Multilingual spoken dialog system

Formal specification of Multilingual Alphanumeric Ordering Systems

This work presents the development of the translation component in a multistage, multilevel, multimode, multilingual and dynamic deliberative (M4D2) system, built to facilitate automated moderation and translation in the languages of five European countries: Italy, Ireland, Germany, France and Poland. Two main topics were to be addressed in the deliberation process: (i) the environment and climate change; and (ii) the economy and inequality. In this work, we describe the development of neural machine translation (NMT) models for these domains for six European languages: Italian, English (included as the second official language of Ireland), Irish, German, French and Polish. As a result, we generate 30 NMT models, initially baseline systems built using freely available online data, which are then adapted to the domains of interest in the project by (i) filtering the corpora, (ii) tuning the systems with automatically extracted in-domain development datasets and (iii) using corpus concatenation techniques to expand the amount of data available. We compare our results produced by the domain-adapted systems with those produced by Google Translate, and demonstrate that fast, high-quality systems can be produced that facilitate multilingual deliberation in a secure environment.

This paper describes the ADAPT-DCU machine translation systems built for the WMT 2020 shared task on Similar Language Translation.\n\nWe explored several set-ups for NMT for Croatian–Slovenian and Serbian–Slovenian language pairs in both translation directions. Our experiments focus on different amounts and types of training data: we first apply basic filtering on the OpenSubtitles training corpora, then we perform additional cleaning of remaining misaligned segments based on character n-gram matching.\n\nFinally, we make use of additional monolingual data by creating synthetic parallel data through back-translation. Automatic evaluation shows that multilingual systems with joint Serbian and Croatian data are better than bilingual, as well as that character-based cleaning leads to improved scores while using less data.\n\nThe results also confirm once more that adding back-translated data further improves the performance, especially when the synthetic data is similar to the desired domain of the development and test set. This, however, might come at a price of prolonged training time, especially for multitarget systems.

The application of nlp techniques to improve the results of information retrieval is still considered as a controversial issue, whereas in cross-language information retrieval (clir) linguistic processing is already well established. In this paper, the clir component - Mpro-IR - which is presented has been developed as the core module of a multilingual information system in a legal domain. This component uses not only the lexical base form for indexing but also derivational information and, for German, information about the decomposition of compounds. This information is provided by a sophisticated morpho-syntactic analyser and is exploited not only for query translation but also for query expansion as well as the search and the document ranking. The objective of the clef evaluation was to assess this linguistic based retrieval approach in an unrestricted domain. The focus of the investigation was on how derivation and decomposition can contribute to improve the recall.KeywordsInformation RetrievalMachine TranslationQuery ExpansionLinguistic ProcessingLegal DomainThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Anthony Ferritto, Sara Rosenthal, Mihaela Bornea, Kazi Hasan, Rishav Chakravarti, Salim Roukos, Radu Florian, Avi Sil. Proceedings of the 28th International Conference on Computational Linguistics: System Demonstrations. 2020.

This book presents an analysis of Spanish prepositional clauses (< P + CP >) - complement and adverbial clauses. The goal is to examine the syntax and evolution of those clauses and their components in Spanish, contrasting them with other European languages. Prepositional argument and adjunct clauses are grammatical in present-day Spanish. However, Medieval Spanish only attests the latter; the former were not frequent until the 16th/17th centuries. Both types are examined in their syntactic evolution and properties, including clausal nominality, argumenthood, nature of prepositions, and optionality. Latin and Portuguese, French, and Italian - both in their present-day and past forms - are studied and compared to Spanish. Likewise, several Germanic languages are surveyed. These languages show variable grammatical degrees of < P + CP >. The comparison reveals aspects which challenge the commonly accepted conclusions about the clausal patterns of each language. This study offers a novel approach to the analysis of Spanish prepositional clauses by looking at its properties and formation not only from within but also in contrast with other languages. It argues for cross-linguistically valid categories and explanations in order to comprehend the properties of human language.

This chapter addresses the discussion about the origin of Afrikaans and the political context of this discussion. It focuses on the way settlers introduced Dutch at the Cape and how their descendants later forged the national language Afrikaans from the meanwhile changed language. It also discusses the long-denied influence of the so-called Coloured People on Afrikaans. Two views opposed each other. The first claims that Afrikaans is a European language and invokes, through careful philological research, similar phenomena in Dutch, Dutch dialects, or other Germanic languages and is therefore called the Philological School. The other emphasizes influences from native languages spoken at the Cape and from languages of imported slaves and could therefore be called the Creole School. The Philological School reached its peak during Apartheid. The Creole School only managed to gain the upper hand in the aftermath of Apartheid and initially especially outside South Africa. Finally, three examples are presented that illustrate how the discussion was conducted.

This paper explores features of the new lexical and phraseological units of German political discourse and their function in other languages. German media discourse, reflecting the policy of one of the major European states, generates many new lexemes and idioms. An important component of the neologisms is vocabulary based on politicians’ names. Deonyms are widespread in all European languages; however, their formation and function have certain differences, both morphological and semantic. The article studies the peculiarities of such new words as merkeln and Merkelism in the political discourse in different languages. While the German language is characterised by the use of verb deonyms, they are not characteristic of Slavic languages. In turn, the neologism Merkelism is being actively used by the media of almost all European countries, although its semantics vary. Relevance is the main condition for a new lexical or phraseological unit to enter the active vocabulary of another language, which is demonstrated through the neophraseme Europa der verschiedenen Geschwindigkeiten. The connotational meaning of phraseology in a given language depends on the political situation and linguocultural characteristics.

Statistical approaches in speech technology, whether used for statistical language models, trees, hidden Markov models or neural networks, represent the driving forces for the creation of language resources (LR), e.g., text corpora, pronunciation and morphology lexicons, and speech databases. This paper presents a system architecture for the rapid construction of morphologic and phonetic lexicons, two of the most important written language resources for the development of ASR (automatic speech recognition) and TTS (text-to-speech) systems. The presented architecture is modular and is particularly suitable for the development of written language resources for inflectional languages. In this paper an implementation is presented for the Slovenian language. The integrated graphic user interface focuses on the morphological and phonetic aspects of language and allows experts to produce good performances during analysis. In multilingual TTS systems, many extensive external written language resources are used, especially in the text processing part. It is very important, therefore, that representation of these resources is time and space efficient. It is also very important that language resources for new languages can be easily incorporated into the system, without modifying the common algorithms developed for multiple languages. In this regard the use of large external language resources (e.g., morphology and phonetic lexicons) represent an important problem because of the required space and slow look-up time. This paper presents a method and its results for compiling large lexicons, using examples for compiling German phonetic and morphology lexicons (CISLEX), and Slovenian phonetic (SIflex) and morphology (SImlex) lexicons, into corresponding finite-state transducers (FSTs). The German lexicons consisted of about 300,000 words, SIflex consisted of about 60,000 and SImlex of about 600,000 words (where 40,000 words were used for representation using finite-state transducers). Representation of large lexicons using finite-state transducers is mainly motivated by considerations of space and time efficiency. A great reduction in size and optimal access time was achieved for all lexicons. The starting size for the German phonetic lexicon was 12.53 MB and 18.49 MB for the morphology lexicon. The starting size for the Slovenian phonetic lexicon was 1.8 MB and 1.4 MB for the morphology lexicon. The final size of the corresponding FSTs was 2.78 MB for the German phonetic lexicon, 6.33 MB for the German morphology lexicon, 253 KB for SIflex and 662 KB for the SImlex lexicon. The achieved look-up time is optimal, since it only depends on the length of the input word and not on the size of the lexicon. Integration of lexicons for new languages into the multilingual TTS system is easy when using such representations and does not require any changes in the algorithms used for such lexicons.

Word prediction is a process that tries to guess the word a user is writing, at the same time he/she is doing it. It is mainly used to decrease the effort needed to write a text in applications devoted to people with disabilities. In this paper, we describe and evaluate the architecture of a multilingual word prediction system. The proposed architecture is modular and flexible, with common interfaces between the modules to allow the use of different prediction algorithms or even the prediction in different languages. The current system consists of a general lexicon for each language, the possibility to create and store personal lexicons, prediction methods based on words and POS (parts of speech) probabilistic grammars (when available). The system has been trained and evaluated for English, Portuguese, Spanish and Swedish. The Spanish version is currently included in a technical aid widely used for people with communication disabilities.KeywordsPrediction MethodPrediction ProcessGeneral LexiconSpelling MistakeWord SuffixThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

When comparing different literature sources, differences in the terminology concerning soil deformation due to seepage may be found. The distinctions are caused by language barrier (Slavic and Germanic language groups), by the view on the mechanisms and by the way of understanding individual soil deformation processes. These differences in terminology are related to the naming of the phenomenon itself or to translation issues. Clear terminology is important when comparing stability criteria applied in different countries. Here “western” and “eastern” schools may be distinguished. This study provides an overview on terminology for soil deformation due to seepage in English, French, German, Russian, Polish and Czech languages, some suggestions are presented with discussion. The paper is an initial phase for further comparison of criteria for individual soil deformation modes.

The provision of automatic systems that can provide conversational practice for beginners would make a valuable addition to existing aids for foreign language teaching. To achieve this goal, the SCILL (Spoken Conversational Interaction for Language Learning) project is developing a spoken dialogue system that is capable of maintaining interactive dialogues with non-native students in the target language. However, the effective realisation of the intelligent language understanding and dialogue management needed for such a system, requires robust recognition of poorly articulated non-native speech. This paper studies several popular techniques for robust acoustic modelling including HLDA, MAP and CMLLR on non-native speech data within a specific dialogue domain. In addition, a novel approach for using cross language speech data to adapt the acoustic models is described and shown to be useful when very limited non-native adaptation data is available. The experimental results provide a clear story of how to improve recognition performance on non-native speech for a specific task, and this will be of interest more generally for those developing multi-lingual spoken dialogue systems.

This paper introduces a technique for adapting GF (Grammatical Framework)-based dialogue systems. This technique can be used to adapt dialogue systems in two aspects: user adaptation and self-adaptation. By user adaptation, users can customize the system to their own needs and define alternatives for a series of information to use in later utterances. By self-adaptation, the system can update GF grammar to keep the system adequate when new situations occur. This technique is demonstrated by a multi-lingual transport query system, which allows users to find up-to-date travel plans. Adapting GF-based dialogue systems improves the functionality of speech recognizers by defining alternatives for specific phrases and also keeps the dialogue system always updated.

Few European languages have in the course of their histories undergone as radical changes as English did in the medieval period. The earliest documented variety of the language, Old English (c. 450 to 1100 ce), was a synthetic language, typologically similar to modern German, with its three genders, relatively free word order, rich case system, and verbal morphology. By the beginning of the Middle English period (c. 1100 to 1500), changes that had begun a few centuries earlier in the Old English period had resulted in a remarkable typological shift from a synthetic language to an analytic language with fixed word order, very few inflections, and a heavy reliance on function words. System-internal pressures had a role to play in these changes, but arguably they were primarily due to intensive contacts with other languages, including Celtic languages, (British) Latin, Scandinavian languages, and a little later, French. As a result, English came to diverge from its Germanic sister languages, losing or reducing such Proto-Germanic features as grammatical gender; most inflections on nouns, adjectives, pronouns, and verbs; verb-second syntax; and certain types of reflexive marking. Among the external influences, long contacts with speakers of especially Brittonic Celtic languages (i.e., Welsh, Cornish, and Cumbrian) can be considered to have been of particular importance. Following the arrival of the Angles, Saxons, and Jutes from around 450 ce onward, there began an intensive and large-scale process of language shift on the part of the indigenous Celtic and British Latin speaking population in Britain. A general wisdom in contact linguistics is that in such circumstances—when the contact is intensive and the shifting population large enough—the acquired language (in this case English) undergoes moderate to heavy restructuring of its grammatical system, leading generally to simplification of its morphosyntax. In the history of English, this process was also greatly reinforced by the Viking invasions, which started in the late 8th century ce, and brought a large Scandinavian-speaking population to Britain. The resulting contacts between the Anglo-Saxons and the Vikings also contributed to the decrease of complexity of the Old English morphosyntax. In addition, the Scandinavian settlements of the Danelaw area left their permanent mark in place-names and dialect vocabulary in especially the eastern and northern parts of the country. In contrast to syntactic influences, which are typical of conditions of language shift, contacts that are less intensive and involve extensive bilingualism generally lead to lexical borrowing. This was the situation following the Norman Conquest of Britain in 1066 ce. It led to an influx of French loanwords into English, most of which have persisted in use up to the present day. It has been estimated that almost one third of the present-day English vocabulary is of French origin. By comparison, there is far less evidence of French influence on “core” English syntax. The earliest loanwords were superimposed by the French-speaking new nobility and pertained to administration, law, military terminology, and religion. Cultural prestige was the prime motivation for the later medieval borrowings.

For our third participation in the CLEF evaluation campaign, our first objective was to propose more effective and general stopword lists for the Swedish, Finnish and Russian languages, along with an improved, more efficient and simpler stemming procedure for these three languages. Our second goal was to suggest a combined search approach based on a data fusion strategy that would work with various European languages. Included in this combined approach is a decompounding strategy for the German, Dutch, Swedish and Finnish languages.