Scientists are working to teach computers to recognize not only words but also the myriad meanings, subtleties and attitudes those words can convey. This type of “stance recognition” could help improve all kinds of speech recognition systems, says Valerie Freeman, a Ph.D. candidate in the department of linguistics at the University of Washington. She presented her team’s research on the Automatic Tagging and Recognition of Stance project at an Acoustical Society of America meeting in October.
“We’re trying to understand what is it about the way we talk that makes our attitude clear while speaking the words, but not necessarily when we type the same thing,” Freeman told McKnight’s.
The work could impact customer service and allow for speedier meetings.
The researchers recorded the voices of people of different ages and backgrounds, all from the Pacific Northwest. They found that when pairs of people worked together on a task, they tended to speak faster, louder and with “more exaggerated pitches” when they had a strong opinion. This was the case both when people were arguing, and when they were having regular conversations. For example, people talked quicker when they worked on engaging tasks, such as balancing an imaginary budget, Freeman said.
In addition, when people were engaged in a topic, they spoke less fluently, tending toward false starts, repeating themselves and saying “um” more often than when they were less interested.
The researchers plan to continue analyzing the conversations for subtler cues and patterns, such as different pronunciations for positive versus negative opinions, men versus women and older people versus younger people.
“A few potential improvements that come to mind include smoother customer service calls, more accurate and adaptable speech-to-text and even automatic summarizing of meetings,” Freeman said.
From the December 01, 2014 Issue of McKnight's Long-Term Care News
