How do traditional machine translation systems work…
0 reactions
How do traditional machine translation systems work?
Recently I've been researching machine translation software for a project and came by this repository: https://github.com/SakiRinn/LiveCaptions-Translator
It has a list of translation services and strongly recommends using LLM-based ones, but also lists a number of "traditional" ones, like Google Translate, DeepL or LibreTranslate.
I intuitively understand how an LLM operates on high level - it's fed the existing text as tokens and it predicts the most likely next token (or set of tokens with varying probabilities). All the learning is within its deep neural model.
However I have no idea how does a "traditional" machine translation system work? Is it analogous in some way - neural net predicting tokens/output based on input, just trained purely on translating language?
Or is it something very different? Are are multiple techniques?
#science
Recently I've been researching machine translation software for a project and came by this repository: https://github.com/SakiRinn/LiveCaptions-Translator
It has a list of translation services and strongly recommends using LLM-based ones, but also lists a number of "traditional" ones, like Google Translate, DeepL or LibreTranslate.
I intuitively understand how an LLM operates on high level - it's fed the existing text as tokens and it predicts the most likely next token (or set of tokens with varying probabilities). All the learning is within its deep neural model.
However I have no idea how does a "traditional" machine translation system work? Is it analogous in some way - neural net predicting tokens/output based on input, just trained purely on translating language?
Or is it something very different? Are are multiple techniques?
#science
| 0 mlx | |
| 2 | |
| 0 |