Master Thesis Defense: Alexandra Panagiotopoulos

Speaker: Alexandra Panagiotopoulos

Supervisor: Dr. S. Bergler

Examining Committee: Drs. L. Kosseim, O. Ormandjieva,R. Jayakumar (Chair)

Title: Feature Potential vs Performance Contribution, A Study
           of Verb Features Including Tense and Aspec

Date: Friday, April 10, 2015

Time: 14:00

Place: EV 11.119

ABSTRACT

Recent interest in second language acquisition (SLA) has resulted in studying the relationship between linguistic indices and writing proficiency in English. In this thesis we obtain an understanding of how basic linguistic notions introduced at an early stage of English, behave in automatic writing evaluation task. We discuss their predictive potential word level n-grams (unigrams, bigrams, trigrams) and verb features (tense, aspect, voice, type and degree of embedding) in writing proficiency. We conducted four experiments using standard language corpora that differ in authors' cultural backgrounds and essay topic variety. Our experiments have shown that features when examined individually can form a promising baseline in writing proficiency prediction. But better performance comes from looking into feature combinations and not at those features individually. Finally, we addressed how the corpus homogeneity can affect the performance of our features. The essays we examined in their majority contain present tense, indefinite aspect and passive voice. This uniform distribution affected the performance of our verb features especially tense, aspect and voice.