Paper Image

Generalizing morpheme glossing models for endangered languages

Published on:

5 November 2023

Primary Category:

Computation and Language

Paper Authors:

Michael Ginn,

Alexis Palmer


Key Details

Models tested on texts of unseen genres to evaluate generalization ability

Weight decay optimization improved out-of-distribution performance

Output denoising helped handle unknown morphemes

Iterative pseudo-labeling further adapted models to new domains

AI generated summary

Generalizing morpheme glossing models for endangered languages

This paper investigates strategies to improve the ability of neural models to accurately predict grammatical gloss labels for morphemes in texts of an endangered language. The models are evaluated on texts of genres not seen during training, to test their generalization ability. Techniques like weight decay, output denoising, and iterative pseudo-labeling are applied, achieving a 2% improvement in performance on out-of-distribution test data.

Answers from this paper


No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up