metadata

license: apache-2.0
task_categories:
  - text-classification
  - token-classification
language:
  - tr
tags:
  - synturk
  - sentagram
  - sentence
  - nlp
  - cümle
  - elements
  - cümlenin ögeleri
pretty_name: SYNTÜRK SENTAGRAM Dataset
size_categories:
  - 1K<n<10K

Dataset Card for SYNTÜRK SENTAGRAM Dataset

Dataset Summary

The SYNTÜRK SENTAGRAM Dataset is a specialized dataset designed to aid in the development and evaluation of NLP models focused on Turkish syntax and grammar. The dataset consists of sentences annotated with grammatical elements such as subjects, predicates, objects, and adjuncts. This dataset is intended for educational purposes and research in natural language processing (NLP), particularly in the context of the Turkish language.

Languages

The dataset is in Turkish (ISO code: tr).

Dataset Structure

Data Fields

Each entry in the dataset consists of the following fields:

id: Unique identifier for the sentence.
text: The original sentence in Turkish.
elements: A list of grammatical elements within the sentence, each annotated with:
- word: The specific word or phrase in the sentence.
- type: The grammatical role of the word, including in last_dataset.json:
  - ozne (subject)
  - yuklem (predicate)
  - d-tum (complement)
  - z-tum (adjunct)
  - e-tum (prepositional complement)
  - bsiz-nesne (bare object)
  - bli-nesne (marked object)
  - c-dıs (out-of-sentence element)
- start: Start index of the word in the sentence.
- end: End index of the word in the sentence.

After creating our last_dataset.json dataset file, we performed data augmentation and we got the aug_aug_last_dataset.json file. In this way, we increased the number of datasets by 2.5 times. Afterwards, we pre-processed our dataset and changed its types as follows:

type:
- B-ozne
- I-ozne
- B-yuklem
- I-yuklem
- B-dtum
- I-dtum
- B-ztum
- I-ztum
- B-etum
- I-etum
- B-blinesne
- I-blinesne
- B-bsiznesne
- I-bsiznesne
- B-cdıs
- I-cdıs

After all of these steps, we got the revised_aug_aug_last_dataset.json file and we trained our model SENTAGRAM

Example of last_dataset.json

{
  "id": 1,
  "text": "Aşığa Bağdat sorulmaz.",
  "elements": [
    {
      "word": "Aşığa",
      "type": "d-tum",
      "start": 0,
      "end": 6
    },
    {
      "word": "Bağdat",
      "type": "ozne",
      "start": 7,
      "end": 13
    },
    {
      "word": "sorulmaz",
      "type": "yuklem",
      "start": 14,
      "end": 23
    }
  ]
}

Example of revised_aug_aug_last_dataset.json

{
  "id": 2,
  "text": "Aşık olduğun zaman ben sana söylerim.",
  "elements": [
      {
          "word": "Aşık",
          "type": "B-ztum",
          "start": 0,
          "end": 3
      },
      {
          "word": "olduğun",
          "type": "I-ztum",
          "start": 5,
          "end": 11
      },
      {
          "word": "zaman",
          "type": "I-ztum",
          "start": 13,
          "end": 17
      },
      {
          "word": "ben",
          "type": "B-ozne",
          "start": 19,
          "end": 21
      },
      {
          "word": "sana",
          "type": "B-dtum",
          "start": 23,
          "end": 26
      },
      {
          "word": "söylerim",
          "type": "B-yuklem",
          "start": 28,
          "end": 35
      }
  ]
}

Dataset Creation

Initial Data Collection

The data was manually curated from various Turkish literary sources, including news websites, educational materials, and original compositions. Each sentence was annotated by linguists with expertise in Turkish grammar to ensure the accuracy and quality of the annotations.

Preprocessing

The dataset was preprocessed to align with the BERTürk model input format. The sentences were tokenized, and the annotations were adjusted accordingly to fit the tokenized structure.

Annotations

The dataset includes human-annotated grammatical elements, focusing on syntactic and grammatical correctness. The annotations were verified through multiple rounds of review to ensure high quality.

Usage

Intended Use

This dataset is intended for use in developing and evaluating models for:

Part-of-speech tagging
Grammatical structure analysis
Educational tools for Turkish language learners

Load the Dataset

To load the dataset in Hugging Face, use the following code:

from datasets import load_dataset

dataset = load_dataset("synturk/turkish-sentence-elements")

Performance Metrics

Models trained on this dataset should aim to achieve high accuracy in correctly identifying the grammatical elements of the sentences. Performance can be evaluated using standard NLP metrics like precision, recall, F1 score, and accuracy.

Limitations

While the dataset covers a wide range of grammatical structures, it may not include all possible variations found in the Turkish language. Future updates to the dataset will include additional sentence structures and more diverse linguistic elements.

Future Work

We plan to expand the dataset with more complex sentences and additional grammatical annotations, including semantic roles and discourse relations. This will enable the development of even more sophisticated NLP tools for the Turkish language.

License

This dataset is licensed under the Apache 2.0 License.

Citation

If you use this dataset in your work, please cite it as:

@dataset{synturk_sentagram_2024,
  author    = {SYNTÜRK Team},
  title     = {SYNTÜRK SENTAGRAM Dataset},
  year      = {2024},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/synturk/turkish-sentence-elements},
}

Contributions

The dataset was created and annotated by the SYNTÜRK Team. Contributions to the dataset are welcome. If you have suggestions or improvements, please submit a pull request on GitHub.

Follow SYNTÜRK Team on,