Citation Context

...s the best results. One of the most successful techniques, the benefits of which we can confirm, is pseudo in-domain data acquisition and selection, described, e.g., by Mansour et al. (2011). 3244 For genre, similar techniques can be used as in the case of domain adaptation, but typically much less data is available, especially within the domain of interest. Moreover, while domain adaptation concentrates on lexical coverage, genre differences are mostly in the structure of the texts in question, such as grammar, length, and use of punctuation. As the most relevant related work, we can mention Nikoulina et al. (2012), who worked on a similar problem of SMT adaptation to short queries in the CLEF 2009 task. 3. The data sets 3.1. Data source The source side of the presented data set consists of randomly selected 749 real English queries asked by the general public through the Health on the Net Foundation website3, and another 759 queries asked in English by healthcare professionals in the Trip database4 (Meats et al., 2007). All the queries have been manually cleaned of random artifacts (e.g., texts which could not be interpreted as search queries or contained nonsensical strings such as asdfghj etc.). Spel...

Citation Context

...after each interaction is too costly, online adaptation after each interaction has become the learning protocol of choice for CAT. Online learning has been applied in generative SMT, e.g., using incremental versions of the EM algorithm (Ortiz-Martınez et al., 2010; Hardt and Elming, 2010), or in discriminative SMT, e.g., using perceptron-type algorithms (Cesa-Bianchi et al., 2008; Martınez-Gomez et al., 2012; Waschle et al., 2013; Denkowski et al., 2014). In a similar way to deploying human feedback, extrinsic loss functions have been used to provide learning signals for SMT. For example, Nikoulina et al. (2012) propose a setup where an SMT system feeds into cross-language information retrieval, and receives feedback from the performance of translated queries with respect to cross-language retrieval performance. This feedback is used to train a reranker on an n-best list of translations order with respect to retrieval performance. In con882 Figure 1: Response-based learning cycle for grounding SMT in virtual trivia gameplay. trast to our work, all mentioned approaches to interactive or adaptive learning in SMT rely on human post-edits or human reference translations. Our work differs from these appro...

Citation Context

...tion (SMT) component of cross-lingual information retrieval (CLIR) systems is often regarded as black box that is optimized for translation quality independent from the retrieval task. In recent work =-=[10]-=-, SMT has been tuned for retrieval by training a reranker on k-best translations ordered according to their retrieval performance. In this paper we propose a decomposable proxy for retrieval quality t...

Citation Context

... queries (Downey et al., 2008). Attempts to inform the SMT system about its use for retrieval by optimizing its parameters towards a retrieval objective have been presented in the form or re-ranking (=-=Nikoulina et al., 2012-=-) or ranking (Sokolov et al., 2014). The most direct integration of SMT and CLIR has been presented by Hieber and Riezler (2015) in an approach to Bag-of-Words Forced Decoding (BOW-FD) where IR featur...

Citation Context

...LIR) 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] m...