WITS (Why Is This Sarcastic?)

该数据集是 MASAC 的扩展，MASAC 是一个多模式、多方、印地语-英语代码混合对话数据集，由印度流行电视节目“Sarabhai v/s Sarabhai”编译而成。 WITS 是通过在 MASAC 中为每个讽刺对话添加自然语言解释而创建的。该数据集包括从 55 集电视节目中转录的讽刺对话，以及音频和视频多模态信号。它旨在促进对话中的讽刺解释 (SED)，这是一项新颖的任务，旨在为给定的讽刺对话生成自然语言解释，从而阐明预期的讽刺意味。 WITS 中的每个数据实例都与相应的视频、音频和文本转录相关联，其中最后一个话语本质上是讽刺性的。所有最终选定的解释都包含以下属性： • 讽刺来源：对话中正在讽刺的说话者。 • 讽刺目标：讽刺所针对的人/物。 • 动作词：用于描述讽刺是如何发生的动词/动作。例如嘲弄、侮辱、嘲讽等。 • 描述：对场景的描述，以帮助将讽刺语境化。

This dataset, WITS, is an extension of MASAC—a multimodal, multi-party Hindi-English code-mixed dialogue dataset compiled from the popular Indian television show "Sarabhai v/s Sarabhai". WITS is developed by appending natural language explanations to every sarcastic dialogue within MASAC. This dataset encompasses sarcastic dialogues transcribed from 55 episodes of the aforementioned TV show, paired with multimodal audio and video signals. It is designed to advance Sarcasm Explanation in Dialogue (SED), a novel task that generates natural language explanations for given sarcastic dialogues to clarify the intended sarcastic meaning. Each data instance in WITS is linked to corresponding video, audio, and text transcriptions, where the final utterance is inherently sarcastic. All selected final explanations contain the following attributes:
- Sarcasm Source: The speaker who delivers sarcasm in the dialogue.
- Sarcasm Target: The person or object that the sarcasm is directed at.
- Action Verb: The verb or action describing how the sarcasm is manifested, such as mock, insult, taunt, etc.
- Description: A description of the scene to help contextualize the sarcasm.

搜集汇总

背景与挑战

背景概述

WITS是一个多模态讽刺解释数据集，包含印地语-英语代码混合的讽刺对话及其自然语言解释，来源于印度电视节目55集的转录，支持音频、视频和文本多模态信号，旨在促进讽刺解释任务的研究。