The extent to which a sentence is coherent in its local context affects how it is processed. A major source of local coherence is co-reference relations between an anaphoric phrase in the current sentence and its antecedent in a preceding sentence (Clifton & Dufy, 2001; Charolles, 2002). The appropriateness of particular forms of reference depends on the cognitive status of referents in the mental representations of both the producer and the listener/reader: it depends on recency of antecedents, as well as topichood, subjecthood, focus, goal status and parallelism (Arnold J. E., 1998). As for plural anaphora, the use of the connective and puts into discourse focus a Complex Reference Object (Eschenbach et al. 1989) to the detriment of its constituent atomic objects (Sanford & Lockhart, 1990). Koh & Clifton (2002) hypothesize a principle of equivalence between discourse entities, with respect to a property, as a factor predisposing their regrouping. Questions arise on whether the equivalent property is syntactic, discourse structural, semantic or pragmatic. Moxey et al. (2004) find that in case of split antecedents, the sharing of a thematic role predisposes the referents to form a CRO. Moxey et al. (2010) further suggest that the concept of equivalence can be defined in terms of the relationship between the roles that two characters can play in a discourse. This paper concerns the "mirror image" relationship introduced between the participants of reciprocal predicates, a property that could support the use of plural pronouns after split antecedents. A small class of verbs appears to involve natural reciprocal events such as “meet” or “fight (Kemmer, 1993) without specific reciprocal markers. These so called “lexical reciprocal verbs” have the property of symmetry (Borillo, 1971) and can be constructed discontinuously like the bi-transitive predicates. As reciprocal, they represent a situation from the point of view of the entire set of "mutuants" (Haspelmath, 2007), expressing the joint participation in a single coherent situation. We hypothesize this property predisposes the reciprocal lexical predicates to generate a CRO more often than distributive predicates in discontinuous constructions. To test this hypothesis, a 2 (reciprocal vs. distributive predicate) x 2 (female vs male grammatical subject) within subjects design was used in a sentence-continuation task with items such as "Brenda sympathise/grimpe dans le métro avec Samy en un dixième de seconde. (Brenda sympathizes/gets in/on the tube with Samy in a split second.)" where NPs subject and prepositional were counterbalanced. Native French-speaking students from a master class of Paris3-Sorbonne University participated. They were required to complete every one of 58 written sentences beginning with a pronoun given orally:“il”. As there is no oral difference in french between the singular (il/he) and plural (ils/they) third person pronoun, participants had the choice between singular and plural. Of the 58 sentences, 28 were the tested items, 10 were discontinuous constructions with only male characters (5 with reciprocal and 5 with distributive verbs), 10 were constructions with a conjoined subject (5 with reciprocal and 5 with distributive verbs) and 10 were fillers. Results confirm our hypothesis when the subject gender is female but not when it is male.