Challenge: Existing reference-free metrics have obvious limitations for evaluating controlled text generation models.
Approach: They propose an unsupervised reference-free metric which evaluates controlled text generation from different aspects by formulating each aspect into multiple text infilling tasks.
Outcome: The proposed metric has higher correlations with human judgments while obtaining better generalization of evaluating generated texts from different models and with different qualities.

Similar Papers

On the Limitations of Reference-Free Evaluations of Generated Text (2022.emnlp-main)

Copied to clipboard

Challenge: a recent study has shown that evaluation metrics which accurately estimate the quality of generated text are limited in their ability to evaluate generated text.
Approach: They argue that reference-free metrics are limited in their ability to evaluate generated text . they recommend that they be used as diagnostic tools for analyzing and understanding model behavior .
Outcome: The proposed evaluation metrics are limited in their ability to evaluate generated text . they can be optimized at test time, can be biased against models with similar outputs .
SESCORE2: Learning Text Generation Evaluation via Synthesizing Realistic Mistakes (2023.acl-long)

Copied to clipboard

Challenge: Existing learned metrics perform unsatisfactory across text generation tasks or require human annotations for training on specific tasks.
Approach: They propose a self-supervised approach to train a model-based metric for text generation evaluation using sentences retrieved from a corpus.
Outcome: The proposed model outperforms all prior unsupervised metrics on four text generation evaluation benchmarks, with an average Kendall improvement of 0.158.
INSTRUCTSCORE: Towards Explainable Text Generation Evaluation with Automatic Feedback (2023.emnlp-main)

Copied to clipboard

Challenge: Existing methods to evaluate the quality of language generation do not provide explicit explanation of their verdicts.
Approach: They propose a fine-grained explainable evaluation metric for text generation that harnesses human instruction and implicit knowledge of GPT-4 to fine-tune it.
Outcome: The proposed model outperforms all other unsupervised metrics on translation, captioning, data-to-text, and commonsense generation tasks.
NLG Evaluation Metrics Beyond Correlation Analysis: An Empirical Metric Preference Checklist (2023.acl-long)

Copied to clipboard

Challenge: a systematic review of automatic evaluation metrics for Natural Language Generation (NLG) shows that task-agnostic metrics have a weak correlation with human .
Approach: They propose a framework to assess the effectiveness of automatic metrics in three NLG tasks . they propose task-agnostic and human-aligned metrics to be used for evaluation .
Outcome: The proposed framework provides access to the evaluation tools for three NLG tasks.
Spurious Correlations in Reference-Free Evaluation of Text Generation (2022.acl-long)

Copied to clipboard

Challenge: Recent work suggests that reference-free evaluation metrics may rely on spurious correlations with human judgments.
Approach: They propose to use model-based, reference-free evaluation metrics to evaluate natural language generation systems.
Outcome: The proposed metrics achieve high correlations with human judgments, but they may not be robust enough to evaluate their efficacy and robustness.
REFeREE: A REference-FREE Model-Based Metric for Text Simplification (2024.lrec-main)

Copied to clipboard

Challenge: Existing methods for text simplification lack a universal standard of quality and require a small number of human annotations.
Approach: They propose to introduce a reference-free model-based metric with a 3-stage curriculum that can be applied to any quality standard with fewer annotations.
Outcome: The proposed metric outperforms existing reference-based metrics in predicting ratings while requiring no reference simplifications at inference time.
Not All Errors are Equal: Learning Text Generation Metrics using Stratified Error Synthesis (2022.findings-emnlp)

Copied to clipboard

Challenge: Existing learning metrics are limited to tasks where large human ratings are available.
Approach: They propose a model-based natural language generation (NLG) evaluation metric that is highly correlated with human judgements without requiring human annotation.
Outcome: The proposed metric outperforms all prior unsupervised metrics on multiple NLG tasks including translation, image captioning, and WebNLG text generation.
HypoEval: Hypothesis-Guided Evaluation for Natural Language Generation (2026.acl-long)

Copied to clipboard

Challenge: Existing frameworks for LLM-as-a-judge use zero-shot setting without consulting any human input, which leads to low alignment, or fine-tune LLMs on labeled data, which requires a non-trivial number of samples.
Approach: They propose a hypothesis-guided evaluation framework that uses a small corpus of human evaluations to generate more detailed rubrics for human judgments and incorporates a checklist-like approach to combine LLM’s assigned scores on each decomposed dimension to acquire overall scores.
Outcome: The proposed framework outperforms existing frameworks in both human rankings and human scores with 30 human evaluations and fine-tunes LLMs on labeled data with 3 times more human evaluation by 11.95%.
A Tutorial on Evaluation Metrics used in Natural Language Generation (2021.naacl-tutorials)

Copied to clipboard

Challenge: This tutorial presents the evolution of automatic evaluation metrics to their current state along with emerging trends in this field.
Approach: This tutorial presents the evolution of automatic evaluation metrics to their current state . it aims to assess the extent of scientific progress made and identify areas/components that need improvement .
Outcome: This tutorial presents the evolution of automatic evaluation metrics to their current state along with emerging trends in this field.
T5Score: Discriminative Fine-tuning of Generative Evaluation Metrics (2023.findings-emnlp)

Copied to clipboard

Challenge: Existing methods for evaluating text quality are discriminative and generative . current methods use manual annotation of human judgements to train them .
Approach: They propose a framework that combines the best of both worlds by using supervised and unsupervised signals from whatever data we have available.
Outcome: The proposed method outperforms existing metrics on 5 datasets, 19 languages and 280 systems.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations