Language Modelling with Pixels
Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Standard
Language Modelling with Pixels. / Rust, Phillip; Lotz, Jonas F.; Bugliarello, Emanuele; Salesky, Elizabeth; Lhoneux, Miryam de; Elliott, Desmond.
The Eleventh International Conference on Learning Representations. arXiv.org, 2023.Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - Language Modelling with Pixels
AU - Rust, Phillip
AU - Lotz, Jonas F.
AU - Bugliarello, Emanuele
AU - Salesky, Elizabeth
AU - Lhoneux, Miryam de
AU - Elliott, Desmond
PY - 2023
Y1 - 2023
N2 - Language models are defined over a finite set of inputs, which creates a vocabulary bottleneck when we attempt to scale the number of supported languages. Tackling this bottleneck results in a trade-off between what can be represented in the embedding matrix and computational issues in the output layer. This paper introduces PIXEL, the Pixel-based Encoder of Language, which suffers from neither of these issues. PIXEL is a pretrained language model that renders text as images, making it possible to transfer representations across languages based on orthographic similarity or the co-activation of pixels. PIXEL is trained to reconstruct the pixels of masked patches instead of predicting a distribution over tokens. We pretrain the 86M parameter PIXEL model on the same English data as BERT and evaluate on syntactic and semantic tasks in typologically diverse languages, including various non-Latin scripts. We find that PIXEL substantially outperforms BERT on syntactic and semantic processing tasks on scripts that are not found in the pretraining data, but PIXEL is slightly weaker than BERT when working with Latin scripts. Furthermore, we find that PIXEL is more robust than BERT to orthographic attacks and linguistic code-switching, further confirming the benefits of modelling language with pixels.
AB - Language models are defined over a finite set of inputs, which creates a vocabulary bottleneck when we attempt to scale the number of supported languages. Tackling this bottleneck results in a trade-off between what can be represented in the embedding matrix and computational issues in the output layer. This paper introduces PIXEL, the Pixel-based Encoder of Language, which suffers from neither of these issues. PIXEL is a pretrained language model that renders text as images, making it possible to transfer representations across languages based on orthographic similarity or the co-activation of pixels. PIXEL is trained to reconstruct the pixels of masked patches instead of predicting a distribution over tokens. We pretrain the 86M parameter PIXEL model on the same English data as BERT and evaluate on syntactic and semantic tasks in typologically diverse languages, including various non-Latin scripts. We find that PIXEL substantially outperforms BERT on syntactic and semantic processing tasks on scripts that are not found in the pretraining data, but PIXEL is slightly weaker than BERT when working with Latin scripts. Furthermore, we find that PIXEL is more robust than BERT to orthographic attacks and linguistic code-switching, further confirming the benefits of modelling language with pixels.
KW - cs.CL
KW - cs.AI
KW - cs.CV
KW - cs.LG
M3 - Article in proceedings
BT - The Eleventh International Conference on Learning Representations
PB - arXiv.org
T2 - 11h International Conference on Learning Representations - ICLR 2023
Y2 - 1 May 2023 through 5 May 2023
ER -
ID: 379722288