The paraphrase identification task involves measuring semantic similarity
between two short sentences. It is a tricky task, and multilingual paraphrase
identification is even more challenging. In this work, we train a bi-encoder
model in a contrastive manner to detect hard paraphrases