Translation Output Research Articles

A case study of improving and evaluating consumer survey translation

The inherent challenge of managing the translation of multilingual materials for global research is to develop a systematic translation process considering the varying translation expertise and resources available locally. In the United States, consumer survey materials are typically first designed in English (the source language) and then translated into the target language(s) with culture-specific tailoring for markets outside of the United States. In an effort to standardize the translation method globally, three experimental approaches were developed to assess the quality of translation tailoring as a function of the varying resources of expertise and tools available in the local markets. These three approaches involved different translator qualifications: a centralized professional translation service within a large company headquartered in the United States, a local professional translation service, and a company employee who is a native speaker and familiar with consumer surveys. In addition, a toolkit of translation reference materials (including translation validation form, translation input document, terminology bank and/or step-by-step translation procedures) was developed and tailored to the qualification of the translators. The experiment followed the standardized approach of Translation, Review, Adjudication, Pretesting, and Documentation (TRAPD) developed for survey research, and the translation output is evaluated based on the toolkit provided in addition to expertise and resources available for translation and review locally. The face-to-face recruitment materials of a large consumer panel survey in China were used for evaluation of the proposed translation methods from English to Chinese. For each approach, we conducted a review of translation quality based on a qualitative evaluation through a small number of one-on-one interviews in China and a quantitative evaluation of scoring the translation by assigning issue codes that were designed using sociolinguistics principles. This paper will discuss one of the three translation approaches (using a centralized professional translation service), describe the lessons learned therefrom, and share recommendations for adapting and implementing translation standards as well as areas for future research.

Postediting machine translation output: subject-matter experts versus professional translators

ABSTRACT This study compares the quality of postediting performed by subject-matter experts as opposed to professional translators. A total of 10 professional translators and 10 engineers postedited a 482-word technical text pre-translated from English into Turkish using data-based machine translation system, Google Translate. The findings suggest that, for this particular task (technical translation), translators’ and engineers’ postediting quality is similar as far as the categories of mistranslation, accuracy, and consistency are concerned. Engineers performed significantly better than translators only in the terminology category. In the language category, translators made significantly fewer (minor) errors than engineers. The qualitative data analysis revealed that, for this particular task, a degree in translation does not directly correlate with postediting quality, unless it is combined with subject-matter knowledge and professional experience in translation. Finally, the present study indicates that – both for the engineers and the professional translators – expertise and experience in the subject matter are important factors determining postediting quality.

To Swap or Not to Swap? Exploiting Dependency Word Pairs for Reordering in Statistical Machine Translation

Reordering poses a major challenge in machine translation (MT) between two languages with significant differences in word order. In this paper, we present a novel reordering approach utilizing sparse features based on dependency word pairs. Each instance of these features captures whether two words, which are related by a dependency link in the source sentence dependency parse tree, follow the same order or are swapped in the translation output. Experiments on Chinese-to-English translation show a statistically significant improvement of 1.21 BLEU point using our approach, compared to a state-of-the-art statistical MT system that incorporates prior reordering approaches.

Context as Achilles’ heel of translation technologies

The tools of translation memories and machine translation can be viewed as (not) being able to draw on different aspects of context that are relevant to a particular translation project, such as bilingual text, portions of a text, versions of a text, related text, or extralinguistic context. The aspects of context that the tools can indeed draw on and how well they do so highlight their most important benefits, whereas the aspects of context that the tools may fail to draw on reveal their weaknesses. In cases where, depending on the context, a piece of text being translated may have more than one meaning or a translation of a piece of text may be rendered in more than one way, the contexts the tools are able to draw on are critical. An analysis of different kinds of context in different applications highlights the circumstances in which users of TM and MT tools may risk accepting semantically and lexically undesirable output, as well as when TM tool users may risk inputting contextually inappropriate translations. Further, context is an important consideration when integrating MT output into TM tools. Although TM and MT technologies differ in how they produce translation output, end-users of both tool types may face similar contextual challenges.

Classification of Machine Translation Outputs Using NB Classifier and SVM for Post-Editing

Machine translation outputs are not correct enough to be used as it is, except for the very simplest translations. They only give the general meaning of a sentence not the exact translation. As Machine Translation (MT) is gaining a position in the whole world, there is a need for estimating the quality of machine translation outputs. Many prominent MT-Researchers are trying to make the MT-System, that produces very good and accurate translations and that also covers maximum language pairs. If good translations out of all translations can be categorized then the time and cost can be saved to a great extent. Now, Good quality translations will be sent for post-editing and rest will be sent for pre-editing or retranslation. In this paper, Kneser Ney smoothing language model is used to calculate the probability of machine translated output. But a translation cannot be said good or bad. Based on its probability score there are many other parameters that effect its quality. The quality of machine translation is made easier to estimate for post-editing by using two different predefined famous algorithms for classification.

Modelling and Optimizing on Syntactic N-Grams for Statistical Machine Translation

The role of language models in SMT is to promote fluent translation output, but traditional n-gram language models are unable to capture fluency phenomena between distant words, such as some morphological agreement phenomena, subcategorisation, and syntactic collocations with string-level gaps. Syntactic language models have the potential to fill this modelling gap. We propose a language model for dependency structures that is relational rather than configurational and thus particularly suited for languages with a (relatively) free word order. It is trainable with Neural Networks, and not only improves over standard n-gram language models, but also outperforms related syntactic language models. We empirically demonstrate its effectiveness in terms of perplexity and as a feature function in string-to-tree SMT from English to German and Russian. We also show that using a syntactic evaluation metric to tune the log-linear parameters of an SMT system further increases translation quality when coupled with a syntactic language model.

Integrated Transcriptome and Proteome Analyses Reveal Organ-Specific Proteome Deterioration in Old Rats

SummaryAging is associated with the decline of protein, cell, and organ function. Here, we use an integrated approach to characterize gene expression, bulk translation, and cell biology in the brains and livers of young and old rats. We identify 468 differences in protein abundance between young and old animals. The majority are a consequence of altered translation output, that is, the combined effect of changes in transcript abundance and translation efficiency. In addition, we identify 130 proteins whose overall abundance remains unchanged but whose sub-cellular localization, phosphorylation state, or splice-form varies. While some protein-level differences appear to be a generic property of the rats’ chronological age, the majority are specific to one organ. These may be a consequence of the organ’s physiology or the chronological age of the cells within the tissue. Taken together, our study provides an initial view of the proteome at the molecular, sub-cellular, and organ level in young and old rats.

Translation, Pdcd4 and eIF4A.

Pdcd4 (programmed cell death 4) has received considerable attention as a tumor suppressor protein in recent years, however, its molecular function is still poorly understood. Pdcd4 is a nuclear-cytoplasmic shuttling and RNA-binding protein, which is involved in the control of translation of specific mRNAs. Pdcd4 interacts with the eukaryotic translation initiation factor eIF4A, an RNA helicase that plays a critical role in cap-dependent translation by melting stable RNA secondary structures in the 5′-untranslated regions (UTRs) of mRNAs [1,2]. It has been shown that Pdcd4 inhibits the helicase activity of eIF4A, suggesting that it suppresses translation of mRNAs with highly structured 5′-UTRs [3]. This idea was supported by analyzing the effect of Pdcd4 on artificial RNA constructs containing stable hairpin structures in the 5′-UTR and, more recently, confirmed by demonstrating that translation of p53 mRNA (whose 5′-UTR forms very stable secondary structures) is suppressed by Pdcd4 via an eIF4A-dependent mechanism [4] (Figure ​(Figure1a1a). Figure 1 Suppression of translation initiation (a) and translation elongation (b) by Pdcd4 Recent findings indicate that the role of Pdcd4 in translation is more complex and involves an additional, entirely different inhibitory mechanism. It was shown that Pdcd4 suppresses the translation of the c-myb and A-myb mRNAs even when the eIF4A binding site was destroyed by mutation [5,6]. Instead, the RNA-binding activity of Pdcd4 was required to suppress translation of these RNAs, suggesting that direct RNA-binding by Pdcd4 plays a key role. Additional work revealed that the nucleotide sequences responsible for Pdcd4-induced translation suppression were located in the coding regions of c-myb and A-myb mRNAs. Furthermore, in vitro RNA-binding studies demonstrated that the “Pdcd4 response regions” of the c-myb and A-myb mRNAs are able to form secondary structures which were preferentially bound by Pdcd4. Overall, these experiments suggested that Pdcd4 is able to suppress translation by a novel mechanism, which involves direct binding of Pdcd4 to specific target RNAs. How does Pdcd4 suppress translation from within the coding region? When the c-myb coding region was placed under the control of the Hepatitis C virus internal ribosomal entry site (HCV-IRES) Pdcd4 failed to suppress translation [5]. Because the HCV-IRES does not depend on the translation initiation factors required for cap-dependent initiation, this suggested that Pdcd4 suppresses translation of c-myb mRNA by interfering with one of these factors (except eIF4A) at translation initiation. However, when the c-myb and A-myb “Pdcd4 response regions” were fused to GFP RNA to ask if they are able to convey Pdcd4-responsiveness onto a heterologous RNA, a surprising observation was made. It was indeed found that Pdcd4 was able to suppress translation of the recombinant RNA, but only when a continuous open reading frame extended from the GFP coding sequence into the added c-myb or A-myb sequences. In other words, Pdcd4 suppressed the translation of the recombinant RNAs only when the binding region for Pdcd4 was itself part of the open reading frame. This observation was confirmed by introducing in-frame translational stop codons into the authentic c-myb and A-myb coding regions upstream of the “Pdcd4 response regions”. This completely abolished the inhibitory effect of Pdcd4, again indicating that Pdcd4 supresses translation only when the sequence to which it binds is part of the translated region [6]. Thus, truncating the coding region by a single stop codon is sufficient to abrogate Pdcd4-dependent inhibition. A straightforward explanation for this observation is that Pdcd4 suppresses translation of these RNAs at the elongation step (Figure ​(Figure1b1b). How can this be reconciled with the fact that the translation of the c-myb coding region was not suppressed by Pdcd4 when translation was initiated at the HCV-IRES? A possible explanation comes from the observation that the efficiency of IRES-dependent translation initiation was much lower than cap-dependent translation [5]. If the inhibition of translation elongation by Pdcd4 is augmented with increasing translation efficiency, it suggests that Pdcd4 acts like a self-adjusting controller, that limits the translation output when the translation rate is high but has no effect when it is low. How Pdcd4 actually suppresses translation at elongation is currently unknown. Binding of Pdcd4 could stabilize RNA secondary structures and thereby hinder the passage of approaching ribosomes. Pdcd4 also interacts with the poly(A) binding protein which, in turn, could stabilize the binding of Pdcd4 to the response region [7]. Elongating ribosomes might also be blocked in an active manner. Such a mechanism has been described for the cytoplasmic polyadenylation element binding protein CPEB2, which interacts with the elongation factor eEF2 and reduces eEF2/ribosome-triggered GTP hydrolysis, thereby slowing down translation elongation of CPEB2-bound RNAs [8]. In any case, the work discussed here has led to a new paradigm for translational suppression and recognition of specific target RNAs by Pdcd4. Exploring its relevance to the function of Pdcd4 as a tumor suppressor will now be an important task.

Adaptación en español del nuevo sistema de codificación de enfermedad renal primaria de la European Renal Association-European Dialysis and Transplant Association (ERA-EDTA)

BackgroundThe European Renal Association and the European Dialysis and Transplant Association (ERA-EDTA) have issued an English-language new coding system for primary kidney disease (PKD) aimed at solving the problems that were identified in the list of “Primary renal diagnoses” that has been in use for over 40 years. PurposeIn the context of Registro Español de Enfermos Renales (Spanish Registry of Renal Patients, [REER]), the need for a translation and adaptation of terms, definitions and notes for the new ERA-EDTA codes was perceived in order to help those who have Spanish as their working language when using such codes. MethodsBilingual nephrologists contributed a professional translation and were involved in a terminological adaptation process, which included a number of phases to contrast translation outputs. Codes, paragraphs, definitions and diagnostic criteria were reviewed and agreements and disagreements aroused for each term were labelled. Finally, the version that was accepted by a majority of reviewers was agreed. ResultsA wide agreement was reached in the first review phase, with only 5 points of discrepancy remaining, which were agreed on in the final phase. ConclusionsTranslation and adaptation into Spanish represent an improvement that will help to introduce and use the new coding system for PKD, as it can help reducing the time devoted to coding and also the period of adaptation of health workers to the new codes.

A Spanish version for the new ERA-EDTA coding system for primary renal disease

BackgroundThe European Renal Association and the European Dialysis and Transplant Association (ERA-EDTA) have issued an English-language new coding system for primary kidney disease (PKD) aimed at solving the problems that were identified in the list of “Primary renal diagnoses” that has been in use for over 40 years. PurposeIn the context of Registro Español de Enfermos Renales (Spanish Registry of Renal Patients [REER]), the need for a translation and adaptation of terms, definitions and notes for the new ERA-EDTA codes was perceived in order to help those who have Spanish as their working language when using such codes. MethodsBilingual nephrologists contributed a professional translation and were involved in a terminological adaptation process, which included a number of phases to contrast translation outputs. Codes, paragraphs, definitions and diagnostic criteria were reviewed and agreements and disagreements aroused for each term were labelled. Finally, the version that was accepted by a majority of reviewers was agreed. ResultsA wide agreement was reached in the first review phase, with only 5 points of discrepancy remaining, which were agreed on in the final phase. ConclusionsTranslation and adaptation into Spanish represent an improvement that will help to introduce and use the new coding system for PKD, as it can help reducing the time devoted to coding and also the period of adaptation of health workers to the new codes.

From grammar-based MT to post-processed SL representations

This paper proposes the implementation of a post-processing stage to a grammar-based machine translation (MT) system from written into sign language for the pair written Greek---Greek sign language (GSL), where translation output is visualized by an avatar. Post-processing is applied on the output of the MT system's transfer module by activating an editing environment for SL which utilizes a dictionary of signs coded for HamNoSys transcriptions and non-manual elements along with grammar knowledge. This post-processing environment may also be used as a standalone editing environment for expert users allowing for a considerable range of editing actions, as well as an SL editing tool for the general public who are not familiar with the underlying technology. In both cases, this environment provides the opportunity of on-the-fly structuring of new signed utterances going far beyond current limitations in composition with new signed language content.

Towards accurate predictors of word quality for Machine Translation: Lessons learned on French–English and English–Spanish systems

This paper proposes some ideas to build effective estimators, which predict the quality of words in a Machine Translation (MT) output. We propose a number of novel features of various types (system-based, lexical, syntactic and semantic) and then integrate them into the conventional (previously used) feature set, for our baseline classifier training. The classifiers are built over two different bilingual corpora: French–English (fr–en) and English–Spanish (en–es). After the experiments with all features, we deploy a “Feature Selection” strategy to filter the best performing ones. Then, a method that combines multiple “weak” classifiers to constitute a strong “composite” classifier by taking advantage of their complementarity allows us to achieve a significant improvement in terms of F-score, for both fr–en and en–es systems. Finally, we exploit word confidence scores for improving the quality estimation system at sentence level.

Topic-Based Coherence Modeling for Statistical Machine Translation

Coherence that ties sentences of a text into a meaningfully connected structure is of great importance to text generation and translation. In this paper, we propose topic-based coherence models to produce coherence for document translation, in terms of the continuity of sentence topics in a text. We automatically extract a coherence chain for each source text to be translated. Based on the extracted source coherence chain, we adopt a maximum entropy classifier to predict the target coherence chain that defines a linear topic structure for the target document. We build two topic-based coherence models on the predicted target coherence chain: 1) a word level coherence model that helps the decoder select coherent word translations and 2) a phrase level coherence model that guides the decoder to select coherent phrase translations. We integrate the two models into a state-of-the-art phrase-based machine translation system. Experiments on large-scale training data show that our coherence models achieve substantial improvements over both the baseline and models that are built on either document topics or sentence topics obtained under the assumption of direct topic correspondence between the source and target side. Additionally, further evaluations on translation outputs suggest that target translations generated by our coherence models are more coherent and similar to reference translations than those generated by the baseline.

Interactive translation prediction versus conventional post-editing in practice: a study with the CasMaCat workbench

We conducted a field trial in computer-assisted professional translation to compare interactive translation prediction (ITP) against conventional post-editing (PE) of machine translation (MT) output. In contrast to the conventional PE set-up, where an MT system first produces a static translation hypothesis that is then edited by a professional (hence post-editing), ITP constantly updates the translation hypothesis in real time in response to user edits. Our study involved nine professional translators and four reviewers working with the web-based CasMaCat workbench. Various new interactive features aiming to assist the post-editor/translator were also tested in this trial. Our results show that even with little training, ITP can be as productive as conventional PE in terms of the total time required to produce the final translation. Moreover, translation editors working with ITP require fewer key strokes to arrive at the final version of their translation.

Depfix, a Tool for Automatic Rule-based Post-editing of SMT

Abstract We present Depfix, an open-source system for automatic post-editing of phrase-based machine translation outputs. Depfix employs a range of natural language processing tools to obtain analyses of the input sentences, and uses a set of rules to correct common or serious errors in machine translation outputs. Depfix is currently implemented only for English-to-Czech translation direction, but extending it to other languages is planned.

Online discriminative learning for machine translation with binary-valued feedback

Viewing machine translation (MT) as a structured classification problem has provided a gateway for a host of structured prediction techniques to enter the field. In particular, large-margin methods for discriminative training of feature weights, such as the structured perceptron or MIRA, have started to match or exceed the performance of existing methods such as MERT. One issue with these problems in general is the difficulty in obtaining fully structured labels, e.g. in MT, obtaining reference translations or parallel sentence corpora for arbitrary language pairs. Another issue, more specific to the translation domain, is the difficulty in online training and updating of MT systems, since existing methods often require bilingual knowledge to correct translation outputs online. The problem is an important one, especially with the usage of MT in the mobile domain: in the process of translating user inputs, these systems can also receive feedback from the user on the quality of the translations produced. We propose a solution to these two problems, by demonstrating a principled way to incorporate binary-labeled feedback (i.e. feedback on whether a translation hypothesis is a "good" or understandable one or not), a form of supervision that can be easily integrated in an online and monolingual manner, into an MT framework. Experimental results on Chinese---English and Arabic---English corpora for both sparse and dense feature sets show marked improvements by incorporating binary feedback on unseen test data, with gains in some cases exceeding 5.5 BLEU points. Experiments with human evaluators providing feedback present reasonable correspondence with the larger-scale, synthetic experiments and underline the relative ease by which binary feedback for translation hypotheses can be collected, in comparison to parallel data.

Measuring Semantic Preservation in Machine Translation with HCOMET: Human Cognitive Metric for Evaluating Translation

Human ranking of machine translation output is a commonly used method for comparing different innovations in machine translation research. Theoretically simple, the comparison of multiple translations is, in effect, cognitively complex, requiring judges to balance the weight of different types of translation errors in the context of the whole sentence. This cognitive complexity is made evident through low intra- and inter- annotator agreements, which call into question the reliability of such ranking metrics. HMEANT (Lo and Wu, 2011) attempted to decrease the complexity of ranking by dividing sentences into smaller semantic units whose translation alignments were more objective, rendering the task cognitively simpler. However, HMEANT does not discern how these semantic units are related and relies heavily on language-dependent verb frames – a significant problem for a translation metric. This project defines a new set of human metrics focusing on HCOMET (Human COgnitive Metric for Evaluating Translation). HCOMET, attempting to overcome the limitations of HMEANT, employed a new cognitively-informed annotation scheme and new scoring guidelines. While the inter-annotator agreement did not surpass that of HMEANT, the conceptual framework of HCOMET allows for a much more detailed analysis of semantic adequacy in machine translation.

Crowdsourced Monolingual Translation

An enormous potential exists for solving certain classes of computational problems through rich collaboration among crowds of humans supported by computers. Solutions to these problems used to involve human professionals, who are expensive to hire or difficult to find. Despite significant advances, fully automatic systems still have much room for improvement. Recent research has involved recruiting large crowds of skilled humans (“crowdsourcing”), but crowdsourcing solutions are still restricted by the availability of those skilled human participants. With translation, for example, professional translators incur a high cost and are not always available; machine translation systems have been greatly improved recently but still can only provide passable translation; and crowdsourced translation is limited by the availability of bilingual humans. This article describes crowdsourced monolingual translation, where monolingual translation is translation performed by monolingual people. Crowdsourced monolingual translation is a collaborative form of translation performed by two crowds of people who speak the source or the target language, respectively, with machine translation as the mediating device. This article describes a general protocol to handle crowdsourced monolingual translation and analyzes three systems that implemented the protocol. These systems were studied in various settings and were found to supply significant improvement in quality over both machine translation and monolingual editing of machine translation output (“postediting”).

The impact of cultural background knowledge in the processing of metaphorical expressions

This study aims to explore the role and possible impact of cultural background knowledge (CBK) on performance in sight translation, specifically the translation of metaphorical expressions (MEs). A between-subjects experiment was designed for 68 interpreting students who were assigned to a control group (CG) and an experimental group (EG). They were asked to sight translate a speech containing ten MEs, with only the EG members given relevant CBK beforehand. The study triangulates data from the recordings of sight translation outputs, the transcriptions of the recordings, and the subjects’ guided interviews. The paper concludes with two main findings: (1) CBK markedly alleviates the cognitive load imposed by MEs and thus facilitates the process of translation by shortening the processing time and improving the translation quality; (2) although CBK does not exert a significant influence on the choice of translation strategies as a whole, it discernibly reduces the use of omission as a coping strategy to deal with inadequacies in the understanding of metaphorical meanings.

AS PHILOSOPHISCHEN BEMERKUNGEN DE WITTGENSTEIN

Philosophische Bemerkungen. By Ludwig Wittgenstein. Edited by r. Rhees. (Frankfurt am Main, Suhrkamp Verlag, 1964; Oxford, Basil Blackwell, 1964, p. 348). Published in the original German version without English translation. [Translator's Note: I chose to translate all the passages cited by Malcolm, the ones from the Bemerkungen as well as those of the Untersuchungen, directly from German. To this end, I used the editing of Suhrkamp Verlag, Frankfurt am Main, Werkausgabe Band 1 and 2, 1984. In time, I am very grateful to numerous suggestions for translation outputs and relevant corrections made by an anonymous reviewer of this magazine.]