Implementation av adaptiv evaluering för feedback i chatbot för yrkessvenska

Sammanfattning: The report aims to study whether the implementation of language models in a chatbot for Swedish can make language learning more effective. A fundamental factor for an individual to have the opportunity to be integrated into the labor market and into society, is language. By developing tools for language learning, individuals and society can benefit. The work included the implementation of two different language models and machine learning frameworks to improve a chatbot's ability to adaptively evaluate user responses in a conversation between chatbot and human. Word-embedding was used here with the help of the BERT framework, as well as sentence similarity. The results indicate that the use of language models for adaptive evaluation has good potential and can be further developed to be applied to applications intended for language learning. However, further development is required to determine whether these language models are applicable for the development of adaptive feedback generation in a chatbot. The work has been based on the chatbot that Lingio, VINNOVA and KTH have developed.