As part of their senior project hasan.ozturk
boun.edu.tr (Hasan Öztürk) and alperen.degirmenciboun.edu.tr (Alperen Değirmenci) developed a web application to compare the predictions of a neural network model with that of humans in solving the commonly encountered Turkish spelling error -- namely whether ‘de/da’ is to be spelled as a suffix (locative case) or separately (as a conjunction). This work entails two main tasks: to provide an explanation of a previously developed neural network model for correcting such errors and to develop a games-with-a-purpose style web application for eliciting human input in detecting such errors (for validation purposes).
The outcome of this work (the web application and the accompanying spelling corrector) was deployed in June 2020. It has received significant attention due to the pesky nature of the `de/da’ spelling error that is a persistent topic of general interest. Within 10 days nearly 450 people participated in the game providing input on 15 thousand sentences. The spelling corrector gained even more attention with users having tested over 92 thousand sentences.
The data collection is ongoing. Our preliminary analysis reveals a consistency among the predictions of the neural model and humans.
Spelling game with "de/da" spelling corrector: http://dedatakintisi.derlem.com/
Project advisors: onurguboun.edu.tr (Onur Güngör), PhD candidate and Dr. suzan.uskudarliboun.edu.tr (Suzan Uskudarli).
