Transfer of Structural Knowledge from Synthetic Languages

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Budnikov, Mikhail, Yamshchikov, Ivan
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912385988558848
author Budnikov, Mikhail
Yamshchikov, Ivan
author_facet Budnikov, Mikhail
Yamshchikov, Ivan
contents This work explores transfer learning from several synthetic languages to English. We investigate the structure of the embeddings in the fine-tuned models, the information they contain, and the capabilities of the fine-tuned models on simple linguistic tasks. We also introduce a new synthetic language that leads to better transfer to English than the languages used in previous research. Finally, we introduce Tiny-Cloze Benchmark - a new synthetic benchmark for natural language understanding that is more informative for less powerful models. We use Tiny-Cloze Benchmark to evaluate fine-tuned models in several domains demonstrating that fine-tuning on a new synthetic language allows for better performance on a variety of tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2505_15769
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Transfer of Structural Knowledge from Synthetic Languages
Budnikov, Mikhail
Yamshchikov, Ivan
Computation and Language
Machine Learning
This work explores transfer learning from several synthetic languages to English. We investigate the structure of the embeddings in the fine-tuned models, the information they contain, and the capabilities of the fine-tuned models on simple linguistic tasks. We also introduce a new synthetic language that leads to better transfer to English than the languages used in previous research. Finally, we introduce Tiny-Cloze Benchmark - a new synthetic benchmark for natural language understanding that is more informative for less powerful models. We use Tiny-Cloze Benchmark to evaluate fine-tuned models in several domains demonstrating that fine-tuning on a new synthetic language allows for better performance on a variety of tasks.
title Transfer of Structural Knowledge from Synthetic Languages
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2505.15769