In our project, by typing questions in the terminal, users can let the agent answer their questions based on agent’s perception of the game world. Specifically, our project will focus on environment description. Questions will include but not limit to “how many pigs can you see in your view”, “what is the river” and “How many black sheep can you see?”. Our project is expected to use Natural Language Processing (NLP) models to parse semantics and action functions for agent to answer questions. In detail, we expect our agent to describe the range of scenery in front of him by answering the user input questions typed in terminal. We will also handle the situation that different semantics with one meaning in our project.
AllenNLP - constituency parsing Gensim
We will evaluate our success rate based on if our agent gives the answer with correct information, as well as the corresponding time. Since we also handle synonyms in users’question, we will manually check how accurate our model performs for synonyms.
We will expect the agent answering questions about environment with correct information.
2pm, Wednesday, Oct 28th, 2020