Graduate Student Fellow Helps Computers Interpret Human Languages

BY: ALEX JOHNSON

Computers don’t speak the same language as humans, but Afra Feyza Akyürek uses machine learning to help computers process and interpret human languages better.  Akyürek, a PhD student in Computer Science at Boston University, aims to improve the dialogue between humans and computers and refine programs like Siri or Alexa.

Akyürek’s primary school teacher supported her early interest in math and prepared her for computational and quantitative research.  Emine Karaoğlu picked up that Akyürek had both an interest and a knack for numbers, and created a more advanced curriculum for her.  “I would wake up at five a.m everyday before school to solve the advanced problems that she would give me,” Akyürek states.  “It helped me improve my mental math, which paved the way to success.”   

Akyürek’s fascination with math led her to study engineering at Koç University. She first focused on industrial engineering to optimize computer algorithms, but soon discovered that she was more interested in teaching the very machines she worked to create. As a computer engineering major, Akyürek learned different computer languages and how to create programs for things like data security and processing. “Once you learn a new programming tool or technique, the number of ways to use it increases exponentially,” she says.  Akyürek decided to focus her graduate studies on natural language processing, or giving computers the ability to understand words.  

As a Graduate Student Fellow at the Hariri Institute, Akyürek intends to use language processing to teach computers how to understand human speech and texts.  This process consists of feeding a program lots of data so the computer can pick up on patterns and context to then be able to make its own decisions. “There is a big difference between being able to replicate a conversation, and having a conversation that a machine is actually able to understand,”  Akyürek says.  It is similar to talking with a parrot.  They may be able to mimic certain words, and even carry a conversation, but the bird may not understand anything it is saying. True comprehension can be hard to quantify, and Akyürek’s task is to give computers the right tools to be able to understand human language, not just mimic it.  Creating a computer program that understands language is one of the goals of machine learning.  “You can have conversations with advanced machine learning systems,” Akyürek says,  “They can learn how to identify hate speech, identify fake news, or translate between two languages.”  

Akyürek wants to inform the ways that computers learn and interpret human languages through her research. Her research could help improve other language processing programs, like translation softwares that “understand” multiple human languages and engage in interactions with humans.  “If we can find a way for this software to actually learn from the user, like when a user corrects the machine’s spelling, then they might feel more realistic to humans,” she says. 

To learn more about the Hariri Institute’s transformational research, click here to sign up for our newsletter.