Self-supervised Analogical Learning using Language Models

Ben Zhou, Sarthak Jain, Yi Zhang, Qiang Ning, Shuai Wang Yassine Benajiba Dan Roth, Amazon, University of Pennsylvania · 2025

Read Self-supervised Analogical Learning using Language Models by Ben Zhou, Sarthak Jain, Yi Zhang, Qiang Ning, Shuai Wang Yassine Benajiba Dan Roth, Amazon, University of Pennsylvania (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00996.txt.

Description

Ben Zhou, Sarthak Jain, Yi Zhang, Qiang Ning, Shuai Wang, Yassine Benajiba, and Dan Roth wrote this paper in two thousand twenty-five. It teaches language models by analogy, so a fix learned on a known case can be carried to a rare twin.

Models often fail on odd forms of a task they ace in its common form. The frame, SAL, trains the model to state the abstract plan of a solved case in symbols, then lay that plan on a hard case. No hand marks are needed, since the model mines its own good cases. Gains show across question, math, and multi-hop test sets. Checks show the skill is more portable and can be steered.

Catalog record

CREATORBen Zhou, Sarthak Jain, Yi Zhang, Qiang Ning, Shuai Wang Yassine Benajiba Dan Roth, Amazon, University of Pennsylvania
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-00996

Browse the full library · About the Tower of Babel