Volume 50, Issue 2

Linguistic complexity is the result of the two motivations of explicitness and economy. Most approaches focus on the exlpicitness side of complexity (overt complexity) but there is also an explicitness-oriented side to complexity (hidden complexity). The aim of the paper is to introduce hidden complexity as the neglected side of complexity and to discuss the issues of trade-offs, global complexity and equal complexity from a more encompassing perspective that integrates overt and hidden complexity.

Complexity trade-offs are often considered as evidence for the hypothesis that all languages are equally complex; simplicity in one component of grammar is balanced by complexity in another. According to Shosted (2006), this "negative correlation hypothesis", as he calls it, was never validated using quantitative methods. The present paper recalls, in a first step, our previously found significant negative cross-linguistic correlations between syllable complexity and number of syllables per clause and per word, as well as an almost significant negative correlation between syllable complexity and number of morphological cases. All these correlations indicate complexity trade-offs between subsystems of language, as do the positive correlations found between syllable complexity, number of syllable types, and number of monosyllabic words. In a second step we argue against the view of such complexity trade-offs as proof of the equal complexity hypothesis. This hypothesis is hardly testable for several reasons: As long as it is impossible to quantify the overall complexity of a single language, it is also impossible to compare different languages with respect to that quantity. Secondly, it could – because of its character as a null hypothesis – never be corroborated for principal reasons.

This article examines the history of two examples from American Indian languages as an illustration of the ways in which the ambivalent and controversial nature of linguistic complexity has shaped the attitudes to "exotic" languages. In particular, I trace the life cycles of the Cherokee verbs for washing and the Eskimo words for snow from an unbiased origin in the works of John Pickering (1777-1846) and Franz Boas (1858-1942) to ideologically oriented uses as evidence for claims advanced within late 19th century anthropological racism and 20th century cultural relativism, respectively. The history of these examples illustrates the instrumental treatment of linguistic data in secondary references in linguistics and the social sciences as well as contradictory interpretations of complexity, as shown by the overemphasis on lexical overabundance and the disregard for morphological complexity of the languages in question. More generally, I show that the notion of complexity has been a key aspect of discussions concerning such issues as cross-linguistic differences in linguistic structure, evaluations of languages regarded as "exotic" or "primitive" as well as the assumed cognitive, cultural and social correlates of linguistic structure.

Two kinds of objection are made to the claim that some languages are simpler than others. Many linguists have asserted that, as a matter of empirical observation, all languages are roughly equal in complexity, but very little evidence has been cited. A more weighty objection is that the claim is meaningless because the complexity of different languages is incommensurable. It may not be numerically quantifiable; but a comparison with the evolution of legal systems shows that that does not make claims of differential complexity meaningless.

It is sometimes assumed in research on language complexity that if languages tended towards some optimal level of global complexity, so that all languages would be roughly equally complex, then local complexity trade-offs should be a general principle in language. Drawing evidence from computer simulations I show that in equally complex systems the proportion of trade-offs (significant negative correlations) is higher than in random systems but far from being a general principle in language. In addition, it may be impossible to determine whether a certain correlation-set comes from random systems or equally complex systems. Based on these results a correlational approach on a handful of typological variables cannot be used to validate, or even falsify, the assumption that all languages are equally complex and, therefore, complexity trade-offs should be kept separate from that assumption. The typological distribution of complexity, drawn from the World Atlas of Language Structures, is further shown to differ from both random systems and equally complex systems.

The physical, mathematical, and information sciences have developed a number of ways to measure complexity and complex systems in the social, biological, and physical domains. One way of measuring complex systems that might be useful to language scientists is the set of tools from the interdisciplinary field known as network science. A number of studies that have used the tools of network science to examine various aspects of language and language processing are summarized. It is acknowledged that much work must be done to use the tools of network science to address the debate about the (equal) complexity of languages. However, this work may prove useful to language scientists interested in the (equal) complexity of languages, as well as in other topics about language. Furthermore, the distinct structural characteristics observed in networks of several languages to date may also prove useful to network scientists as they try to understand how certain structural characteristics influence network dynamics in other domains. Language scientists are urged to embrace the techniques of network science to address the question of the complexity of languages.