AI and the Evolving Challenge of Language

Augmented Intelligence Brings New Powers to Tackle Language

In the previous blog, Artificial or Augmented: Intelligence Is in the Application, we discussed the importance of precision and vocabulary when talking to people about augmented intelligence (AI). While we as humans need to be clear in our speech with each other, we also need to commit sufficient resources into improving how we communicate with our machine counterparts. 

This is where the discipline of natural language processing (NLP) comes in. It is concerned with the data-rich, resource heavy interactions between computers and human languages. Fortunately, AI and machine learning (ML), are allowing for the processing and analysis of large amounts of natural language data created from human speech. Take Alexa for example, a “robot” who is human-like in her ability to understand and carry out basic commands or answer questions. However, with the increasing sophistication of AI capabilities, this technology has the potential to do much more than simply understand a basic declarative sentence so that it can play a song or tell you a joke. But to jump to the next level of understanding and translation, we need the power of AI to help in areas like language translation.

Past approaches to translation focused on ML. “If you look at a large enough corpus of the language, you just get millions and millions and millions of articulations,” says Dr. Anthony Scriffignano, Chief Data Scientist at Dun & Bradstreet. A system based on the machine learning approach could ingest that information and do all kinds of fancy things to get good at translating that language. 

This approach has shortcomings, however, because language is always evolving. For example, the word “cool” in the distant past would just mean not hot, but in contemporary times it can also mean hip or trendy. “I don’t know whether cool is cool anymore or not,” Scriffignano said.

Modern English has many more examples. “I don’t know whether “good” is bad and “bad” is good,” said Scriffignano. “You also have neologisms that either stick or don’t, like ‘hashtag,’ or even people speaking in hashtags. [Or] people saying something like ‘L-O-L’ and using it as a word.” A traditional ML algorithm would fail to cope with these semantic nuances. 

Branch of AI
Chief Data Scientist Anthony Scriffignano, like many other business leaders, is inundated by emails, requests, articles, etc. There are augmented intelligence solutions being developed that can help business leaders decide what information is important, so less time must be spent filtering out things that are not so impactful.
 

“We’re studying some of the confounding characteristics of language like neologism, intentional manipulations of spelling and grammar, and sarcasm,” Scriffignano continued. “And we are building methods to assess the degree of that confounding characteristic in a corpus of articulations … the degree to which these confounding characteristics of language will confound whatever it is that we’re trying to do, like entity extraction or sentiment attribution. This is on the edge of computational linguistics.”

The need to understand words in their context highlights another distinction between other ML and current AI. For instance, when moving between two languages, ML would be able to do a straight translation. But AI has the potential to bring a higher capability to the table. Rather than straight word-for-word translation, a better approach would also account for the subtle nuances that can sometimes be really important. 

Natural language processing plays a role in all of this. But Scriffignano warns, “Let’s be careful [to remember] that just because the software can turn the phonemes that are coming out of your mouth into letters of the alphabet, there isn’t necessarily any recognition going on there.” 

…when looking for patterns or actionable information in large datasets, augmented intelligence could complement the work of an expert to find new insights in a faster time.…
Dr. Anthony Scriffignano, Senior Vice President and Chief Data Scientist
 

“There’s another step called ‘computational linguistics,’ where you make the language computable or consumable to an algorithm, and that’s where natural language processing comes in,” said Scriffignano. He explained how AI principles applied to understanding the spoken word have applications in many other fields. For example, “identifying fraud often requires both spotting anomalies and understanding intent. Similarly, when looking for patterns or actionable information in large datasets, augmented intelligence could complement the work of an expert to find new insights in a faster time.” 

 

The previous example is one of many areas where AI-powered NLP can have a significant positive impact by increasing their human counterparts’ ability to quickly and efficiently identify patterns and spot irregularities. However, our success depends on the precision and pragmatism of our expectations, as well as the words we choose. “We often design things, we use them for purposes they were not intended to be used for, and then we get upset at them for not functioning the way we intended them to work when they were never designed for that in the first place,” muses Scriffignano. 

As the role of augmented intelligence in everyday human processes becomes more prevalent, humans will need to develop new skills and knowledge to work effectively alongside their new assistants. To achieve the maximum benefit of employing these new technologies, we need to truly understand their capabilities and what our role is in implementing them. 

To read more of Dr. Scriffognano’s thought leadership on AI, check an earlier article entitled, It’s Not About AI: A Conversation Worth Having After AI World 2018.