In this talk, I will review studies which show, directly or
indirectly, how linguistic properties of queries and corpora can
impact the effectiveness of search over those corpora. The review is
motivated by our frustrating observation that linguistically-based
techniques that have been found to significantly improve QA over the
web can nevertheless fail to produce similar improvements in QA over
the AQUAINT corpus. The studies provide initial evidence that search
over specialised corpora can benefit from adjusting to common
linguistic features of the query and the corpus -- from lexical
features to discourse features.
Copyright. All Rights Reserved.