Mind the Gap: Cross-Lingual Information Retrieval with Hierarchical Knowledge Enhancement

Cross-Lingual Information Retrieval (CLIR) aims to rank the documents written\nin a language different from the user's query. The intrinsic gap between\ndifferent languages is an essential challenge for CLIR. In this paper, we\nintroduce the multilingual knowledge graph (KG) to the CLIR task due to the\nsufficient information of entities in multiple languages. It is regarded as a\n"silver bullet" to simultaneously perform explicit alignment between queries\nand documents and also broaden the representations of queries. And we propose a\nmodel named CLIR with hierarchical knowledge enhancement (HIKE) for our task.\nThe proposed model encodes the textual information in queries, documents and\nthe KG with multilingual BERT, and incorporates the KG information in the\nquery-document matching process with a hierarchical information fusion\nmechanism. Particularly, HIKE first integrates the entities and their\nneighborhood in KG into query representations with a knowledge-level fusion,\nthen combines the knowledge from both source and target languages to further\nmitigate the linguistic gap with a language-level fusion. Finally, experimental\nresults demonstrate that HIKE achieves substantial improvements over\nstate-of-the-art competitors.\n