AI Challenges Are Being Met but... Who Broke Ground Here?

While AI is one of the most important trends nowadays, there are still challenges to overcome. Apart from common technical issues such as a lack of quality data, there is much  beyond its abilities for an AI to effectively understand and react when it comes to human-machine interaction.

It is an undeniable fact that humans speak as they wish using natural language. What if a user wants to ask two things at once? What if a user changes their mind during a conversation? Oddly enough, simple structures of human communication as coordination or negation are posing big challenges for traditional AI systems. To solve this, all the queries including such phenomena must be simplified making it easier for a bot to understand everything.


Let us take a look at the behavior of an AI-powered virtual bot of the Wall Street Journal. Here you have a real conversation where a user asks for news on two different topics using the conjunction ‘and’:



As you can see in the example above, such a simple utterance wasn’t properly understood by the virtual agent. While some developers are now finding a way out, Bitext solved this problem long ago.


Thanks to our Deep Linguistic Platform, the query can be easily rewritten into different sentences with one intent each. Here there is no need for existing chatbot frameworks to redesign their architectures to properly process both intents. Instead, Bitext system takes the sentence ‘show me news on Apple and China’ and produces two different sentences that will be sent to the platform: ‘show news Apple’ and ‘show news China’ can be processed sequentially by existing frameworks. Splitting a complex query into different simplified sentences allows for multiple intents to be recognized. Take a look at our post “How to Solve the Double-Intent Issue” published last year and you’d see how it works.


What also can be a pain in the neck for bot developers is to make their bots detect negation in a sentence. Something that usually seems not to be that easy to target but… See what happened when we tried to look for restaurants but not Chinese, through the Trip Advisor bot:



Here you can clearly see how the bot detects the keywords ‘restaurant’ and ‘Chinese’ and ignores the real sense of the sentence, which is crucial. Not long ago, IBM Watson developers brought to light that they were working on negation and disambiguation. If such a tech giant is worrying about these issues, it means that there is a growing need to solve them. That’s not bad at all considering the difficulties that it can pose. Nevertheless, Bitext technology has been once again well ahead of them, as you can see on our publication from last year “How to Make Your Chatbot More Human-Like”. At that time, we realized that many bots didn’t understand negation in a sentence because they were built based on a keyword approach. Bitext approach considers the linguistic context in which a keyword appears helping the bot react accordingly.


Bitext Query Simplification + Negation Detection service’ takes a text input and outputs a simplified version, by splitting it into smaller sentences, when needed, reducing it to its essential components and including extra information about its polarity (positive or negative). This simplified version speeds up your bot training and improves, at the same time, its understanding skills.


Coordination and negation are constituent parts of human language and Bitext linguistic technology is a perfect choice to target them. After years of experience in the field, no linguistic challenge is too complex for us. Make your life easier and contact us to enhance your AI solutions as easy as clicking a button.


Subscribe Here!