 What are sentence transformers? Computers don't understand language, but they understand numbers, so sentences are converted into numeric representations called vectors. Early methods converted sentences to n-gram vectors and tf-idf vectors, but these were sparse and too large to process by computers, so dense vector methods were used. But still even taking the average of these vectors to generate sentences delivered vectors that were super low quality. And so to deal with this, sentence transformers were introduced. They were networks that are built on top of BERT that can better capture the meaning of a sentence. For more information, check out this accompanying video.