CoMuMDR: Code-mixed Multi-modal Multi-domain corpus for Discourse paRsing in conversations

Introduction

We present a novel code-mixed multi-modal multi-domain corpus for discourse parsing, named CoMuMDR. This corpus is designed to facilitate research in discourse parsing across multiple languages, modalities and domains, specifically focusing on code-mixed (Hindi + English = Hinglish) data. This repository contains the code for our paper on CoMuMDR: Code-mixed Multi-modal Multi-domain corpus for Discourse paRsing in conversations. The paper presents and compares various baseline modeling approaches for discourse parsing on CoMuMDR and popular discourse datasets, used in prior research.

Abstract

Discourse parsing is an important task useful for NLU applications such as summarization, machine comprehension, and emotion recognition. The current discourse parsing datasets based on conversations consists of written English dialogues restricted to a single domain. In this resource paper, we introduce CoMuMDR: Code-mixed Multi-modal Multi-domain corpus for Discourse paRsing in conversations. The corpus (code-mixed in Hindi and English) has both audio and transcribed text and is annotated with nine discourse relations. We experiment with various SoTA baseline models; the poor performance of SoTA models highlights the challenges of multi-domain code-mixed corpus, pointing towards the need for developing better models for such realistic settings.

Dataset

This is a gated dataset. To access please fill the above form and we shall provide access to the dataset. The dataset has been manually inspected and verified for quality and all privacy concerns have been addressed.

Format

The dataset is in JSON format, with each entry containing the following fields:

[
  {
    "id": "<conversation_id>",
    "edus": [
      {
        "text": "<string>",
        "speaker": "<string>",
        "audio_embeddings": "<filepath>"
      }
    ],
    "relations": [
      {
        "type": "<relation_type>",
        "x": "<edu_index_1>",
        "y": "<edu_index_2>"
      }
    ],
  }
  .
  .
  .
]

Usage

Clone the repository from huggingface and move the comumdr_data folder to the root directory of the code repository. This directory contains the main datafiles to run the implementations present in the code repo. The other directories in the repo contain the pre-processed data files to run/evaluate the pretrained models.

git clone https://huggingface.co/datasets/Exploration-Lab/CoMuMDR.git
cd CoMuMDR
mv comumdr_data ../