International Workshop on Computational Semantics International Workshop on Computational Semanticsà 5 . Embedding and Speaker Justiica - Tion

the y-core conditions in L j have to be satis-ed. Note that there is a subtle issue here. As deened, an embedding of L i tests the conditions of remote markers on which local markers depend. However, it is not forced to choose the same value for the remote marker as would a local embedding of the remote LDRS choose. This is so in order to allow for remote anaphoric references even in case the whole discourse contribution of the other participant fails to be embeddable. When attempting a simultaneous embedding of all LDRSs of D, an additional constraint on choice agreement is imposed (see below). It may be possible to impose this condition always and disallow anaphoric reference to antecedents that themselves fail to refer. Now we are in a position to give a straightforward deenition for the notions distinguishing MDRT from DRT: An MDRS D is S-justifying (for S in S) if the S-labelled LDRS is embeddable. D is harmonious ii it is S-justifying for every S in S in such a way, that whenever a dependency condition (x; y) occurs in LDRS L i and x is a marker in L j , e i and e j agree on the choice for y. Other discursive notions, such as connict (or argument), can now similarly be deened. 6. Conclusions In this paper we made some rst steps towards a multi-agent discourse representation theory. Our theory, MDRT, has beed designed as an extension of DRT. The main issues handled and consequences drawn are as follows: Cross-speaker anaphora is studied as a central feature of MDRT. The issue of what is a speaker committed to when using cross-speaker anaphoric reference was identiied as an important factor of the emerging theory. MDRT supports a non-degenerate solution, whereby a speaker commits only to some core conditions. A split DRS (MDRS) is proposed as a representation supporting the required commitment. A ner notion of embedding, inducing speaker justiication instead of global truth conditions, is proposed as the meaning of an MDRS. Acknowledgements We would like to thank Shalom Lappin for useful discussions of our proposed extensions to DRT. The initiation of this work was partially supported by a grant from the joint fund from the Technion-IIT and Haifa university. It was completed by a partial suuport of the rst author by a grant from the Israeli Ministry of Science and Technology "Programming …