Update README.md

README.md

@@ -15,13 +15,22 @@ configs:
     path: test_copal_colloquial.csv
 ---
 
+## About COPAL-ID
 
-## Paper
+COPAL-ID is an Indonesian causal commonsense reasoning dataset that captures local nuances. It provides a more natural portrayal of day-to-day causal reasoning within the Indonesian (especially Jakartan) cultural sphere. Professionally written and validatid from scratch by natives, COPAL-ID is more fluent and free from awkward phrases, unlike the translated XCOPA-ID.
 
-URL Arxiv: https://arxiv.org/abs/2311.01012
+COPAL-ID is a test set only, intended to be used as a benchmark.
+For more details, please see [our paper](https://arxiv.org/abs/2311.01012).
 
+### Local Nuances Categories
+Our dataset consists of 3 subcategories: local-term, culture, and language reasoning.
+ - Local-term captures common knowledge for Indonesians that is most likely unknown or uncommon for non-natives, e.g., local foods, public figures, abbreviations, and other local concepts.
+ - Culture captures norms used in Indonesia.
+ - Language captures the reasoning for the language itself, for example, local idioms, figures of speech, as well as ambiguous words.
+Specifically, the distribution of COPAL-ID across these categories is:
 
-We present publicly available COPAL-ID, a novel Indonesian language common sense reasoning dataset. Unlike the previous Indonesian COPA dataset (XCOPA-ID), COPAL-ID incorporates Indonesian local and cultural nuances, and therefore, provides a more natural portrayal of day-to-day causal reasoning within the Indonesian cultural sphere. Professionally written by natives from scratch, COPAL-ID is more fluent and free from awkward phrases, unlike the translated XCOPA-ID. In addition, we present COPAL-ID in both standard Indonesian and in Jakartan Indonesian--a dialect commonly used in daily conversation. COPAL-ID poses a greater challenge for existing open-sourced and closed state-of-the-art multilingual language models, yet is trivially easy for humans. Our findings suggest that even the current best open-source, multilingual model struggles to perform well, achieving 65.47% accuracy on COPAL-ID, significantly lower than on the culturally-devoid XCOPA-ID (79.40%). Despite GPT-4's impressive score, it suffers the same performance degradation compared to its XCOPA-ID score, and it still falls short of human performance. This shows that these language models are still way behind in comprehending the local nuances of Indonesian.
+### Colloquial vs Standard Indonesian
+In daily scenarios, almost no one in Indonesia uses purely formal Indonesian. Yet, many NLP datasets use formal Indonesian. This surely causes a domain mismatch with real-case settings. To accommodate this, COPAL-ID is written in two variations: Standard Indonesian and Colloquial Indonesian. If you use COPAL-ID to benchmark your model, we suggest testing on both variants. Generally, colloquial Indonesian is harder for models to handle.
 
 ## How to Use
 
@@ -32,6 +41,17 @@ copal_id_dataset = load_dataset('haryoaw/COPAL', 'id', subset='test')
 copal_id_colloquial_dataset = load_dataset('haryoaw/COPAL', 'id', subset='test_colloquial')
 ```
 
+## Data Collection and Human Performance
+
+COPAL-ID was created through a rigorous data collection pipeline. Each example is written and checked by natives accustomed to Jakartan culture. Lastly, we have run a human benchmark performance test across native Jakartans, in which they achieved an average accuracy of ~95% in both formal and colloquial Indonesian variants, noting that this dataset is trivially easy for those familiar with the culture and local nuances of Indonesia, especially in Jakarta.
+
+For more details, please see our paper.
+
+## Limitation
+
+Indonesia is a vast country with over 700+ languages and rich in culture. Therefore, it is impossible to pinpoint a singular culture. Our dataset is specifically designed to capture Jakarta's (the capital) local nuances. Expanding to different local nuances and languages across Indonesia is a future work.
+
+
 ## Cite Our Work
 
 ```