notesum.ai

Published at November 12

Derivational Morphology Reveals Analogical Generalization in Large Language Models

cs.CL

cs.AI

cs.LG

Released Date: November 12, 2024

Authors: Valentin Hofmann¹, Leonie Weissweiler², David Mortensen³, Hinrich Schütze², Janet Pierrehumbert⁴

Aff.: ¹Allen Institute for AI; ²LMU Munich; ³Carnegie Mellon University; ⁴University of Oxford

Arxiv: http://arxiv.org/abs/2411.07990v1

Refer to caption

		Examples		Counts		MGL		GCM
Regularity	Suffix	Real	Nonce	-ity	-ness	Type	Token	Type	Token
High	-able	available	tegornable	11,081	1,034	.893	.893	.893	.893
High	-ish	selfish	friquish	0	1,502	.997	.997	.997	.997
Low	-ive	sensitive	cormasive	4,508	2,438	^†.658	.662	^†.622	.688
Low	-ous	luminous	momogorous	1,372	2,450	^†.657	^†.613	^†.610	.703