Gavagai, a Swedish company that began at the Swedish Institute for Computer Science, will be partnering with KTH Royal Institute of Technology to find out what dolphins are talking about.

The research is meant to demonstrate Gavagai’s Word Space Technology, which utilizes artificial intelligence and machine learning to independently decipher and learn languages. So far, with over 20 years of development, Gavagai’s software has learned 40 human languages, modeling languages on meaning instead of their structure. 

The company will challenge its technology by applying it to dolphin communication. It is suspected that dolphins communicate using complex patterns of clicking and whistles, and it's even been posited that dolphins have what appear to be conversations. However, the meaning behind this communication is—so far—not known, and research so far has been based on structure and repeated patterns in communication.

 

 

In this context, the language will be completely unknown. The success of Word Space Technology in deciphering dolphin language would not only be a huge advancement in the study of marine biology but be a breakthrough in AI and machine learning, as well.

 

Intelligent Text Analytics

What makes Gavagai’s Word Space Technology unique is that, instead of analyzing text data to look for patterns, it analyzes and predicts the meaning of a word based on its context and use. This is more similar to how humans are able to discern the meaning of a word they have never seen before based on how it is being used.

An example would be how the software handles the recently trending hashtag “#covfefe”— the abrupt ending word to what is an otherwise cryptic Tweet by US President Donald Trump, leaving many on Twitter unsure of the intended meaning.

In the Gavagai Living Lexicon, you can see words commonly used before or after “covfefe”, semantically similar words (words that are used in similar context), n-grams (words in which it appears in), and associated words. You can likely discern that the word has something to do with Twitter based on its right-side neighbor being “Tweet”, and some other semantically similar words being other Twitter jargon. If “covfefe” continues to be used, the parameters mentioned before would continue to be updated to eventually discern a meaning.

 

The Gavagai Living Lexicon [Click image to enlarge]. Image courtesy of Gavagai

 

Linguistics and Computer Science Research

Gavagi is a powerhouse when it comes to research in the fields of linguistics and computer science. Since its founding, over 400 academic papers have been published by the company. Gavagi also has a long, impressive list of partners and supporters in the private and academic realms.

In 2016, Gavagi presented “Dead Man Tweeting” at the RE-WOCHAT 2016 workshop on “Collecting and Generating Resources for Chatbots and Conversational Agents Development and Evaluation”. The concept involves analyzing the text of someone who is deceased and then creating a chatbot who replicates that person’s language style to create a “living avatar”. The avatar is modeled to generate unique sentences and Tweets while remaining coherent.

So far, the results tend to vary depending on the length of the sentence (longer sentences run a greater risk of being incoherent), as well as lacking any sort of correction method when something is Tweeted that makes no sense. 

 

Experimental Tweets from a Dostoyevsky and Churchill avatars. Figures courtesy of Gavagi

 

In another publication from 2015 titled “Inferring the Location of Authors From Words Used in Their Texts”, the researchers attempt to identify where the author of text is from based on their language use in blogs. Words are given a placeness and locationality. All this information together is modeled to identify where the author might be from.

 


 

As AI and machine learning develop, they continue to find more and more uses in disparate fields. Gavagai shows that they will be instrumental in furthering studies of linguistics and communication—even of the dolphin variety.

 

Comments

0 Comments