A corpus-based investigation of definite description use

摘要: We present the results of a study use definite descriptions in written texts aimed at assessing feasibility annotating corpora with information about description interpretation. ran two experiments, which subjects were asked to classify uses corpus 33 newspaper articles, containing total 1,412 descriptions. measured agreement among annotators classes assigned descriptions, as well antecedent those definites that classified being related an text. The most interesting result this from annotation perspective was rather low (K = 0.63) we obtained using versions Hawkins's and Prince's classification schemes; better 0.76) simplified scheme proposed by Fraurud includes only classes, first-mention subsequent-mention. antecedents also not complete. These findings raise questions concerning starategy evaluating systems for interpretation comparing their standardized annotation. From linguistic point view, observations great number discourse-new our (in one 50% collection discourse-new, 30% anaphoric, 18% associative/bridging) presence did seem require complete disambiguation.