Combining Signals for Cross-Lingual Relevance Feedback

We present a new cross-lingual relevance feedback model that improves a machine-learned ranker for a language with few training resources, using feedback from a better ranker for a language that has more training resources. The model focuses on linguistically non-local queries, such as [world cup] and [copa mundial], that have similar user intent in different languages, thus allowing the low-resource ranker to get direct relevance feedback from the high-resource ranker. Our model extends prior work by combining both queryand document-level relevance signals using a machine-learned ranker. On an evaluation with web data sampled from a real-world search engine, the proposed cross-lingual feedback model outperforms two state-of-the-art models across two different low-resource languages.