As English continues to grow as the most used language on an international level, many non-native speakers are sharpening their language skills in pursuit of higher education and business opportunities.
Technology plays a big role in providing training and testing for these individuals, and Keelan Evanini, pictured at right, expects its role to grow.
Evanini, a managing senior research scientist at Educational Testing Service (ETS) will address this topic in a presentation titled “Using Automated Speech Recognition and Natural Language Processing to Assess Non-Native English” on Thursday, March 17, at 8 p.m.. Refreshments and networking begin at 7:30 p.m. Sponsored by the Princeton ACM/IEEE-CS chapters, the free event takes place in Room CS 105 of the Computer Science Building at Princeton. Students and their parents are welcome. For more information, call 908-285-1066 or visit the Princeton ACM website: http://PrincetonACM.acm.org.
Attendees are invited to a pre-meeting dinner at Ruby Tuesday’s Restaurant on Route 1 at 6 p.m. To be included in the dinner meeting, E-mail firstname.lastname@example.org.
Evanini will introduce a program he and his team have been developing, SpeechRater, an application that can score spontaneous spoken responses in addition to restricted responses. His presentation will also cover several technologies including Automated Speech Recognition (ASR), Natural Language Processing (NLP), and Spoken Dialog Systems.
Although computers have been used for educational applications since the 1960s, speech and language processing technologies were not a part of the picture until the 1990s. By then, desktop and multimedia computers were available to a broad population. Research related to SpeechRater began in the early 2000s and was first used in 2006. To insure that the program’s scoring emulates human scoring as closely as possible, its structure is informed by content experts and a database of previously observed responses scored by human raters.
Used primarily by international students applying to universities, the program helps individuals prepare for ETS’s TOEFL (Test Of English as a Foreign Language) exam which tests reading, listening, speaking, and writing. The speaking section includes independent speaking tasks, — requiring you to draw entirely on your own ideas and experiences — and integrated speaking tasks, requiring you to integrate listening, reading and speaking. While SpeechRater is used for study and preparation purposes, the actual TOEFL test is rated by humans.
ETS says that TOEFL scores are accepted by more than 9,000 organizations in over 130 countries. In addition to colleges and universities, TOEFL is used by immigration departments to issue residential and work visas; medical and licensing agencies for professional certification purposes; and individuals who want to improve their English.
There are several benefits of test preparation using ASR and SpeechRater, Evanini says. It is a cost efficient method for learning and practicing skills and it speeds up the turnaround period for receiving scores. It provides an option for individuals who do not have access to live tutors, which is especially helpful for people living in other countries.
The current drawback, he says, is that it is hard for the computer to understand everything about language with all it nuances. However, SpeechRater is a big improvement over earlier computer programs. “We’ve done a lot of work in this area,” Evanini says. SpeechRater was developed to work with open-ended interactions like answering questions based on your experience and knowledge as opposed to simply repeating a sentence you had read, known as a restricted response.
As a kid, growing up in Union County, Evanini was interested in different languages, solving word games and logic puzzles, he says. He picked up an interest in computing from his parents, who were computer scientists.
He earned a bachelor’s degree in linguistics and classical languages from the University of California, Berkeley; and a doctorate in linguistics from the University of Pennsylvania. His advisor, Bill Labov, is credited as the founder of variationist socio-linguistics, and has written several books, including “The Atlas of North American English,” “Dialect Diversity in America,” and “Principles of Linguistic Change.”
Throughout Evanini’s academic studies, he learned several languages, including German, Spanish, Japanese, Latin, and Greek, and he has taught English in Japan and Mexico. Before joining ETS in 2009, he worked as a consultant for SpeechCycle and as a programmer for Linguistic Data Consortium.
For a deeper understanding of Evanini’s work on automated scoring and linguistics, visit his website (www.evanini.com/keelan.html), which includes a link to his dissertation, “The permeability of dialect boundaries: A case study of the region surrounding Erie, Pennsylvania.” His site also includes links to several of his publications and downloadable links to programs he has written for his research. He also offers solutions to speech technology research on a WordPress blog with posts through 2014: speechtechie.wordpress.com.
Today Evanini lives in Pennington with his wife and three children. In addition to his dedication to helping individuals for whom English is a second language, Evanini is a unicyclist and an advocate for the environment.
Looking ahead, Evanini envisions an interactive dialog system that would allow you to have a conversation with your computer. As a learning option, it would be more interesting than reading a book, he says. He acknowledges that it will be a challenging task but believes it can be developed within the next several years.