[Paper Review] Learning to Exploit Different Translation Resources for Cross Language Information Retrieval

Hosein Azarbonyad, Azadeh Shakery+1 more|arXiv (Cornell University)|May 20, 2014

Algorithms and Data Compression22 references3 citations

TL;DR

This paper proposes a Learning to Rank (LTR) approach to combine multiple translation resources for Cross-Language Information Retrieval (CLIR), using both translation relation-based and context-based features to rank query word translations. The method significantly improves retrieval performance in English-Persian CLIR compared to single-resource baselines, demonstrating the effectiveness of fusing diverse translation sources via supervised ranking.

ABSTRACT

One of the important factors that affects the performance of Cross Language Information Retrieval(CLIR)is the quality of translations being employed in CLIR. In order to improve the quality of translations, it is important to exploit available resources efficiently. Employing different translation resources with different characteristics has many challenges. In this paper, we propose a method for exploiting available translation resources simultaneously. This method employs Learning to Rank(LTR) for exploiting different translation resources. To apply LTR methods for query translation, we define different translation relation based features in addition to context based features. We use the contextual information contained in translation resources for extracting context based features.The proposed method uses LTR to construct a translation ranking model based on defined features. The constructed model is used for ranking translation candidates of query words. To evaluate the proposed method we do English-Persian CLIR, in which we employ the translation ranking model to find translations of English queries and employ the translations to retrieve Persian documents. Experimental results show that our approach significantly outperforms single resource based CLIR methods.

Motivation & Objective

• To address the challenge of integrating multiple translation resources with varying quality and characteristics in Cross-Language Information Retrieval (CLIR).
• To improve translation quality for query terms by leveraging diverse translation sources rather than relying on a single resource.
• To develop a unified framework that ranks translation candidates using both linguistic and contextual features.
• To evaluate the effectiveness of the proposed method in a real-world CLIR scenario, specifically English-to-Persian retrieval.

Proposed method

• The method employs Learning to Rank (LTR) to train a model that scores translation candidates based on a combination of translation relation-based features and context-based features.
• Translation relation-based features capture the reliability and consistency of translations across different resources.
• Context-based features utilize surrounding lexical and syntactic context to assess the appropriateness of a translation in a given query context.
• The model is trained using labeled data where relevant translations are identified, enabling the system to learn which features best predict high-quality translations.
• The final translation ranking model is applied to re-rank candidate translations of English queries before retrieving Persian documents.
• The approach is evaluated in an English-Persian CLIR setup, where translated queries are used to retrieve relevant Persian documents.

Experimental results

Research questions

• RQ1 Can combining multiple translation resources through a learning-to-rank framework improve translation quality in CLIR compared to using a single resource?
• RQ2 Which types of features—translation relation-based or context-based—are most effective in distinguishing high-quality translations?
• RQ3 How does the performance of the proposed multi-resource fusion method compare to baseline single-resource CLIR systems in a real retrieval task?
• RQ4 To what extent does contextual information enhance translation selection in low-resource or ambiguous query contexts?

Key findings

• The proposed LTR-based method significantly outperforms single-resource CLIR methods in English-Persian cross-lingual retrieval.
• The integration of both translation relation-based and context-based features leads to better translation ranking than using either feature type alone.
• The model achieves improved retrieval effectiveness by selecting more accurate and contextually appropriate translations for query terms.
• The experimental results demonstrate that fusing multiple translation resources via learning-to-rank yields measurable gains in retrieval performance metrics.