PrExMe! Large Scale Prompt Exploration of Open Source LLMs for Machine Translation and Summarization Evaluation

TL;DR

This study contributes to understanding the impact of different prompting approaches on LLM-based metrics for MT and summarization evaluation, highlighting the most stable prompting patterns and potential limitations.

Abstract

Large language models (LLMs) have revolutionized NLP research. Notably, in-context learning enables their use as evaluation metrics for natural language generation, making them particularly advantageous in low-resource scenarios and time-restricted applications. In this work, we introduce PrExMe, a large-scale Prompt Exploration for Metrics, where we evaluate more than 720 prompt templates for open-source LLM-based metrics on machine translation (MT) and summarization datasets, totalling over 6.6M evaluations. This extensive comparison (1) benchmarks recent open-source LLMs as metrics and (2) explores the stability and variability of different prompting strategies. We discover that, on the one hand, there are scenarios for which prompts are stable. For instance, some LLMs show idiosyncratic preferences and favor to grade generated texts with textual labels while others prefer to return numeric scores. On the other hand, the stability of prompts and model rankings can be susceptible to seemingly innocuous changes. For example, changing the requested output format from "0 to 100" to "-1 to +1" can strongly affect the rankings in our evaluation. Our study contributes to understanding the impact of different prompting approaches on LLM-based metrics for MT and summarization evaluation, highlighting the most stable prompting patterns and potential limitations.

Figures (16)

• Figure 1: Schematic overview of our prompt exploration methodology, featuring a grid search across datasets, task descriptions, output formats, and base prompts.
• Figure 2: Distribution of the (top 2% of every unique task) base prompts across all datasets, format requirements, task descriptions and tasks for Orca and Tower.
• Figure 3: Distribution of the top (top 2% of every unique task) format requirements across all datasets, format requirements, task descriptions and tasks for Orca and Tower.
• Figure 4: Distribution of the top (top 2% of every unique model) base prompts across all, format requirements, task descriptions and tasks for the ZS Eval4NLP train set evaluation and the ZS WMT23 evaluation.
• Figure 5: Correlation of the task description (left) and format requirement (right) ranking when changing the base prompt. The correlations across tasks, models and format requirement resp. task description are aggregated with the median. ZS-CoT is abbreviated with ZSC and ZS-CoT-EM is abbreviated with ZSCE.