Note: This is a re-post from my old blog (with no longer exists)
Chat Bots are quickly becoming a global trend. And if you are in a company who faces customers (e.g. B2C business or one which simply handles customer support calls), chances are that your business has thought of implementing a chat bot. But what if English (or other supported languages) is not good enough for you? What if your language isn’t supported nor is it in the product roadmap? This post is about a possible workaround using text-to-text translation.
Chat Bots are not just about doing a closest-match text search and returning an answer. Language Understanding is important. For Microsoft Azure, this cognitive service is called Language Understanding Intelligent Service, or LUIS for short. LUIS is designed to identify valuable information in conversations, LUIS interprets user goals (intents) and distills valuable information from sentences (entities), for a high quality, nuanced language model. LUIS integrates seamlessly with the Azure Bot Service, making it easy to create a sophisticated bot.
At the time of this writing, LUIS supports 13 languages so far.
Use Text Translation (you probably saw this coming). Now, machine translation is ugly and not very human reader friendly. But it works because we are simply after the intent. Your chat bot can still respond with canned local-language messages.
Yes, the utterance may be ugly, so this may require more training, but it works!
Here is a sample output of this workaround at work.
You may download my RazType™ source code here.
Note: As this was I demo that I created last year, the source code still uses the previous Bot Framework v3. At the time of this writing, Bot Framework v4 has just announced general availability.