MIPE: A Metric Independent Pipeline for Effective Code-Mixed NLG Evaluation

TL;DR

This paper introduces MIPE, a new evaluation pipeline for code-mixed NLG that enhances the correlation between automatic metrics and human judgments, addressing the challenges posed by linguistic diversity in code-mixed text.

Contribution

MIPE is a metric-independent evaluation framework that improves the assessment of code-mixed NLG outputs, adaptable to various language pairs and tasks.

Findings

MIPE significantly improves correlation with human judgments.

Demonstrated effectiveness on Hinglish sentences from HinGE corpus.

Applicable to multiple code-mixed language pairs and NLG tasks.

Abstract

Code-mixing is a phenomenon of mixing words and phrases from two or more languages in a single utterance of speech and text. Due to the high linguistic diversity, code-mixing presents several challenges in evaluating standard natural language generation (NLG) tasks. Various widely popular metrics perform poorly with the code-mixed NLG tasks. To address this challenge, we present a metric independent evaluation pipeline MIPE that significantly improves the correlation between evaluation metrics and human judgments on the generated code-mixed text. As a use case, we demonstrate the performance of MIPE on the machine-generated Hinglish (code-mixing of Hindi and English languages) sentences from the HinGE corpus. We can extend the proposed evaluation strategy to other code-mixed language pairs, NLG tasks, and evaluation metrics with minimal to no effort.