Do these semantic sentence embeddings produce better results than semantic word embeddings?
If yes, why are semantic sentence embeddings not used more often? They're so much more compact.
If no, what was the point of making these? Just for research purposes?
What type of embedding is best for accurately replicating a person's speech pattern?
It purely depends on our use case.
Feel free to scan the article below for an overview of Sentence vs Word embeddings. I'm happy to discuss more about the same as well.
https://spotintelligence.com/2022/12/17/sentence-embedding/
Listing "Limitations on context" under "Disadvantages of sentence embeddings" seems quite odd, as one of the primary benefits of sentence embeddings over word embeddings is that it can understand context much much better than word embeddings, for which the same word in different contexts will always have the same embedding. This is often problematic: e.g. "crane" can't be contextually understood to be the bird or the large lifting equipment with word embeddings.
@siddhant-kumar lol yeah tomaarsen is right, that wasn't a very informative article. The use case I have is replicating my speech patterns because I want to finetune a chatbot model to talk like me