AMR-eseResearch

• Conducted this research under the supervision of Dr. Justin DeBenedetto from the Computer Science Department of Villanova University.
• Developed and implemented 4 quantitative metrics in Python to evaluate the presence of translationese, the decrease in the natural-sounding of the text, in an AMR-to-text generation system’s output. Used AMR 2.0 as text input for human written vs computer generated text.
• The methods are developed based on these research articles: https://arxiv.org/pdf/2304.11501.pdf https://arxiv.org/pdf/2304.11501v1.pdf
• Based on the metrics designed, measured that the provided computer-generated text contains 12.94% more characteristics typical of translationese compared to human-written text.