SUBJECT: &NAME &NAME : &NAME &NAME : &NAME task versus Filtering / Routing task : No , You &NAME Just Take Your Pick &NAME RESEARCH LECTURE SPEAKER : &NAME &NAME &NUM HOST : &NAME &NAME DATE : Thursday 16th January &NUM TIME : &NUM : &NUM - &NUM : &NUM PLACE : &NAME Theatre , &NAME Research Ltd &NUM &CHAR &CHAR &NAME Avenue ( Off &NAME Road ) &NAME TITLE : &NAME task versus Filtering / Routing task : No , You &NAME Just Take Your Pick This talk focuses on &NUM aspect of document-query duality . In this talk , ' duality ' refers to the fact that given a data set , one might arbitrarily choose to perform either query-oriented estimation or document-oriented estimation . I explain that this duality is not a choice , but a problem , for almost any conceivable approach to parameter estimation in &NAME , and it is rarely adequately addressed . Early Language &NAME efforts made this problem worse , in some respects , but a consensus is emerging about how to solve the problem . I will present a probabilistic solution , whose principles can be adapted to a wide variety of &NAME models ( i.e. even non-probabilistic &NUM ) . Experiments using the proposed solution show that , at least in &NUM respect , queries are as informative as documents are , contrary to a widely held belief . There are also important implications for construction of &NAME data sets . You are currently subscribed to msrclectures as : &EMAIL To unsubscribe send a blank email to &EMAIL