Comparing Chat GPT and Claude Vestluse postitaja: Philippe Locquet
|
|---|
Philippe Locquet 
Portugal
Local time: 15:13
Liige (2013)
inglise - prantsuse
+ ...
I’m starting this thread because I don’t think there is information on this anywhere at ProZ.
Test conditions : 1 file (docx). English to French. No editing. Neutral translation prompt for both. File pre-translated as a whole before segmentation (Using full context before segmentation improves text quality, not many tools do this. I used ... See more I’m starting this thread because I don’t think there is information on this anywhere at ProZ.
Test conditions : 1 file (docx). English to French. No editing. Neutral translation prompt for both. File pre-translated as a whole before segmentation (Using full context before segmentation improves text quality, not many tools do this. I used Plus Tools, see this forum to learn more: https://www.proz.com/forum/wordfast_support/370853-tools_plus_tools_a_new_style_of_cat_free_now_very_refreshing_approach_imo.html).
Once the translation was done, I opened MS Word and compared both files. The screenshots show the following markup: Original text = Chat GPT; added editions = Claude.
I attach screenshots below of the results so you can check them out if you are fluent in these languages.
My observations:
Claude applies the US English capitalizing words style to French, bad.
Claude handles better spaces around tags, good
Writing style: Chat GPT very formal. Claude less formal, sometimes uses wording that feels more modern.
Choice of words: Chat GPT, usually good meaning wise. Claude: has a tendency to use the wrong words.
Both fail to understand some idiomatic expressions.
Pricing: according to https://context.ai/compare/gpt-4o/claude-3-sonnet “GPT-4o is roughly 66.7% more expensive compared to Claude 3 Sonnet for input tokens and the same price for output tokens.” In my experience, 5 USD (minimum in Claude) goes quite a long way.
Speed: Claude is 2x faster than Chat GPT
Conclusion: for en2fr, Claude would require more word replacement than Chat GPT. Chat GPT would require more space fixing than Claude. When prompts are neutral, Chat GPT feels more formal than Claude.
It may be possible to improve output a little by tuning the prompts, but I doubt it will change things a lot.
So, if you are looking to use one of these as an MT (never ever do this on proprietary or sensitive content), I would prefer Claude because of speed and because I feel since we will edit the output anyway because of style, I feel Claude will get me there faster than Chat GPT, especially when using my own glossary.
Of-course things will change dramatically depending on the language pair you will be testing.
I hope this will be useful to some of you. 😊
My bests 👍😎
 ▲ Collapse
| | | | James McVay
Ameerika Ühendriigid
Local time: 10:13
vene - inglise
+ ...
| Thoughtful prompting is key to good results | Dec 18, 2024 |
I'm a mostly retired translator at this point in my life, so I don't have an axe to grind over use of AI as a translation aid. I think you will find that many people here are very resistant to the idea.
I've been experimenting with use of LLM AIs as translation assistants for two years now. I initially found that, with ChatGPT, a neutral prompt yielded translations much like those done by Google Translate or DeepL, so I began testing various prompts intended to give me translations ... See more I'm a mostly retired translator at this point in my life, so I don't have an axe to grind over use of AI as a translation aid. I think you will find that many people here are very resistant to the idea.
I've been experimenting with use of LLM AIs as translation assistants for two years now. I initially found that, with ChatGPT, a neutral prompt yielded translations much like those done by Google Translate or DeepL, so I began testing various prompts intended to give me translations in more natural English. I've experimented with ChatGPT, Claude, CoPilot, and Gemini as those models have been released. One thing I've found is that not all AIs are equally proficient at translation.
But my point here is that the content and wording of the prompt makes a big difference in the quality of the result. i've found that prompting the AI to maintain the voice, register, tone, and style of the original are essential. Other instructions relevant to the source should be added. For example, it's a good idea to tell an AI what field a document is in. If experience reveals that an AI has difficulty with something like capitalization, a reminder of the target language's rules will usually correct the problem.
Sometimes, a simple correction can yield lasting results. In Russian, dialogue is normally indicated by an initial dash. At one point, ChatGPT began reading Russian dialogue as bullet points and treating them as such in the English translation. I pointed that out in a reply prompt, ChatGPT added it to its memory of our interactions, and ceased making that error in future translations.
I consider the prompts I use for paid translations proprietary, but I will share the simple prompt I use for pro bono literary translation I do as a favor for an author friend. It yields translations I treat as a first draft, and they do need light editing, but they are a great time saver. Frankly, I wouldn't be doing this favor for my friend if it weren't quick and easy.
Here is that prompt:
You are an accomplished literary translator. Translate the following Russian short story into colloquial English suitable for publication. The translation should capture the voice, tone, style, and register of the original . Avoid overly formal language or archaic expressions. Follow normal English word-order conventions. Aim for an engaging narrative voice that resonates with contemporary readers. When translating idioms, use a corresponding English idiom wherever possible. If not possible, create a metaphor that conveys the same meaning. Here is the first part of the story:
[Edited at 2024-12-18 23:53 GMT] ▲ Collapse
| | | | Sellele foorumile pole määratud ühtegi moderaatorit. To report site rules violations or get help, please contact site staff » Comparing Chat GPT and Claude | Trados Studio 2022 Freelance | The leading translation software used by over 270,000 translators.
Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop
and cloud solution, empowering you to work in the most efficient and cost-effective way.
More info » |
| | TM-Town | Manage your TMs and Terms ... and boost your translation business
Are you ready for something fresh in the industry? TM-Town is a unique new site for you -- the freelance translator -- to store, manage and share translation memories (TMs) and glossaries...and potentially meet new clients on the basis of your prior work.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
