New Features of Dialogflow

Introduction:

Google Dialogflow is a platform which can be used to develop chatbots. It also helps developers integrate with many applications like Google Assistant, Skype, Telegram, Slack, etc. On July 23, 2018, Dialogflow announced few new features that can help the developers build chatbots more efficiently.

Here is the List of Features:

Knowledge Base
Phone Gateway
Environment
Speech
Automatic spell correction

Knowledge Base:

Before Knowledge Base, it was difficult for a developer to add each and every FAQ questions one by one, including different utterances. This required too much manual work for a very simple task.

After Knowledge Base for a developer, it is as easy as clicking a button to add any number of FAQ type questions without missing out any questions. The only Limitation here is that as of now it requires data to be in text and structured format like HTML page containing only questions and answers (ex - https://www.samplestore.com/site/faq), or an excel with two columns as ‘Question’ and ‘Answer’. HTML file has to be publically open i.e it should not require any authentication.

Phone Gateway:

Letting a bot answer your phone call seemed like quite far away, but after an introduction on ‘Phone Gateway’, it’s quite possible now. Phone Gateway has made easy integration of Dialogflow bot to a phone number. This means any bot currently working on Dialogflow can now be accessed over the phone. Integration with the phone makes sense only for a bot that is completely text specific and does not use any rich UI as a response.

Google is also working on WaveNet Technology, which will be very beneficial for phone Gateway as the bot might sound so ‘human-like’ over the phone that it will be hard to differentiate it from a human. Large companies can take benefits from this by letting a bot answer to customer queries instead of an actual human.

Phone Gateway as of now still struggles to understand user utterance (it is still in BETA phase). On integration with Phone Gateway you get a free number, but it is a temporary number and it is valid only for 30 days. You can get a permanent number by upgrading to Enterprise Edition.

Environments:

Version control was a very big and important feature that was missing from Dialogflow, they have introduced it as ‘Environments’. Environments help us build a different version of our bot depending on which platform we want to release which features like Testing, Staging, Production. Earlier it was difficult to maintain a version and developers ended up making 2-3 clones of the same bot.

Speech:

Dialogflow converts Text to Speech while reading the response, but it is a simple conversion of word to speech without understanding the context of the sentence.

Examples:

When we say “I work in Xoriant” it understands that ‘Xoriant’ is a company and correctly pronounces it, which was not the case earlier, as it pronounced as ‘X orient’.
When we say “You can call me at 9495559999. It understands that it is a phone number and pronounces it correctly “nine four nine five five five nine nine nine nine”, where earlier it used to pronounce as “9 billion 49 million ….”
When we say “My birthday is on 14th March.”, it pronounces as “My birthday is on the 14th of March”, this sound more human-like.

This can help developers, as there will be no need to use SSML while forming the responses.

Automatic spell correction:

With immense data that Google has, it can understand the words with the spelling mistake. Automatic spell correction corrects any spelling mistake and then queries dialogflow with the correct spelling. This is very helpful for developers as due to wrong spelling provided by the user, correct intent doesn’t get selected. Here developers were helpless as they could not do anything about this problem.

Examples

“Qualifications” can be written as “qualication”.
“How are you” can be written as “howare you”.
“I want an apple” can be written as “I want an aple”.

The only limitation to Automatic spell correction is that it only works on general user queries and fails to correct the spelling if the words are ‘domain-specific’.

Conclusion:

“Language is the most natural interface humans understand, and that’s the interface that bots use.” - Matt Schlicht (CEO of Octane AI, Founder of Chatbots Magazine)

We can already see Google making efforts in exposing the Bot over the phone, and with growth in wavenet technology, we might see chatbots which can imitate human speech.