Please use this identifier to cite or link to this item:Volltext verfügbar? / Dokumentlieferung
|Title:||On Training Instance Selection for Few-Shot Neural Text Generation|
|Year of Publication:||2021|
|DDC notations:||400 Language, linguistics|
|Abstract:||Large-scale pretrained language models have led to dramatic improvements in text generation. Impressive performance can be achieved by finetuning only on a small number of instances (few-shot setting). Nonetheless, almost all previous work simply applies random sampling to select the few-shot training instances. Little to no attention has been paid to the selection strategies and how they would affect model performance. In this work, we present a study on training instance selection in few-shot neural text generation. The selection decision is made based only on the unlabeled data so as to identify the most worthwhile data points that should be annotated under some budget of labeling cost. Based on the intuition that the few-shot training instances should be diverse and representative of the entire data distribution, we propose a simple selection strategy with K-means clustering. We show that even with the naive clustering-based approach, the generation models consistently outperform random sampling on three text generation tasks: data-to-text generation, document summarization and question generation. We hope that this work will call for more attention on this largely unexplored area.|
|URL of the first publication:||https://arxiv.org/abs/2107.03176|
|Link to this record:||urn:nbn:de:bsz:291--ds-386450|
|Date of registration:||3-Jan-2023|
|Faculty:||MI - Fakultät für Mathematik und Informatik|
|Department:||MI - Informatik|
|Professorship:||MI - Prof. Dr. Vera Demberg|
|Collections:||SciDok - Der Wissenschaftsserver der Universität des Saarlandes|
Files for this record:
There are no files associated with this item.
Items in SciDok are protected by copyright, with all rights reserved, unless otherwise indicated.