In this thesis I investigate the degree of transparency of 22 languages. Transparency is defined as the extent to which a language maintains one-to-one relations between units at different levels of organisation, i.e. pragmatics, semantics, morphosyntax and phonology. Languages are compared not only as regards the amount of non-transparent phenomena in their grammars, so as to rank them from relatively transparent to relatively non-transparent, but also to uncover an implicational pattern in the distribution of non-transparent features across languages.
Transparency is discussed in connection to the notions of simplicity and learnability. Simplicity and transparency are shown to be different concepts that are both relevant in accounting for acquisition data. Whereas most definitions of simplicity take it to apply to particular domains of grammar, transparency is defined as a property of interfaces between levels within the grammar. The term is further operationalised using the framework of Functional Discourse Grammar.
It turns out that all languages have at least some non-transparent features, most notably of the redundancy type. Fusion and domain disintegration are less commonly attested, but the non-transparent features that are found only in the least transparent languages are so-called form-based forms: highly syntacticised forms and structures, that have no pragmatic or semantic motivation. Furthermore, non-transparent relations at the interfaces of the phonological and pragmatic levels are found in many languages, whereas it is less
common to violate transparency at morphosyntactic and semantic interfaces. The attested implicational hierarchy of transparency proves that transparency is a relevant typological notion.