Language Modelling with Pixels

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F25%3ATYUVJKMT" target="_blank" >RIV/00216208:11320/25:TYUVJKMT - isvavai.cz</a>

Result on the web

<a href="http://arxiv.org/abs/2207.06991" target="_blank" >http://arxiv.org/abs/2207.06991</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.48550/arXiv.2207.06991" target="_blank" >10.48550/arXiv.2207.06991</a>

Result language

angličtina

Original language name

Language Modelling with Pixels

Original language description

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.

Czech name

—

Czech description

—

Type

J_ost - Miscellaneous article in a specialist periodical

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

—

Continuities

—

Publication year

2023

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Name of the periodical

ArXiv

ISSN

2331-8422

e-ISSN

—

Volume of the periodical

2023

Issue of the periodical within the volume

2023-04-26

Country of publishing house

US - UNITED STATES

Number of pages

32

Pages from-to

1-32

UT code for WoS article

—

EID of the result in the Scopus database

—