Europarl Corpus Research Articles

Exploring variation in student translation

AbstractThis paper explores the issue of variation in translation, as well as its connection with the concept of “literal translation” and translator experience, on the basis of a multiple student translation corpus containing concurrent Italian versions of the same English source text produced by 35 undergraduate and postgraduate trainee translators. Translation paradigms for preselected lexical items expected to trigger different degrees of variation are extracted and analysed to identify both recurrent and sporadic solutions, whose acceptability in the target language is assessed using the source text’s official translation, alternative professional translations and theEuroparl Corpusas reference. The analysis shows that variation is most remarkable with respect to idiomatic/metaphorical and evaluative items than for non-idiomatic items, but also when a literal translation would not be possible in the target language. Translators are found to generally prefer literal translations whenever acceptable in the target language, irrespective of their degree of experience.

Sentence Boundary Disambiguation for Tibetan Based on Attention Mechanism at the Syllable Level

Tibetan is a low-resource language with few existing electronic reference materials. The goal of Tibetan sentence boundary disambiguation (SBD) is to segment long text into sentences, and it is the foundation for downstream tasks corpora building. This study implemented the Tibetan SBD at the syllable level to avoid word segmentation (WS) errors affecting the accuracy of SBD. Specifically, the attention mechanism is introduced based on a recurrent neural network (RNN) to study Tibetan SBD. The primary objective is to determine, using a trained model, whether the shad contained in Tibetan text is the ending of the sentence, and implement experiments on syllable embedding and component embedding to measure the model's performance. The highest accuracy for Tibetan syllable embedding and component embedding is 96.23% and 95.40 %, respectively, and the F1 score reaches 96.23% and 95.37%, respectively. The experimental results demonstrate that the proposed method can achieve better results than the established rule-based and statistical methods without considering various syntactic and part-of-speech (POS) tagging rules. German and English data from the Europarl corpus and Thai data from the IWSLT2015 corpus are validated to prove the models’ reliability and generalizability. The results demonstrate that this method is efficient not only for low-resource languages but also for high-resource languages. More importantly, we can formally apply the experimental results of this study to the research of downstream tasks, such as machine translation and automatic summarization.

Revisiting simplification in corpus-based translation studies: Insights from readability research

Ever since the publication of Laviosa’s (1998a; 1998b) pioneering work, the study of lexico-syntactic simplification has held centre stage in corpus translation research concerned with the typical features of translated texts. The simplification hypothesis states that translated texts are simpler than non-translated texts. The convergence hypothesis, also discussed by Laviosa (1998a; 1998b), but less so in follow-up studies, is that translated texts are more homogeneous than original texts, that is they display less variance. To date, simplification has mostly been operationalised in CBTS as type-token ratio, lexical density, core vocabulary coverage, list head coverage and average sentence length. Relying on these parameters, previous research has produced mixed results, with simplification varying across translation modalities, language pairs and registers. The present article sets out to revisit the simplification and convergence hypotheses through the lens of NLP-informed readability research. In particular, we rely on a larger set of simplification indicators and make use of multivariate statistical techniques. We present a simplification study of Europarl corpus data in French translated from English and in non-translated French. The results show that translated French is simpler than original French, lexically and syntactically. We also find evidence of convergence that shows that translators smooth out cross-speaker lexical heterogeneity in translated parliamentary proceedings.

Counterexpectational Translations: The Dutch Markers Toch and Eigenlijk Contrasted with Their English Correspondents

Abstract The Dutch particle toch and modal adverb eigenlijk both belong to the broad domain of ‘contrast’: their prototypical functions have been described as ‘concession’ and ‘counterexpectation’, respectively, and both have been attributed emphatic discourse uses (e.g. toch as a common ground reminder and eigenlijk as a reinforcement marker). This study analyses the multifunctional scope of toch and eigenlijk in Dutch and maps their translation equivalents in English through an in-depth analysis of the Dutch-English component of the Europarl corpus. For eigenlijk the most common correspondents (in fact, really, actually) belong to the core of the semantic domain of (counter)expectation, whereas none of the English correspondents of toch is able to capture the marker’s duality as both a concessive and a common ground marker. The high degree of zero correspondences in English source texts suggests the translators’ strong awareness of the markers’ pragmalinguistic contribution.

Not…Until across European Languages: A Parallel Corpus Study

We present a parallel corpus study on the expression of the temporal construction ‘not…until’ in a sample of European languages. We use data from the Europarl corpus and create semantic maps by multidimensional scaling, in order to analyze cross-linguistic and language-internal variation. This paper builds on formal semantic and typological work, extending it by including conditional constructions, as well as connectives of the type as long as. In an investigation of 7 languages, we find that (i) languages use many more different constructions to convey this meaning than was expected from the literature; and (ii) the combination of polarity marking (negation/assertion) strongly correlates with the type of connective. We corroborate our results in a larger sample of 21 European languages. An analysis of clusters and dimensions of the semantic maps based on the enlarged dataset shows that connectives are not randomly distributed across the semantic space of the ‘not…until’-domain.

POS-Tagging based Neural Machine Translation System for European Languages using Transformers

The interaction between human beings has always faced different kinds of difficulties. One of those difficulties is the language barrier. It would be a tedious task for someone to learn all the syllables in a new language in a short period and converse with a native speaker without grammatical errors. Moreover, having a language translator at all times would be intrusive and expensive. We propose a novel approach to Neural Machine Translation (NMT) system using interlanguage word similaritybased model training and Part-Of-Speech (POS) Tagging based model testing. We compare these approaches using two classical architectures: Luong Attention-based Sequence-to-Sequence architecture and Transformer based model. The sentences for the Luong Attention-based Sequence-to-Sequence were tokenized using SentencePiece tokenizer. The sentences for the Transformer model were tokenized using Subword Text Encoder. Three European languages were selected for modeling, namely, Spanish, French, and German. The datasets were downloaded from multiple sources such as Europarl Corpus, Paracrawl Corpus, and Tatoeba Project Corpus. Sparse Categorical CrossEntropy was the evaluation metric during the training stage, and during the testing stage, the Bilingual Evaluation Understudy (BLEU) Score, Precision Score, and Metric for Evaluation of Translation with Explicit Ordering (METEOR) score were the evaluation metrics.

Neues zuüberhauptundsowieso

AbstractThis article investigates the use ofüberhauptandsowiesoin German and Dutch. These two words are frequently classified as particles, if only because of their pragmatic functions. The frequent use of particles is considered a specific trait common to German and Dutch, and the description of their semantics and pragmatics is notoriously difficult. It is unclear whether both particles have the same meaning in Dutch (where they are loanwords) and German, whether they can fulfil the same syntactic functions and to what extent the (semantic and pragmatic) functions ofüberhauptundsowiesooverlap. There has already been linguistic research onüberhauptandsowiesoby Fisseni (2009) using the world-wide web and by Bruijnen and Sudhoff (2013) using the EUROPARL corpus. In the present study we critically evaluated the corpus study, integrating information on original utterance language and discussing the adequacy of this corpus. Moreover, we conducted an experimental survey collecting subjective-intuitive judgements in three dimensions, thus gathering more data on sparse and informal constructions.By using these complementary methods, we obtain a more nuanced picture of the use ofüberhauptandsowiesoin both languages: On the one hand, the data show where the use of both words is more similar and on the other hand, differences between the languages can also be discerned.

Tradução de Marcadores Discursivos portugueses para alemão: ‘na verdade’, o que se ganha e o que se perde?

Discourse markers signal discourse coherence and become very important items in the translation process although only recently the problems raised by their translation have begun to capture the attention of researchers. This study intends to contribute to this research area by analyzing the translations of the discourse marker na verdade into German. Drawing on examples from contemporary European Portuguese, the study shows that the marker has three main functions: (i) confirmation; (ii) reformulation; and (iii) contrast. The contrastive analysis carried out in the Europarl corpus reveals that the confirmation value is the most common in this corpus and that the translators opted for a very wide range of equivalents. Although the German marker eigentlich, which scarcely appears in the corpus, could be the most appropriate equivalent to translate na verdade, another discourse marker, tatsachlich, seems to be the preferred choice of professionals, and this option raises pertinent questions in the field of Translation.

Data Categorization and Model Weighting Approach for Language Model Adaptation in Statistical Machine Translation

Language model encapsulates semantic, syntactic and pragmatic information about specific task. Intelligent systems especially natural language processing systems can show different results in terms of performance and precision when moving among genres and domains. Therefore researchers have explored different language model adaptation strategies in order to overcome effectiveness issue. There are two main categories in language model adaptation techniques. The first category includes the techniques that based on the data selection where task-oriented corpus can be extracted and used to train and generate models for specific translations. While the second category focuses on developing a weighting criterion to assign the test data to specific model corpus. The purpose of this research is to introduce language model adaptation approach that combines both categories (data selection and weighting criterion) of language model adaptation. This approach applies data selection for specific-task translations by dividing the corpus into smaller and topic-related corpora using clustering process. We investigate the effect of different approaches for clustering the bilingual data on the language model adaptation process in terms of translation quality using the Europarl corpus WMT07 that includes bilingual data for English-Spanish, English-German and English-French. A mixture of language models should assign any given data to the right language model to be used in the translation process using a specific weighting criterion. The proposed language model adaptation has achieved better translation quality compare to the baseline model in Statistical Machine Translation (SMT).

Optimising the Europarl corpus for translation studies with the EuroparlExtract toolkit

ABSTRACTThe freely available European Parliament Proceedings Parallel Corpus, or Europarl, is one of the largest multilingual corpora available to date. Surprisingly, bibliometric analyses show that it has hardly been used in translation studies. Its low impact in translation studies may partly be attributed to the fact that the Europarl corpus is distributed in a format that largely disregards the needs of translation research. In order to make the wealth of linguistic data from Europarl easily and readily available to the translation studies community, the toolkit ‘EuroparlExtract’ has been developed. With the toolkit, comparable and parallel corpora tailored to the requirements of translation research can be extracted from Europarl on demand. Both the toolkit and the extracted corpora are distributed under open licenses. The free availability is to avoid the duplication of effort in corpus-based translation studies and to ensure the sustainability of data reuse. Thus, EuroparlExtract is a contribution to satisfy the growing demand for translation-oriented corpora.

Factored Statistical Machine Translation for German-English

Machine Translation is a machine that is going to automatically translate given sentences in a language to other particular language. This paper aims to test the effectiveness of a new model of machine translation which is factored machine translation. We compare the performance of the unfactored system as our baseline compared to the factored model in terms of BLEU score. We test the model in German-English language pair using Europarl corpus. The tools we are using is called MOSES. It is freely downloadable and use. We found, however, that the unfactored model scored over 24 in BLEU and outperforms the factored model which scored below 24 in BLEU for all cases. In terms of words being translated, however, all of factored models outperforms the unfactored model.

Cross-Lingual Syntactic Transfer with Limited Resources

We describe a simple but effective method for cross-lingual syntactic transfer of dependency parsers, in the scenario where a large amount of translation data is not available. This method makes use of three steps: 1) a method for deriving cross-lingual word clusters, which can then be used in a multilingual parser; 2) a method for transferring lexical information from a target language to source language treebanks; 3) a method for integrating these steps with the density-driven annotation projection method of Rasooli and Collins (2015). Experiments show improvements over the state-of-the-art in several languages used in previous work, in a setting where the only source of translation data is the Bible, a considerably smaller corpus than the Europarl corpus used in previous work. Results using the Europarl corpus as a source of translation data show additional improvements over the results of Rasooli and Collins (2015). We conclude with results on 38 datasets from the Universal Dependencies corpora.

Resumptive anaforer. En komparativ undersøgelse af en lidt overset sproglig størrelse

Resumptive anaphors encapsulate and hypostatise a whole text segment. This survey compares and categorises Italian and Danish resumptive anaphors in a comparable subcorpus of the “Europarl Corpus” (political speeches held in the European Parliament). The anaphors are morphologically subdivided in NPs and pro-forms and the NP anaphors are semantically categorised in nominal derivatives and synonyms, generalising NPs, specifying NPs, axiological NPs, and performative NPs. With a pro-form anaphor, a similar categorisation of the anaphorised text segment is conveyed by the co-text, typically the predicate, of the anaphor. Pro-form anaphors are particularly frequent in Danish and NP anaphors in Italian, thus confirming the predominantly verbal style in Scandinavian and nominal style in Romance.

中間言語情報を記憶するピボット翻訳手法

Pivot translation allows for translation of language pairs with little or no parallel data by introducing a third language for which data exists. In particular, the triangulation method, which translates by combining source-pivot and pivot-target translation models into a source-target model, is known for its high translation accuracy. However, in the conventional triangulation method, information of pivot phrases is forgotten and not used in the translation process. In this paper, we propose a novel approach torememberthe pivot phrases in the triangulation stage, and use a pivot language model as an additional information source at translation time. Experimental results on the Europarl corpus showed gains of 0.4-1.2 BLEU points in all tested combinations of languages 1 .

Towards a corpus-based analysis of evaluative scales associated with <i>even</i>

Scalar focus operators like even, only, etc. interact with scales, i. e., ordered sets of alternatives that are referenced by focus structure. The scaling dimensions interacting with focus operators have been argued to be semantic (e. g. entailment relations, probability) in earlier work, but it has been shown that purely semantic analyses are too restrictive, and that the specific scale that a given operator interacts with is often pragmatic, in the sense of being a function of the context. If that is true, the question arises what exactly determines the (types of) scales interacting with focus operators. The present study addresses this question by investigating the distributional behaviour of the additive scalar particle even relative to scales whose focus alternatives are ordered in terms of evaluative attitudes (positive, negative). Our hypothesis is that such evaluative attitudinal scales are at least partially functions of the lexical material in the sentential environment. This hypothesis is tested by determining correlations between sentence-level attitudes and lexically encoded attitudes in the relevant sentences. We use data from the Europarl corpus, a corpus of scripted and highly elaborated political speech, which is rich in argumentative discourse and thus lends itself to the study of attitudes in context. Our results show that there are in fact significant correlations between (manual) sentence-level evaluations and lexical evaluations (determined through machine learning) in the textual environment of the relevant operators. We conclude with an outlook on possible extensions of the method applied in the present study by identifying attitudinal patterns beyond the sentence, showing that positively and negatively connotated instances of even differ in terms of their argumentative function, with positive even often marking the climax and endpoint of an argument, while negative even often occurs in qualifying insertions like concessive parentheses. While we regard our results as valid, some refinements and extensions of the method are pointed out as necessary steps towards the establishment of an empirical sentence semantics, in the domain of scalar additive operators as well as more generally speaking.

Optimizing Instance Selection for Statistical Machine Translation with Feature Decay Algorithms

We introduce FDA5 for efficient parameterization, optimization, and implementation of feature decay algorithms (FDA), a class of instance selection algorithms that use feature decay. FDA increase the diversity of the selected training set by devaluing features (i.e., n-grams) that have already been included. FDA5 decides which instances to select based on three functions used for initializing and decaying feature values and scaling sentence scores controlled with five parameters. We present optimization techniques that allow FDA5 to adapt these functions to in-domain and out-of-domain translation tasks for different language pairs. In a transductive learning setting, selection of training instances relevant to the test set can improve the final translation quality. In machine translation experiments performed on the 2 million sentence English-German section of the Europarl corpus, we show that a subset of the training set selected by FDA5 can gain up to 3.22 BLEU points compared to a randomly selected subset of the same size, can gain up to 0.41 BLEU points compared to using all of the available training data using only 15% of it, and can reach within 0.5 BLEU points to the full training set result by using only 2.7% of the full training data. FDA5 peaks at around 8M words or 15% of the full training set. In an active learning setting, FDA5 minimizes the human effort by identifying the most informative sentences for translation and FDA gains up to 0.45 BLEU points using 3/5 of the available training data compared to using all of it and 1.12 BLEU points compared to random training set. In translation tasks involving English and Turkish, a morphologically rich language, FDA5 can gain up to 11.52 BLEU points compared to a randomly selected subset of the same size, can achieve the same BLEU score using as little as 4% of the data compared to random instance selection, and can exceed the full dataset result by 0.78 BLEU points. FDA5 is able to reduce the time to build a statistical machine translation system to about half with 1M words using only 3% of the space for the phrase table and 8% of the overall space when compared with a baseline system using all of the training data available yet still obtain only 0.58 BLEU points difference with the baseline system in out-of-domain translation.

Improved recovery and reconstruction of DEFLATEd files

This paper presents a method for recovering data from files compressed with the DEFLATE algorithm where short segments in the middle of the file have been corrupted, yielding a mix of literal bytes, bytes aligned with literals across the corrupted segment, and co-indexed unknown bytes. An improved reconstruction algorithm based on long byte n-grams increases the proportion of reconstructed bytes by an average of 8.9% absolute across the 21 languages of the Europarl corpus compared to previously-published work, and the proportion of unknown bytes correctly reconstructed by an average of 20.9% absolute, while running in one-twelfth the time on average. Combined with the new recovery method, corrupted segments of 128–4096 bytes in the compressed bit-stream result in reconstructed output which differs from the original file by an average of less than twice the number of bytes represented by the corrupted segment. Both new algorithms are implemented in the trainable open-source ZipRec 1.0 utility program.

Cardinality pruning and language model heuristics for hierarchical phrase-based translation

In this article we present two novel enhancements for the cube pruning and cube growing algorithms, two of the most widely applied methods when using the hierarchical approach to statistical machine translation. Cube pruning is the de facto standard search algorithm for the hierarchical model. We propose to adapt concepts of the source cardinality synchronous search organization as used for standard phrase-based translation to the characteristics of cube pruning. In this way we will be able to improve the performance of the generation process and reduce the average translation time per sentence to approximately one quarter. We will also investigate the cube growing algorithm, a reformulation of cube pruning with on-demand computation. This algorithm depends on a heuristic for the language model, but this issue is barely discussed in the original work. We analyze the behaviour of this heuristic and propose a new one which greatly reduces memory consumption without costs in runtime or translation performance. Results are reported on the German---English Europarl corpus.

Annotating abstract anaphora

In this paper, we present first results from annotating abstract (discourse-deictic) anaphora in German. Our annotation guidelines provide linguistic tests for identifying the antecedent, and for determining the semantic types of both the antecedent and the anaphor. The corpus consists of selected speaker turns from the Europarl corpus. To date, 100 texts have been annotated according to these guidelines. The annotations show that anaphoric personal and demonstrative pronouns differ with respect to the distance to their antecedents. A semantic analysis reveals that, contrary to suggestions put forward in the literature, referents of anaphors do not tend to be more abstract than the referents of their antecedents.

Negation and lexical morphology across languages: Insights from a trilingual translation corpus

This paper proposes an exploratory cross-linguistic bird’s eye-view of negative lexical morphology by examining English, French and Italian negative derivational affixes. More specifically, it aims to uncover the French and Italian equivalents of the English affixes de , dis , in , non , un and less . These include morphological equivalents (i.e. negative prefixes in French and Italian) as well as non-morphological equivalents (i.e. single words devoid of negative affixation, multi-word units or paraphrases). The study relies on a nine-million-word trilingual translation corpus made up of texts from the Europarl corpus and shows that the systematic analysis of translation data makes it possible to identify the major morphological dissimilarities between the three languages investigated. The frequent use of non-morphological translations in French and Italian reflects fundamental differences between the source language (English) and the two target lan-guages (French and Italian), hence pointing to possible translation difficulties. Morphological translations, on the other hand, bring to light cross-linguistic similarities in the use of negative affixes.