Issues in Cross-Language Retrieval from Document Image Collections

Over the past decade, broad-coverage crosslanguage text retrieval has progressed from isolated experiments on small collections to establish credible performance in large-scale evaluations. Extending this capability to document image collections presents some additional challenges that have not yet been well explored. This paper presents a general framework for cross-language retrieval, specializes that framework to retrieval from document image collections, and identifies opportunities for closer integration of the key enabling technologies and resources. 1 Introduction Information retrieval systems seek to help users obtain information objects from large collections [2]. Early systems typically relied on manually assigned indexing terms, and such "controlled vocabulary" techniques were widely used in libraries to support the retrieval of printed documents. As storage costs declined and processing power improved, "free text" searching became cost effective and was widely deployed. ..