Marija Brkić

Croatian speech technologies

Speech technologies deal with designing computer systems that can recognize spoken words, comprehend human language and generate intelligible speech. There is a wide range of applications speech technology systems were successfully implemented in. One of the most complex applications in speech technology is a spoken dialog system, which can be used for information inquiry services. In the paper we present the work in the development of a spoken dialog system for Croatian language. We propose an approach for development of modules in a spoken dialog system for the limited domain which uses the same acoustic model for speech recognition and speech synthesis. For the linguistic analysis an approach is proposed which is based on qualitative language modelling, while the development of the dialog module is based on the formalism of the object-oriented frame logic language. Some experimental results for Croatian speech recognition and understanding are presented and discussed.

Data mining of computer game assisted e/m-learning systems in higher education

One of the major challenges in the implementation of e-learning systems is motivating students to use it. E-learning systems are mostly unadapted to mobile platforms because they do not use responsive web design and progressive enhancement based on browser, device or feature detection. This paper presents the evolutionary cycle, current status and potential development of new platforms for e/m-learning in higher education. Educational computer games that serve as a motivational element in e/m-learning are presented, as well as the data structures that can complement the existing models of e/m-learning. The University of Rijeka uses MudRi e-learning system, which is based on Moodle open source software. This system collects data on the activity of participants in the educational process, which is suitable for data mining. A preliminary research on the correlation between the activity in forum discussions and the course pass or fail grade is carried out. The possibility of using data mining based on the data structure of the Moodle system and the models of e/m-learning and learning through play is described, aiming at finding optimal ways of learning and optimal learning results. Finally, a model of integrated system user interaction on which our future research will be based is presented.

Development of an online multimedia course

At the University of Rijeka the theoretical knowledge is mainly transferred by the traditional model consisting of professor explaining material to a group of students in front of a blackboard. We believe that this model dates from the period of the industrial revolution when it was necessary to educate large number of people to work on new machines, but there was no technology for more efficient knowledge transfer. Nowadays, such a model cannot keep up with the speed of generating new knowledge and technology. A larger number of participants cannot be educated due to time and space constraints, and the knowledge usually becomes obsolete by the end of studies. The result of this model are students with skills unadapted to the labor market requirements. We believe this is an extremely important issue for our university. One of the possible solutions might be digitizing theoretical parts of courses. Using web technologies, theoretical lectures can be held online and in that way more efficient methods of teaching on campus can be used. In this paper, the process of digitizing the university course Introduction to Psychology is presented, as well as the benefits that can be achieved by creating and using multimedia online courses.

Can we beat google translate

This paper presents a machine translation evaluation study for Croatian-English language pair. In-domain and out-of-domain translations from Croatian into English have been obtained from Google Translate, our own statistical machine translation system LegTran, and from a professional translator. These translations have been evaluated by six different automatic metrics. The gains obtained from increasing the number of reference translations have been explored and measured. System level correlation between automatic evaluation metrics is given and the significance of the results is discussed. Bootstrapping, approximate randomization and the sign test have been used for confidence intervals and hypothesis testing.

VoiceXML for Slavic languages application development

Since Croatian language, as well as other Slavic languages, is essentially very different from English, we are in the need of developing language specific tools and systems. In this paper we point out the benefits that speech applications have and describe our future platform. Furthermore, we create domain-specific dialogues using VoiceXML. VoiceXML is a language for creating voice user interfaces. We opted for VoiceXML because it simplifies application development by permitting developers to use familiar Web techniques, tools and infrastructure.

Automatic and human evaluation on english-croatian legislative test set

This paper presents work on the manual and automatic evaluation of the online available machine translation (MT) service Google Translate, for the English-Croatian language pair in legislation and general domains. The experimental study is conducted on the test set of 200 sentences in total. Human evaluation is performed by native speakers, using the criteria of fluency and adequacy, and it is enriched by error analysis. Automatic evaluation is performed on a single reference set by using the following metrics: BLEU, NIST, F-measure and WER. The influence of lowercasing, tokenization and punctuation is discussed. Pearsons correlation between automatic metrics is given, as well as correlation between the two criteria, fluency and adequacy, and automatic metrics.

Automatic and Human Evaluation on English-Croatian Legislative Test Set

In this paper, we introduce feature adaptation, an unsupervised method for cross-domain natural language processing (NLP). Feature adaptation adapts a supervised NLP system to a new domain by recomputing feature values while retaining the model and the feature definitions used on the original domain. We demonstrate the effectiveness of feature adaptation through cross-domain experiments in compositionality grading and show that it rivals supervised target domain systems when moving from generic web text to a specialized physics text domain.This paper presents work on the manual and automatic evaluation of the online available machine translation (MT) service Google Translate, for the English-Croatian language pair in legislation and general domains. The experimental study is conducted on the test set of 200 sentences in total. Human evaluation is performed by native speakers, using the criteria of fluency and adequacy, and it is enriched by error analysis. Automatic evaluation is performed on a single reference set by using the following metrics: BLEU, NIST, F-measure and WER. The influence of lowercasing, tokenization and punctuation is discussed. Pearsons correlation between automatic metrics is given, as well as cor relation between the two criteria, fluency and adequacy, and automatic metrics.

Corpora building and processing

Creativity is a basic feature of a language. Therefore, it is perfectly possible to create a completely new context that has never occurred before. This feature allows us to express our ideas, thoughts, knowledge and fears, but it also complicates the idea of human-machine communication. Since it became obvious that natural languages cannot be formalized and described as a whole, the idea of combining linguistic knowledge and corpora has arisen. The combination of these techniques has proven to give the best results and our research is based on that notion. Since data sparsity poses a huge problem, this work presents a practical solution in overcoming data sparsity problem and gives a detailed account of an advanced data processing technique.

VoiceXML Platform for Minority Languages

Not only that the use of different services is faster and easier in a voice-based user interface, but it also has benefits for people with physical impairments. This paper investigates the possibilities that minority languages have in developing spoken dialogue applications. We present our future platform and discuss difficulties that we as Croatian language speakers face in the planning and development phases. Furthermore, we set forth our Dialogue Manager Strategy, discuss properties which affect it and create weather-forecast dialogues. We have opted for VoiceXML as a dialogue modeling language since it is a language for creating voice user interfaces which simplifies application development, especially for minority languages.