HackerPapers

arXiv:2401.05749cs.CL

A Shocking Amount of the Web is Machine Translated: Insights from Multi-Way Parallelism

Brian Thompson, Mehak Preet Dhaliwal, Peter Frisch, Tobias Domhan, Marcello Federico

Abstract

We show that content on the web is often translated into many languages, and the low quality of these multi-way translations indicates they were likely created using Machine Translation (MT). Multi-way parallel, machine generated content not only dominates the translations in lower resource languages; it also constitutes a large fraction of the total web content in those languages. We also find evidence of a selection bias in the type of content which is translated into many languages, consistent with low quality English content being translated en masse into many lower resource languages, via MT. Our work raises serious concerns about training models such as multilingual large language models on both monolingual and bilingual data scraped from the web.

arXiv comments
Accepted at ACL Findings 2024
Journal reference
DOI
Subjects
cs.CLComputation and Language cs.AIArtificial Intelligence

On Hacker News

Linked by 3 identifiable commenters in 3 comments across 2 threads, from Jan 2024 to Dec 2025. Found in 1 HN submission from Jan 2024 to Jan 2024. Open submission discussions.

Comments linking this paper, per year.

Discussed in comments2 threads

    • easily 30-40%. you could see trends even as far back as 2020. same patterns as with reddit. e.g. bunch of old accounts with random bursts of activity, often posting to old / stale threads (e.g. 2-3 days old, when no one is looking at them), and often saying generic one liners or…

      red-iron-pineView on HNParent context

    • The study they refer to [1] seems to be about machine-translated versions of human-written pages (that exist on a large scale for a ~decade already). The article somehow blows it out of proportion, it's not like most of what you're reading is generated by the current crop of…

      orbital-decayView on HNParent context

    • The title is clickbait. The actual study[0] reads as follows: > Multi-way parallel, machine generated content not only dominates the translations in lower resource languages; it also constitutes a large fraction of the total web content in those languages This is talking…

      chacham15View on HNParent context

Submitted to Hacker News

Connected through shared HN threads, not inferred similarity or a recommendation.