One of my friend taught me that if I want to speak more like a German, I should try saying with the German modal particle 'halt' because it carries the subtle nuance of annoyment but never being taught in language schools. He told me that if I say 'Der Bus kommt nicht.(The bus is not coming...)' might be good, however, if I say 'Der Bus kommt halt nicht(The bus is not coming 😒)', the Germans would be amazed.
However, it was difficult for me to understand the subtle nuance of 'halt' as a native speaker of Korean, so I thought it would be helpful if I can do a research based on this.
My idea was simple: German 'halt' carries emotions more than an actual meanings. So what if I use emojis for the loss in translation?
From this perspective I explored if the modal particles can be replaced by scoring their sentiments and matching emojis with similar sentiments scores. I conducted 2 research; based on surveys in German and Korean(my mother language), and an experiment to determine which model best approximates the sentiment estimates of modal particles from the survey among BERT, HuggingFace and FastText models. What was interesting was that, contrary to my expectations, the FastText model's sentiment scores were similar to those obtained from actual surveys.
However, my experiment is based on only a very small number of sample data(~100 participants), so the results may be limited. Please keep in mind the basic research nature of this term paper :)
Sentiments in Modal Particles: https://github.com/bakeyeon/modalparticles_sentiment_analysis/blob/main/Modalparticles_Sentiment_Analysis_HyeyeonPark.pdf
Sentiments - Emojis Research: https://github.com/bakeyeon/Modalparticles_Emojis/blob/main/Modalpartikeln_Can_Sentiments_Survive_Translation_with_Emojis.pdf
Thank you for reading this long post. Feedback and comments are always welcome! 😃