An Investigation on Text-Based Cross-Language Picture Retrieval Effectiveness through the Analysis of User Queries

Purpose: This paper describes a study of the queries generated from a user experiment for cross-language information retrieval (CLIR) from a historic image archive. Italian speaking users generated 618 queries for a set of known-item search tasks. The queries generated by user’s interaction with the system have been analysed and the results used to suggest recommendations for the future development of cross-language retrieval systems for digital image libraries.

Methodology: A controlled lab-based user study was carried out using a prototype Italian-English image retrieval system. Participants were asked to carry out searches for 16 images provided to them, a known-item search task. User’s interactions with the system were recorded and queries were analysed manually quantitatively and qualitatively.

Findings: Results highlight the diversity in requests for similar visual content and the weaknesses of Machine Translation for query translation. Through the manual translation of queries we show the benefits of using high-quality translation resources. The results show the individual characteristics of user’s whilst performing known-item searches and the overlap obtained between query terms and structured image captions, highlighting the use of user’s search terms for objects within the foreground of an image.

Limitations and Implications: This research looks in-depth into one case of interaction and one image repository. Despite this limitation, the discussed results are likely to be valid across other languages and image repository.

Value: The growing quantity of digital visual material in digital libraries offers the potential to apply techniques from CLIR to provide cross-language information access services. However, to develop effective systems requires studying user’s search behaviours, particularly in digital image libraries. The value of this paper is in the provision of empirical evidence to support recommendations for effective cross-language image retrieval system design.