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

2024/01/11 by Brian Thompson, Brian J. Thompson, Thompson, Brian +8 · 7 voices · 13 citations
Computer Science · #Natural Language Processing Techniques #Text Readability and Simplification #Topic Modeling #cs.AI #cs.CL

paper · pdf · doi:10.48550/arxiv.2401.05749

openalex publication_date 2024/01/11 · openalex created_date 2024/01/13 · openalex updated_date 2026/07/28

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.

Cited by

• A question of alignment – AI, GenAI and applied linguistics
• Large Language Models as Search Engines: Societal Challenges
• Generative AI in Sociological Research: State of the Discipline
• Automatic Machine Translation Detection Using a Surrogate Multilingual Translation Model
• KletterMix: Climbing Toward High-Quality German Pretraining Data - The Full Report
• The Art of Asking: Multilingual Prompt Optimization for Synthetic Data
• Linguistic Characteristics of AI-Generated Text: A Survey
• The Even Sheen of AI: Kitsch, LLMs, and Homogeneity
• Mussolini and ChatGPT. Examining the Risks of A.I. writing Historical Narratives on Fascism
• How Persuasive Could LLMs Be? A First Study Combining Linguistic-Rhetorical Analysis and User Experiments
• Lessons from complex systems science for AI governance
• Why Do Decision Makers (Not) Use AI? A Cross-Domain Analysis of Factors Impacting AI Adoption
• Generative AI [wikipedia]

Discussions

• Much of the Web Is Machine Translated: Insights from Multi-Way Parallelism [hn, 95 points, 75 comments]
• Fun fact: 57% of all of the text content on the entire web was either generated or translated by AI, according to an August 2024 study by AWS AI Lab. Sleep well. 😵 arxiv.org/pdf/2401.05749 [bsky, 18 points, 2 comments]
• I ain't gonna let misinfo spread from my shit even if it's blowing up so here's actual study article is citing arxiv.org/pdf/2401.05749 It's just translated content from 1 or 2 languages which while a [bsky, 6 points, 1 comments]
• The original study (arxiv.org/pdf/2401.05749) suggests that 57% of text based translated content on the Internet is AI-translated. The headline we have now is 2-3 steps down of game of clickbait telep [bsky, 2 points, 2 comments]
• We all knew it was happening but: AI is eating it's own shit and generating more shit arxiv.org/abs/2401.05749 [bsky, 2 points, 1 comments]
• the paper in which they’ve both misquoted/misunderstood: arxiv.org/pdf/2401.05749 [bsky, 0 points, 1 comments]
• found on the web. Paper [pdf]: “A Shocking Amount of the Web is Machine Translated: Insights from Multi-Way Parallelism” https://arxiv.org/pdf/2401.05749.pdf via https://sfba.social/@drahardja/1118 [bsky, 0 points, 0 comments]

Related

• How Multilingual Are Large Language Models Fine-Tuned for Translation?
• Building Machine Translation Systems for the Next Thousand Languages
• Towards the Next 1000 Languages in Multilingual Machine Translation: Exploring the Synergy Between Supervised and Self-Supervised Learning
• The Reality of Multi-Lingual Machine Translation
• Languages Still Left Behind: Toward a Better Multilingual Machine Translation Benchmark