Domain Adaptation of Statistical Machine Translation Models with Monolingual Data for Cross Lingual Information Retrieval

Abstract

This work proposes to adapt an existing general SMT model for the task of translating queries that are subsequently going to be used to retrieve information from a target language collection. In the scenario that we focus on access to the document collection itself is not available and changes to the IR model are not possible. We propose two ways to achieve the adaptation effect and both of them are aimed at tuning parameter weights on a set of parallel queries. The first approach is via a standard tuning procedure optimizing for BLEU score and the second one is via a reranking approach optimizing for MAP score. We also extend the second approach by using syntax-based features. Our experiments show improvements of 1-2.5 in terms of MAP score over the retrieval with the non-adapted translation. We show that these improvements are due both to the integration of the adaptation and syntax-features for the query translation task.

Adaptation of machine translation for multilingual information retrieval in the medical domain

“…One highly relevant study which explicitly deals with genre adaptation is by Nikoulina et al [38]. They adapt the Moses SMT toolkit to translate queries for cross-lingual IR applied on the CLEF Ad Hoc TEL 2009 test collection of bibliography entries.…”

Learning to translate queries for CLIR

“…This usually restricts the search space for oracle translations to the k-best list of derivations [10]. To alleviate this problem, we abstract away from the ranking problem and approximate retrieval quality of a translation q with its relevance score S rel (q, C + f ) to the set of relevant documents C…”

“…We address this problem by discriminative training techniques which are widely used in the SMT community, and use automatically constructed relevance judgments from linked data. We show that a decomposable proxy for retrieval quality in training alleviates the problem of a costly intermediate retrieval step in reranking frameworks [10], and allows us to make use of the full, and lexically more diverse, decoder search space to optimize query translations for the CLIR task.…”

Flat vs. hierarchical phrase-based translation models for cross-language information retrieval

“…Magdy et al [17] showed that preprocessing text consistently for MT and IR systems is beneficial. Nikoulina et al [19] built MT models tailored to query translation by tuning model weights with queries and reranking the top n translations to maximize effectiveness on a held-out query set. While improvements were more substantial using the latter method, another interesting finding was the low correlation between translation and retrieval quality.…”

“…Despite the prevalence of context-independent word-based approaches for cross-language information retrieval (CLIR) derived from the IBM translation models [4], recent studies have shown that exploiting ideas from machine translation (MT) for context-sensitive query translation produces higher-quality results [17,19,24]. State-of-the-art MT systems take advantage of sophisticated models with "deeper" representations of translation units, e.g., phrase-based [13], syntax-based [25,27], and even semantics-based [11] models.…”