Translationese
Writing that sounds like a direct translation, not natural speech in the new language.
Translationese is when translated text sounds awkward or unnatural in the target language. It often happens because the translator sticks too closely to the grammar or common phrases of the original language.
For example, if a sentence in French uses a common structure that doesn't sound right when directly translated into English, the result might be translationese. Good translation aims to avoid this by making the text sound like it was originally written in the new language.
Translationese refers to a linguistic style in translated text that retains structural or idiomatic features of the source language, deviating from natural target language usage. This can manifest as unusual word order, literal translations of idioms, or overuse of grammatical constructions common in the source but rare in the target.
While early machine translation systems often produced noticeable translationese, modern large language models are increasingly capable of producing fluent and natural-sounding translations. However, subtle forms can still appear, especially in low-resource language pairs or when translating highly specialized or culturally nuanced content.
The term is relevant as advanced AI models like GPT-4 are being used to preserve and adapt languages. This work involves careful translation and adaptation, highlighting the ongoing challenge of avoiding translationese and ensuring natural language output.
- Preserving languages for the futureOpenAI · OpenAI News ·
- OpenAI ChatGPT
- OpenAI's GPT-4 was used in a project to preserve Icelandic, a task that requires careful translation to avoid unnatural language patterns.
- Idiom
- An idiom is a phrase whose meaning cannot be deduced from the literal mea
- Fluency
- Fluency refers to the smooth and natural flow of language. Translatione
- Source Language
- The original language of a text. Translationese is a characteri
The term translationese has been used for decades in translation studies to describe the characteristic style of translated texts. It's a descriptive term rather than one tied to a specific paper or company.
Ask an AI model to translate a common idiom from English into another language, then ask it to translate the result back into English. Compare the original idiom to the final translation to see if it sounds natural or awkward.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- How do AI models avoid translationese?
- Modern AI models are trained on massive amounts of diverse text, including both original content and high-quality translations. This helps them learn natural patterns of language and avoid simply mirroring the source text's structure.
- Is translationese always bad?
- Generally, yes, as it makes the text sound unnatural. However, in some specific contexts, retaining certain source language structures might be intentional for stylistic effect, though this is rare.
First seen on Subvolts Mar 14, 2023. Updated Sep 15, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.