Reflexives, Impersonals and Their Kin: a Classification Problem

Despite the fact that true reflexives always require a local antecedent, attempting an automatic referential resolution is often far from trivial: in many languages, reflexives are morphologically indistinguishable from impersonals and both particles are sensitive to the syntactic structure in a non-trivial sense. Focusing on Italian, we annotated part of the Repubblica Corpus to attempt an automatic classification of the reflexive and impersonal si constructions. In this preliminary study we show that the accuracy of the automatic classification methods that do not use any relevant structural information are rather modest. A thoughtful discussion of the structural analysis required to distinguish among different contexts is provided, in the end suggesting that these structural configurations are not easily recoverable using a purely distributional approach.