Automated Essay Scoring (AES) is a task of automatically grading the essays each of which is written on a specific prompt. This task has gained attention as an alternative to the human grading system, which has issues with fairness and expense. Our goal is to design a model that can score the essays of new prompts and can provide feedback by evaluating multiple trait scores.
Automatic Document Generation
Chatbot + Text generation = Interview-based document generation. The chatbot should deliver the contents of the interview to the generator as accurately as possible. In addition, the generator should naturally generate the document based on what it receives. The whole system provides a convenient function to create documents based on natural conversations with users.
Grammatical Error Correction
GEC is the task of finding and correcting grammatical errors in text. It is mainly used to give language learners a better educational experience. It is also used to correct the input of models used for other tasks, or to improve the quality of generated text. Modern approaches view GEC as a text-to-text task and employ an encoder-decoder structure.
Intent/Emotion classification in conversations can be utilized by various products such as Human-AI communication, conversation synthesis, and risk detection.  Our goal is (1) to classify the intent/emotion of each utterance of conversation in real-time and (2) to deliver helpful information such as detected risk based on classification results to the user or other programs while the conversation is ongoing.
Semantic parsing is a task of mapping natural language to logical forms, which can be evaluated on given contexts to produce corresponding denotations. For example, a question answering system can exploit a semantic parser to convert a user's question to a query (logical form), then the query derives an answer (denotation) from a knowledge base (context). Therefore, semantic parsers become natural language interfaces to computer systems that execute complex logical operations, which are hidden to end users.