Macro- and micro-comparative linguistics

In his seminal paper from 1963 Joseph Greenberg formulated a large number of cross-linguistic generalizations, called ‘universals’, concerning syntax, morphology, and the interplay between them on the basis of comparison of 30 languages from various parts of the world. Although Greenberg’s method has many pitfalls, the research program is essentially sound: Formulate cross-linguistic generalizations concerning grammatical properties on the basis of comparison of a set of languages sampled carefully so as to avoid genetic and areal bias. Later research has validated many, but not all, of Greenberg’s original generalizations (see especially Dryer 1992, who tested a subset of Greenberg’s universals against a much larger and more carefully sampled set of languages).

Greenberg’s universals typically have the form ‘If a language has property P, it also has property Q’, in some cases unexceptionally, in other cases with a certain probability. These generalizations capture interdependencies in the system underlying natural language. The next, crucial, step is explaining these generalizations: Why do these interdependencies hold? What is the nature of the cognitive system of combining primitive elements into complex linguistic expressions (which we call grammar), which yields the Greenbergian universals as one of its consequences?

One of the stumbling blocks of the Greenbergian method is the lack of adequate, sufficiently detailed descriptions of most languages. A related weak point is that the comparison must be restricted to easily observable, salient properties of the sort which are likely to be mentioned even in fairly rough grammatical descriptions. Therefore Greenbergian macro-comparison needs to be supplemented by detailed studies of individual languages, and micro-comparative studies of small sets of languages. A particularly fruitful approach has been comparison of closely related languages in order to study a specific variable property P. As long as the languages in the set are alike in most or all other respects than P, most or all other variables are controlled, making it possible to isolate and study the effects of P. These studies deal with issues such as negation in Romance languages, possessive constructions in Scandinavian languages, syllable structure in Slavic languages, consonant gradation in Saami dialects, etc.

Another approach is to study a particular construction or phenomenon known to be found in many, or all languages, and then select a small set of languages from different corners of the world, and investigate how this phenomenon is instantiated in these languages. If the languages are selected so as to avoid genetic or areal relationships, this will be one way of isolating and studying those properties of human language that are not the result of historical accident, but are conditioned by the language faculty alone. This approach has not been pursued as much as the more standard macro- or micro-comparative approaches. We believe it should be, and that a Centre such as the one we have in mind is ideal for this form of research.

A notable property of natural languages which has emerged from macro- as well as micro-comparative work is that, although there are left- as well as right-oriented dependency relations in grammar (prefixes as well as suffixes, OV as well as VO order etc.), the distribution is asymmetric. To name but some asymmetries: Suffixes are more common than prefixes, subject-VP order is much more common than VP-subject, wh-preposing is common but wh-postposing is nonexistant, Aux-VP order often combines with OV order, but VP-Aux order never combines with VO order (in the same construction), and Comp-S order often combines with VP-Aux, but S-Comp never combines with Aux-VP. It has been claimed, in particular by Kayne 1994, that these asymmetries are a consequence of a fundamental asymmetry in grammar: All syntactic (including word-syntactic) structure is binary and underlyingly right-branching, all movement is leftwards, and all adjunction (including word-building) is to the left. These initially surprising hypotheses have turned out to shed new light on a number of universals. Much work remains to be done, though, to test these hypotheses against a wider range of languages and constructions.

It is well known that information structure plays an important part in syntax, that it obscures underlying grammatical relations, and that it may explain some of the asymmetries mentioned above. Yet, in spite of decades of research, the interplay of information structure and grammatical structure across languages remains rather poorly understood, perhaps mainly because the grammatical models assumed have been too coarse. However, by combining insights from semantic studies of topic and focus with insights from comparative studies of word order and prosody, and recent findings within syntactic theory, we count on being able to contribute to progress in this field of research.

Language change, language contact, language acquisition

It has become increasingly clear (through work by Matthew Dryer, among others) that areal features are more prevalent than we used to think. That is to say, language contact over the millennia has more to say for the distribution of linguistic features than envisaged by Greenberg, for instance. One implication of this is that Greenberg style classification of languages into major types (for instance SOV, SVO, or V-initial) each characterized by a set of correlating properties, assumed to be a consequence of the setting of a few major parameters, may be fundamentally mistaken. The high correlation of P and P' may, after all, be a consequence of historical accident rather than universal grammar, at least in certain cases. In those cases there is nothing to explain for formal grammar, nor do the exceptions to the correlation need any explanation.

To a certain extent these findings are consistent with recent findings within formal grammar, according to which, say, SOV order can be derived in several ways, so that there is no reason to expect all languages exhibiting surface SOV order to exhibit the same grammatical properties. Instead, since there is more than one way that this order can be derived, we expect languages with SOV order to fall into a number of subtypes, some of which may fall together with subtypes of languages with SVO order. That is to say, there are no ‘exceptions to SOV grammar’, or ‘languages in transition from SOV to SVO’ but a number of subtypes with properties predicted by formal linguistic theory.

The existence of wide spread areal features indicates the need for supplementing comparative linguistics with study of language change. Language change and linguistic variation are obviously intertwined; in fact, language change is the reason that there is linguistic variation. Therefore diachronic linguistics has a natural place within our research program. So does the study of language acquisition: As is well known (through the works of David Lightfoot among others) language change is crucially tied up with language acquisition: change mainly occurs between generations. We could mention that the study of language acquisition has established itself at our faculty in recent years: It is the topic of several master’s theses, and two dissertations are already underway.

An important question is to what extent a set of correlating surface properties in two languages L and L', say OV order, postpositions, and absence of prefixes, presupposes that L and L' have the same grammar, i.e. derive these structures in the same manner. OV order is apparently easily taken over from L by L' under language contact, and postpositions and various suffixes can be taken over as well. But at which point does the grammar of L' become the same as that of L? Answering this question requires combining insights from comparative linguistics, formal grammar, language acquisition, and studies of language contact.



Aim | Why? | How | Research Plan | Applications