Parallel Corpora

"This article presents a corpus-based study of the metaphorical and metonymical use of the words "head" and "heart," together with the Norwegian correspondents "hode" and "hjerte." The continuum between metaphor and metonymy is explored, as well as the extent to which the non-literal expressions containing "head"/"heart" and "hode"/"hjerte" correspond to each other. The data analyzed is gathered from the English-Norwegian Parallel Corpus.
It is necessary to differentiate various meanings of these expressions in order to pinpoint cross-linguistic similarities and differences. Although a relatively high degree of mutual correspondence is found, the relationship is said to be asymmetrical, especially as regards "head"/"hode", since English seems to employ more metaphorical meanings of this lexeme."

This paper proposes a new approach for collecting lexical and grammatical data: one that meets the need to control the features to be elicited, while ensuring a fair level of idiomaticity. The method, called conversational questionnaires, consists in eliciting speech not at the level of words or of isolated sentences, but in the form of a chunk of dialogue. Ahead of fieldwork, a number of scripted conversations are written in the area’s lingua franca, each anchored in a plausible real-world situation – whether universal or culture-specific. Native speakers are then asked to come up with the most naturalistic utterances that would occur in each context, resulting in a plausible conversation in the target language. Experience shows that conversational questionnaires provide a number of advantages in linguistic fieldwork, compared to traditional elicitation methods. The anchoring in real-life situations lightens the cognitive burden on consultants, making the fieldwork experience easier for all. The method enables efficient coverage of various linguistic structures at once, from phonetic to pragmatic dimensions, from morphosyntax to phraseology. The tight-knit structure of each dialogue makes it an effective tool for cross-linguistic comparison, whether areal, historical or typological. Conversational questionnaires help the linguist make quick progress in language proficiency, which in turn facilitates further stages of data collection. Finally, these stories can serve as learning resources for language teaching and revitalization. Five dialogue samples are provided here as examples of such questionnaires. Every linguist is encouraged to write their own dialogues, adapted to a region’s linguistic and cultural profile. Ideally, a set of such texts could be developed and made standard among linguists, so as to create comparable or parallel corpora across languages – a mine of data for typological comparison.

François, Alexandre. 2019. A proposal for conversational questionnaires. In Aimée Lahaussois & Marine Vuillermet (eds.), Methodological Tools for Linguistic Description and Typology. Special issue of Language Documentation & Conservation 16, 155-196.

Contrastive methods have long been employed in lexicography, in particular in bi-and multilingual dictionary projects. The main rationale for this is the necessity to comprehensively study, i.e. compare and contrast, two or more linguistic systems that are to be presented in one way or another in respective dictionaries. Similarly, the contrastive approach is of paramount importance in terminographic undertakings, on account of the need to draw a distinction between terminological (conceptual) systems existing in various languages and across cultures. It must be emphasised, however, that the contrastive element is not only a part of terminographic practice, but also of the theory of terminography. This article aims to present the role of contrastive research across various spheres of specialised (=LSP) lexicography.

We present HindEnCorp, a parallel corpus of Hindi and English, and HindMonoCorp, a monolingual corpus of Hindi in their release version 0.5. Both corpora were collected from web sources and preprocessed primarily for the training of statistical machine translation systems. HindEnCorp consists of 274k parallel sentences (3.9 million Hindi and 3.8 million English tokens). HindMonoCorp amounts to 787 million tokens in 44 million sentences. Both the corpora are freely available for non-commercial research and their preliminary release has been used by numerous participants of the WMT 2014 shared translation task.

In this thesis we describe and evaluate a tool for automatic generation of translations for multiword English terms into Spanish from a monolingual specialized Spanish corpus, compiled by means of web crawling. The resulting translations may be further used to expand or to revise the existing lexicographic resources, i.e. language for specific purposes dictionaries, terminological databases and translation memories. We evaluate the output with the precision and recall metrics and apply our approach to the small English-Spanish glossary in the legal domain that serves both as a reference dictionary and a source for input multiword terms. We analyze the obtained results and suggest possible solutions to the detected problems based on the theoretical part of this research devoted to the issues of machine translation, the use of dictionaries and corpora (non-parallel corpora in particular) for machine translation tasks, and the works on automatic terminology extraction.

We report on a project to annotate biblical texts in order to create an aligned multilingual Bible corpus for linguistic research, particularly computational linguistics, including automatically creating and evaluating translation lexicons and semantically tagged texts. The output of this project will enable researchers to take advantage of parallel translations across a wider number of languages than previously available, providing, with relatively little effort, a corpus that contains careful translations and reliable alignment at the near-sentence level. We discuss the nature of the text, our annotation process, preliminary and planned uses for the corpus, and relevant aspects of the Corpus Encoding Standard (CES) with respect to this corpus. We also present a quantitative comparison with dictionary and corpus resources for modern-day English, confirming the relevance of this corpus for research on present day language.

This paper describes the first phase of the CEXI project at the University of Bologna in Forlì, involving the selection of the texts to be included in the corpus and decisions about the processing of these texts. The aim of the project is to create a resource which can be used by both students and researchers to learn about translation and translating. The English Italian Translational Corpus can be described as a bi-directional, parallel, translation-driven corpus, which in its core component will consist of about 4.6 million words, or 368 text samples of 10 to 15 thousand words each, from Italian original texts and their translations into English and vice versa, published between 1975 and 2000. The core corpus will subsequently be flanked with additional unidirectional parallel collections which will better reflect the specific characteristics of the two very different translation populations sampled, as well as expanded to include full texts and text types excluded from the core component. The paper deals with issues such as representativeness, balance and directionality with respect to the Italian and English language book publishing sector, detailing the composition of the different sub-components of CEXI, and showing that the creation of a corpus involves a series of compromises between what is ideally desirable and what is possible given practical and theoretical limitations.

Automatic extraction of bilingual lexicons from parallel corpora has been recently exploited to overcome the knowledge acquisition bottleneck in a number of research areas in natural language processing, such as machine translation (MT) and cross-language information retrieval. In this paper the author presents a method for automatic extraction of bilingual lexicons from a parallel Arabic-English corpus annotated with part-of-speech tags for potential use in a full-scale MT system. The extraction method, which does not make use of an initial bilingual dictionary, is based on the statistical technique of co-occurrence frequency in the parallel corpus. In addition, dependency relations for some parallel syntactic constructions are made use of to automatically extract head-dependent translation pairs which are then filtered to obtain one-word translation seeds. These seeds are used as anchor points to resegment the parallel corpus in order to bootstrap the lexicon extraction process. Experimental results show that the accuracy of the extracted lexicons was improved after applying the bootstrapping techniques.

The feature selection or extraction is the most important task in Opinion mining and Sentimental Analysis
(OSMA) for calculating the polarity score. These scores are used to determine the positive, negative, and
neutral polarity about the product, user reviews, user comments, and etc., in social media for the purpose
of decision making and Business Intelligence to individuals or organizations. In this paper, we have
performed an experimental study for different feature extraction or selection techniques available for
opinion mining task. This experimental study is carried out in four stages. First, the data collection process
has been done from readily available sources. Second, the pre-processing techniques are applied
automatically using the tools to extract the terms, POS (Parts-of-Speech). Third, different feature selection
or extraction techniques are applied over the content. Finally, the empirical study is carried out for
analyzing the sentiment polarity with different features.

The paper presents an attempt to propose an exact method for identifying the so-called "language-specific" lexicon, a controversial notion often reasonably questioned. An aligned bilingual parallel corpus is chosen as an instrument for finding " specificity" , and statistical entropy and other indices are used as markers of the dispersion of translation patterns (viz. stimuli). For example, a word can be deemed (maximally) language-specific if it occurs multiple times in a given bilingual corpus and is translated each time in a different way. A word is minimally (or simply not) language-specific if it is translated each time identically. Some problems relative to the application of this method are discussed. These data can be explicitly used in bilingual dictionaries.

The aim of this paper is to investigate Polish equivalents of English phrasal verbs as found in an English-Polish (E-P) parallel corpus PHRAVERB. Given the semantic idiosyncrasy exhibited by phrasal verbs, it is assumed that the equivalents generated by PHRAVERB will often differ from those found in E-P dictionaries. The qualitative corpus analysis aims to show that arriving at the desirable Polish counterpart involves a detailed semantic breakdown of the English structure, a careful analysis of the context in which it is used, as well as linguistic and translation skills, necessary to detect the nuances and subtleties of meaning in both languages. PHRAVERB is used to analyze the lexicographic potential (LP) of corpus equivalents. Four levels of LP have been established – high, average, low and zero – to evaluate which corpus-derived equivalents are eligible for inclusion in E-P dictionaries. To this end, 2,514 occurrences of PVs in the parallel corpus, with their equivalents, have been identified and analyzed.

This paper concentrates on the verbal moods used after Spanish adverbs expressing potentiality (quizá(s), tal vez, probablemente, posiblemente). With the use of the corpus CREA, we sought to determine whether there is a preference for either the indicative or subjunctive after each adverb, and, consequently, whether a reason for the preference of either mood exists. With the assistance of the parallel corpus InterCorp we have been able to present a summary of Czech translations of these grammatical constructions, in order to show the possibilities of reflecting the semantic difference between Spanish subjunctive and indicative in Czech. Based upon these observations it is possible to create a scale that compares potentiality in Czech and Spanish.

The sentences in the RNC are aligned sentence -by -sentence. The texts kindly offered for the use in the RNC by Adrian Barentsen and included into the Amsterdam Slavic Parallel Aligned Corpus multilingual corpus are already aligned paragraph -by -paragraph. This segmentation has been additionally refined by us semi -automatically, introducing boundaries between the sentences within these pre -defined paragraphs. For the majority of the texts the alignment is undertaken completely by the RNC team. The original text's sentence segmentation overrides the segmentation in the translation. So, each single element of parallel tagging corresponds to the original sentence, whilst its translated counterpart(s) may well be either a part of a sentence starting with a space or a comma, or more than one sentence.

The paper describes three studies concerned with inner-Slavic variation in the use of different functional categories, two of which involve verbal aspect and one of which involves reflexive coding. The leading interest behind this research is (a) to understand convergences and divergences between the Slavic languages from a usage based perspective and (b) to develop a method for empirically approaching this issue using a parallel corpus. This research focuses on variation in a set of closely related standard language varieties of a single genus, and may be considered intermediate between a dialectological and a typological approach.

In this paper we present a method for term extraction that can be used in classroom with translation students. The terms are extracted from a multilingual parallel corpus with the aid of a parallel concordancer, AntPConc. Our work is focused on English-Greek medical terminology translation, but the method can be transferred to other language pairs and other fields of terminology.

This study investigates formal and functional variation of analytic causatives (ACs) in eighteen European languages from the Indo-European and Uralic language families. Employing the comparative concept approach to language comparison, the paper presents a probabilistic semantic map of the main functions of ACs on the basis of a multilingual parallel corpus of film subtitles. This method enables us to detect common dimensions of semantic variation of ACs and to pinpoint cross-linguistic commonalities in the form-meaning mapping. The paper also presents several case studies, which test previous hypotheses about the grammaticalization clines in Romance and Germanic and facts of language contact between German and Slavic languages. The role of language contact is further explored in quantitative analyses which compare how the languages ‘carve up’ the semantic space of causation. The results of this comparison suggest a hypothesis that frequently occurring semantically vague ACs may be regarded as a feature of Standard Average European.

Nous proposons une méthode de découverte et de compilation des normes de traduction des concepts spécialisés employés dans des termes simples et complexes attestés dans un corpus parallèle bilingue. Les normes de traduction mises au jour par cette méthode ont la particularité d’être fondées sur l’usage et prennent appui sur des solutions de traduction éprouvées. Celles-ci sont essentielles à l’enseignement des compétences en traduction spécialisée telles que proposées par le groupe PACTE et généralement acceptées en traductologie. Notre méthode consiste à analyser la traduction des occurrences spécialisées de business en économie et en finance réunies dans un corpus bilingue constitué d’un échantillon d’occurrences aléatoires obtenues au moyen d’un concordancier en ligne. L’analyse repose sur trois catégories d’annotations et leurs corrélations : l’acception de business, la fonction de business dans le syntagme nominal et les modalités de traduction de business. Cette méthode d’analyse peut être facilement étendue à l’ensemble des concepts spécialisés de nature nominale qui sont des unités distinctives des textes de spécialité.


We propose a method for the discovery and compilation of translation standards of specialized concepts used in simple and complex terms attested in a parallel bilingual corpus. The standards of translation developed by this method have the characteristic of being based on usage and are based on proven translation solutions. These are essential to the teaching of specialized translation skills, as proposed by the PACTE group and generally accepted in translation studies. Our method consists in analysing the translation of specialized tokens of business in economics and finance collected in a bilingual corpus made up of a sample of random occurrences generated with an online concordancer. The analysis is based on three categories of annotations and their correlations: senses of business, functions of business in the noun phrase and translation modes of business. This method of analysis can be easily extended to all specialized concepts of a nominal nature, which are distinctive units of specialized texts.

La lingüística histórica, en su camino hacia la consagración como disciplina autónoma, no ha podido, o no ha querido, distanciarse de las corrientes anejas que transitan y evolucionan en el seno de una lingüística más general y abarcadora. Dentro de esa corriente se encuentra la denominada Lingüística de corpus, marco en el que, desde hace no mucho más de veinte años, viene desarrollándose una nueva corriente de estudio de raigambre anglosajona denominada Corpus-based Translation Studies (CTS). Los estudios fundamentados en esta relativamente nueva disciplina se orientan a la consideración y análisis de hechos de lengua contemporáneos, tal y como cabría esperar dada su relación con la lingüística de corpus. En este artículo se describirán los postulados más interesantes de los trabajos amparados en las teorías y métodos de los CTS y, desde una perspectiva crítica, se insistirá en la dimensión histórica de las lenguas y en los aspectos discursivo-tradicionales muchas veces olvidados o interesadamente obviados en pos del establecimiento de leyes o reglas universales que den validez al paradigma teórico propuesto.

The present study investigates the cross-linguistic differences in the use of so-called T/V forms (e.g. French tu and vous, German du and Sie, Russian ty and vy) in ten European languages from different language families and genera. These constraints represent an elusive object of investigation because they depend on a large number of subtle contextual features and social distinctions, which should be cross-linguistically matched. Film subtitles in different languages offer a convenient solution because the situations of communication between film characters can serve as comparative concepts. I selected more than two hundred contexts that contain the pronouns you and yourself in the original English versions, which are then coded for fifteen contextual variables that describe the Speaker and the Hearer, their relationships and different situational properties. The creators of subtitles in the other languages have to choose between T and V when translating from English, where the T/V distinction is not expressed grammatically. On the basis of these situations translated in ten languages, I perform multivariate analyses using the method of conditional inference trees in order to identify the most relevant contextual variables that constrain the T/V variation in each language.

Multilingual corpora, containing the same documents in a variety of languages, are becoming an essential resource for natural language processing. Clustering multilingual corpora provides us with an insight into the differences between languages when term frequency based information retrieval (IR) tools are used. It also allows one to use the natural language processing (NLP) and IR tools in one language to implement IR for another language. For instance, in this way, the most relevant articles to be translated from language X to language Y can be selected after studying the clusters of abstracts in language Y. In this article, we report on our work on applying hierarchical agglomerative clustering (HAC) to a large corpus of documents where each appears both in Bulgarian and English. We cluster these documents for each language and compare the results both with respect to the shape of the tree and content of clusters produced. On the data available, the results of clustering one language resemble the other, provided the number of clusters required is relatively small. Further, we study the effects of reducing the set of terms used for clustering. This step appears a viable strategy for English, but is not acceptable for Bulgarian. Finally, we describe an experiment employing a genetic algorithm to fine-tune the individual term weights in order to reproduce more closely a predefined set of clusters. In this way, clustering becomes a supervised learning technique that is trained to better reproduce known clusters in language X when applied to the corresponding documents in language Y. Other possible applications include training the algorithm on a hand-clustered set of documents, and subsequently applying it to a superset, including unseen documents, incorporating in this way expert knowledge about the domain in the clustering algorithm.

The merging of corpus linguistic methods and digital technology can provide new ways of representing medieval digital texts. In this paper, we introduce a multi-layered parallel Old Occitan-English corpus. We show how parallel alignment can help overcome some challenges associated with historical manuscripts. Furthermore, we apply a resource-light method of building an emotion annotation via parallel alignment. Finally, using visualization tools such as ANNIS and GoogleViz, we demonstrate how our parallel corpus can be queried and visualized dynamically via modern language.

Accessing historical texts is often a challenge because readers either do not know the historical language, or they are challenged by the technological hurdle when such texts are available digitally. Merging corpus linguistic methods and digital technology can provide novel ways of representing historical texts digitally and providing a simpler access. In this paper, we describe a  multi-dimensional parallel Old Occitan-English corpus, in which word alignment serves as the basis for search capabilities as well as for the transfer of annotations.  We show how parallel alignment can help overcome some challenges of historical manuscripts.  Furthermore, we apply a resource-light method of building an emotion annotation via parallel alignment, thus showing that such annotations are possible without speaking the historical language. Finally, using visualization tools, such as ANNIS and GoogleViz, we demonstrate how the emotion analysis can be queried  and visualized dynamically in our parallel corpus, thus showing that such information can be made accessible with low technological barriers.

Demand for Chinese-to-English translation has increased over recent years. In contrast, resources for training translators for Chinese-to-English are few although increasing now, relative to English-to-Chinese for example. Corpus-based techniques are now more widely acknowledged as being appropriate for the study of translation. A number of Chinese/English parallel translation corpora have been built and applied to the research of translation practice.

Canonical question tags feature prominently in spoken English, where they display great versatility. At face value they are meant to elicit a response from a co-participant in the form of (dis)agreement with the proposition to which the tag has been added. Their pragmatic scope is, however, considerably broader: they serve as politeness strategies but also emphasize the speaker's convictions or mark accusations. Like many other languages, Dutch does not have a similar structure, which raises questions as to what devices Dutch employs to serve the same purpose as question tags. This contrastive study examines such correspondents in a parallel corpus of English novels and their Dutch translations. Three structures can be identified: pragmatic markers, clause-final parentheticals and combinations of these. The data indicate a preference for pragmatic markers (most notably "hè" and "toch"), which predominantly appear either as invariant tags or clause-medially, indicating subtle shifts in utterance interpretation.

This paper describes the characteristics of Brazilian Portuguese (BP) thetic sentences by means of a parallel corpus study consisting of the original dialogues of two Argentinean movies from 2004 and 2010 and the corresponding doubling into BP. The notion of theticity used comes from Sasse (1987), with some contributions from Lambrecht (2000), Sasse (2006) and Smit (2010). The criteria used to identify this kind of assertion also comes from these authors. The use of a parallel corpus to identify thetic sentences in BP comes from the fact that Spanish is considered a thetic-marking language and in Brazilian Portuguese the identification of these sentences is not clear. The results show that theticity and word order are not related in Brazilian Portuguese and that subjects of thetic sentences in this language are mainly preverbal and bear nominative case (as well as subjects of categorical sentences). These results suggest that the detopicalization strategy reported for thetic sentences in the literature may be related to prosody (not studied in this paper) or it may not be relevant in this language.

Generating source code API sequences from an English query using Machine Translation (MT) has gained much interest in recent years. For any kind of MT, the model needs to be trained on a parallel corpus. In this paper we clean STACKOVERFLOW, one of the most popular online discussion forums for programmers, to generate a parallel English-Code corpus from Android posts. We contrast three data cleaning approaches: standard NLP, title only, and software task extraction. We evaluate the quality of the each corpus for MT. To provide indicators of how useful each corpus will be for machine translation, we provide researchers with measurements of the corpus size, percentage of unique tokens, and per-word maximum likelihood alignment entropy. We have used these corpus cleaning approaches to translate between English and Code [22, 23], to compare existing SMT approaches from word mapping to neural networks [24], and to reexamine the "natural software" hypothesis [29]. After cleaning and aligning the data, we create a simple maximum likelihood MT model to show that English words in the corpus map to a small number of specific code elements. This model provides a basis for the success of using StackOverflow for search and other tasks in the software engineering literature and paves the way for MT. Our scripts and corpora are publicly available on GitHub [1] as well as at https://search.datacite.org/works/10.5281/zenodo.2558551.

Anotace
Several parallel corpora built from European Union language resources are presented here. They were processed by state-of-the-art tools and made available for researchers in the Sketch Engine corpus management system. A completely new resource is introduced: EUR-Lex corpus, being one of the largest parallel corpus available at the moment, containing 840 million tokens of English and having the largest language pair (English-French) with more than 25 million aligned segments (paragraphs).