For a long time, the Lithuanian language was not available in many of the technologies used daily. Why can’t you talk to your smart devices in Lithuanian, why doesn’t a robot vacuum respond to Lithuanian commands, and why are voice assistants in our native language so limited? Researchers at Vilnius University (VU) have been working for over ten years to make technologies as accessible as possible in Lithuanian.
Currently, funding has been received for the project “Creating the Great Lithuanian Language Speaker” (LIEPA-3), which is one of the most important steps in the field of Lithuanian language technologies. LIEPA-3 will provide new opportunities for implementing our language in modern smart systems. The project aims to make the Lithuanian language an integral part of technology, which will be as accessible as major languages such as English or German.
“We live in a world where language technology is becoming increasingly important in everyday life. If the Lithuanian language cannot survive in this area, we will be left behind. “LIEPA-3 is our chance to keep the Lithuanian language alive and integrated into modern technologies,” says Gediminas Navickos, a researcher at the Faculty of Mathematics and Informatics at VU and one of the initiators of the LIEPA-3 project.
According to G. Navickos from the Faculty of Mathematics and Informatics (MIF) at VU, this project not only creates technologies, but also records the unique sound and distinctiveness of our language for future generations, which is especially important for the development of language technologies and research into the language itself.
The future of the Lithuanian language in technology
LIEPA-3 is an ongoing project aimed at ensuring the survival and adaptation of the Lithuanian language in a modern digital environment, strengthening the country’s digitalization capabilities. This project is a result of the success of the previous LIEPA and LIEPA-2 projects, taking over their achievements and expanding the possibilities of Lithuanian language technologies.
The LIEPA and LIEPA-2 projects were also implemented by researchers from the Faculty of Mathematics, Informatics and Philology at VU. Both projects include two main directions: IT solutions – new services for the public and infrastructure solutions related to speakers of spoken Lithuanian, speech synthesizers and speech recognition systems.
A large team of scientists is already working on the new LIEPA-3 project. Unlike its predecessors, this project focuses on several directions, but on one main one – the creation of a large-scale annotated Lithuanian speech. An annotated speaker is a structurally described audio recording (examples of Lithuanian speech), consisting of audio recordings and their corresponding texts with timestamps. The newly created speaker will be ten times larger than the largest existing Lithuanian speaker and will cover 10 thousand hours.
With a speaker of this size, we will be able to get Lithuanian speech recognition tools (when computers and other devices understand what we say to them in Lithuanian) that meet modern quality standards.
Real estate agents answer the question of what are the three most important components of the value of a property: location, location, location. To rephrase the question: what are the three most important components of the value of language technologies, the answer will be: data, data, data. So far, a versatile, comprehensive, and most importantly, large-scale Lithuanian dictionary has not been created.
The 1000-hour speaker developed during the LIEPA-2 project is very small compared to speakers in technologically advanced languages. In addition to the range, the creation of a sound system is also important in another aspect: when developing speech technologies, computer scientists also use linguists and linguists.
“This is a beautiful and meaningful example of interdisciplinary cooperation, which has continued since the beginning of the LIEPA-1 project. A large speaker is not only important for speech technology, but will be an excellent base for various linguistic studies. It is especially important that half of the speaker is spontaneous speech, and this part will well show the state of modern spoken Lithuanian, which, unfortunately, we now see is quite fragmented precisely due to the lack of detailed and extensive data. Garsyn is important not only in the context of speech technologies and linguistic research – it would probably not be bold to compare it with the Great Lithuanian Dictionary, where not only our language is preserved, but also identity markers – in this case, “not written down, but spoken in living words”, says VU Philology Professor Vytautas Kardelis.
The project is being implemented by the VU Faculty of Mathematics and Informatics together with the Faculty of Philology in cooperation with partners Vytautas the Great University and the Lithuanian Language Institute. “The time allotted for the project implementation is very short – just over a year and a half, although such work usually takes at least three years. However, the project is being implemented by the strongest consortium of experts in this field, gathered from experienced organizations, so we believe that the project will be completed successfully”, says the head of the LIEPA-3 project, researcher at VU MIF Dr. Gordijk.
The project results are awaited
“The main goal of the LIEPA-3 project is to create 10 thousand hours of annotated audio in the Lithuanian language, which will be composed according to the age, gender and dialect region of the announcers. The speaker will reflect the phonetic, morphological, grammatical, stylistic and dialectal diversity of the speech content, differences in different recording devices and the acoustic background of the environment.” G. Courville.
The process of creating the speaker itself requires a lot of work: from collecting audio data, processing it and assessing its accuracy to the final uploading of the data to open access platforms so that the project results are available to all interested parties. According to the project leader, with a comprehensive and publicly available speaker, scientists will be able to develop advanced methods of speech recognition, synthesis and natural language processing, which will open up opportunities for the development of artificial intelligence in Lithuania. In addition, the speaker will be of great value in research focused on increasing social inclusion – we will be able to take into account people with special needs and create technologies that are intuitive and accessible to everyone.
After the project is implemented, there will be many opportunities to apply its results in practice. First of all, Lithuanian scientists and technology developers will be able to develop advanced language technologies and create innovative e-services in the Lithuanian language. In addition, publicly available audio resources in the Lithuanian language will encourage researchers from other countries working in the field of language technologies to include the Lithuanian language in their research, thus ensuring the activity of the Lithuanian language in the digital space and opening up opportunities for international cooperation. According to G. Navickos, the project aims to contribute to the implementation of the State Program for the Development of Digitalization of the Ministry of Economy and Innovation, increase the availability of language technologies, ensure their operation in the Lithuanian language and help update the digital skills of society: “In the next two years, a loudspeaker will be created, which will become public and freely accessible and will be used in scientific research and development of digital solutions. This will contribute to the creation of high-quality e-services and the development of digitalization in Lithuania.”
