Tailoring Self-Rationalizers with Multi-Reward Distillation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ramnath, Sahana, Joshi, Brihi, Hallinan, Skyler, Lu, Ximing, Li, Liunian Harold, Chan, Aaron, Hessel, Jack, Choi, Yejin, Ren, Xiang
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909208593563648
author Ramnath, Sahana
Joshi, Brihi
Hallinan, Skyler
Lu, Ximing
Li, Liunian Harold
Chan, Aaron
Hessel, Jack
Choi, Yejin
Ren, Xiang
author_facet Ramnath, Sahana
Joshi, Brihi
Hallinan, Skyler
Lu, Ximing
Li, Liunian Harold
Chan, Aaron
Hessel, Jack
Choi, Yejin
Ren, Xiang
contents Large language models (LMs) are capable of generating free-text rationales to aid question answering. However, prior work 1) suggests that useful self-rationalization is emergent only at significant scales (e.g., 175B parameter GPT-3); and 2) focuses largely on downstream performance, ignoring the semantics of the rationales themselves, e.g., are they faithful, true, and helpful for humans? In this work, we enable small-scale LMs (approx. 200x smaller than GPT-3) to generate rationales that not only improve downstream task performance, but are also more plausible, consistent, and diverse, assessed both by automatic and human evaluation. Our method, MaRio (Multi-rewArd RatIOnalization), is a multi-reward conditioned self-rationalization algorithm that optimizes multiple distinct properties like plausibility, diversity and consistency. Results on five difficult question-answering datasets StrategyQA, QuaRel, OpenBookQA, NumerSense and QASC show that not only does MaRio improve task accuracy, but it also improves the self-rationalization quality of small LMs across the aforementioned axes better than a supervised fine-tuning (SFT) baseline. Extensive human evaluations confirm that MaRio rationales are preferred vs. SFT rationales, as well as qualitative improvements in plausibility and consistency.
format Preprint
id arxiv_https___arxiv_org_abs_2311_02805
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Tailoring Self-Rationalizers with Multi-Reward Distillation
Ramnath, Sahana
Joshi, Brihi
Hallinan, Skyler
Lu, Ximing
Li, Liunian Harold
Chan, Aaron
Hessel, Jack
Choi, Yejin
Ren, Xiang
Computation and Language
Large language models (LMs) are capable of generating free-text rationales to aid question answering. However, prior work 1) suggests that useful self-rationalization is emergent only at significant scales (e.g., 175B parameter GPT-3); and 2) focuses largely on downstream performance, ignoring the semantics of the rationales themselves, e.g., are they faithful, true, and helpful for humans? In this work, we enable small-scale LMs (approx. 200x smaller than GPT-3) to generate rationales that not only improve downstream task performance, but are also more plausible, consistent, and diverse, assessed both by automatic and human evaluation. Our method, MaRio (Multi-rewArd RatIOnalization), is a multi-reward conditioned self-rationalization algorithm that optimizes multiple distinct properties like plausibility, diversity and consistency. Results on five difficult question-answering datasets StrategyQA, QuaRel, OpenBookQA, NumerSense and QASC show that not only does MaRio improve task accuracy, but it also improves the self-rationalization quality of small LMs across the aforementioned axes better than a supervised fine-tuning (SFT) baseline. Extensive human evaluations confirm that MaRio rationales are preferred vs. SFT rationales, as well as qualitative improvements in plausibility and consistency.
title Tailoring Self-Rationalizers with Multi-Reward Distillation
topic Computation and Language
url https://arxiv.org/abs/2311.02805