Search

In this project, we have investigated the use of advanced machine learning methods, specifically fine-tuned large language models, for pre-annotating data for a lexical extension task, namely adding descriptive words (verbs) to an existing (but incomplete, as of yet) ontology of event types. Several research questions have been focused on, from the investigation of a possible heuristics to provide at least hints to annotators which verbs to include and which are outside the current version of the ontology, to the possible use of the automatic scores to help the annotators to be more efficient in finding a threshold for identifying verbs that cannot be assigned to any existing class and therefore they are to be used as seeds for a new class. We have also carefully examined the correlation of the automatic scores with the human annotation. While the correlation turned out to be strong, its influence on the annotation proper is modest due to its near linearity, even though the mere fact of such pre-annotation leads to relatively short annotation times., Comment: Accepted to LAW-XVII @ ACL 2023

This report describes guidelines that can be applied for training specialized neural machine translation (NMT) systems aimed at translation in a narrow textual domain, namely the domain of social surveys, requiring a specialized MT model that is able to handle domain-specific terminology. The work presented in this report demonstrates how relatively low-resource in-domain corpora can be used to prepare these specialized models. All described models are compatible with the packaged MT framework described in Deliverable D4.5 and the best performing models are available at the Lindat repository. The code used in the training pipeline (for experiment reproduction) is available on GitHub distributed under Mozilla Public License 2.0.1 Partners also describe the full translation pipeline including file sharing and preprocessing that was used to help with automatic translation of the Covid-19 surveys into English. While the description of the pipeline is general enough to be used in other future projects, the code published by partners on GitHub serves only as an example of a task-specific solution., Approved by EC - 27 April 2022

We present an extension of the SynSemClass event-type ontology, originally conceived as a bilingual Czech-English resource. We added German entries to the classes representing the concepts of the ontology. Having a different starting point than the original work (unannotated parallel corpus without links to a valency lexicon and, of course, different existing lexical resources), it was a challenge to adapt the annotation guidelines, the data model and the tools used for the original version. We describe the process and results of working in such a setup. We also show the next steps to adapt the annotation process, data structures and formats and tools necessary to make the addition of a new language in the future more smooth and efficient, and possibly to allow for various teams to work on SynSemClass extensions to many languages concurrently. We also present the latest release which contains the results of adding German, freely available for download as well as for online access.

LLOD (Linguistic Linked Open Data) is a generic name for a set of mutually connected language resources, using ontological relations. The connections between concepts and between concepts and their expressions in natural language make them suitable for both research and industrial applications in the area of content analysis, natural language understanding, (language- and knowledge-based) inferencing and other tasks. In the presented task, the concrete work will be on converting the SynSemClass project dataset (in part as a result of a previous Humane AI Net microproject called META-O-NLU) into LLOD, connecting it to the huge amount or interlinked data already available. A partner is involved in the Prêt-à-LLOD H2020 project, making this project synergistic in nature and multiplicative in terms of results in previous projects. Partners are also involved in the COST Action “European network for Web-centered linguistic data science” (NexusLinguarum).

This report presents a guideline for including German synonyms into the multilingual SynSemClass lexicon.

We present a richly annotated and genre-diversified language resource, the Prague Dependency Treebank-Consolidated 1.0 (PDT-C 1.0), the purpose of which is - as it always been the case for the family of the Prague Dependency Treebanks - to serve both as a training data for various types of NLP tasks as well as for linguistically-oriented research. PDT-C 1.0 contains four different datasets of Czech, uniformly annotated using the standard PDT scheme (albeit not everything is annotated manually, as we describe in detail here). The texts come from different sources: daily newspaper articles, Czech translation of the Wall Street Journal, transcribed dialogs and a small amount of user-generated, short, often non-standard language segments typed into a web translator. Altogether, the treebank contains around 180,000 sentences with their morphological, surface and deep syntactic annotation. The diversity of the texts and annotations should serve well the NLP applications as well as it is an invaluable resource for linguistic research, including comparative studies regarding texts of different genres. The corpus is publicly and freely available., Comment: Accepted at LREC 2020 (Proceedings of Language Resources and Evaluation, Marseille, France)

Discourse Representation Theory (DRT) is a formal account for representing the meaning of natural language discourse. Meaning in DRT is modeled via a Discourse Representation Structure (DRS), a meaning representation with a model-theoretic interpretation, which is usually depicted as nested boxes. In contrast, a directed labeled graph is a common data structure used to encode semantics of natural language texts. The paper describes the procedure of dressing up DRSs as directed labeled graphs to include DRT as a new framework in the 2020 shared task on Cross-Framework and Cross-Lingual Meaning Representation Parsing. Since one of the goals of the shared task is to encourage unified models for several semantic graph frameworks, the conversion procedure was biased towards making the DRT graph framework somewhat similar to other graph-based meaning representation frameworks., Comment: 10 pages, 4 figures, 4 tables, CoNLL 2020 Shared Task

Europe is a multilingual society with 24 EU Member State languages and dozens of additional languages including regional and minority languages and languages spoken by immigrants and trade partners. While languages are an essential part of our cultural heritage, language barriers continue to be unbreachable in many situations. The only option to enable and to benefit from multilingualism is through Language Technologies (LT) including Natural Language Processing (NLP) and Speech Technologies. The commercial European LT landscape is dominated by hundreds of SMEs. While the academic and industrial European LT community is world class, it is also massively fragmented. We describe the European Language Grid (ELG), Release 2. The system is targeted to evolve into the primary platform and marketplace for LT in Europe by providing one umbrella platform for the European LT landscape, including research and industry, enabling all stakeholders to showcase, share and distribute their services, products and resources. At the end of our EU-funded project, which will establish a legal entity in late 2021, the ELG will provide access to approx. 1300 commercial and non-commercial services for all European languages, as well as thousands of language resources (LRs). ELG will enable the whole European LT community to upload their technologies and resources and to deploy them through the ELG. The ELG is also meant to support digital language equality in Europe, i.e., to foster a situation in which all languages are supported through technologies equally well. We currently still have an extreme imbalance, characterised by a stark predominance of LTs for English.

With 24 official EU and many additional languages, multilingualism in Europe and an inclusive Digital Single Market can only be enabled through Language Technologies (LTs). European LT business is dominated by hundreds of SMEs and a few large players. Many are world-class, with technologies that outperform the global players. However, European LT business is also fragmented – by nation states, languages, verticals and sectors, significantly holding back its impact. The European Language Grid (ELG) project addresses this fragmentation by establishing the ELG as the primary platform for LT in Europe. The ELG is a scalable cloud platform, providing, in an easy-to-integrate way, access to hundreds of commercial and non-commercial LTs for all European languages, including running tools and services as well as data sets and resources. Once fully operational, it will enable the commercial and non-commercial European LT community to deposit and upload their technologies and data sets into the ELG, to deploy them through the grid, and to connect with other resources. The ELG will boost the Multilingual Digital Single Market towards a thriving European LT community, creating new jobs and opportunities. Furthermore, the ELG project organises two open calls for up to 20 pilot projects. It also sets up 32 national competence centres and the European LT Council for outreach and coordination purposes.

While studying verbal synonymy, we have investigated the relation between syntax and semantics in hope that the exploration of this relationship will help us to get more insight into the question of synonymy as the relationship relating (similar) meanings between different lexemes. Most synonym lexicons (Wordnets and similar thesauri) are based on an intuition about the similarity of word meanings, or on notions like “semantic roles.” In some cases, syntax is also taken into account, but we have found no annotation and/or evaluation experiment to see how strongly can syntax contribute to synonym specification. We have prepared an annotation experiment for which we have used two treebanks (Czech and English) from the Universal Dependencies (UD) set of parallel corpora (PUDs) in order to see how strong correlation exists between syntax and the assignment of verbs in context to pre-determined (bilingual) classes of synonyms. The resulting statistics confirmed that while syntax does support decisions about synonymy, such support is not strong enough and that more semantic criteria are indeed necessary. The results of the annotation will also help to further improve rules and specifications for creating synonymous classes. Moreover, we have collected evidence that the annotation setup that we have used can identify synonym classes to be merged, and the resulting data (which we plan to publish openly) can possibly serve for the evaluation of automatic methods used in this area.

Every year, the Conference on Computational Natural Language Learning (CoNLL) features a shared task, in which participants train and test their learning systems on the same data sets. In 2018, one of two tasks was devoted to learning dependency parsers for a large number of languages, in a real-world setting without any gold-standard annotation on test input.

Chimera is a machine translation system that combines the TectoMT deep-linguistic core with phrase-based MT system Moses. For English–Czech pair it also uses the Depfix post-correction system. All the components run on Unix/Linux platform and are open source (available from Perl repository CPAN and the LINDAT/CLARIN repository). The main website is https://ufal.mff.cuni.cz/tectomt. The development is currently supported by the QTLeap 7th FP project (http://qtleap.eu).

This paper presents the work on Machine Translation (MT) that has been conducted within the KConnect project, funded under the H2020 programme and focused on development and commercialization of cloud-based services for multilingual Semantic Annotation, Semantic Search and Machine Translation of electronic health records and medical publications. We first present the main goal and role of MT in the project and then briefly describe the main methods and components developed in the project, including training data acquisition, methods of domain adaptation, deployment of MT as cloud-based web-service, and the training toolkit allowing easy adaptation of the MT service to new languages.

Meaning Representation is a newly developed formalism for representing meaning, which abstracts from syntax and some other phenomena but is (still) language-dependent. It has been developed by a consortium of mostly U.S. universities, with team members including Martha Palmer, Kevin Knight, Philipp Koehn, Ulf Hermjakob, Kathy McKeown, Nianwen Xue and others. In the talk, the basic facts about the AMR will be presented, and then comparison will be made for Czech and English as carried out in detail on a small 100-sentence corpus; some examples from Chinese-English comparison will also be shown. In addition, AMR will be compared to the deep syntactic representation used in the set of Prague Dependency Treebanks, and observations will be made about the level of abstraction used in these two formalisms. There is an ongoing work on possible conversion between the Prague deep syntactic representation and the AMR representation, and the main issues of such a conversion will also be described, as well as plans for future studies and possible corpus annotation work in the nearest future.

Machine Translation is as old as the field of Computational Linguistics itself. It is also a problem that has been predicted to „be solved in the next five years“ many times, but in fact it is not yet solved today. Machine translation has been naively considered a simple problem solvable by simple statistical means, then studied in depth by complicated but unsuccessful detailed sets of rules trying to describe all details of natural language use, only to return to statistical approach on a completely different level, using a combination of linguistic analysis and powerful machine learning algorithms.

The talk presented activities in the area of language technology and specifically machine translation in the medical domain in the context of EC-funded projects.

Fundamentals of Machine Translation technology using Deep language analysis using the traditional "Vauquois triangle" have been presented, which now use advanced statistical and machine learning techniques. In addition, latest WMT 2015 Shared Task competition results for the en-cs pair have been shown in which the Chimera system created at UFAL MFF UK won.

LINDAT/CLARIN, as a node of the pan-European research infrastructure Clarin ERIC, has been presented. Its repository has been featured together with data archivation techniques and related web services and applications.

This article presents the results of one of the stages of the user-centered evaluation conducted in a framework of the EU project Khresmoi. In a controlled environment, users were asked to perform health-related tasks using a search engine specifically developed for trustworthy online health information. Twenty seven participants from largely the Czech Republic and France took part in the evaluation. All reported overall a positive experience, while some features caused some criticism. Learning points are summed up regarding running such types of evaluations with the general public and specifically with patients.

In this work, we investigate machine translation (MT) of search queries in the context of cross-lingual information retrieval (IR) in the domain of medicine. The main focus is on MT adaptation techniques to increase translation quality, however we also explore MT adaptation to improve cross-lingual IR directly. The experiments described herein have been performed and thoroughly evaluated for MT quality on the datasets created within the Khresmoi project and for IR performance on the CLEF eHealth 2013 datasets on three language pairs: Czech–English, German–English, and French–English. The search query translation results achieved in our experiments are outstanding – our systems outperformed not only our strong baselines, but also the Google Translate and Microsoft Bing Translator in direct comparison carried out on all the language pairs. In terms of the retrieval performance on this particular test collection, a significant improvement over the baseline has been achieved only for French–English. Throughout the article, we provide discussion and details on the contribution of the state-of-the-art features and adaptation techniques under exploration and provide future research directions.

There is increasing interest in and need for innovative solutions to medical search. In this paper we present the EU-funded Khresmoi medical search and access system, currently in year 3 of 4 of development across 12 partners. The Khresmoi system uses a component-based architecture housed in the cloud to allow for the development of several innovative applications to support target users' medical information needs. The Khresmoi search systems based on this architecture have been designed to support the multilingual and multimodal information needs of three target groups: the general public, general practitioners and consultant radiologists. In this paper we focus on the presentation of the systems to support the latter two groups using semantic, multilingual text and image-based (including 2D and 3D radiology images) search.

The Khresmoi project is developing a multilingual multimodal search and access system for medical and health information and documents. This scientific demonstration presents the current state of the Khresmoi integrated system, which includes components for text and image annotation, semantic search, search by image similarity and machine translation. The flexibility in adapting the system to varying requirements for different types of medical information search is demonstrated through two instantiations of the system, one aimed at medical professionals in general and the second aimed at radiologists. The key innovations of the Khresmoi system are the integration of multiple software components in a flexible scalable medical search system, the use of annotation cycles including manual correction to improve semantic search, and the possibility to do large scale visual similarity search on 2D and 3D (CT, MR) medical images.

Khresmoi is a European Integrated Project developing a multilingual multimodal search and access system for medical and health information and documents. It addresses the challenges of searching through huge amounts of medical data, including general medical information available on the internet, as well as radiology data in hospital archives. It is developing novel semantic search and visual search techniques for the medical domain. At the MIE Village of the Future, Khresmoi proposes to have two interactive demonstrations of the system under development, as well as an overview oral presentation and potentially some poster presentations.

This white paper presents the state of language technology support for the Czech language. It is a part of a series that analyses the available language resources and technologies for 30 European languages.

We describe our attempt to improve on previous English to Hindi machine translation results, using two open-source phrase-based MT systems: Moses and Joshua. We use several approaches to morphological tagging: from automatic word classes, through stem-suffix segmentation, to a POS tagger. We also experiment with factored language models. We evaluate various combinations of training data sets and other existing English-Hindi resources. To our knowledge, the BLEU score we obtained is currently the best published result for the IIIT-TIDES dataset.

The Universal Dependencies Project1 (Nivre, [9]; Nivre et al., [10]) is an ongoing effort towards creating a set of harmonised dependency treebanks that are annotated and structured according to universal guidelines. This paper reports on the addition of morphological features to the Irish Universal Dependencies Treebank (IUDT). Our feature set subscribes to the feature inventory of the UD Project and has been mapped from Irish morpho-syntactic tags – the output of a Finite State Morphological Analyser for Irish (Uí Dhonnchadha and van Genabith [16]). Irish, a Celtic language, has some relatively unusual morphological features that require language-specific labels not covered by the universal feature set. In this paper, we summarise the Irish-specific features that we have added to this set by explaining the linguistic properties that they each describe. We also report on the first parsing experiments using the IUDT by assessing the effect that the inclusion of morphological features has on parsing accuracy.