作者: Thorsten Ohrstrom-Sandgren , Kenneth K. McKeever , Janyce Wiebe , Thomas P. O'Hara
DOI:
关键词:
摘要: Scheduling dialogs, during which people negotiate the times of appointments, are common in everyday life. This paper reports results an in-depth empirical investigation resolving explicit temporal references scheduling dialogs. There four phases this work: data annotation and evaluation, model development, system implementation evaluation analysis. The were developed primarily on one set data, then applied later to a much more complex set, assess generalizability for task being performed. Many different types methods pinpoint strengths weaknesses approach. Detailed instructions intercoder reliability study was performed, showing that naive annotators can reliably perform targeted annotations. A fully automatic has been evaluated unseen test with good both sets. We adopt pure realization recency-based focus identify precisely when it is not adequate addressed. In addition results, itself presented, based detailed manual few errors occur specifically due used, anaphoric relations defined low ambiguity