Cross-lingual Retrieval Research Articles

Chinese-Vietnamese cross-lingual event retrieval method based on knowledge distillation

Cross-lingual event retrieval is an information retrieval task aimed at cross-lingual event retrieval among multiple languages to find text or documents related to a specific event. Specific to Chinese-Vietnamese cross-language event retrieval, it involves using Chinese as a query to retrieve Vietnamese documents related to the query event. The critical issue is how to efficiently align query and document representations with limited resources. Existing cross-language pre-training models are trained on large-scale multilingual corpora, but their training goals do not include explicit language alignment tasks. Due to the uneven distribution of training corpora between different languages, these models have The problem of language bias. Therefore, this linguistic bias is also inherited in cross-lingual retrieval based on these models. To solve this problem, this paper proposes a Chinese-Vietnamese cross-lingual event retrieval method based on knowledge distillation. This approach enables the model to learn good query-document matching features from monolingual retrieval by transferring knowledge from high-resource to low-resource languages. By enhancing the alignment between queries and documents in different languages in a shared semantic space, the method improves the performance of Chinese-Vietnamese cross-lingual event retrieval.

CL2CM: Improving Cross-Lingual Cross-Modal Retrieval via Cross-Lingual Knowledge Transfer

Cross-lingual cross-modal retrieval has garnered increasing attention recently, which aims to achieve the alignment between vision and target language (V-T) without using any annotated V-T data pairs. Current methods employ machine translation (MT) to construct pseudo-parallel data pairs, which are then used to learn a multi-lingual and multi-modal embedding space that aligns visual and target-language representations. However, the large heterogeneous gap between vision and text, along with the noise present in target language translations, poses significant challenges in effectively aligning their representations. To address these challenges, we propose a general framework, Cross-Lingual to Cross-Modal (CL2CM), which improves the alignment between vision and target language using cross-lingual transfer. This approach allows us to fully leverage the merits of multi-lingual pre-trained models (e.g., mBERT) and the benefits of the same modality structure, i.e., smaller gap, to provide reliable and comprehensive semantic correspondence (knowledge) for the cross-modal network. We evaluate our proposed approach on two multilingual image-text datasets, Multi30K and MSCOCO, and one video-text dataset, VATEX. The results clearly demonstrate the effectiveness of our proposed method and its high potential for large-scale retrieval.

Dual-view Curricular Optimal Transport for Cross-lingual Cross-modal Retrieval.

Current research on cross-modal retrieval is mostly English-oriented, as the availability of a large number of English-oriented human-labeled vision-language corpora. In order to break the limit of non-English labeled data, cross-lingual cross-modal retrieval (CCR) has attracted increasing attention. Most CCR methods construct pseudo-parallel vision-language corpora via Machine Translation (MT) to achieve cross-lingual transfer. However, the translated sentences from MT are generally imperfect in describing the corresponding visual contents. Improperly assuming the pseudo-parallel data are correctly correlated will make the networks overfit to the noisy correspondence. Therefore, we propose Dual-view Curricular Optimal Transport (DCOT) to learn with noisy correspondence in CCR. In particular, we quantify the confidence of the sample pair correlation with optimal transport theory from both the cross-lingual and cross-modal views, and design dual-view curriculum learning to dynamically model the transportation costs according to the learning stage of the two views. Extensive experiments are conducted on two multilingual image-text datasets and one video-text dataset, and the results demonstrate the effectiveness and robustness of the proposed method. Besides, our proposed method also shows a good expansibility to cross-lingual image-text baselines and a decent generalization on out-of-domain data.

Kannada to English Agricultural Cross-Lingual Retrieval: Enhancing Knowledge Access in Farming Practices

Abstract: This paper presents the development and implementation of a specialized cross-lingual information retrieval system tailored for agriculture-related queries in the Kannada language. The primary objective of the system is to facilitate accurate translation of Kannada queries into English, the target language, and to retrieve relevant documents containing vital agricultural information. The proposed system addresses key challenges including incorporating effective query preprocessing techniques, designing an efficient document retrieval mechanism, and establishing optimal data indexing strategies as well as strategies have been introduced to mitigate the challenges posed by the diversity of Kannada dialects. By overcoming the language barrier, this system enables seamless and effective knowledge dissemination in the agriculture domain.

On bridging consumer health search across languages using cross-lingual word space

Information search is a lucrative activity for electronic commerce, contributing to significant revenues. As search engine becomes the primary source for seeking health information, the search quality concerns consumers. Specifically, the health vocabulary used by general consumers differs from the professionals, leading to unsatisfied retrieval performance and misleading information. The situation is increasingly complex when consumers search health information across languages. To study consumer health search (CHS) across languages, we propose a cross-lingual retrieval framework comprising semantic space construction and information retrieval modules. The semantic space construction module adopts a weakly-supervised approach to determine a cross-lingual word space (CLWS) from collected consumer-generated health content, which helps capture consumers’ vocabulary and identify medical expression translations. The information retrieval module suggests strategies that utilize the translations of CLWS for the subsequent retrieval tasks. By evaluating the performance on two health information search engines, we found that Google Translate (GT), a widely-adopted translation service, is prone to generate mistranslations to the colloquial medical expressions. The CLWS helps identify GT’s mistranslations and filter out the results retrieved by those. More importantly, our framework demonstrates a strategy to integrate translations of CLWS and GT to reach the best retrieval performance for the cross-lingual CHS task.

On cross-lingual retrieval with multilingual text encoders

Pretrained multilingual text encoders based on neural transformer architectures, such as multilingual BERT (mBERT) and XLM, have recently become a default paradigm for cross-lingual transfer of natural language processing models, rendering cross-lingual word embedding spaces (CLWEs) effectively obsolete. In this work we present a systematic empirical study focused on the suitability of the state-of-the-art multilingual encoders for cross-lingual document and sentence retrieval tasks across a number of diverse language pairs. We first treat these models as multilingual text encoders and benchmark their performance in unsupervised ad-hoc sentence- and document-level CLIR. In contrast to supervised language understanding, our results indicate that for unsupervised document-level CLIR—a setup with no relevance judgments for IR-specific fine-tuning—pretrained multilingual encoders on average fail to significantly outperform earlier models based on CLWEs. For sentence-level retrieval, we do obtain state-of-the-art performance: the peak scores, however, are met by multilingual encoders that have been further specialized, in a supervised fashion, for sentence understanding tasks, rather than using their vanilla ‘off-the-shelf’ variants. Following these results, we introduce localized relevance matching for document-level CLIR, where we independently score a query against document sections. In the second part, we evaluate multilingual encoders fine-tuned in a supervised fashion (i.e., we learn to rank) on English relevance data in a series of zero-shot language and domain transfer CLIR experiments. Our results show that, despite the supervision, and due to the domain and language shift, supervised re-ranking rarely improves the performance of multilingual transformers as unsupervised base rankers. Finally, only with in-domain contrastive fine-tuning (i.e., same domain, only language transfer), we manage to improve the ranking quality. We uncover substantial empirical differences between cross-lingual retrieval results and results of (zero-shot) cross-lingual transfer for monolingual retrieval in target languages, which point to “monolingual overfitting” of retrieval models trained on monolingual (English) data, even if they are based on multilingual transformers.

Cross-lingual embedding for cross-lingual question retrieval in low-resource community question answering

In today’s digital world people are keen on finding the knowledge they need by surfing the internet to find the answers to their questions. To this aim, many Community Question Answering (CQA) systems are established, in which people can ask their question and receive the required information. The gathered data in such systems is a rich repository for people to search through the available questions that have been answered before. CQA users, however, are not always successful in finding their answers in their native CQA systems. One solution to enrich the searching process is translating input questions and searching them in other CQA systems. This solution is useless as the process of translating each question is time-consuming. To make the non-English CQA systems richer in finding the available answers, the systems can develop a model to find similar English questions. To help Persian CQA systems in providing the answers to the questions, we propose a cross-lingual question retrieval model to retrieve relevant English questions to any input Persian question. In the proposed model, we benefit from a translation model-based retrieval using neural cross-lingual word embedding. Our experiment shows that the proposed model achieves 71.4% MRR and 83.5% success@5 using supervised cross-lingual word embedding.

Cross-lingual text similarity exploiting neural machine translation models

This article studies cross-lingual text similarity using neural machine translation models. A straightforward approach based on machine translation is to use translated text so as to make the problem monolingual. Another possible approach is to use intermediate states of machine translation models as recently proposed in the related work, which could avoid propagation of translation errors. We aim at improving both approaches independently and then combine the two types of information, that is, translations and intermediate states, in a learning-to-rank framework to compute cross-lingual text similarity. To evaluate the effectiveness and generalisability of our approach, we conduct empirical experiments on English–Japanese and English–Hindi translation corpora for a cross-lingual sentence retrieval task. It is demonstrated that our approach using translations and intermediate states outperforms other neural network–based approaches and is even comparable with a strong baseline based on a state-of-the-art machine translation system.

Cross-lingual multi-keyword rank search with semantic extension over encrypted data

The emergence of searchable encryption technology sheds new light on the problem of secure search over encrypted data, which has been a research hotspot in the past few years. However, most existing search schemes based on searchable encryption only support queries in a certain language. The few searchable encryption schemes that have implemented multilingual search fail to achieve automated cross-lingual retrieval, which significantly impacts users search experience. Additionally, most of these schemes only support the exact matching and can-not provide semantic search. Hence, the implementation of cross-lingual and semantic search together remains an open topic for searchable encryption. To the best of our knowledge, no previous research has investigated the problem of cross-lingual ranked search over encrypted cloud data. To address this issue, we propose a cross-lingual multi-keyword rank search (CLRSE) scheme upon the Open Multilingual Wordnet. Our CLRSE scheme can break the barrier of languages, and realizes intelligent and personalized search through flexible keyword and language preference settings. Additionally, an improved scheme is developed to speed up the sorting process. We evaluate the performance of our scheme including security, functionality, precision, and efficiency, with extensive experiments.

COCO-CN for Cross-Lingual Image Tagging, Captioning, and Retrieval

This paper contributes to cross-lingual image annotation and retrieval in terms of data and baseline methods. We propose COCO-CN, a novel dataset enriching MS-COCO with manually written Chinese sentences and tags. For more effective annotation acquisition, we develop a recommendation-assisted collective annotation system, automatically providing an annotator with several tags and sentences deemed to be relevant with respect to the pictorial content. Having 20,342 images annotated with 27,218 Chinese sentences and 70,993 tags, COCO-CN is currently the largest Chinese-English dataset that provides a unified and challenging platform for cross-lingual image tagging, captioning and retrieval. We develop conceptually simple yet effective methods per task for learning from cross-lingual resources. Extensive experiments on the three tasks justify the viability of the proposed dataset and methods. Data and code are publicly available at https://github.com/li-xirong/coco-cn

Building a multi-domain comparable corpus using a learning to rank method

AbstractComparable corpora are key translation resources for both languages and domains with limited linguistic resources. The existing approaches for building comparable corpora are mostly based on ranking candidate documents in the target language for each source document using a cross-lingual retrieval model. These approaches also exploit other evidence of document similarity, such as proper names and publication dates, to build more reliable alignments. However, the importance of each evidence in the scores of candidate target documents is determined heuristically. In this paper, we employ a learning to rank method for ranking candidate target documents with respect to each source document. The ranking model is constructed by defining each evidence for similarity of bilingual documents as a feature whose weight is learned automatically. Learning feature weights can significantly improve the quality of alignments, because the reliability of features depends on the characteristics of both source and target languages of a comparable corpus. We also propose a method to generate appropriate training data for the task of building comparable corpora. We employed the proposed learning-based approach to build a multi-domain English–Persian comparable corpus which covers twelve different domains obtained from Open Directory Project. Experimental results show that the created alignments have high degrees of comparability. Comparison with existing approaches for building comparable corpora shows that our learning-based approach improves both quality and coverage of alignments.

Multilingual information retrieval in the language modeling framework

Multilingual information retrieval (MLIR) provides results that are more comprehensive than those of mono- and cross-lingual retrieval. Methods for MLIR are categorized as: (1) Fusion-based methods that merge results from multiple retrieval runs, and (2) Direct methods that build a unique index for the entire collection. Merging results of individual runs reduces the overall effectiveness, while more effective direct methods suffer from either time complexity and memory overhead, or over-weighting of index terms. In this paper, we propose a direct MLIR approach by using the language modeling framework that includes a novel multilingual language model estimation for documents, and a new way to globally estimate word statistics. These contributions enable ranking documents in multiple languages in one retrieval phase without having the problems of the previous direct methods. Moreover, our approach has the advantage of accommodating multilingual feedback information which helps to prevent query drift, and consequently to improve the performance. Finally, we effectively address the common case of incomplete coverage of translation resources in our proposed estimation methods. Experimental results show that the proposed approach outperforms the previous MLIR approaches.

Building Structured Query in Target Language for Vietnamese – English Cross Language Information Retrieval Systems

Abstract — Query translation is the most important component in Cross Language Information Retrieval systems using dictionary-based approach. In this paper, we present a method to build structured query in target language from a given query in source language. The method is based on constructing bi-lingual dictionaries, keyword extraction from source query, getting translation candidates for each keyword using mutual information and finally building structured query in target language. By combining several translations for each query term in target language, we overcome shortcomings of selecting only one translation and therefore improve system performance. Results for Vietnamese-English cross-lingual retrieval show improvements of building structured query over other methods using dictionary to produce query translation in target language by combining single selection translation for each keyword. Keywords — CLIR; dictionary-based; Vietnamese; keyword extraction; structured query; disambiguation; mutual information

Building Subject-aligned Comparable Corpora and Mining it for Truly Parallel Sentence Pairs

Parallel sentences are a relatively scarce but extremely useful resource for many applications including cross-lingual retrieval and statistical machine translation. This research explores our methodology for mining such data from previously obtained comparable corpora. The task is highly practical since non-parallel multilingual data exist in far greater quantities than parallel corpora, but parallel sentences are a much more useful resource. Here we propose a web crawling method for building subject-aligned comparable corpora from Wikipedia articles. We also introduce a method for extracting truly parallel sentences that are filtered out from noisy or just comparable sentence pairs. We describe our implementation of a specialized tool for this task as well as training and adaption of a machine translation system that supplies our filter with additional information about the similarity of comparable sentence pairs.

Cross-lingual latent semantic analysis

Cross-lingual information retrieval is a difficult task typically involving query translation into multiple languages followed by monolingual retrieval in each language. Latent Semantic Analysis allows cross-lingual retrieval without translating queries by working from an already existing corpus of translations. Thus, collecting such a corpus obviates the need to construct complicated translation tools, making this technique particularly applicable to querying less commercially appealing languages. First, we extend work on retrieval from an English-French corpora split into training and test sets to examine the effects of training on a corpus from a completely different. Success is measured by the proportion of direct translations correctly considered most similar by Latent Semantic Analysis. Secondly, an English only similarity task from the literature is also extended to train on a different corpus to the one being tested on. Here the degradation in performance is measured through examining the variation in the correlations between the inter-document similarity judgements calculated by Latent Semantic Analysis and an experimentally derived baseline of human judgements of inter-document similarity. Higher order indexing schemes discarding uncommon terms, sparse matrix representations and the removal of factors with very low eigenvalues are used to enhance efficiency. Performance degradation from exogenous training is shown in both cases. The best results occur using stopping, log-entropy weighting and over 500 factors. References K. Boerner. Extracting and visualizing semantic structures in retrieval results for browsing. In Peter J. Nuernberg, David L. Hicks and Richard Furuta, editors, Proceedings of the fifth ACM conference on Digital libraries, pages 234--235. ACM 2000. doi:http://doi.acm.org/10.1145/336597.336672 Deerwester, S. C., Dumais, S. T., Landauer, T. K., Fernas, G. W. and Harshman, R. A., Indexing by Latent Semantic Analysis, Journal of the American Society of Information Science, 41, 1990, 391--407. doi:10.1002/(SICI)1097-4571(199009)41:6<391::AID-ASI1>3.0.CO;2-9 S. T. Dumais, T. K. Landauer and M. L. Littman. Automatic cross-linguistic information retrieval using Latent Semantic Indexing. In SIGIR'96 - Workshop on Cross-Linguistic Information Retrieval, pages 16--23. ACM, 1996. T. K. Landauer and M. L. Littman. Fully automatic cross-language document retrieval using latent semantic indexing. In Gregory Grefenstette, editor, Proceedings of the Sixth Annual Conference of the UW Centre for the New Oxford English Dictionary and Text Research, pages 31--38. UW Centre for the New OED and Text Research, Waterloo Ontario, 1990. Landauer, T. K., Littman, M. L. and Stornetta, W. S., A statistical method for cross-language information retrieval. Unpublished manuscript, 1992. Landauer, T. K., Foltz, P. W. and Laham, D., Introduction to Latent Semantic Analysis, Discourse Processes, textbf{25}, 1998, 259--284. Lloyd, R. and Shakiban, C., Improvements in Latent Semantic Analysis, American Journal of Undergraduate Research, 3, 2004, 29--34. http://www.ajur.uni.edu/v3n2 B. Pincombe. Comparison of Human and Latent Semantic Analysis (LSA) Judgements of Pairwise Document Similarities for a News Corpus. Research Report DSTO-RR-0278. DSTO, 2004. http://dspace.dsto.defence.gov.au/dspace/bitstream/1947/3334/1/DSTO-RR-0278%0PR.pdf P. G. Young. Cross-language information retrieval using latent semantic indexing. Technical Report UT-CS-94-259. University of Tennessee, 1994. M. D. Lee, B. M. Pincombe and M. B. Welsh. An empirical evaluation of models of text document similarity. In Bruno G. Bara, Lawrence Barsalou and Monica Bucciarelli, editors, Proceedings of the 27th Annual Conference of the Cognitive Science Society, pages 1254--1259. Lawrence Erlbaum Associates, Mahwah, NJ, 2005. http://hdl.handle.net/2440/28910

Korean to English Patent Automatic Translation (K2E-PAT) and cross lingual retrieval on KIPRIS

Korean to English Patent Automatic Translation (K2E-PAT) and cross lingual retrieval on KIPRIS

Translation Probabilities in Cross-Language Information Retrieval

Translation ambiguity is a major problem in dictionary-based cross-language information retrieval. To attack the problem, indirect disambiguation approaches, which do not explicitly resolve translation ambiguity, rely on query-structuring techniques such as a structured Boolean model and Pirkola's method. Direct disambiguation approaches try to assign translation probabilities to translation equivalents, normally by employing co-occurrence statistics of target language terms from target documents as disambiguation clues. Thus far, translation probabilities have not been well explored in terms of statistical query translation models, query formulation, or cross-lingual retrieval models, etc. In order to study the impact of translation probabilities on retrieval effectiveness in direct disambiguation approaches, this paper empirically investigates the following issues: different disambiguation factors affecting the calculation of translation probabilities, the comparison of cross-lingual query formulation techniques involving translation probabilities, the relationship between the accuracy of translation disambiguation and retrieval effectiveness, and the relationship between top n translations and retrieval effectiveness.

Structured queries, language modeling, and relevance modeling in cross-language information retrieval

Two probabilistic approaches to cross-lingual retrieval are in wide use today, those based on probabilistic models of relevance, as exemplified by INQUERY, and those based on language modeling. INQUERY, as a query net model, allows the easy incorporation of query operators, including a synonym operator, which has proven to be extremely useful in cross-language information retrieval (CLIR), in an approach often called structured query translation. In contrast, language models incorporate translation probabilities into a unified framework. We compare the two approaches on Arabic and Spanish data sets, using two kinds of bilingual dictionaries––one derived from a conventional dictionary, and one derived from a parallel corpus. We find that structured query processing gives slightly better results when queries are not expanded. On the other hand, when queries are expanded, language modeling gives better results, but only when using a probabilistic dictionary derived from a parallel corpus. We pursue two additional issues inherent in the comparison of structured query processing with language modeling. The first concerns query expansion, and the second is the role of translation probabilities. We compare conventional expansion techniques (pseudo-relevance feedback) with relevance modeling, a new IR approach which fits into the formal framework of language modeling. We find that relevance modeling and pseudo-relevance feedback achieve comparable levels of retrieval and that good translation probabilities confer a small but significant advantage.

Fuzzy conceptual indexing for concept-based cross-lingual text retrieval

Cross-lingual text retrieval (CLTR) is a technique for locating relevant documents in different languages. The authors have developed fuzzy conceptual indexing (FCI) to extend CLTR to include documents that share concepts but don't contain exact translations of query terms. In FCI, documents and queries are represented as a function of language-independent concepts, thus enabling direct mapping between them across multiple languages. Experimental results suggest that concept-based CLTR outperforms translation-based CLTR in identifying conceptually relevant documents.

A multilingual text mining approach to web cross-lingual text retrieval

Abstract To enable concept-based cross-lingual text retrieval (CLTR) using multilingual text mining, our approach will first discover the multilingual concept–term relationships from linguistically diverse textual data relevant to a domain. Second, the multilingual concept–term relationships, in turn, are used to discover the conceptual content of the multilingual text, which is either a document containing potentially relevant information or a query expressing an information need. When language-independent concepts hidden beneath both document and query are revealed, concept-based matching is made possible. Hence, concept-based CLTR is facilitated. This approach is employed for developing a multi-agent system to facilitate concept-based CLTR on the Web.