Paper Title:
IndicGenBench: A Multilingual Benchmark to Evaluate Generation Capabilities of LLMs on Indic Languages
Published on:
25 April 2024
Primary Category:
Computation and Language
Paper Authors:
Harman Singh,
Nitish Gupta,
Shikhar Bharadwaj,
Dinesh Tewari,
Partha Talukdar
New benchmark, IndicGenBench, for evaluating language models on generation tasks in 29 Indic languages
Extends existing datasets by adding test data in under-resourced Indic languages through human translation
Contains summarization, translation, multilingual QA, and cross-lingual QA evaluation sets
Tests major language models and finds significant gaps compared to English performance
Released under open license to facilitate research into more inclusive language models
Benchmark for evaluating language models on generation tasks in Indian languages
The authors release IndicGenBench, a new benchmark to measure the ability of language models to perform generation tasks like summarization, translation, and question answering across 29 languages native to India. It extends existing datasets, providing multi-way parallel test data in many under-resourced Indic languages for the first time.
Benchmarking language models on software development
Building Indonesian Task-Oriented Dialogue Systems
Benchmark for long context language model understanding
Testing language models' understanding of linguistic features
Komodo excels at Indonesian and regional languages
Portuguese language understanding benchmark
No comments yet, be the first to start the conversation...
Sign up to comment on this paper