abstract The objective of this work is to analyze transformer-based machine translation models. It aims to test the feasibility of using trained models based on specialized corpus. For the training of such model, a parallel English-French corpus was built with seven texts related to the Convention of 25 October 1980 on the Civil Aspects of International Child Abduction. The translation results obtained by the trained model were compared with those produced by Google Translate. For the evaluation stage, sacreBLEU automatic evaluation and human evaluation methods were used. The outcome of the automatic evaluation of sentences produced by the trained model was, on average, higher than those generated by the non-trained model. The human evaluation of the sentences revealed that there were adequacy errors in the use of language specific to the subject matter of the 1980 Hague Convention both in sentences generated by the trained model and in sentences generated by the Google Translate model. transformerbased transformer EnglishFrench English French 2 198 Abduction stage used average nontrained non 19 1

Resumo O presente trabalho possui como objetivo a análise dos modelos de tradução automática baseados em Transformer. Em específico, a pesquisa visa ao teste da viabilidade do uso de modelos treinados a partir de corpus especializado. Para o treinamento do modelo, foi construído um corpus paralelo inglês-francês a partir de sete textos da Convenção de 25 de outubro de 1980 sobre os Aspectos Civis do Rapto Internacional de Crianças. Os resultados de tradução obtidos pelo modelo treinado foram comparados com aqueles produzidos pelo Google Tradutor. Para a etapa de avaliação foram utilizados os métodos de avaliação automática sacreBLEU e avaliação humana. Os resultados da avaliação automática de frases produzidas pelo modelo treinado foram, em média, mais positivos que aqueles gerados pelo modelo não treinado. A avaliação humana das frases revelou que houve erros de adequação no uso da linguagem específica à matéria da Convenção da Haia de 1980 tanto em frases geradas pelo modelo treinado, quanto em frases geradas pelo modelo do Google Tradutor. Transformer específico especializado inglêsfrancês inglês francês 2 198 Crianças Tradutor média 19 1

Abstract The current demand for specialized translation, technological advances, use of CAT tools, and delivery times have impacted both product and process, thereby fostering post-editing services. Post-editing refers to the adjustments of raw products by an automatic translator (Alves et al., 2016). Consequently, there has been an interest in studying product quality and various linguistic features including syntactic, morphosyntactic, and semantic aspects (Popovic et al., 2014; Koponen et al., 2012; Temnikova, 2010) without an examination of the cognitive processes involved in reading for post-editing. This paper aims to describe specialized text comprehension for post-editing purposes. Data was collected using an eye-tracker and a retrospective protocol from three translators. Acting as post-editors, participants were assigned to post-edit a 227-word text from the health area. Findings reveal that specialized text comprehension involves attention and working memory (Cowan, 1999, 2005; Cowan et al., 2021) during the processing of terminological units (TUs). Additionally, factors such as the field of knowledge, expertise, experience, and level of education might also influence the post-editing process. translation advances tools process postediting post editing services Postediting Post Alves al al. 2016. 2016 . 2016) Consequently syntactic morphosyntactic Popovic 2014 2012 Temnikova 2010 postediting. editing. purposes eyetracker eye tracker translators posteditors, posteditors editors, editors post-editors postedit edit 227word word 227 area Cowan, (Cowan 1999 2005 2021 TUs. TUs (TUs) Additionally knowledge expertise experience 201 22 199 200 202 (TUs 20 2 19 1

Resumen La demanda actual de traducción especializada, los avances tecnológicos, el uso de herramientas y los plazos de entrega han influido en el producto y el proceso, fomentando los servicios de posedición. La posedición se refiere a los ajustes realizados a los productos arrojados por un traductor automático (Alves et al., 2016). Por consiguiente, ha surgido un interés por estudiar la calidad del producto y cuestiones lingüísticas como aspectos sintácticos, morfosintácticos y semánticos (Popovic et al., 2014; Koponen et al., 2012; Temnikova, 2010), sin investigar los procesos cognitivos implicados en la lectura con fines de posedición. El objetivo de este artículo es describir la comprensión de un texto especializado con fines de posedición. La recolección de la información se realizó mediante seguimiento ocular y protocolo retrospectivo, a tres traductores. Los participantes poseditaron un texto de 227 palabras del ámbito de salud. Los resultados mostraron un involucramiento de la atención y la memoria de trabajo (Cowan, 1999, 2005; Cowan et al., 2021) durante el procesamiento de unidades terminológicas (UTs). Además, factores como el campo de conocimiento, la experiencia y el nivel de escolaridad también podrían afectar el proceso de posedición. especializada tecnológicos Alves al al. 2016. 2016 . 2016) consiguiente sintácticos Popovic 2014 2012 Temnikova 2010, 2010 , 2010) retrospectivo traductores 22 salud Cowan, (Cowan 1999 2005 2021 UTs. UTs (UTs) Además conocimiento 201 2 199 200 202 (UTs 20 19 1

Abstract Automation technologies have altered media localisation workflows as much as practitioners’ workstations and habits. Subtitling systems and streaming services now often integrate built-in automatic speech recognition (ASR) engines, sometimes even combined with machine translation engines, to produce subtitles from audio tracks. The rise of post-editors in the audiovisual translation (AVT) sector, specifically subtitling, has been a reality for some time, thus triggering the need for up-to-date training methods and academic curricula. This article examines the uses and applications of editing practices for machine-generated timed transcriptions in subtitler training environments. A situated learning experience was designed for an international team of eight AVT trainees and three educators to edit raw machine-generated subtitles (both inter- and intra-lingually) for educational videos. The publication of an accessible video book by a publishing house was the ultimate objective of this project, undertaken by an international team of English- and Spanish-speaking postgraduate students and graduates. The feedback collated after this experience through an online questionnaire proved paramount to understanding the use of subtitle post-editing for ASR-produced templates in AVT education. Interestingly, most respondents believed that subtitle post-editing training, be it intralingual or interlingual, should be further embedded in translation curricula while also identifying bottlenecks that AVT educators may find useful when developing activities of this nature. practitioners habits builtin built ASR (ASR engines tracks posteditors post editors (AVT sector subtitling time uptodate up date machinegenerated generated environments both inter intralingually intra lingually intra-lingually videos project English Spanishspeaking Spanish speaking graduates postediting ASRproduced produced education Interestingly interlingual nature

Abstract: Machine translation deals with automatic translation from one natural language to another. Neural machine translation is a widely accepted technique of the corpus-based machine translation approach. However, an adequate amount of training data is required, and there is a need for the domain-wise parallel corpus to improve translational performance that shows translational coverages in various domains. In this work, a domain-specific parallel corpus is prepared that includes different domain coverages, namely, Agriculture, Government Office, Judiciary, Social Media, Tourism, COVID-19, Sports, and Literature domains for low-resource English-Assamese pair translation. Moreover, we have tackled data scarcity and word-order divergence problems via data augmentation and prior alignment concept. Also, we have contributed Assamese pretrained LM, Assamese word-embeddings by utilizing Assamese monolingual data, and a bilingual dictionary-based post-processing step to enhance transformer-based neural machine translation. We have achieved state-of-the-art results for both forward (English-to-Assamese) and backward (Assamese-to-English) directions of translation.

Resumen El objetivo de este artículo fue generar un modelo reconocimiento automático de voz (RAV) basado en la traducción de la voz humana a texto, a este proceso se lo ha considerado como una de las ramas de la inteligencia artificial. El análisis de voz permite identificar información sobre la acústica, fonética, sintáctica, semántica de las palabras, entre otros elementos que pueden identificar ambigüedad en términos, errores de pronunciación, sintáctica similar pero semántica diferente, que representan características propias del lenguaje humano. El modelo se centró en el análisis acústico de las palabras, proponiendo la generación de una metodología para reconocimiento acústico, a partir de transcripciones del habla de audios que contienen voz humana. Se utilizó la tasa de error por palabra para identificar la precisión del modelo. Los audios son llamadas de emergencia registrados por el Servicio Integrado de Seguridad ECU 911. El modelo fue entrenado con la herramienta CMUSphinx para idioma español sin conexión a internet. Los resultados mostraron que la tasa de error por palabra varía en relación con la cantidad de audios; es decir a mayor cantidad de audios menor cantidad de palabras erróneas y mayor exactitud del modelo. La investigación concluyó haciendo énfasis en la duración de cada audio como variable que afecta la precisión del modelo.

Abstract The objective of the article was to generate an Automatic Speech Recognition (ASR) model based on the translation from human voice to text, being considered as one of the branches of artificial intelligence. Voice analysis allows identifying information about the acoustics, phonetics, syntax, semantics of words, among other elements where ambiguity in terms, pronunciation errors, similar syntax but different semantics can be identified, which represent characteristics of the language. The model focused on the acoustic analysis of words proposing the generation of a methodology for acoustic recognition from speech transcripts from audios containing human voice and the error rate per word was considered to identify the accuracy of the model. The audios were taken from the Integrated Security Service ECU 911 that represent emergency calls registered by the entity. The model was trained with the CMUSphinx tool for the Spanish language without internet connection. The results showed that the word error rate varies in relation to the number of audios; that is, the greater the number of audios, the smaller number of erroneous words and the greater the accuracy of the model. The investigation concluded by emphasizing the duration of each audio as a variable that affects the accuracy of the model.

The Georgian language, the official language of Georgia, is the only written member of the Kartvelian language family, the indigenous language family of the Caucasus region. Georgian philology and lexicography have long-standing tradition, English-Georgian lexicography being no exception. Given the increasing use of ample electronic text corpora for lexicographical purposes, the team of Georgian lexicographers, working on the Comprehensive English-Georgian Dictionary (CEGD), subsequently the Comprehensive English-Georgian Online Dictionary (CEGOD), decided to compile an English-Georgian Parallel Corpus (EGPC). The aim of the project was to develop the methodology of building a parallel corpus of Georgian and assess its efficiency for Georgian bilingual lexicography. The work on the corpus is going on for over a decade. The ultimate aim is to create a standard for Georgian bilingual corpora that will be compiled in future. The article describes the content and composition of the EGPC, its structure, functionalities, search engines and so on. The article also deals with various studies conducted over years in order to assess and enhance the value, applicability and efficiency of the EGPC for the automatic or semiautomatic recognition, tagging and extraction of terminology, the compilation of terminological entries, as well as the entries for the English-Georgian Dictionary and those for the Georgian-English Learner's Dictionary, etc. Particular emphasis is laid upon the actual or potential applicability of the corpus for the lexicographical activities and for the machine translation projects. The findings of the study may be interesting for other under-resourced languages like Georgian.

Georgies, die amptelike taal van Georgie, is die enigste geskrewe lid van die Kartveliaanse taalfamilie, die inheemse taalfamilie van die Kaukasiese streek Die Georgiese taalwetenskap en leksikografie het 'n lang verbintenis waarvan die Engels-Georgiese leksikografie geen uitsondering is nie. In die lig van die toenemende gebruik van uitgebreide elektroniese tekskorpora vir leksiko-grafiese doeleindes, het die Georgiese span leksikograwe wat aan die Comprehensive English-Georgian Dictionary (CEGD), later die Comprehensive English-Georgian Online Dictionary (CEGOD), werk, besluit om 'n Engels-Georgiese Parallelle Korpus (EGPK) saam te stel. Die doel van die projek was die ontwikkeling van die metodologie vir die bou van 'n parallelle Georgiese korpus en die bepa-ling van die effektiwiteit daarvan vir die Georgiese tweetalige leksikografie. Daar word al meer as 'n dekade aan die korpus gewerk. Die uiteindelike doel is om 'n standaard vir Georgiese tweetalige korpora wat in die toekoms saamgestel sal word, te skep. Die artikel beskryf die inhoud en samestelling van die EGPK, die struktuur, funksionaliteit en soekenjins daarvan, ensovoorts. Die verskillende studies wat oor die jare uitgevoer is om die waarde, toepaslikheid en effektiwiteit van die EGPK rakende die outomatiese of semi-outomatiese herkenning, etikettering en onttrekking van terminologie, die samestelling van terminologiese inskrywings asook inskrywings vir die English-Georgian Dictionary en die Georgian-English Learner's Dictionary, ens., te bepaal en te verbeter, word in die artikel uiteengesit. Daar word spesifiek klem gelê op die werklike of potensiêle toepaslikheid van die korpus vir die leksikografiese aktiwiteite en masjienvertalingsprojekte. Die bevindings van die studie mag ook van waarde wees vir ander hulpbronskaars tale soos Georgies.

This is a report from the real world. It informs about the outcome of a project, which the author conducted during a months-long research stay at the Danish company Ordbogen where he integrated its research and development (R) team. The first part of the project was to test machine translation and find out to what extent it is usable in the compilation of bilingual lexicographical databases. The hypothesis was that the technology was not yet mature. But surprisingly, it turned out that the accuracy rate is already so high that it is worth considering how to implement it. The second part of the project aimed at further developing an idea formulated by Fuertes-Olivera et al (2018) on how to invert a dictionary without losing semantic content. The new vision is to compile a monolingual L2 database, bilingualize it to an L2-L1 database using machine translation, and then invert the relationship between L2 lemmata and L1 equivalents using the L1 definitions of the L2 lemmata as the axis. The third part of the project was to test this idea using a specially designed ad hoc program. The program automatically uploads relevant data from existing lexicographical databases, translates L2 definitions and example sentences into L1, suggests adequate L1 equivalents, and eventually inverts the relationship between the two languages. It worked, but the methodology still needs further refinement to be implementable on a large scale. The report concludes by listing some of the remaining challenges and defining the new role of the lexicographer in this type of project.

verslag uit die praktyk Daar word verslag gelewer oor die uitkoms van 'n projek, wat die outeur gedurende 'n maandelange navorsingstydperk by die Deense maatskappy Ordbogen aangevoer het waartydens hy die navorsings- en ontwikkeling- (N)-span geïntegreer het. Die eerste deel van die projek het die toets van masjienvertaling behels en om die bruikbaarheid daarvan in die saam-stel van tweetalige leksikografiese databasisse te bepaal. Die hipotese was dat die tegnologie nog nie gevorderd genoeg was nie. Dit het egter, verrassend genoeg, geblyk dat die akkuraatheidsyfer reeds so hoog was dat dit die moeite werd was om die implementering daarvan te oorweeg. Die tweede deel van die projek het die verdere ontwikkeling van 'n idee, geformuleer deur Fuertes-Olivera et al. (2018) oor die omskakeling van 'n woordeboek sonder verlies van semantiese inhoud, ten doel gehad. Die nuwe visie is om 'n eentalige L2-databasis saam te stel, dan met behulp van masjienvertaling te omskep in 'n L2-L1-databasis, en daarna die verhouding tussen L2-lemmata en L1-ekwivalente om te skakel deur die L1-definisies van die L2-lemmata as die spil te gebruik. Die derde deel van die projek was die toets van hierdie idee met 'n spesiaal ontwerpte ad hoc-program. Hierdie program laai outomaties relevante data vanuit leksikografiese databasisse, vertaal L2-definisies en -voorbeeldsinne in L1, stel gepaste L1-ekwivalente voor, en skakel uiteindelik die verhouding tussen die twee tale om. Dit was geslaagd, maar die metodologie moet nog verder verfyn word voordat dit op groot skaal geïmplementeer kan word. Die verslag word afgesluit met die lys van sommige van die oorblywende uitdagings en met die definiëring van die nuwe rol van die leksikograaf in hierdie tipe projek.

Resumo Devido a seus sistemas e subsistemas morfofonológicos, morfossintáticos, lexicais e outros de seu funcionamento, a língua árabe representa um sistema de não-catenação (ou não-cadeia de morfemas). Difere-se nesse sentido das línguas indo-européias - com sistemas de concatenação - e permanece, por assim dizer, uma linguagem bastante complexa para se gerenciar no campo do Processamento Linguístico Automático (PNL). Isto se dá especialmente quando se trata da tradução automática de fatos linguísticos que carregam elementos culturais específicos daquela língua. Os dados examinados neste artigo são indicativos da lacuna resultante de uma tradução automática dos textos jurídicos árabes para outros idiomas, como o francês ou o inglês. A natureza geneticamente diferente das línguas em questão não só coloca problemas de natureza linguística na passagem de uma língua para outra, mas também - e sobretudo - dificulta o entendimento de textos jurídicos em árabe, que carregam cargas semânticas, culturais, religiosas e civilizacionais que nem sempre refletem os mesmos referentes ou gestalt das línguas-alvo. Conclui-se que a intervenção humana nesse processo de tradução é mais do que necessária, como revela o estudo de textos legais na Arábia Saudita.

Abstract Due to its morphophonological, morphosyntactic, lexical and other systems and subsystems of its functioning, the Arabic language represents a system of non-concatenation (or non-chaining of morphemes). It differs in this sense from Indo-European languages - with concatenation systems - and remains, so to speak, a fairly complex language to manage in the field of Automatic Language Processing (NLP). This happens especially when it comes to automatically translating language facts that carry cultural elements specific to that language. The data examined in this article is indicative of the gap resulting from an automatic translation of Arabic legal texts into other languages such as French or English. The genetically different nature of the languages in question not only poses problems of linguistic nature in the passage from one language to another, but also and above all it makes the comprehension of legal texts in Arabic diffcult, since they carry semantic, cultural, religious and civilizational loads that do not always reflect the same referents or gestalt of the target languages. It is concluded that human intervention in this translation process is more than necessary as revealed by the study of legal texts in Saudi Arabia.

Résumé De par ses caractéristiques morpho-phonologiques, morphosyntaxiques, lexicales et autres systèmes et sous-systèmes de son fonctionnement, la langue arabe représente un système de non-concaténation (ou non-enchaînement des morphèmes). Elle diffère dans ce sens des langues Indo-européennes - à systèmes de concaténation - et demeure pour ainsi dire, une langue assez complexe à gérer dans le domaine du Traitement Automatique des Langues(TAL). Ceci, surtout lorsqu’il s’agit de traduire automatiquement des faits de langue porteurs d’éléments culturels propres à cette langue. Les données examinées dans cet article sont révélatrices de hiatus quant à la traduction automatique de textes juridiques arabes vers d’autres langues telles que le français ou l’anglais. L’apparentement génétique différent des langues en question pose non seulement des problèmes d’ordre linguistique dans le passage d’une langue vers une autre, mais aussi et surtout que les textes juridiques en langue arabe sont porteurs de poids sémantiques, culturels, religieux et civilisationnels qui ne reflètent pas toujours les mêmes référents ou gestalt des langues cibles. Il en est conclu que l’intervention humaine dans ce processus de traduction est plus que nécessaire comme le révèle l’étude des cas de textes juridiques en Arabie Saoudite.

RESUMO Após 64 anos de seu surgimento, a Inteligência Artificial (IA) enfrenta o complexo e o incerto dos nossos dias. Nesse período de sua existência, podem-se destacar dois “invernos” (1980 e 1993) em que as aplicações eram interessantes, mas a IA não conseguiu dar respostas adequadas: na compreensão da linguagem e no diagnóstico médico. Por outro lado, a IA apresentou duas décadas de progressos (2000 e 2010), em que as aplicações se mantiveram, mas as respostas da IA apresentaram avanços na tradução automática (Google), no reconhecimento de imagens (a primeira a utilizar a tecnologia foi a Apple, no iPhone 10), diagnóstico do câncer (IBM Watson) e carros autônomos (dentre outras, a Tesla). Este texto é fruto da revisão sistemática da literatura e da experiência da autora na área.

ABSTRACT Sixty-four years after its emergence, Artificial Intelligence (AI) faces the complexity and uncertainty of our day. During this period, two “winters” (1980 and 1993) can be highlighted where applications were interesting, but AI was not able to provide adequate responses, for example, in understanding language and in medical diagnosis. On the other hand, AI enjoyed two decades of progress (2000 and 2010), where the applications remained, but AI responses advanced in automatic translation (Google), image recognition (Apple was the first to use the technology, on its iPhone X), cancer diagnosis (IBM’s Watson) and autonomous cars (among others, Tesla). This text is the result of a systematic review of the literature and of the author’s experience in the area.

Resumo Essa pesquisa pretende analisar as estratégias usadas na pós-edição de provérbios traduzidos automaticamente do francês para o árabe pelo Francês como Língua Estrangeira (FLE). Apresentaremos primeiro as bases teóricas da tradução automática de provérbios e, em seguida, o estado da arte. O estudo utiliza uma abordagem metodológica mista. O método quantitativo inclui um questionário distribuído aos estudantes de FLE para descobrir a ferramenta de tradução mais usada na tradução automática. O método qualitativo se concentra na coleta de dados sobre erros de tradução, bem como na observação direta das estratégias implementadas na pós-edição de provérbios traduzidos pelo Google Tradutor. Os resultados do estudo mostram que o Google Tradutor é a ferramenta mais usada na tradução e que os erros na tradução automática são sintáticos, léxicos, semânticos e estilísticos. As estratégias implementadas na pós-edição são variadas, como dividir provérbios em unidades linguísticas para a busca de seus significados, usando uma linguagem pivô, o inglês, para conhecer o significado de provérbios e buscar equivalentes em recursos de referência. O estudo conclui que, embora a tradução automática de provérbios em pares de idiomas distantes, como o par franco-árabe, pareça complexa, o Google Tradutor tem sido uma ferramenta para ajudar um aprendiz de FLE a entender o significado de expressões proverbiais, a encontrar seus equivalentes e a pós-editá-los.

Abstract This research aims to analyze the strategies used in the post-editing of proverbs translated automatically from French into Arabic by French as a Foreign Language (FFL) learners. We will first present the theoretical foundations of machine translation of proverbs as well as the state of the art. The study uses a mixed methodological approach. The quantitative method includes a questionnaire distributed for FFL students to find out the most used translation tool in machine translation. The qualitative method focuses on collecting data on translation errors as well as on directly observing the strategies implemented in the post-editing of proverbs translated by Google Translate. The results of the study show that Google Translate is the most widely used tool in translation and that errors in machine translation are syntactic, lexical, semantic, and stylistic. The strategies implemented in the post-edition are varied such as splitting proverbs into linguistic units for the search of their meanings, using a pivot language, namely English to know the meaning of proverbs and searching for equivalents in reference resources. The study concludes that although the automatic translation of proverbs into distant language pairs such as the French-Arabic pair appears complex, Google Translate has been a tool to help an FFL learner to understand the meaning of proverbial expressions, to find their equivalents and to post-edit them.

Résumé Cette recherche a pour objectif d’analyser les stratégies utilisées dans la post-édition des proverbes traduits automatiquement du français en arabe par des apprenants du Français Langue Étrangère (FLE). Nous présenterons d’abord les fondements théoriques de la traduction automatique des proverbes ainsi que l’état de l’art. L’étude utilise une approche méthodologique mixte. La méthode quantitative comprend un questionnaire distribué sur les étudiantes de FLE pour connaître l’outil de traduction le plus utilisé dans la traduction automatique. La méthode qualitative est axée sur la collecte des données concernant les erreurs de traduction ainsi que sur l’observation directe des stratégies mises en oeuvre dans la post-édition de proverbes traduits par Google Translate. Les résultats de l’étude montrent que Google Translate est l’outil le plus utilisé en traduction et que les erreurs de traduction automatique sont d’ordre syntaxique, lexical, sémantique et stylistique. Les stratégies mises en oeuvre dans la post-édition sont variées : découpage des proverbes en unités linguistiques pour la recherche de sens, recours à une langue-pivot (l’anglais), et recherche des équivalents dans des ressources de référence. L’étude conclut que bien que la traduction automatique des proverbes dans les paires de langues éloignées telles que le français et l’arabe semble complexe, Google Translate a constitué un outil d’aide à la compréhension des mots constitutifs des proverbes pour des apprenants FLE, ce qui a permis la compréhension du sens des expressions proverbiales, de trouver leurs équivalents et de les post-éditer.

This paper describes a Luganda text normalization module, a crucial component needed for a Luganda Text to Speech system. We describe the use of a rule-based approach for detection, classification and verbalization of Luganda text. At the core of this module are the Luganda grammar rules that were hand-built to normalize Non-Standard Words (NSWs) from different semiotic and noun classes. Input text is first analyzed, matched against handcrafted patterns developed using regular expressions to detect any NSWs. Upon detection, NSWs are tokenized and classified into one of the semiotic classes and then if necessary, into one of the Luganda noun classes. These are subsequently verbalized, each according to its semiotic as well as noun class, and a new text file is produced. We tested the module with 7 datasets and achieved average detection and normalization rates of 82% and 77.7% respectively.

Abstract: In this paper, we present a study of an automatic error analysis in the context of machine translation into Arabic. We have created a pipeline tool allowing evaluation of machine translation outputs and identification of errors. A statistical analysis based on cumulative link models is performed also in order to have a global overview about errors of statistical machine translation from English to Arabic, and to investigate the relationship between encountered errors and the human perception of machine translation quality. As expected, this analysis demonstrates that the impact of lexical, semantic and reordering errors is more significant than other errors related to the fluency of the machine translation outputs.

ABSTRACT The aim of this work was to develop a web application for analysisand automatic translation Guaraní - Spanish and Spanish - Guaraní. It was need to design a lexical, syntactic and semantic analyzer, develop the necessary modules for the proposed web application, implement a translation scheme directed by lexical, semantic and synthetic analyzers, add a database with a designed dictionary, integrate a simple interface to the web application. For this purpose, the WAMPSERVER tools with the MySQL and Eclipse database was used for the JAVA programming language. For the analyzers there was an algorithm with a translation approach based on corpus, automatic translation implementation based on examples. The lexical corpus used in the database was obtained from texts suggested by the Secretaria de Políticas Lingüísticas (SPL), textbooks suggested by the Ministerio de Educación y Ciencias, glossaries and unofficial dictionaries printed or published on the internet. Six essential grammar of the Guarani language distinguished by the director of Ateneo de Lengua Guaraní, used to empirically verify the phrases included in the database were distinguished. Finally, a web application could be developed where it can be obtained automatic translation of words, phrases or sentences from Spanish to Guaraní and from Guaraní to Spanish.

RESUMEN Esta investigación tuvo como objetivo desarrollar una aplicación web de análisis y traducción automática Guaraní - Español y Español - Guaraní. Se necesitó diseñar un analizador léxico, sintáctico y semántico, elaborar los módulos necesarios para la aplicación web propuesta, implementar un esquema de traducción dirigida mediante los analizadores léxico, semántico y sintáctico, agregar una base de datos con un diccionario propuesto, integrar una interfaz sencilla a la aplicación web. Para dicho objetivo se utilizaron las herramientas WAMPSERVER con la base de datos MySQL y Eclipse para el lenguaje de programación JAVA. Para los analizadores se utilizó un algoritmo con enfoque de traducción basada en corpus, implementando traducción automática basada en ejemplos. El corpus léxico utilizado en la base de datos fue obtenido de textos sugeridos por la Secretaría de Políticas Lingüísticas (SPL), libros de textos sugeridos por el Ministerio de Educación y Ciencias, glosarios y diccionarios no oficiales impresos o publicados en internet. Se logró distinguir seis gramáticas esenciales de la Lengua Guaraní aprobadas por el director de la Ateneo de Lengua Guaraní, utilizadas para comprobar de manera empírica las frases incluidas en la base de datos. Finalmente se logró desarrollar una aplicación web donde de acuerdo al usuario se puede obtener traducción automática de palabras, frases u oraciones del Español al Guaraní y del Guaraní al Español.

Resumo Este artigo apresenta o COPA-TRAD Versão 2.0, um sistema com base em corpus paralelo desenvolvido na Universidade Federal de Santa Catarina (UFSC) para a pesquisa, ensino e prática da tradução. O COPA-TRAD permite que o usuário investigue as práticas de tradutores profissionais por meio da identificação de padrões tradutórios relacionados a um determinado elemento ou padrão linguístico. Além disso, o sistema permite a comparação entre a tradução humana e a tradução automática fornecida por serviços amplamente conhecidos na Internet (Google Translate, Microsoft Translator e Yandex). Atualmente, o COPA-TRAD prevê cinco subcorpora (Literatura Infantojuvenil, Textos Literários, Metadiscurso em Tradução, Legendas e Textos Jurídicos) e disponibiliza as seguintes ferramentas: concordanciador paralelo, concordanciador monolíngue, lista de palavras e, uma DIY Tool[,] que permite ao usuário criar seu próprio corpus descartável. O sistema apresenta também uma interface que oferece uma ferramenta de POS-tagging que analisa e classifica as categorias gramaticais de um texto.

Abstract This paper describes COPA-TRAD Version 2.0, a parallel corpus-based system developed at the Universidade Federal de Santa Catarina (UFSC) for translation research, teaching and practice. COPA-TRAD enables the user to investigate the practices of professional translators by identifying translational patterns related to a particular element or linguistic pattern. In addition, the system allows for the comparison between human translation and automatic translation provided by three well-known machine translation systems available on the Internet (Google Translate, Microsoft Translator and Yandex). Currently, COPA-TRAD incorporates five subcorpora (Children's Literature, Literary Texts, Meta-Discourse in Translation, Subtitles and Legal Texts) and provides the following tools: parallel concordancer, monolingual concordancer, wordlist and a DIY Tool that enables the user to create his own parallel disposable corpus. The system also provides a POS-tagging tool interface to analyze and classify the parts of speech of a text.

Abstract The essay provides a critical assessment of the performance of two distinct machine translation systems, Systran and Google Translate. First, a brief overview of both rule-based and statistical machine translation systems is provided followed by a discussion concerning the issues involved in the automatic and human evaluation of machine translation outputs. Finally, the German translations of Mark Twain’s The Awful German Language translated by Systran and Google Translate are being critically evaluated highlighting some of the linguistic challenges faced by each translation system.