Before we begin with Seq2Seq modeling, I would like to share an anecdote that I witnessed at Bengaluru Airport in India. A traveler from China was trying to order a meal at one of the airport restaurants and the butler was unable to comprehend Mandarin. An onlooker stepped in and used Google Translate to convert the English being spoken by the store owner into Mandarin and vice versa. Seq2Seq modeling has helped build applications such as Google Translate, which made the conversation between these folks possible.
When we try to build chatbots or language translating systems, we essentially try to convert a sequence of text of some arbitrary length into another sequence of text of some unknown length. For example, the same chatbot might respond with one word or multiple words depending on the conversational prompts coming from the other party involved in the conversation. We do not always respond with text of the same length. We saw this as one of the many-to-many variants...