As Amazon makes its first foray into the Asian market by releasing Amazon Echo in India and Japan, we can’t help but ask ourselves: Does Alexa work in India? What is the present and future of Chatbots and VAs?
With the rise of Chatbot usage most marketing departments either already have a bot or are looking to implement one within the next few years. Chatbots offer great possibilities: they can save companies money, help clients get services faster and let’s be honest, they are just cool to chat with.
However, many companies are finding the training process to be arduous and they are unable to achieve basic rates of success. Coming back to the release of Amazon Echo in India and Japan, we want to tackle the problem of creating a chatbot successful for all fields and countries.
What are the steps to take to improve the quality of human-chatbot interactions?
There are two action fronts, one is on the side of the bot platform (in this case Amazon). So far, we know they have taken several steps toward customizing the bot for the Indian public:
- Including a new English voice with Indian English pronunciation and intonation
- New local skills by Indian developers
- Local knowledge such as Indian names, cities, music/movie titles etc.
The second front is the linguistic one: chatbots are still struggling to understand American English so how can we make them understand other variants of English, let alone other languages? American English is probably the variant with the most training data available, and most likely the mother tongue (or at least second language) of the majority of the workers creating and training chatbots.
How can Bitext technology help?
The technology at Bitext can help with the latter challenge in two main ways:
- The training process can be shortened and ameliorated with Bitext Generation of Variants. Imagine you have a corpus of 200 sentences that represents the main questions/requests your clients can ask your bot. It is not tagged and it is not enough to train a functional bot because clients can ask one thing in dozens of different ways. At Bitext, we can use those sentences and create dozens of already-tagged variants for each one, which translates into a sizeable tagged training corpus ready in a couple days rather than many months.
Our technology can include customized real world knowledge, which means that if a person from India asks “where is the departmental store closest to my colony” our system understands it just as well as if an American asks “where is the department store closest to my neighborhood.”
- Once your training is complete, you can add Bitext rewriting technology to handle those structures that are syntactically too complicated for a bot platform to interpret.
- Coordination: “I want a pizza with pepperoni and cheese and a coke”
2 rewritings for this sentence:
I want a PIZZA with [pepperoni and cheese]
I want a COKE
- Negation: “I want a pizza with pepperoni but no cheese, please”
Pizza - with: pepperoni
Pizza - without: cheese
In the context of this booming market, companies do not want to be the last to integrate Chatbot technology in their customer relations, with the technology offered at Bitext, you can be the first to have a fully functional bot ready in a week. Get to know more about this solution!