HinGE

HinGE数据集是由TCS Research和IIT Gandhinagar创建，专注于Hinglish（印度语和英语混合）文本生成与评估。该数据集包含1978条由人类和两种规则算法生成的Hinglish句子，对应于平行印度语-英语句子。数据集的创建过程涉及从IIT Bombay英语-印度语平行语料库中随机选择句子，并由五名精通英语和印度语的注释者进行注释。HinGE旨在支持代码混合自然语言生成研究，解决现有资源稀缺的问题，特别是在评估代码混合文本生成任务中。

The HinGE dataset was developed by TCS Research and IIT Gandhinagar, focusing on Hinglish (a blend of Hindi and English) text generation and evaluation. This dataset comprises 1978 Hinglish sentences generated by humans and two rule-based algorithms, paired with parallel Hindi-English sentence pairs. The dataset construction process involves randomly selecting sentences from the IIT Bombay English-Hindi Parallel Corpus, followed by annotation from five annotators proficient in both English and Hindi. HinGE aims to support research on code-mixed natural language generation, addressing the scarcity of existing resources, particularly for evaluating code-mixed text generation tasks.

搜集汇总

背景与挑战

背景概述

HinGE数据集是一个专注于Hinglish（印度语和英语混合）文本生成与评估的资源，由TCS Research和IIT Gandhinagar创建。它包含1978条由人类和规则算法生成的句子，基于平行印度语-英语语料库，并由双语注释者处理，旨在支持代码混合自然语言生成研究，解决该领域资源稀缺和评估挑战。