Automated Scoring: from Essays to Speeches and finally to Videos

Lei Chen
Educational Testing Service (ETS)

Friday 4/11, 2:15pm, Graduate Center (365 Fifth Ave), Science Center (Rm. 4102).

More assessments have been used recently to track students' skill levels and to
support their learning processes. Accordingly, scoring such rapidly increasing
assessments in a timely and cost efficient way becomes an important challenge. A
solution is provided by Automated assessment (AA), a technique that uses natural
language processing (NLP), speech processing, and machine learning, to simulate
human raters' behaviors in order to automatically rate students' essays or
speech responses.

In this talk, I will first provide a brief overview of the research efforts
carried out in the AA area by the ETS NLP & Speech group. Then, focusing on
speech scoring, I will introduce my previous research on AA's two major areas:
(a) finding useful features and (b) building accurate machine learning models.
With respect to the former, I will talk about the use of the acoustic features
widely investigated in Phonetics, such as vowel space. With respect to the
latter, I will talk about using feature bagging in order to achieve more robust
and accurate scoring models. Finally, I will present my recent work of using
multimodal signal processing technology - such as body tracking, for example -to
extend the scoring capability to nonverbal communication.

Bio

Lei Chen received his Ph.D. in Electrical and Computer Engineering from Purdue
University. He worked as an intern in the Palo Alto Research Center (PARC)
during the summer of 2007, and he is currently a research scientist in the R&D
division at Educational Testing Service (ETS) in Princeton, NJ. Before joining
ETS, his research focused on using non-verbal communication cues, such as
gestures and eye gazes, to support language processing. He has been involved in
the NSF KDI project and ARDA VACE project to investigate multimodal signals,
including speech, gestures, and eye gazes, used in human-to-human conversations.
In the 2009 International Conference of Multimodal Interface (ICMI), he won the
Outstanding Paper Award sponsored by Google. At ETS, his research focuses on the
automated assessment of spoken language by using speech recognition, natural
language processing, and machine learning technologies. Since 2013, he has been
working on multimodal signal processing technology for assessing video-based
performance tests in areas such as public-speaking.