CADIA-LVL present a lunchtime talk by Helga Svala Sigurðurðardóttir based on her ongoing Ph.D. work titled Learning Classifier Systems for Text Normalization using Regular Expressions as Input.
Time: Monday 15/11/21, 12:20-13:00
The talk focuses on automatic Text Normalization (TN), an important task for achieving speech synthesis with proper pronunciation involving expanding so-called non-standard words. Contemporary automatic TN systems employ Deep Neural Networks (DNNs), which improve upon the traditional method of using hand-coded rules but are prone to misleading errors, are data-hungry, and can be difficult to interpret. The current trend for TN systems is a hybrid of DNNs and hand-crafted rules, resulting in a highly complex procedure.
Helga’s Ph.D. work focuses on a solution to this challenge with a system that improves itself based on experience over time, which neither the rule-based systems nor DNNs can accomplish. Helga will discuss how a system for TN called Regína, with rules from regular expressions, could learn autonomously using an eXtended Classifier System, which returns transparent rules and learns learn cumulatively over time, constantly improving with more usage.
Please join us via Zoom at the following link: https://eu01web.zoom.us/j/65818539418