Disentangling Chat with Local Coherence Models

‡e ev—lu—te sever—l popul—r models of lo™—l dis™ourse ™oheren™e for dom—in —nd t—sk genE er—lity ˜y —pplying them to ™h—t disent—ngleE mentF …sing experiments on syntheti™ multiE p—rty ™onvers—tionsD we show th—t most modE els tr—nsfer well from text to di—logueF goE heren™e models improve results over—ll when good p—rses —nd topi™ models —re —v—il—˜leD —nd on — ™onstr—ined t—sk for re—l ™h—t d—t—F yne property of — wellEwritten do™ument is ™oE heren™eD the w—y e—™h senten™e ¢ts into its ™ontext– senten™es should ˜e interpret—˜le in light of wh—t h—s ™ome ˜eforeD —nd in turn m—ke it possi˜le to interE pret wh—t ™omes —fterF wodels of ™oheren™e h—ve prim—rily ˜een used for textE˜—sed gener—tion t—sksX ordering units of text for multido™ument summ—riz—E tion or inserting new text into —n existing —rti™leF sn gener—lD the ™orpor— used ™onsist of inform—tive writingD —nd the t—sks used for ev—lu...

• Type: Journal
• Year: 2011
• Where: ACL
• Authors: Micha Elsner, Eugene Charniak
• Keywords: ACL 2011, Computational Linguistics, Diof, Nnels, Public Implementation