Is query translation a distinct task from search?

INTRODUCTION

The University of Sheffield participated in iCLEF 2002 using, as a test-bed, the prototype under development in the Clarity project. Clarity is an EU funded project aimed at developing a system for cross-language information retrieval for so-called low density languages, those with few translation resources. Currently translation between English and Finnish is supported; soon Swedish will be added and in the near future Latvian and Lithuanian.

Clarity is being developed in a user-centred way with user involvement from the beginning. The design of the first user interface was based on current best practise, particular attention was paid to empirical evidence for a specific design choice. Six paper-based interface mock-ups representing important points in the cross-language search task were generated and presented for user assessment as a part of an extensive user study. The study (reported in Petrelli et al. 2002) was conducted to understand users and uses of cross-language information retrieval systems. Many different techniques were applied: contextual enquiry, interviews, questionnaires, informal evaluation of existing cross-language technology, and participatory design sessions with the interface mock-ups mentioned above. As a result, a user class profile was sketched and a long list of user requirements was compiled. As a followup, a redesign session took place and the new system was designed for users whoknow the language(s) they are searching (polyglots); • search for writing (journalists, translators business analysts); • have limited searching skills; • know the topic in advance or will learn/read on it while searching; • use many languages in the same search session and often swap between them.

New system features were listed as important and the user interface was redesigned. Considering the result of the study the new interface allowed the user to dynamically change the language setting from query to query, hid the query translation and showed the retrieved set as ranked list primary.

Despite the fact that this new design was considered to be more effective, a comparison between the first layout based on the relevant literature and the new one based on the user study was considered an important research question. In particular, the choice of hiding the query translation was considered an important design decision, against the common agreement to allow and support the user in controlling the system actions. Thus the participation of Sheffield in iCLEF was organized around the idea of checking if the user should validate the query translation before the search is run or instead if the system should perform the translation and search in a single step without any user’s supervision.