Search

ATLASLang is a machine translation system from Arabic text language into Arabic sign language (ArSL). The first version of the system (Brour and Benabbou, 2019) is based on two approaches: rule-based Interlingua and example-based approaches. It is a classical machine translation system, its limitations are the linguistic knowledge necessary to develop the rules, also many rules and exceptions required. In the last few years, a neural machine translation has achieved notable results, and several well-known companies including Google (Wu et al., 2016) and Systran (Crego et al., 2016) have starting to exploit it. In this paper, we present a new version of ATLASLang that is implemented using a feed-forward back-propagation Artificial Neural Network. In this version, we have started to translate simple sentences composed from a limited number of words since generally communication with deaf person uses short sentences. We have used the same ATLASLang MTS database of signs and we have utilized morphological characteristics to derive the maximum information from each word in the input of the neural translation system. The two versions of the system are compared using n-gram BLEU score (Papineni et al., 2002), and the results demonstrate that neuron network approaches outperform the other classical approaches.

One of the key problems in comparative studies based on frame semantics is the question whether frames can become an interlingua. This paper argues that not only single frames, but their systems or frame semantic domain representations consisting of frames and their relations are also useful in comparative studies. Such a system of frames helps one explain why seemingly unrelated expressions in different languages find a common denominator in higher-order frames, thus becoming semantic-pragmatic equivalents. To support this argument, an analysis of Polish, English and German lease agreements as parallel texts is conducted and the benefits of this approach to comparative studies are presented. The study is in line with the recent FrameNet initiatives, such as the Global FrameNet and automatic translation studies. However, it differs in some methodological aspects. Instead of using FrameNet as the given lexical resource, domain specific frames are defined starting from common general concepts of the analyzed semantic domain. A text-based approach rather than a comparison of bi-sentences or phrases is adapted. The work thus introduces a new approach to comparative studies based on frame semantics and frame semantic research. It also follows the recent research trend of adding a pragmatic dimension to frame semantic analysis by analyzing frames in context.

syntax is an interlingual representation used in compilers. Grammatical Framework (GF) applies the abstract syntax idea to natural languages. The development of GF started in 1998, first as a tool for controlled language implementations, where it has gained an established position in both academic and commercial projects. GF provides grammar resources for over 40 languages, enabling accurate generation and translation, as well as grammar engineering tools and components for mobile and Web applications. On the research side, the focus in the last ten years has been on scaling up GF to wide-coverage language processing. The concept of abstract syntax offers a unified view on many other approaches: Universal Dependencies, WordNets, FrameNets, Construction Grammars, and Abstract Meaning Representations. This makes it possible for GF to utilize data from the other approaches and to build robust pipelines. In return, GF can contribute to data-driven approaches by methods to transfer resources from one language to others, to augment data by rule-based generation, to check the consistency of hand-annotated corpora, and to pipe analyses into high-precision semantic back ends. This article gives an overview of the use of abstract syntax as interlingua through both established and emerging NLP applications involving GF.

ARTEMIS (Automatically Representing Text Meaning via an Interlingua-based System), is a natural language processing device, whose ultimate aim is to be able to understand natural language fragments and arrive at their syntactic and semantic representation. Linguistically, this parser is founded on two solid linguistic theories: the Lexical Constructional Model and Role and Reference Grammar. Although the rich semantic representations and the multilingual character of Role and Reference Grammar make it suitable for natural language understanding tasks, some changes to the model have proved necessary in order to adapt it to the functioning of the ARTEMIS parser. This paper will deal with one of the major modifications that Role and Reference Grammar had to undergo in this process of adaptation, namely, the substitution of the operator projection for feature-based structures, and how this will influence the description of function words in ARTEMIS, since they are strongly responsible for the encoding of the grammatical information which in Role and Reference Grammar is included in the operators. Currently, ARTEMIS is being implemented for the controlled natural language ASD-STE100, the Aerospace and Defence Industries Association of Europe Simplified Technical English, which is an international specification for the preparation of technical documentation in a controlled language. This controlled language is used in the belief that its simplified nature makes it a good corpus to carry out a preliminary testing of the adequacy of the parser. In this line, the aim of this work is to create a catalogue of function words in ARTEMIS for ASD-STE100, and to design the lexical rules necessary to parse the simple sentence and the referential phrase in this controlled language.

The present study aims at creating a level of collaboration between MT and HT in order to obtain more production in translation projects and save time, money and effort for translators, businessmen and any other non-specialist MT users. This can be carried out by applying both pre-editing and post-editing on the ST before entering to MT, in a specific way. Hence, the present study is based on approaching the techniques in which MT runs and to what extent we can make it more effective and more productive. MT users should understand that MT includes two types; direct and indirect. Direct type works on the word level while indirect types work on higher levels (i.e. phrase, sentence levels). On the other hand, indirect levels are divided into two approaches; transfer approach which is language -dependent and interlingua approach which is non-linguistic specific. Indirect approaches are capable of translating better than direct approach. A MT user should pay heed to the fact that these approaches run better because they are based on ST analysis and then TT generation. In order to benefit more from MT, facilitate translation process, and make the TT seem as natural as possible, each MT user should pay heed to the fact that the more he simplifies the analysis process, the more he gets more natural translation. A MT user can achieve this simplicity in the analysis process by means of pre-editing ST before submitting it to MT. In fact, pre-editing cannot be performed randomly. Hence, the present study encourages MT users to use the rules of controlling language to simplify and adjust STs according to their business requirements. This, in turn, alleviates the load on post-editing because the ST to be entered to MT becomes easy enough to be understood by MT.

Arabic Sign Language (ArSL) is a visual language used by deaf Arabs and some hard of hearing to translate their thinking. It is a language in its own right as well as spoken languages such as Arabic or English. It is produced by body, face, and gestures of the hands. Hundreds of thousands of deaf people around the world currently practice it. Each country has its own sign language that deaf people use so it is not universal, but deaf people from different countries communicate easily with each other after a short period of adaptation. This communication becomes difficult when a deaf want to communicate with a normal person with no interpreter available. In this way, building a translation system that can generate real-time statements via a signing avatar is helpful. The system to be developed is a machine translation system from Arabic text to the Arabic sign language, it will allow hearing persons to communicate more easily, using a text written in Arabic by normal person, with people suffering from hard hearing and knowing only sign language. The system must perform a morpho-syntactic analysis of the text in the input and convert it to video sequences phrases playing by a 3D human avatar who expresses the usual signs used by deaf people. In this paper, we present a first version of our machine translation system ATLASLang MTS 1 based on rule-based Interlingua and example-based approaches. It uses SAFAR Platform [11] and ALKHALIL morpho system [1] to extract the morphological properties of each word of the input sentence. Then it generates a video sequence representing the sentence in Arabic sign language based on well-established translation rules and the database of signs. During the translation, if a word of sentence is a proper noun or does not have a correspondence in the database, it will be finger spelled. The main constraint imposed by our machine translation system is the assumption that the sentence given in the input must be completely vowelized.

The strategic borderless knowledge sharing and development of communication interacts with dialects. Significant factors such as education, medical, business, research and others are vastly diffused over the world based on various lingoes. Bilingual or multilingual expression is the standard of having unknown/new linguistic along with its resources. The initial endeavor of the study is to implement the MT (machine translation) approaches for English to Bangla language processing and vice-versa. The emphasis of the study is the distinct ambiguities are identified along with their best solutions. Certain machine translation approaches such as word-to-word, direct, transfer, interlingua, corpus-based and statistical translation are surviving and, few of them are deployed in this Smart Natural Language Processing (SNLP) for dispatching the source to the target language and vice-versa. Two different dictionaries (bilingual and monolingual) are developed for the execution process. An eminent resource Stanford POS Tagger (as a toolkit) is used for identifying the grammatical structure of the source (English) dialect. This research also focuses on output acquiring through performance analyzing of different translation models.

For most of human history, language was a spoken phenomenon. Only recently in cosmic terms have we developed systems for written communication, which has enabled us to communicate over long distances and to preserve information over long periods of time. Our written languages are mostly derived from spoken language and shared experiences, so it is very unlikely that a natural language or its written form will be mutually understandable with an alien species. We should not expect anything like spoken language in interstellar communication, except maybe when providing examples of animal communication. On the other hand, one can communicate a great deal using sets and symbolic networks and can do so using a minimal amount of data. This sort of symbolic communication system also forms the basis for more abstract language, with common mathematical terms and logic as its foundation. This is known as a constructed language or interlingua.

Compact representation of sentences like feature vectors, offer better understanding of the sentence formation. Majority applications in natural language processing often requires the help of such meaningful representations. If an interlingua is constructed for the same, it will be useful for the applications like machine translation, text summarization etc. Here we are proposing a deep learning based Interlingua whereby creating a structured meaningful representation for a Malayalam document.

Multilingual neural machine translation (NMT) has led to impressive accuracy improvements in low-resource scenarios by sharing common linguistic information across languages. However, the traditional multilingual model fails to capture the diversity and specificity of different languages, resulting in inferior performance compared with individual models that are sufficiently trained. In this paper, we incorporate a language-aware interlingua into the Encoder-Decoder architecture. The interlingual network enables the model to learn a language-independent representation from the semantic spaces of different languages, while still allowing for language-specific specialization of a particular language-pair. Experiments show that our proposed method achieves remarkable improvements over state-of-the-art multilingual NMT baselines and produces comparable performance with strong individual models.

State-of-the-art multilingual machine translation relies on a universal encoder-decoder, which requires retraining the entire system to add new languages. In this paper, we propose an alternative approach that is based on language-specific encoder-decoders, and can thus be more easily extended to new languages by learning their corresponding modules. So as to encourage a common interlingua representation, we simultaneously train the N initial languages. Our experiments show that the proposed approach outperforms the universal encoder-decoder by 3.28 BLEU points on average, while allowing to add new languages without the need to retrain the rest of the modules. All in all, our work closes the gap between shared and language-specific encoderdecoders, advancing toward modular multilingual machine translation systems that can be flexibly extended in lifelong learning settings.

Machine Translation is considered as integral part of natural language processing. The translation from one language to another has always been a challenging issue in terms of percentage of the sentences that can be translated and the rate of accuracy of translated sentences. Rule-based machine translation is a definitive approach toward machine translation. It helps to translate statements from one language (source) to other (target) language on the basis of the defined rules. The software tool ‘EtranS’ provides translation of English to Sanskrit language. It has so far reported more than ninety percent of accuracy. The accuracy obtained in translation is large because of mathematical approach of Sanskrit grammar. Since it is a formulized language; therefore, it has great potential to act as Interlingua. This explores the potential of Sanskrit language to act as Interlingua. A generic framework for multiple language translation is presented here. Sanskrit can act as Interlingua for many Indian as well as foreign languages. Sanskrit language is mother of eighty-five percent of Indian languages and its grammar has resemblance with foreign languages like German, Russian, etc. The ‘EtranS’ can be further extended in order to obtain multilingual translation.

We propose a simple solution to use a single Neural Machine Translation (NMT) model to translate between multiple languages. Our solution requires no changes to the model architecture from a standard NMT system but instead introduces an artificial token at the beginning of the input sentence to specify the required target language. Using a shared wordpiece vocabulary, our approach enables Multilingual NMT systems using a single model. On the WMT’14 benchmarks, a single multilingual model achieves comparable performance for English→French and surpasses state-of-theart results for English→German. Similarly, a single multilingual model surpasses state-of-the-art results for French→English and German→English on WMT’14 and WMT’15 benchmarks, respectively. On production corpora, multilingual models of up to twelve language pairs allow for better translation of many individual pairs. Our models can also learn to perform implicit bridging between language pairs never seen explicitly during training, showing that transfer learning and zero-shot translation is possible for neural translation. Finally, we show analyses that hints at a universal interlingua representation in our models and also show some interesting examples when mixing languages.

Introduction: the process of globalisation has strengthened the position of the English language as a means of communication in all spheres of life, including scientific communication. The expansion of one language not only necessitates changes in the status of other national languages and the emergence of a hierarchical relationship between them, but also significantly affects the political and economic balance of power in the world. The global dominance of English in science not only confers distinct advantages on its native speakers but also discriminates against scholars from non-Anglophone societies. As a result, a threat arises concerning the loss to humanity of unique, culture-specific ways of understanding reality. Materials and Methods: on the basis of an analysis of modern trends and literature review, such mani¬festations of linguistic imperialism in the field of academic communication as the IMRaD format, CLIL teaching technologies and English academic writing centres are revealed. Subsequently, these phenomena are investigated using empirical sociological methods: in-depth expert interviews, participant observation and the content study of chemistry papers indexed in Scopus. Results: it is demonstrated that the Anglophone societies use the global distribution of the English language to advance their competitive advantage in the field of science. The implementation of English language instruction in higher education and Anglophone communicative patterns in scholarly communication – particularly with regard to the representation of research results – might have a negative effect both on the development of researchers’ competencies and their future effectiveness in advancing science. Discussion and Conclusions: it is concluded that an increased awareness of potential threats caused by the dominance of the English language in scientific communication is needed among all the participants of scientific communication, including higher school lecturers. This can be achieved by using bilingual and bicultural educational approaches.

This paper briefly describes the top-scoring submission to the First TextWorld Problems: A Reinforcement and Language Learning Challenge. To alleviate the partial observability problem, characteristic to the TextWorld games, we split the Agent into two independent components: Observer and Actor, communicating only via the Interlingua of the RDF* graph database. The RDF* graph database serves as the “world model” memory incrementally updated by the Observer via FrameNet informed Natural Language Understanding techniques and is used by the Actor for the efficient exploration and planning of the game Action sequences. We find that the deep-learning approach works best for the Observer component while the Actor policy is better served by backtracking over the set of rules.

Automatic detection of sentiments is considered among complex problems in social applications. In information security, emojis are used in several interfaces for user authentication, antropomorphic secure access and remote communication. The use of emojis in multimodal information triggers new challenges in complex networks and mobile security applications. The fast growth of social media, microblogs, floods expands the definition of sentimental analysis where the extraction of emotions from user posts becomes a cutting edge. Therefore, the opinion mining becomes a crucial step for the analysis of social behaviour in individuals or groups for the detection of trends. In current applications, the language of emojis is considered as a common way or an interlingua to express the ideas or intensify feelings. However, there are few studies to reveal its effects on Turkish context for overlapped and separate senses. In this study, emojis have been classified as a parameter of textual descriptions for the emotions in Turkish language. The emotion analysis has been performed by Support Vector Machines (SVM) and multinomial Naive Bayes (NB) using test and train sets derived from Twitter corpus. The preparation and preprocessing of the corpus have been accomplished by generating the classifiers; groups and emotions. The neutral emotion state has been also added to compare the accuracy levels in classification. The use of corpus in a generic domain present a promising field where different emotion states have been measured. The evaluation scores indicate that SVM would perform better and neutral emotional emojis might decrease total accuracy in Turkish language.

Multilingual transfer learning can benefit both high- and low-resource languages, but the source of these improvements is not well understood. Cananical Correlation Analysis (CCA) of the internal representations of a pre- trained, multilingual BERT model reveals that the model partitions representations for each language rather than using a common, shared, interlingual space. This effect is magnified at deeper layers, suggesting that the model does not progressively abstract semantic con- tent while disregarding languages. Hierarchical clustering based on the CCA similarity scores between languages reveals a tree structure that mirrors the phylogenetic trees hand- designed by linguists. The subword tokenization employed by BERT provides a stronger bias towards such structure than character- and word-level tokenizations. We release a subset of the XNLI dataset translated into an additional 14 languages at https://www.github.com/salesforce/xnli_extension to assist further research into multilingual representations.

Mobile applications become the basics of human’s daily life. People worldwide face difficulties in communication due to different languages and culture barriers. So it is important to enable people to communicate faster by bringing people together from across the globe. عrbeng is an Arabic/English Android instant messaging translator application to communicate worldwide that makes the world just got a lot smaller. With the feature of live translation capabilities, عrbeng allows exchanging text messages, voice and video chat as well as online and offline file transmission with anyone regardless of language barriers. عrbeng is an Interlingual based machine translation system. With Interlingua عrbeng becomes an unambiguous and international translation application that based on a client/server Socket programming. عrbeng uses an integrated language platforms that offers comprehensive online communication functions. عrbeng performs live translation with high translation performance, less time consuming and with more enhanced quality of translation.

In this paper, we present our hybrid methodology for building a bidirectional Amazigh-English machine translation. The architecture of the proposed system is based on both Interlingua-Based Machine Translation (IBMT) and Statistical-Based Machine Translation (SBMT) approaches. Amazigh is a less-resourced language. It does not have parallel corpora with enough size. So, using statistical approach for such language will not be a good choice, because this approach requires large parallel corpora to well train probabilistic models, and to ensure a translation of good quality. Since we dispose of an Amazigh IBMT based deconverter, we thought, firstly, to use it in building an Amazigh-English parallel corpus. This latter have been exploited to train the necessary models in SBMT.

This paper addresses an open challenge in educational data mining, i.e., the problem of automatically mapping online courses from different providers (universities, MOOCs, etc.) onto a universal space of concepts, and predicting latent prerequisite dependencies (directed links) among both concepts and courses. We propose a novel approach for inference within and across course-level and concept-level directed graphs. In the training phase, our system projects partially observed course-level prerequisite links onto directed concept-level links; in the testing phase, the induced concept-level links are used to infer the unknown course-level prerequisite links. Whereas courses may be specific to one institution, concepts are shared across different providers. The bi-directional mappings enable our system to perform interlingua-style transfer learning, e.g. treating the concept graph as the interlingua and transferring the prerequisite relations across universities via the interlingua. Experiments on our newly collected datasets of courses from MIT, Caltech, Princeton and CMU show promising results.

This paper presents challenges being faced in designing automatic translation software. There are many approaches to automatic translation like Direct, Rule based, Transfer based, Statistical based and Interlingua. A brief description of all the approaches and their advantages and drawbacks are discussed. Universal Networking Language (UNL) based on Interlingua approach can be used especially for translation among multiple languages because it requires knowledge of UNL and of the language which user wants UNL to support. User can then get translated text in any of the languages supported by UNL without even being oblivious to any other language. It is less expensive approach also. This paper also gives brief introduction to UNL and how it can overcome many of the challenges in translation.

This paper presents the different machine translation approaches. We distinguish two classes: the classical approaches (Direct, Transfer-Based and Interlingua) and the corpus-based approaches (Memory, Example and Statistical). We describe languages challenges such as morphology, syntax and structure. We illustrate a previous systems related to foreign (American, British and European) sign language machine translation. We list a review of the most important works related to Arabic Sign Language (ArSL) machine translation.

In this paper, we explore the learning of neural network embeddings for natural images and speech waveforms describing the content of those images. These embeddings are learned directly from the waveforms without the use of linguistic transcriptions or conventional speech recognition technology. While prior work has investigated this setting in the monolingual case using English speech data, this work represents the first effort to apply these techniques to languages beyond English. Using spoken captions collected in English and Hindi, we show that the same model architecture can be successfully applied to both languages. Further, we demonstrate that training a multilingual model simultaneously on both languages offers improved performance over the monolingual models. Finally, we show that these models are capable of performing semantic cross-lingual speech-to-speech retrieval., to appear at ICASSP 2018

Lojban, a constructed interlingua, has a small number of predicates–totaling 1342. In our prior independent work, involving lojban for machine translation in parallel, when converting English into lojbanic English (our extensions), the number of test cases was too small, because of the lack of a rich lojban vocabulary. There is a procedure for creating new, composite, lojban predicates in terms of existing predicates. Although workable, it lacks an architectural framework that is consistent with English word-sense usage, and enables efficient production of thousands of new predicates.

We incorporate an explicit neural interlingua into a multilingual encoder-decoder neural machine translation (NMT) architecture. We demonstrate that our model learns a language-independent representation by performing direct zero-shot translation (without using pivot translation), and by using the source sentence embeddings to create an English Yelp review classifier that, through the mediation of the neural interlingua, can also classify French and German reviews. Furthermore, we show that, despite using a smaller number of parameters than a pairwise collection of bilingual NMT models, our approach produces comparable BLEU scores for each language pair in WMT15., Accepted in WMT 18

As stated in Perinan-Pascual and Arcas Tunez (2010, p. 2671) FunGramKB´s lexical and grammatical levels are grounded in sound linguistic theories, allowing the system to capture syntactic-semantic generalizations which are able to provide both explanations and predictions of language phenomena. The linguistic foundation of FunGramKB is inspired on Role and Reference Grammar (RRG, Van Valin, 2005) and the Lexical Constructional Model (LCM, Ruiz de Mendoza & Mairal Uson, 2008; Mairal Uson & Ruiz de Mendoza, 2008). The close interrelation between these grammatical theories and FungramKB is patent in the design of ARTEMIS (Automatically Representing Text Meaning via an Interlingua-Based System”), an NLP system whose objective is the simulation of natural language understanding. In doing so, ARTEMIS becomes a development of the syntax-to-semantics linking algorithm proposed in RRG which involves a parsing process that starts with a natural language sentence, extracts its morphosyntactic features and provides a representation of these in terms of the so-called layered structure of the clause (LSC) in RRG. Because of its centrality, there is a vital need for a full development of ARTEMIS. This paper aims at contributing to this elaboration by proposing the set of rules and operations necessary for the treatment of Yes/No interrogative sentences within such an NLP system.

The syntactic parser ARTEMIS (Automatically Representing Text Meaning via an Interlingua-based System) is a prototype intended to process natural language within the environment of the Functional Grammar Knowledge Base (FunGramKB) (Perinan-Pascual & Arcas-Tunez, 2010). Different from other parsing devices, ARTEMIS is grounded on two functional linguistic models: Role and Reference Grammar (RRG) (Van Valin & La Polla, 1997; Van Valin, 2005) and the Lexical Constructional Model (LCM) (Ruiz de Mendoza & Mairal Uson, 2008). However, certain adjustments have to be made to both models in order to meet the requirements derived from the fact that ARTEMIS follows the paradigm of Unification Grammars, in such a way that to comply with this paradigm, the GDE (Grammar Development Environment) within ARTEMIS needs to integrate two types of constructs: a catalogue of Attribute-Value Matrixes (AVMs) to describe grammatical units, and a set of production rules (grammatical, lexical and constructional) to allow it to produce a feature based grammar. It is the aim of this paper to give an overview of the investigation carried out so far within ARTEMIS in relation to these two aspects. We will do so by revisiting the literature in relation to the adjustments made to the linguistic models, especially to the RRG, and by reviewing the efforts made to describe the units and design the rules necessary for the parsing of simple sentences in English. Our paper will conclude by pointing at prospective research needed for the completion of this project.

This paper describes the design and implementation of a computational model for Arabic natural language semantics, a semantic parser for capturing the deep semantic representation of Arabic text. The parser represents a major part of an Interlingua-based machine translation system for translating Arabic text into Sign Language. The parser follows a frame-based analysis to capture the overall meaning of Arabic text into a formal representation suitable for NLP applications that need for deep semantics representation, such as language generation and machine translation. We will show the representational power of this theory for the semantic analysis of texts in Arabic, a language which differs substantially from English in several ways. We will also show that the integration of WordNet and FrameNet in a single unified knowledge resource can improve disambiguation accuracy. Furthermore, we will propose a rule based algorithm to generate an equivalent Arabic FrameNet, using a lexical resource alignment of FrameNet1.3 LUs and WordNet3.0 synsets for English Language. A pilot study of motion and location verbs was carried out in order to test our system. Our corpus is made up of more than 2000 Arabic sentences in the domain of motion events collected from Algerian first level educational Arabic books and other relevant Arabic corpora.

A comparison of interlingua and query translation is proposed in a particular cross-language information retrieval (CLIR) application which consists on retrieving a book from the collection by using one of its chapters in a different language as a query. The experiments are performed in three languages (English, Chinese and Spanish) and all the possible combinations. It is shown that interlin- gua is able to outperform the query translation approach in some cross-language tasks. Results are further analysed and it is found that, for this particular task, the quality of translation (in terms of BLEU and PER) is not directly correlated with the query translation performance.

By addressing both text-to-AMR parsing and AMR-to-text generation, SemEval-2017 Task 9 established AMR as a powerful semantic interlingua. We strengthen the interlingual aspect of AMR by applying the multilingual Grammatical Framework (GF) for AMR-to-text generation. Our current rule-based GF approach completely covered only 12.3% of the test AMRs, therefore we combined it with state-of-the-art JAMR Generator to see if the combination increases or decreases the overall performance. The combined system achieved the automatic BLEU score of 18.82 and the human Trueskill score of 107.2, to be compared to the plain JAMR Generator results. As for AMR parsing, we added NER extensions to our SemEval-2016 general-domain AMR parser to handle the biomedical genre, rich in organic compound names, achieving Smatch F1=54.0%., This work was supported in part by the Latvian State research programmes SOPHIS and NexIT, the H2020 project SUMMA (grant No. 688139), and the European Regional Development Fund grant No. 1.1.1.1/16/A/219.

End-to-end neural machine translation has overtaken statistical machine translation in terms of translation quality for some language pairs, specially those with large amounts of parallel data. Besides this palpable improvement, neural networks provide several new properties. A single system can be trained to translate between many languages at almost no additional cost other than training time. Furthermore, internal representations learned by the network serve as a new semantic representation of words -or sentences- which, unlike standard word embeddings, are learned in an essentially bilingual or even multilingual context. In view of these properties, the contribution of the present work is two-fold. First, we systematically study the NMT context vectors, i.e. output of the encoder, and their power as an interlingua representation of a sentence. We assess their quality and effectiveness by measuring similarities across translations, as well as semantically related and semantically unrelated sentence pairs. Second, as extrinsic evaluation of the first point, we identify parallel sentences in comparable corpora, obtaining an F1=98.2% on data from a shared task when using only NMT context vectors. Using context vectors jointly with similarity measures F1 reaches 98.9%., 11 pages, 4 figures

This paper is based on the work towards developing a machine translation system for Sanskrit to English. An interlingua based machine translation system in Paninian framework was developed in this work. As lexical functional grammar can handle semantic information along with syntactic analysis, the proposed system uses lexical functional grammar in different levels of analysis phase. In this system the given Sanskrit text is first converted to an intermediate notation called interlingua. This notation is then used for mapping to the target language, English, and generates the translated output. The proposed system works in the translation of simple and complex sentences in the given corpus. The karaka analysis system of Panini is used for the semantic analysis.

This document shows the lexical and syntactic study made to evaluate the distinguishing features of the Mexican Sign Language (MSL), it also shows the work accomplished in a model for automatic translation from written Spanish to MSL. The aims of this work were to encourage the conditions of a greater social inclusion of deaf and hearing impaired people, it also looked and contributed to the creation of linguistic resources in order to support further studies of Mexican Sign Language. This project allowed the development of a corpora of signs of MSL, an small interlingua collocations dictionary between Spanish an MSL, and a dictionary of synonyms in MSL. There were identified essentials cases of MSL with regard to Spanish. The application was developed using the system of translation by rules, and lexical and syntactic transfer. It required the design of a model with validation of basic syntactic structures and equivalences rules in Spanish.

This is the Lump team participation at SemEval 2017 Task 1 on Semantic Textual Similarity. Our supervised model relies on features which are multilingual or interlingual in nature. We include lexical similarities, cross-language explicit semantic analysis, internal representations of multilingual neural networks and interlingual word embeddings. Our representations allow to use large datasets in language pairs with many instances to better classify instances in smaller language pairs avoiding the necessity of translating into a single language. Hence we can deal with all the languages in the task: Arabic, English, Spanish, and Turkish.

Concepts and relations in ontologies and in other knowledge organisation systems are usually annotated with natural language labels. Most ontology matchers rely on such labels in element-level matching techniques. State-of-the-art approaches, however, tend to make implicit assumptions about the language used in labels (usually English) and are either domain-agnostic or are built for a specific domain. When faced with labels in different languages, most approaches resort to general-purpose machine translation services to reduce the problem to monolingual English-only matching. We investigate a thoroughly different and highly extensible solution based on semantic matching where labels are parsed by multilingual natural language processing and then matched using language-independent and domain aware background knowledge acting as an interlingua. The method is implemented in NuSM, the language and domain aware evolution of the SMATCH semantic matcher, and is evaluated against a translation-based approach. We also design and evaluate a fusion matcher that combines the outputs of the two techniques in order to boost precision or recall beyond the results produced by either technique alone.

This paper presents a deconverting generator for Malayalam language using Universal Networking Language (UNL) for Machine Translation. UNL being an Interlingua representation, conveyed as directed hyper graph with relations and attributes of source language sentence. A set of Universal Words are generated from the source language with its semantic representation, are mapped to UNL features. The work involves identifying the dependent features like syntactic, semantic and lexical features of target language. UNL Relations, UNL Attributes and Universal Word (UW), which are the building blocks of UNL are identified and mapped to the dependent features of Malayalam. Lexical mapping of UWs to root words of Malayalam was done through UNL-Malayalam Word Dictionary. The deconversion is tested against 100 Malayalam Sentences that has achieved an appreciable F-measure score of 0.978. . General Terms Malayalam Deconversion, Universal Networking Language, Interlingua Machine Translation.

Interlingua-based machine translation is probably the most attractive among the three classic approaches to MT. Early pioneers as well as current researchers experimented with this approach and produced some very stimulating methodologies to reaching such a language-independent framework. In this paper, we shall briefly review some of the most renowned endeavours in interlingua-based machine translation and bring into view how the latest of which; the Universal Networking Language (UNL) differs and compares to these other systems.

Made available in DSpace on 2014-09-30T18:18:36Z (GMT). No. of bitstreams: 0 Previous issue date: 2013-06-01Bitstream added on 2014-10-01T14:03:15Z : No. of bitstreams: 1 S1984-63982013000200008.pdf: 178474 bytes, checksum: 546f59bdcc80b366931b6f98a83ddcf0 (MD5) O artigo apresenta, sob o prisma da Teoria da Complexidade, as características do processo de aprendizagem de espanhol como língua estrangeira em Teletandem. Foram coletados dados de dois pares de interagentes português-espanhol, que travaram uma interação regular e sistemática, baseada nos princípios tandem. Verificou-se que a experiência de aprendizagem se desenvolve com peculiaridades decorrentes do contexto, agentes, elementos e suas nuanças, o que revelou a presença de um espaço raso entre o sistema da língua materna e o da língua estrangeira. This article presents, under the perspective of Complexity Theory, the characteristics of the learning process of Spanish as a foreign language in Teletandem. Data were collected from two pairs of Portuguese-Spanish interagents, who were engaged in a systematic and regular interaction, based on the tandem principles. It was found that the learning experience is developed with the peculiarities that arise from the context, agents, members and their nuances, which revealed the presence of a shallow space between the systems of native and foreign languages. Universidade Estadual Paulista Universidade Estadual Paulista

Universal Networking Language (UNL) has been used by various researchers as an Interlingua approach for AMT (Automatic machine translation). The UNL system consists of two main components/tools, namely, EnConverter-IAN (used for converting the text from a source language to UNL) and DeConverter - EUGENE (used for converting the text from UNL to a target language). This paper highlights the DeConversion generation rules used for the DeConverter and indicates its usage in the generation of Punjabi sentences. This paper also covers the results of implementation of UNL input by using DeConverter-EUGENE and its evaluation on UNL sentences such as Nouns, Pronouns and Prepositions.

Artificial international auxiliary languages, particularly the major ones such as Esperanto and Interlingua, have taken relatively few words or grammatical features from Uralic languages (or from other non-Indo-European languages). In this paper I will examine some auxiliary languages which have used one or more Uralic languages as sources. Among these languages are Angos, Ardano, Budinos, Eura, Kweda, Lingwa de Planeta, Neo Patwa, Uusisuom/Lapsi, and Veltparl. In general Uralic items make up only a small part of the language as a whole.

The key for enabling smooth information sharing across channels is the efficient storage of the information in a language independent format so that it can be retrieved across many geographies without having multiple corpus. This paper explains how this complex requirement can be solved by using UNL. In the paper the architecture and implementation of the UNL interlingua based question answering system is given that is capable enough to handle user queries against a language-independent representation of document. Currently the system solves factoid based questions as where, what, why, which etc using english as source language but later can be extended to other languages once their enconvertors and deconvertors are available. The proposed model is tested on ELOSS documents achieving an accuracy of 82.22%.

The derivation of formal specifications from large and complex requirements is a key challenge in systems engineering. In this paper we present an approach that aims to address this challenge by building formal models from OWL ontologies. An ontology is used in the field of knowledge representation to capture a clear view of the domain and to produce a concise and unambiguous set of domain requirements. We harness the power of ontologies to handle inconsistency of domain requirements and produce clear, concise and unambiguous set of domain requirements for Event-B modelling. The proposed approach works by generating Attempto Controlled English (ACE) from the OWL ontology and then maps the ACE requirements to develop Event-B models. ACE is a subset of English that can be unambiguously translated into first-order logic. There is an injective mapping between OWL ontology and a subset of ACE. ACE is a suitable interlingua for producing the mapping between OWL and Event-B models for many reasons. Firstly, ACE is easy to learn and understand, it hides the math of OWL and would be natural to use by everybody. Secondly ACE has a parser that converts ACE texts into Discourse Representation Structures (DRS). Finally, ACE can be extended to target a richer syntactic subset of Event-B which ultimately would facilitate the translation of ACE requirements to Event-B.

The Argument Interchange Format (AIF) has been devised in order to support the interchange of ideas and data between different projects and applications in the area of computational argumentation. The AIF presents an abstract ontology for argumentation which serves as an interlingua between various reifications that consist of more concrete argumentation languages. In this paper, we aim to give a logical reification of the AIF ontology, by defining translations between the ontology’s language and the formal ASPIC framework for argumentation. We thus lay foundations for interrelating formal logic-based approaches to argumentation captured by the general ASPIC framework, and the wider class of AIF reifications, including those that are more informal and user orientated.

Wordnets have been created in many languages, revealing both their lexical commonalities and diversity. The next challenge is to make multilingual wordnets fully interoperable. The EuroWordNet experience revealed the shortcomings of an interlingua based on a natural language. Instead, we propose a model based on the division of the lexicon and a language-independent, formal ontology that serves as the hub interlinking the language-specific lexicons. The ontology avoids the idiosyncracies of the lexicon and furthermore allows formal reasoning about the concepts it contains. We address the division of labor between ontology and lexicon. Finally, we illustrate our model in the context of a domain-specific multilingual information system based on a central ontology and interconnected wordnets in seven languages.

This paper reports the work for the EnConversion of input Punjabi sentences to an interlingua representation called Universal Networking Language (UNL). The UNL system consists of two main components, namely, EnConverter (used for converting the text from a source language to UNL) and DeConverter (used for converting the text from UNL to a target language). This paper discusses the framework for designing the EnConverter for Punjabi language with a special focus on generation of UNL attributes and relations from Punjabi source text. It also describes the working of Punjabi Shallow Parser used for the processing of the input sentence, which performs the tasks of Tokenizer, Morph-analyzer, Part-of-Speech Tagger and Chunker. This paper also considers the seven phases used in the process of EnConversion of input Punjabi text to UNL representation. The paper highlights the EnConversion analysis rules used for the EnConverter and indicates its usage in the generation of UNL expressions. This paper also covers the results of implementation of Punjabi EnConverter and its evaluation on sample UNL sentences available at Spanish Language Server. The accuracy of the developed system has also been presented in this paper.

Given a user question, the goal of a Question Answering (QA) system is to retrieve answers rather than full documents or even best-matching passages, as most Information Retrieval systems currently do. In this paper, we present BRUJA, a QA system for the management of multilingual collections. BRUJ rkstions (English, Spanish and French). The BRUJA architecture is not formed with three monolingual QA systems but instead uses English as Interlingua to make usual QA tasks such as question classifications and answer extractions. In addition, BRUJA uses Cross Language Information Retrieval (CLIR) techniques to retrieve relevant documents from a multilingual collection. On the one hand, we have more documents to find answers from but on the other hand, we are introducing noise into the system because of translations to the Interlingua (English) and the CLIR module. The question is whether the difficulty of managing three languages is worth it or whether a monolingual QA system delivers better results. We report on in-depth experimentation and demonstrate that our multilingual QA system gets better results than its monolingual counterpart whenever it uses good translation resources and, especially, CLIR techniques that are state-of-the-art.