Hello,
I'm here to ask about the meaning of segment embeddings.
From the paper, segment embeddings(Dialog state embeddings) indicates whether it is a profile sentence, utterance of persona 1, or persona 2.
However, in the post, it seems to indicate whether it is persona information, history, or a reply. (From the image with blue/pink/green instances).
In the code, it somehow takes the sequence input's first persona part as speaker1, the utterance of speaker1 as speaker2 and so on.
It will be grateful if you could clarify these terms for me.
Thanks :)
Hello, I'm here to ask about the meaning of segment embeddings. From the paper, segment embeddings(Dialog state embeddings) indicates whether it is a profile sentence, utterance of persona 1, or persona 2. However, in the post, it seems to indicate whether it is persona information, history, or a reply. (From the image with blue/pink/green instances). In the code, it somehow takes the sequence input's first persona part as speaker1, the utterance of speaker1 as speaker2 and so on. It will be grateful if you could clarify these terms for me. Thanks :)
Below is the code I tried to run from the post.