Strategic impact. MOLTO is addressing the task of high-precision translation of restricted language, which in the past has not belonged to the main stream of machine translation, but which is becoming increasingly relevant due to the advent of the Semantic Web. We expect the technology created in MOLTO to help greatly in the multilingual distribution of web content and also in its usage for information access and retrieval. MOLTO translation will be highly interoperable with Semantic Web standards (such as OWL) and adaptive to standard tools (web browsers and translators’ tools). The interoperability with Semantic Web standards will open existing ontologies and entity knowledge bases for the needs of MT tools. In turn grammar-based translation will strongly impact the way humans access structured knowledge, by providing NL query rendering to ontologies. The semantic retrieval results will also be rendered to grammatically flawless textual representations and presented to the end users as a high usability alternative to traditional table and graph based visualizations. Additionally, the grammar/ontology interoperability will empower knowledge extraction directly from text - a powerful metadata acquisition technique strongly desired by the Semantic Web, as a metadata layer struggling to capture the semantics of existing Web content. Translators are easy to build for new domains and to extend to new languages. They can even learn to translate better “on the fly”, by the use of example-based grammar writing, lexicon extension with minimal human intervention, and new statistical/grammar-based hybrid methods. A typical MOLTO translation system will work on a well-defined domain equipped with an ontology. The MOLTO developer’s tools will permit a domain expert, even without training in linguistics, to efficiently build a system that translates between an ontology and natural language. What is needed is a domain-specific lexicon and a set of example sentences describing the key properties of objects in the domain. This is made possible by the GF Resource Grammar Library (RGL) and the technique of example-based grammar writing. Porting the system into a new language is even easier, since the main relations between ontology and natural language tend to be similar in different languages; yet this similarity need not be followed, but can be overridden by transfer rules, most of which can be applied at compile time.
Appears in 1 contract
Sources: Grant Agreement
Strategic impact. MOLTO is addressing the task of high-precision translation of restricted language, which in the past has not belonged to the main stream of machine translation, but which is becoming increasingly relevant due to the advent of the Semantic Web. We expect the technology created in MOLTO to help greatly in the multilingual distribution of web content and also in its usage for information access and retrieval. MOLTO translation will be highly interoperable with Semantic Web standards (such as OWL) and adaptive to standard tools (web browsers and translators’ tools). The interoperability with Semantic Web standards will open existing ontologies and entity knowledge bases for the needs of MT tools. In turn grammar-based translation will strongly impact the way humans access structured knowledge, by providing NL query rendering to ontologies. The semantic retrieval results will also be rendered to grammatically flawless textual representations and presented to the end users as a high usability alternative to traditional table and graph based visualizations. Additionally, the grammar/ontology interoperability will empower knowledge extraction directly from text - a powerful metadata acquisition technique strongly desired by the Semantic Web, as a metadata layer struggling to capture the semantics of existing Web content. Translators are easy to build for new domains and to extend to new languages. They can even learn to translate better “on ―on the fly”fly‖, by the use of example-based grammar writing, lexicon extension with minimal human intervention, and new statistical/grammar-based hybrid methods. A typical MOLTO translation system will work on a well-defined domain equipped with an ontology. The MOLTO developer’s tools will permit a domain expert, even without training in linguistics, to efficiently build a system that translates between an ontology and natural language. What is needed is a domain-specific lexicon and a set of example sentences describing the key properties of objects in the domain. This is made possible by the GF Resource Grammar Library (RGL) and the technique of example-based grammar writing. Porting the system into a new language is even easier, since the main relations between ontology and natural language tend to be similar in different languages; yet this similarity need not be followed, but can be overridden by transfer rules, most of which can be applied at compile time. Once a translation system is there and integrated in a web page, a wiki, or a translator’s tool, its usage is as easy as using a text editor. The predictive parser, generic for all multilingual GF grammars, helps the author in a way that is similar to a T9 system, but it gives a guarantee of grammaticality and semantic well-formedness and not only of spelling. The syntax editor makes it easier than with text editors to maintain the consistency of documents: every change is propagated to all those places that have to be changed in consequence (e.g. due to agreement). In MOLTO, prototype systems will be built to cover 15 languages, which include at least 12 of the 23 official languages of the European Union. However, the technique is readily usable for the addition of more languages. The RGL is being developed in a collaborative project independently of MOLTO, and will in the near future cover the 23 European languages plus a number of other languages. Thus the technology will enable enriched information flows not only within the EU, but also throughout the rest of the world, opening Europe’s culture and its values for the good of all. In contrast to many other technologies within natural language processing, MOLTO is open- source and free software. It will build on open standards and enhance the interoperability between standards and components. We expect our demos and practically oriented documents to make MOLTO an attractive choice for a large population of potential users of the technology. • Automated translation that is more interoperable, more adaptive, better capable of self- learning and more user-friendly. The project produces translation technology that is fully automated for its domains of application, interoperable with current standards and tools, adapted to new domains, languages, and workflows, capable of self-learning from minimal information given by users, and user-friendly in its low demands for both translation system developers and authors of new translatable content. It should be noted that these goals are achieved without compromising the quality of translation, as regards information content, grammaticality, and idiomaticity. • Gaps in language coverage removed, and speed and quality of translation increased. The language coverage exceeds 50% of the official EU languages and is designed for painless growth. The speed for creating multilingual content is unforeseen, due to the full automation of translations and their updates in existing domains and languages, and to the easy adaptation to new domains and languages. The quality of translation is the main criterion of all MOLTO translation, which aims at reaching publishing quality in most case studies, and at least ―useful‖ on the TAUS scale in the most experimental cases involving non-restricted language. The new partners with their case studies will strengthen the expected impact in two ways. As the Zurich team has a leading position in the Controlled Natural Language community, we expect WP11 to show the way to a new standard in the development of CNLs where the CNL system is simultaneously available in multiple different languages. Research in CNLs has up to now focused on English as the base language. The work planned in WP11 results in a radically increased number of controlled natural languages that can all interact with each other in a semantics-preserving way in the same environment. We thus expect an impact to the future research in controlled natural languages. The multilingual semantic wiki engine developed in WP11 will support 15 languages, most of which are official EU languages. The wiki engine will be a modular and open platform, which can be easily extended to support even more languages. Advances compared to existing technologies, ▇.▇. ▇▇▇▇▇▇▇▇ MediaWiki (▇▇▇▇://▇▇▇▇▇▇▇▇- ▇▇▇▇▇▇▇▇▇.▇▇▇), include: the proposed wiki engine will be multilingual in the sense that all wiki articles are available in all the supported languages without any extra human effort (the wiki content must be provided only in one of the languages), the wiki engine will support much richer semantic content (as ACE is roughly equivalent to first-order logic), the wiki engine will completely hide the formal methods used to encode this semantic content behind the familiar natural languages. We expect these properties to have an impact on the way people with different language backgrounds collaboratively create and consume rich semantic content. The results of WP11 will introduce and promote a new type of collaborative environments which are (1) multilingual and thus usable by most EU citizens; (2) highly formal and thus seamlessly interacting with database systems and Semantic Web technologies; (3) user-friendly despite being machine-friendly. As the Be Informed team will perform in-house development of multilingual grammars for rapidly changing needs of interactive systems, we expect WP12 to show the way for a viable practise in companies with related localization needs. Be Informed develops a suite of products that enable its customers to model their business, for instance in terms of what it is they produce or sell, who they want or will sell it to and what requirements their internal processes must meet. Based on these models, Be Informed delivers the actual services required to run the modelled business, such as case based systems, decision services and registrations, or integrates with existing systems based on these models. This provides its customers with large degrees of agility, but at the same time poses challenges to stakeholders who now have to interact with (formal) models, where they dealt with informal texts, such as specifications, in the past. In WP12, Be Informed will use MOLTO Tools to develop a state of the art, multilingual explanation engine that verbalizes models, service behaviour and services traces into natural language. Intended audience of these explanations ranges from business users responsible for modelling their business, employees such as call centre agents that have to explain the products and processes to their clients and ultimately the clients themselves. For generated texts to be acceptable to these different stakeholders, the generated texts must be of high quality. This extension will add to MOLTO a novel multilingual semantic wiki shared as open-source and free software built on open standards, thus making MOLTO an attractive choice for a large population of potential users of the technology. Reinforced cooperation and better exploitation of ICT R&D synergies across the enlarged European Union: the new partners represent related but previously independent research directions where MOLTO technology can grow. On one hand, the CNL community (UZH) and on the other the ▇▇▇▇▇▇ EU project members. Wider participation in EU-supported ICT research projects across all Member States. Two countries are added to the MOLTO Consortium. Paving the way for strategic partnerships in view of gaining access to knowledge developing standards and interoperable solutions and strengthening European competitiveness. The ▇▇▇▇▇▇ project results will become accessible to MOLTO and vice versa, thus promoting their interoperability. With ▇▇▇▇▇▇ in particular, there is synergy at two levels: o Specification level: Lemon, GF concrete grammars and TermFactory are state of the art developments in the area of lexical information in relation to multilingual ontology labels and ontology verbalization. Scientifically, mapping them could benefit both. Having Be Informed as a partner in both will facilitate the technical partners to study and discuss each other’s lexical models across both projects.
Appears in 1 contract
Sources: Grant Agreement