Measuring Attribution in Natural Language Generation Models
Preprint
Abstract
With recent improvements in natural language generation (NLG) models for various applications, it has become imperative to have the means to identify and evaluate whether NLG output is only sharing verifiable information about the external world. In this work, we present a new evaluation framework entitled Attributable to Identified Sources (AIS) for assessing the output of natural language generation models, when such output pertains to the external world. We first define AIS and introduce a two-stage annotation pipeline for allowing annotators to appropriately evaluate model output according to AIS guidelines. We empirically validate this approach on generation datasets spanning three tasks (two conversational QA datasets, a summarization dataset, and a table-to-text dataset) via human evaluation studies that suggest that AIS could serve as a common framework for measuring whether model-generated statements are supported by underlying sources. We release guidelines for the human evaluation studies.Keywords
All Related Versions
This publication has 6 references indexed in Scilit:
- DialFact: A Benchmark for Fact-Checking in Dialogue2022
- Decontextualization: Making Sentences Stand-AloneTransactions of the Association for Computational Linguistics, 2021
- FEQA: A Question Answering Evaluation Framework for Faithfulness Assessment in Abstractive Summarization2020
- Inherent Disagreements in Human Textual InferencesTransactions of the Association for Computational Linguistics, 2019
- Natural Questions: A Benchmark for Question Answering ResearchTransactions of the Association for Computational Linguistics, 2019
- Why the Child's Theory of Mind Really Is a TheoryMind & Language, 1992