Educational Testing Service

Chong Min Lee

Lei Chen

Chee Wee Leong

Gary Feng

Jidong Tao

Christopher Kitchen

Martin S Chodorow

Swapna Somasundaran

Xinhao Wang

Xianyang Chen

Patrick D Houghton

Vinay K Basheerabad

Data is received that encapsulates a video of a subject performing a task. This video is used to generate a transcript using an automatic speech recognition (ASR) system. A plurality of text segments are generated from the transcript and then tokenized. A textual representation of each segment is extracted by a transformer model using the tokenized text segment (i.e., the tokens corresponding to the text segment). Thereafter, for each segment, a fused representation derived from the textual representations and corresponding visual and audio features from the video is generated. A sparse attention machine learning model then selects an optimal slice of the video based on the fused representations. The optimal slice can then be input into one or more machine learning models trained to characterize performance of the task by the subject.

NLP-guided video thin-slicing for automated scoring of non-cognitive, behavioral performance tasks

Systems and methods are provided for scoring video clips using visual feature extraction. A signal including a video clip of a subject is received. For each frame of the video clip, physiological features of the subject visually rendered in the video clip are extracted. A plurality of visual words associated with the extracted physiological features are determined. A document including the plurality of visual words is generated. A plurality of feature vectors associated with the document are determined. The plurality of feature vectors to a regression model for scoring are provided.

Automated scoring of video clips using extracted physiological features

Systems and methods are described for providing a multi-modal evaluation of a presentation. A system includes a motion capture device configured to detect motion an examinee giving a presentation and an audio recording device configured to capture audio of the examinee giving the presentation. One or more data processors are configured to extract a non-verbal feature of the presentation based on data collected by the motion capture device and an audio feature of the presentation based on data collected by the audio recording device. The one or more data processors are further configured to generate a presentation score based on the non-verbal feature and the audio feature.

Systems and methods for providing a multi-modal evaluation of a presentation

Provide automatic assessment of oral recitations during computer based language assessments using a trained neural network to automate the scoring and feedback processes without human transcription and scoring input by automatically generating a score of a language assessment. Providing an automatic speech recognition ('ASR') scoring system. Training multiple scoring reference vectors associated with multiple possible scores of an assessment, and receiving an acoustic language assessment response to an assessment item. Based on the acoustic language assessment automatically generating a transcription, and generating an individual word vector from the transcription. Generating an input vector by concatenating an individual word vector with a transcription feature vector, and supplying an input vector as input to a neural network. Generating an output vector based on weights of a neural network; and generating a score by comparing an output vector with scoring vectors.

Computer-implemented systems and methods for automatically generating an assessment of oral recitations of assessment items

Computer-based systems and methods are provided for generating a narrative computer scoring model for assessing story narratives. In one embodiment, supervised machine learning is used to generate the narrative computer scoring model. For example, a collection of training story narratives with assigned scores may be used to train the model. In one embodiment, each training story narrative is processed to extract features that signify content relevance, collocation of commonly used words, coherency, detailing, and expressions of sentiment. These features, as well as others, may be selectively used to train a narrative computer scoring model. Once trained, the model can be used to automatically evaluate story narratives and assign appropriate scores.

Growing community of inventors

Pennington, NJ, United States of America

Chong Min Lee