Paradigmatic variability

Increasing grammaticalization of a construction or a category means that constraints on its use become tighter. The eligibility of the structural means decreases.

English has a rule of syntax which requires the use of the preposition to preceding an infinitive that depends on a lexical verb, but requires the bare infinitive if it depends on a modal verb or auxiliary ().

The preposition to is in opposition with other local prepositions if the prepositional phrase forms a local adverbial in the clause. There it has an allative function. In the context illustrated by , however, there is no other formative that might take the position occupied by to. It does not even contrast with zero, viz. not in the same context. This is, thus, an example of maximum obligatoriness of a grammatical formative.

A similar example is provided by the English definite article in certain contexts. In constructions like , i.e. preceding a nominal modified by a superlative adjective (and possibly additional modifiers), it is neither replaceable nor omissible.

According to Roman Jakobson (see below), obligatoriness is the central distinctive feature distinguishing grammatical from lexical items in a language.

With respect to a grammatical category represented by a paradigm of formatives like case or articles, constraints operate at two levels:

1. They may restrict the selection of a certain element from among a paradigm. I.e., there may be rules which enforce the use of a certain member of the paradigm as opposed to all the others. Such constraints reduce intraparadigmatic variability.
2. They may restrict representation of the paradigm by any of its members. I.e., there may be rules which require that the syntagmatic position reserved for the category not be left unoccupied. Such constraints reduce transparadigmatic variability.

The selection of the definite article in illustrates reduction of intraparadigmatic variability, since the indefinite article is excluded from this context. In a different context, either of the two articles must be used ().

Here the category of the article is subject to transparadigmatic variability. Another example of this kind is provided by tense in Latin. The finite verb of a clause must be specified for tense. The verb has a morphological slot following its stem and preceding the person ending which must be occupied by a tense morpheme. The morphological details are complicated and of no relevance here. What matters is that while selection of a particular tense is free in many contexts, there is no tenseless finite clause.

A similar analysis may apply to Latin case. Intraparadigmatic variability is constrained in that in many contexts, the case of a noun is determined by rule of grammar. For instance, in the constructiojn [de [N]_NP ]_PrepP ‘down from N’, N must be in the ablative.¹ Transparadigmatic variability is constrained in that every noun in a clause must be specified for case. It is true that, at the morphological level, there are some nouns whose bare stem serves as the nominative; so such forms may appear to be caseless. At the syntactic level, however, every occurrence of a noun is in a certain case, including the nominative. Thus, case as a category is obligatory in this language.

A final example concerns numeral classifiers as they exist in languages like Mandarin, Yucatec Maya and Cabecar. Their transparadigmatic variability is constrained because if a numeral combines with a noun, the numeral obligatorily combines with a classifier; so the entire category is obligatory in this context. Their intraparadigmatic variability is constrained to the extent that choice of a classifier in combination with a given noun is not free, but determined by the class of the noun. In several languages this intraparadigmatic variability is greater than the transparadigmatic variability since choice of the classifier in combination with a given noun is constrained, but not reduced to one which would be obligatory.

The distinction between intraparadigmatic and transparadigmatic variability collapses if a paradigm consists of a single item which is obligatory in certain contexts, which is precisely the case for English to.

What is functionally the same grammatical category may be obligatory to different degrees in different languages. Thus, many languages have number marking on nouns. However, in some languages it is found only on a subset of nouns and may be optional even there, while in other languages it may be obligatory for pretty much any noun. The former is true of Mandarin -men plural. This is a) restricted to human nouns and pronouns, but excluded in some nominal constructions, including importantly numerative phrases. It is b) optional with those former. It is, in this respect, less grammaticalized than the English plural suffix -s, which is obligatory whenever reference is made to a plural object.

Expectability and desemanticization

To the extent that something is obligatory, it carries no information. The distribution of English to in front of the infinitive illustrated by is a simple example. This distribution is due to an obligatory rule of grammar. As a consequence, the occurrence of to in front of the infinitive in constructions like provides no information to the hearer. He is not prompted to wonder what the speaker may have intended by using this item in this place, because he could not have said otherwise. The same is seen in the grammaticalization of the article system or in the grammaticalization of gender in certain Indo-European languages.

There is a law of information theory by which the amount of information provided by an element of a message is inversely related to its probability (Lehmann 1978, § 1.3). Freedom of choice on the part of the sender reduces the expectability of what he is going to say for the hearer. The less expectable a certain item is at a given position in a message, the more information it gives to the recipient. Increasing obligatoriness of a linguistic item promotes its expansion; its expansion leads to higher probability and expectability; and this correlates with loss of semantic substance. If the probability of an item is 1, it gives no information.

While totally obligatory grammatical categories like gender in German or a Romance language or the use of an appropriate preposition in front of an infinitive in a Germanic and Romance language contribute no information to the content of the message, they do provide metalinguistic information. Namely, gender gives information on the nouns thus categorized, the preposition preceding an infinitive gives information on the category of the governing verb, and the article preceding a nominal indicates that what follows is a noun phrase and, thus, potentially referential.

Metalinguistic information increases the redundancy, thus supporting both speech production and speech perception. Language systems provide for a certain amount of redundancy in messages by grammaticalizing some categories to the point of total obligatoriness. These are then semantically empty. Given this, it is of secondary importance which grammatical categories provide this redundancy. Thus, one language, say Latin or German, may develop a paradigm of tenses and make the specification of tense in clauses obligatory, while another language, say Mandarin or Korean, lacks this category but instead grammaticalizes a totally different category, say numeral classification. At this general and abstract level, the tense system of one language is comparable to the numeral classifier system of another language and to the article system of a third language.

This has consequences for the typology:

Thus the true difference between languages is not in what may or may not be expressed but in what must or must not be conveyed by the speakers. (Jakobson 1959:142)

In other words: Whether or not a language has a system of numeral classification or of tense or of articles makes a fundamental difference for its type, but may make very little difference for the messages that may be composed in the language.

This aspect of grammaticalization theory has consequences for the assessment of the role of grammatical categories in messages. Numeral classifiers are, again, an example. They are often thought to indicate the category of the counted noun. However, to the extent they are grammaticalized, the speaker has no choice but to use a particular classifier in front of a noun, whether or not its etymology fits with the meaning of the noun. Consequently in such cases, the (original or basic) meaning of the classifier makes no contribution to the content of the message; and linguists have no reason to speculate on “folk taxonomies” of speech communities.

The most debated example of this area in the past half century is gender in modern Indo-European languages. To the extent that it is grammaticalized in a language, it is devoid of meaning. The supposition that gender codes sex amounts to a re-semanticization of a category that was already desemanticized. If this becomes entrenched in the language system, it will be a real case of degrammaticalization – in this case, one enforced intentionally by members of the speech community.

¹ A variable marked by double underline stands for what is designated by the same variable lacking this mark.