm is.

The demonstrative which is grammaticalized to a definite article is ille ‘that’. Both this demonstrative and the numeral leave their original paradigmatic environments and start contrasting in the determiner position originally occupied by the demonstrative. The new definite and indefinite articles form a tightly-knit binary paradigm.

In Middle English, paradigmatic pressure converts the gender forms sē, sēo, þæt of the distal demonstrative into þe, þeo, þæt. Consequently, this paradigm becomes formally more homogeneous.

An example of the integration of a grammaticalized form into an existent paradigm is provided by the Romance synthetic future. The morphological exponents of this tense are verbal suffixes exactly like the other tenses of the Latin verb. Interestingly, this is, in the history of Romance, the last case of grammaticalization of a verbal tense/aspect/mood category which joins the synthetic conjugation. All further such constructions have the non-finite verb following the auxiliary, which does not lead to coalescence.

Increasing paradigmaticity implies shrinking of the size of the paradigm. For instance, Romance languages like Spanish and Portuguese have a set of at least nine aspect periphrases involving different auxiliaries (ch. 7.3.3), as opposed to only four synthetic tenses, coded by suffixes. Numeral classifier languages like Yucatec Maya, Mandarin, Cabecar and Persian can have dozens of classifiers which are bound or even free roots, while they have only a binary category of number in which an overt plural suffix contrasts with zero in the singular. As regards case relators, many languages including Latin, German, Hungarian and Turkish have a large or even open set of adpositions while they have a small and closed paradigm of case suffixes. Such examples could be multiplied.

The constraints acting on grammatical formatives are increasingly arbitrary, i.e., only conditioned by structural, not by semantic properties of the context. As a consequence, grammatical morphemes develop allomorphy to an extent that goes far beyond the allomorphy found in lexical items. An example is the allomorphy of the English articles, /thə/ ~ /thi/ and /ə/ ~ /ən/, which is phonologically conditioned, but has no counterpart in lexical stems. An extreme example is provided by inflection classes in those languages that have them. Given a word class W of lexical items which inflect for a morphological category C, given the set of values V that C assumes, and given the morphemes representing V with their allomorphs, then an inflection class is a subclass of W defined by the set of allomorphs that it takes in inflecting for C. For instance, English verbs fall into the conjugation classes of weak and strong verbs by the criterion of whether they form the past tense by the suffix -ed or in some other way; and the strong class again has subclasses according to the kind of morphological modification applied in the formation of the past tense and the perfect participle. Examples are in the section on analogical change. The assignment of a verb to one of these conjugation classes is completely arbitrary; in other words, even knowing the present tense form of a verb and its meaning and distribution, there is no way of inferring its conjugation class; but the conjugation itself is entirely subject to obligatory rules.

A side-effect of this aspect of grammaticalization is the fossilization of certain forms. It happens when a subclass constituted by some formal peculiarity is reduced to one member. Middle English declension comprised a set of allomorphs for the plural, including today's default allomorph /(ə)z/, but also the suffix /ən/ appearing in oxen and very few other contemporary nouns. There is, thus, a declension class defined by this plural allomorph. Within it, there is the form brethren, which is the plural of brother when it means ‘member of a religious group’. This not only has the minority plural suffix, but on top metaphony in the root vowel. It is the only surviving member of this subclass, thus, a fossil.

The reduction of a paradigm sometimes operates at two levels: A word showing some inflection may be grammaticalized and thus integrated into a paradigm, and at the same time its own inflection may be reduced. Clear examples of this are provided by the grammaticalization of modals and auxiliaries. Most of the English modals form a present and a past tense, even if it is totally irregular. Thus we have can - could, shall - should etc. However, must lacks a past tense, so it has lost the category of tense. The Latin verb habere when grammaticalized in the various ways summarized in the section on polygrammaticalization loses some of its conjugation categories. When it is grammaticalized to a future suffix, it loses its TAM inflection since only its present indicative subparadigm survives in this future inflection. When it is grammaticalized to an existential verboid such as the a exist in French il y a ‘there is’, it forfeits its person and number conjugation and only impersonal forms survive.

So in all such cases, a word undergoes paradigmaticization in becoming a member of a paradigm of periphrastic and finally affixal inflection, which paradigm may eventually be reduced to one member; and at the same time, its own inflection undergoes paradigmaticization, too, which implies its reduction and often its loss. There is, thus, some kind of lower-level paradigmaticization at work here; and one may ask whether it is a general feature of grammaticalization. This may not be so. Consider the example of pronouns. There are historically documented cases, especially from East Asia, in which a noun is grammaticalized to a pronoun; so these are in a grammaticalization relationship. Both word classes may decline for such morphological categories as number and case. Now if one investigates the distribution of these morphological categories over nouns and pronouns in a language, one finds many languages in which personal pronouns or clitic pronominal forms decline for such categories even if nouns don't. For instance, French nouns do not decline for case, but the clitic personal pronouns je, tu, il ‘I, you, he’ do. Cabecar nouns do not decline for case, either, and only optionally do they bear a number suffix. Nevertheless, the personal pronouns decline for number and, to some extent, for case, too, as some of them show an ergative form which nouns lack (they may be followed by an ergative postposition). There is even an implicational cross-linguistic generalization saying that if nouns decline for number and case, then personal pronouns decline for these categories. Thus, if there is a tendency for the paradigmaticization of a lexical item to be accompanied by this lower-level paradigmaticization of its own inflection, there must be some counter-acting principle accounting for the solid inflection of pronouns.

The effect of lower-level paradigmaticization has been called decategorization (or “decategorialization”, Traugott 2003: 644) and been treated as a mechanism or "phase" of grammaticalization (Heine & Kuteva 2002: 2, Heine & Narrog 2021, ch. 3). The term is a bit misleading, though, since the higher-level item – the auxiliarized verb in the above examples – becomes a member of a paradigm which represents a grammatical and, ultimately, a morphological category. These are familiar categories like aspect, tense, numeral classifier etc. By a well-established word-class theory (Lehmann 2013, § 6), such categories are minor categories whereas the category of their source is a major category. Apparently the prefix de- in the term refers to this transition from major to minor. However, it should be noted that these minor categories are much more rigid, coherent and well-delimited than the categories that the grammaticalized items come from. In this sense, paradigmaticization could even be regarded as increasing categorization. More importantly, if decategorization implies loss of inflection, it must be adapted to the above-mentioned counter-examples.

The shrinking of the paradigm size bears a causal relationship to desemanticization: Paradigmatic contrasts presuppose features by which the members differ. If the set of such features shrinks, possibilities for paradigmatic contrasts diminish, too, which reduces the number of members of the paradigm. For instance, the members of the article paradigm differ only by a feature [± definite], which allows for precisely two members in the paradigm. The members of the Latin demonstrative paradigm differ by the three values of the deictic category [1^st/2^nd/3^rd speech-act participant], which allows for three demonstratives.

¹ It appears that native speakers prefer paradigmaticization, linguistic authors who are not native speakers prefer paradigmatization.