Improving Low-Resource Cross-lingual Document Retrieval by Reranking with Deep Bilingual Representations

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3173–3179 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 3173 Improving Low-Resource Cross-lingual Document Retrieval by Reranking with Deep Bilingual Representations Rui Zhang Caitlin Westerfield Sungrok Shim Garrett Bingham Alexander Fabbri William Hu Neha Verma Dragomir Radev Department of Computer Science, Yale University {r.zhang, dragomir.radev}@yale.edu Abstract In this paper, we propose to boost low- resource cross-lingual document retrieval per- formance with deep bilingual query-document representations. We match queries and doc- uments in both source and target languages with four components, each of which is im- plemented as a term interaction-based deep neural network with cross-lingual word em- beddings as input. By including query like- lihood scores as extra features, our model ef- fectively learns to rerank the retrieved docu- ments by using a small number of relevance labels for low-resource language pairs. Due to the shared cross-lingual word embedding space, the model can also be directly applied to another language pair without any train- ing label. Experimental results on the MA- TERIAL dataset show that our model outper- forms the competitive translation-based base- lines on English-Swahili, English-Tagalog, and English-Somali cross-lingual information retrieval tasks. 1 Introduction Cross-lingual relevance ranking, or Cross-Lingual Information Retrieval (CLIR), is the task of rank- ing foreign documents against a user query (Hull and Grefenstette, 1996; Ballesteros and Croft, 1996; Oard and Hackett, 1997; Darwish and Oard, 2003). As multilingual documents are more acces- sible, CLIR is increasingly more important when- ever the relevant information is in other languages. Traditional CLIR systems consist of two com- ponents: machine translation and monolingual in- formation retrieval. Based on the translation di- rection, it can be further categorized into the doc- ument translation and the query translation ap- proaches (Nie, 2010). In both cases, we first solve the translation problem, and the task is trans- formed to the monolingual setting. However, while conceptually simple, the performance of this Document in Target Query in Target Document in Source Relevance Score Query in Source Source - Target Source - Source Target - Target Target - Source Figure 1: Cross-lingual Relevance Ranking with Bilin- gual Query and Document Representation. modular approach is fundamentally limited by the quality of machine translation. Recently, many deep neural IR models have shown promising results on monolingual data sets (Huang et al., 2013; Guo et al., 2016; Pang et al., 2016; Mitra et al., 2016, 2017; Xiong et al., 2017; Hui et al., 2017, 2018; McDonald et al., 2018). They learn a scoring function directly from the rel- evance label of query-document pairs. However, it is not clear how to use them when documents and queries are not in the same language. Furthermore, those deep neural networks need a large amount of training data. This is expensive to get for low- resource language pairs in our cross-lingual case. In this paper, we propose a cross-lingual deep relevance ranking architecture based on a bilingual view of queries and documents. As shown …

Content truncated to 500 words