CNIMA (Chinease Non-Native Interactivity Measurement and Automation) is a fully annotated spoken dialogue dataset for automated evaluating dialogue quality and dialogue annotations for Chinese Second Language (CSL)conversation dialogue.

It was created for the CNIMA project:

For more details, please read:

Dialogues labelled for three levels (from the above paper) can be found in CNIMA data.

Features exacted from the annotated datasets can be found in feature_label.csv.

Python script data/explore_data.py provides an example of interfacing with the data.

• dataset/CNIMA - Sample Dataset: Full dataset can bedownloaded via: https://drive.google.com/drive/folders/1uOzINQoNLxdina5tJ5bnumn4i1M5lnAf?usp=drive_link

• Dataset Viewing

• To run the notebooks for examining the datasets, please follow the procedures listed below:

• Download the dataset from the Shared folder link.

• Put the data into dataset/CNIMA and extract sample.zip.

• To view the data, one may use preprocessing.ipynb for viewing the examples in the example folder.

• notebooks/a.ipynb - Notebook for preprocessing
• notebooks/b.ipynb - Notebook for main experiments in micro-level prediction
• notebooks/c.ipync - Notebook for added experiments in score prediction
• folder/d. - Notebook for prompt LLM in annotation and score prediction
• notebooks/.js - Codes for data annotation website and platform
• CSL.html - Interface for data annotation website and platform

• figures/ - Contains all figures used for this project

• utils/ - Contains all utility functions for this project

• An Interpretable and Crosslingual Method for Evaluating Second-Language Dialogues