The purpose of this thesis is automatic diacritic restoration with the help of neural networks. Diacritic restoration is the process of adding accents to vowels in a text where the accents are missing.
In the first parts of the thesis I will describe the main characteristics of the Hungarian language which will be required to understand the solutions I present later. This will be followed by the theoretical backgrounds of machine learning and neural networks and the basic technological details which also will be needed for the following chapters.
After these I will describe my solutions for diacritic restoration, the networks I implemented with several different parameters. This will include the basic concepts of the approaches and the description of the architectures as well.
Later I will present the results of these models, evaluate them by different metrics, illustrate them on charts, and compare them to each other.
At last I will implement accenters to demonstrate the performance of the networks with actual examples.