---
license: cc-by-4.0
task_categories:
- text2text-generation
language:
- ru
tags:
- homograph_resolution
- accentuation
pretty_name: Homograph Resulution Evaluation Dataset
size_categories:
- 1K<n<10K
---


# Homograph Resolution Evaluation Dataset

This dataset is designed to evaluate the performance of Text-to-Speech (TTS) systems and Language Models (LLMs) in resolving homographs in the Russian language. It contains carefully curated sentences, each featuring at least one homograph with the correct stress indicated. The dataset is particularly useful for assessing stress assignment tasks in TTS systems and LLMs.

## Key Features

- **Language**: Russian
- **Focus**: Homograph resolution and stress assignment
- **Unique Samples**: All sentences are original and highly unlikely to be present in existing training datasets.
- **Stress Annotation**: Correct stress marks are provided for homographs, enabling precise evaluation.

## Dataset Fields

- `context`: A sentence containing one or more homographs.
- `homograph`: The homograph with the correct stress mark.
- `accentuated_context`: The full sentence with correct stress marks applied.

**Note**: When evaluating, stress marks on words other than the homograph can be ignored.

## Limitations

1. **Single Stress Variant**: Each sample provides only one stress variant for a homograph, even if the homograph appears multiple times in the sentence (though such cases are rare).
2. **Limited Homograph Coverage**: The dataset includes a small subset of homographs in the Russian language and is not exhaustive.

## Intended Use

This dataset is ideal for:
- Evaluating TTS systems on homograph resolution and stress assignment.
- Benchmarking LLMs on their ability to handle ambiguous linguistic constructs.
- Research in computational linguistics, particularly in stress disambiguation and homograph resolution.

## Citing the Dataset

If you use this dataset in your research or projects, please cite it as follows:

```bibtex
@misc{HomographResolutionEval,
  author       = {Ilya Koziev},
  title        = {Homograph Resolution Evaluation Dataset},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/inkoziev/HomographResolutionEval}}
}