Main summary

Business and Technology University has published a new Georgian-language learning resource whose primary Georgian texts contain approximately 150 million tokens, according to the package’s description. The material gives researchers and developers an additional foundation for improving how AI understands, processes and produces Georgian text.

Main conclusions

• BTU’s new Georgian-language package brings together millions of examples and learning tasks prepared according to Georgian grammar.
• Text material, grammatical rules and practical examples help developers work on specific Georgian-language tasks.
• The project’s main opportunity lies in using the resource to develop Georgian-language learning and workplace tools; results must be evaluated in the particular AI system.

When people address artificial intelligence in Georgian, they want to ask a question naturally and receive a clear answer. A student may need a difficult subject explained, an employee may need help drafting a letter, and a customer may want to understand a service’s terms. In each case, language shapes how easy and useful the interaction will be.

Working well in Georgian requires suitable learning material. That material needs to cover sentence construction, different word forms and their use in different situations. Carefully prepared resources allow researchers and developers to focus on specific tasks.

BTU’s new package is a step in this direction. The university has prepared an expanded Georgian-language resource and made it publicly available on GitHub. Its description puts the volume of the primary texts at approximately 150 million tokens. A token is a unit used to process text: it can be a word or part of a word. For readers, the number indicates the material’s scale. [1]

In BTU researchers’ assessment, the resource also matters because it gives teams working on Georgian a shared learning foundation. They can use the material for their own tasks, test different approaches and evaluate what works better.

What the university prepared

The package contains more than six million primary language records. These are examples created according to Georgian-language rules, with variations in participants, word forms, tense and sentence context. Working with this material allows researchers to concentrate on a particular issue, such as how a sentence changes when it moves from the past to the future. [1]

The texts are accompanied by learning tasks. Some help a system distinguish who is acting in a sentence. Others concern correcting an error or working with particular parts of the text. The package also contains pooled linguistic information derived from sources, including information about word forms and words used together.

Text and task are therefore connected. A developer can select material relevant to the problem at hand. A team working on text comprehension might focus on identifying a participant, while a team improving written responses might examine agreement between words and the use of correct forms.

The resource uses examples generated according to rules. This approach presents the same language issue in many different forms. The package may therefore be particularly useful for work on specific Georgian grammar tasks. Its purpose is to provide additional material for that work.

Why this matters in everyday life

Understanding language involves more than recognising words. A small change in a sentence can change its entire meaning. “I sent the assignment” and “I have not sent the assignment yet” concern the same subject but require very different responses. This simple example shows why tense, negation and relationships between words matter.

Similar distinctions occur in student writing, business correspondence, customer questions and everyday conversation. When a system handles these parts of a text correctly, people can explain their requests more precisely and receive more useful answers. Resources prepared for Georgian support work toward that possibility.

One educational application is the development of learning tools. Developers could, for example, work on a system that explains a subject in Georgian, helps a student understand a text or provides feedback on a written assignment. Such a tool needs answers that are both accurate in meaning and clear in language.

Possible business applications include processing correspondence, understanding a customer’s question and extracting relevant information from text. Correctly understanding a request written in Georgian is a practical part of delivering these services. The new package gives teams working in these areas an additional resource.

These examples show where the material could be used. Any improvement in a particular system must be evaluated through training and subsequent testing. The university’s publication provides a foundation for that process.

Continuing the Georgian-language project

The package builds on BTU’s earlier work. According to the university’s public account, two previously published Georgian-language packages had a combined estimated volume of approximately 27 million tokens. A first-stage pilot edition of a Georgian grammar resource for AI is also available. The new resource adds to these materials and expands the possibilities for work on particular language tasks. [2]

The stages share a purpose: to make knowledge and examples needed for Georgian available in a form that technology teams can use. A grammatical description explains a rule, an example shows its application, and a task provides a way to work with that knowledge.

The project’s partner, Palitra Group, supplied the university with a large collection of Georgian-language texts. Together with BTU’s resources, this material formed a basis for processing linguistic data and preparing new learning material. The corresponding part of the public package contains pooled linguistic information derived from sources. [1][2]

The partnership connects text resources with the university’s research work. Its results become available to other researchers with their own questions. One team might develop a learning tool, another a Georgian-language service, and another test a new approach to processing text.

In BTU researchers’ assessment, a shared foundation of this kind creates more scope for cooperation around Georgian. The university can prepare material that other teams use to test new approaches. The result of one project can thus become the starting resource for further work.

The resource is already available

The package and its terms of use can be found on BTU’s GitHub page, together with a description and instructions. It is released under CC BY-NC 4.0: attribution is required, while commercial use requires separate permission. [3]

The next step is to use the material in particular systems and evaluate the results. Researchers need to establish how much it helps with understanding and producing Georgian text in the tasks for which it was used. The public package is prepared for such research; improvements in a specific model’s quality remain to be evaluated separately.

Seeing better Georgian in different AI systems by the end of the year is an important goal for BTU. Creating resources, making them available and enabling other teams to use them supports that goal.

With this stage completed, the university has prepared additional learning material for the development of Georgian-language technology. For Georgia, its practical significance is the opportunity to continue working on AI using the country’s language, its rules and relevant examples. The eventual benefit should be felt by people: through a clearer answer to a question in Georgian, a better learning experience or an easier way to complete a task.

Questions and answers

What is the package’s main purpose

To give researchers and developers Georgian-language examples and tasks that they can use for AI training and evaluation.

What does approximately 150 million tokens mean

It is the text volume stated in the package. A token is a unit used to process text and can be a word or part of a word.

Where could this material be used

In work on understanding, processing and producing Georgian text, including the development of educational and business tools. Users need to comply with the licence’s terms.

Where is the package available

On BTU’s public GitHub page, together with its description and terms of use. [1]