SNLI Indo: A recognizing textual entailment dataset in Indonesian derived from the Stanford Natural Language...

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning

Background:

• Recognizing Textual Entailment (RTE) is crucial for NLP, determining inference relationships between text pairs.
• Neural networks, particularly deep learning, are state-of-the-art for RTE but require large, high-quality datasets.

Purpose of the Study:

• To address the resource gap for Indonesian NLP by creating a large-scale RTE dataset.
• To facilitate the development of optimal deep learning models for RTE in the Indonesian language.

Main Methods:

• Translated the English Stanford Natural Language Inference (SNLI) corpus into Indonesian.
• Utilized Google Cloud Translation API for sentence pair translation.
• Created the SNLI Indo dataset with over 569,000 premise-hypothesis pairs.

Main Results:

• Successfully developed SNLI Indo, a substantial dataset for Indonesian RTE.
• The dataset includes 549,365 training, 9,840 validation, and 9,822 testing sentence pairs.
• SNLI Indo bridges the gap in Indonesian NLP resources.

Conclusions:

• SNLI Indo enables more effective training of deep learning models for RTE in Indonesian.
• This dataset is vital for advancing NLP research and applications for the Indonesian language.