"Machine language deceives readers about authorship and claims to truth. In the context of AI systems, the term 'translation' is used for a form of machine language with no person behind it, completely unrelated to the exacting, thoughtful, responsible work that goes into a human translation."
Today I tried Gemini Advanced using my Esperanto translation test question: "On Thursday, I will go to the park and walk my dog." Many other LLMs get "walk my dog" wrong. They use "promenos" which is an intransitive verb.
Let's compare:
GPT-4 (OpenAI): Ĵaŭde, mi iros al la parko kaj promenos mian hundon.
Gemini Advanced (Google): Ĵaŭde, mi iros al la parko kaj promenigos mian hundon.
Gemini (formerly Bard) (Google): Ĵaŭdon mi iros al la parko kaj promendigos mian hundon.
Claude (Anthropic): Ĝisdaten mi iros al la parko kaj promenos mian hundon.
Congrats to Google's Gemini Advanced for getting it right!
I have to admit, I'm afraid of Docker. Is there an easier way of running #LibreTranslate? Maybe some instance with a limited free API service? I wanted to try some #MachineTranslation on my open markdown teaching materials repo :/
How would commercial translation and transcription models perform when faced with genuine news content, exacting journalists and the pressing demands of a newsroom? We devised a quality assessment to help BBC journalists decide.
Out of curiosity as #exam season will start soon: We currently have a rule that students can use #AI tools. Without acknowledging so when used for grammar, style, spelling, etc. But they have to mention what, when, and how they did if they use it for generating ideas, data, whole paragraphs, etc.
Surprisingly, we don't have regulations for #MachineTranslation (MT) tools. For sit-in exams students are allowed paper dictionaries (!) and that's it.
> Automated translation of web content is now available to Firefox users! Unlike cloud-based alternatives, translation is done locally in Firefox, so that the text being translated does not leave your machine.
I just discovered that if you put Korean text into Google Translate and set the source language to Japanese, sometimes it will output something completely wrong ("hello", "I'm sorry", etc). Whereas when the source language is set to Chinese, it just spits the source text back out. This could be consequential in cases where language auto-detection is wrong. #MachineTranslation#GoogleTranslate
If you know anything about how #AI is changing the game for subtitling and dubbing translation, please share links with me below.
I’d love to know more about stuff such as machine translation post-editing for subtitling. I’ve never done it, so it’d be useful to hear about other translators’ experience.
This is pretty awful. Like, bravo for getting this far, but they should hide Japanese as an option for now. (The Japanese text says, "Understood.") #machineTranslation#translation
Hi everyone! A sort of #introduction seems in order so: I'm a #ComputationalLinguistics and #NLP prof and researchwr working on minority languages and historical languages. My main and favourite activity is to make tools that benefits language communities and linguists but I really love making tools in general so I do a lot of other stuff.
Troer (temp name) is a #Breton-#French translator, so far in early alpha but it is already quite decent and we are very, very eager to have people test it, so if you are a Breton speaker or know one, please hit us up https://huggingface.co/spaces/lgrobol/troer