Software Teaches Computers to Translate Words to Math
University of Illinois News Bureau (01/20/15) Liz Ahlberg
New software developed at the University of Illinois helps computers understand the mathematical reasoning expressed in language, potentially improving search engines and access to data in addition to strengthening math education. The key hurdle the researchers encountered was teaching the computer to identify quantities and units in text regardless of how they are expressed, which humans do unconsciously. The software also had to determine what to do with the identified numbers and find any accompanying equations within the text. For example, the computer needed to ascertain if an amount is exact or approximate, static or dynamic, a range, presented in relation to something else, and detect other contextual cues a reader intuitively understands. The researchers tested the software's abilities to identify and normalize quantities in text, to perform searches regarding monetary currencies, and to understand and solve elementary school-level math word problems. They found the software performed well and even outperformed the average elementary-level student on standardized word problems, says Illinois professor Dan Roth. "As we move forward and want to help kids understand math, it makes sense to use technology," Roth says. "This shows that computers could help people learn in ways that could not be done before."
University of Illinois News Bureau (01/20/15) Liz Ahlberg
New software developed at the University of Illinois helps computers understand the mathematical reasoning expressed in language, potentially improving search engines and access to data in addition to strengthening math education. The key hurdle the researchers encountered was teaching the computer to identify quantities and units in text regardless of how they are expressed, which humans do unconsciously. The software also had to determine what to do with the identified numbers and find any accompanying equations within the text. For example, the computer needed to ascertain if an amount is exact or approximate, static or dynamic, a range, presented in relation to something else, and detect other contextual cues a reader intuitively understands. The researchers tested the software's abilities to identify and normalize quantities in text, to perform searches regarding monetary currencies, and to understand and solve elementary school-level math word problems. They found the software performed well and even outperformed the average elementary-level student on standardized word problems, says Illinois professor Dan Roth. "As we move forward and want to help kids understand math, it makes sense to use technology," Roth says. "This shows that computers could help people learn in ways that could not be done before."
No comments:
Post a Comment