The structure of Romani
ER was unique among the NIA languages in shifting to a verb-medial typology. Constituent order in the verb phrase will have been flexible, but with the dominance of SVO and VSO patterns which we still find today in most Romani dialects of the Balkans and central Europe. This shift in word order will have involved the prepositioning of local relation expressions and the emergence of prepositions in the language. The order of elements in the noun phrase will have remained more or less intact, as it did not conflict with the patterns found in the contact language, Greek: adjectival modifiers (adjectives, numerals, demonstratives) were preposed. ER maintained however the prepositioning of the genitive attribute, an Indic legacy (compared with the postposed genitive in Greek). As in the other NIA languages, the genitive attribute continued to agree with its head. The most striking development in the noun phrase, once again making ER unique among the NIA languages, was the emergence of the preposed definite article, copying the Greek pattern, but drawing on remote (anaphoric) pronouns.
In syntax, the most notable development was the retreat of most non-finite verb forms and the reliance on finite forms and conjunctions for clause combining. It is impossible to tell what exactly the Proto-Romani legacy was, and whether converbs of the type still found today in NIA had developed in the ancestor language, or were lost before the ER period. ER certainly had at least two gerundial constructions, one in ‑indo, possibly reinforced by the Greek model, and one in ‑i. But most adverbial subordinations and other strategies of clause combining seemed to rely on conjunctions, taken most frequently from the inventory of interrogatives. This could have been an earlier development, triggered through contact with Iranian, much like the reduction of the modal infinitive and its replacement by subject agreement in a finite, subjunctive complement clause (lit. ‘I want that I go’). Perhaps the clearest piece of evidence in favour of syntactic convergence between ER and Greek in this domain is the emergence in ER of kaj, originally ‘where’, as a general subordinator and relativiser, and the emergence of a split between factual complementisers, for which kaj was used, and non-factual or subjunctive complementisers, for which the conditional particle te (originally probably a correlative particle) was employed (cf. Matras 2002: 179-185).