Abnormal Detection for Language Assessment

A typical language test usually consists of four parts: speaking, listening, reading, and writing. Both audio and text data from the test takers will be collected and used by the automated scoring system. During the test, some test takers will intentionally/unintentionally provide abnormal answers, which may contain memorized content, repeated sentences, and meaningless or off-topic words, and may affect the accuracy of the scoring system. Pearson would like to develop a more robust system for automated scoring to identify and penalize abnormalities in the audio and text data, which could greatly improve Pearson’s competitiveness in the market and in the meantime contribute to the development of abnormal detection models.

Faculty Supervisor:

Jesse Gronsbell

Student:

Partner:

Pearson Inc

Discipline:

Computer science

Sector:

Information and cultural industries; Professional, scientific and technical services

University:

University of Toronto

Program: