Why is middle voice the way it is?

Abstract The middle voice defies definitive characterization due to its dual nature, straddling both oppositional (grammatical) and non-oppositional (lexical) properties. Data from eight Sinitic languages reveal an emerging middle marker derived from passives, whose distribution is restricted by four features: [+non-agentive], [+subject affected], [+change of state], and [+adverse]. Over time, this marker expands to additional predicate classes while becoming obligatory with some predicates, thus becoming more “grammatical” on the one hand and more “lexical” on the other. An “overlap-and-bleach” mechanism is at work, in which a grammatical element dependent on a semantically similar lexical element undergoes semantic bleaching due to its secondary discourse status and semantic redundancy. Partial semantic overlap preserves some of the marker’s oppositional meaning, while complete overlap results in complete bleaching and yields non-oppositional middles when combined with frequency effects. This diachronic process driven by the overlap-and-bleach mechanism may account for middle voice’s dual nature.