M.S. AAI Capstone Chronicles 2024

A.S.LINGUIST

17

example, “Chris Martin” is not a “British actor” but a “British singer and songwriter” (see

question 17), and answering that next Easter will be celebrated “in the spring” is correct but quite

vague (see question 18). Also, punctuation is often absent and letter cases can be wrong

depending on the question. For instance, city names can be provided in either uppercase, like

“Rome” (question 8), or lowercase, like “san francisco” (question 7), that is grammatically

incorrect. Another issue with our natural language processing model is its difficulty in

distinguishing between the questions "how" and "what." For instance, when asked, "How are you

doing?" the model responds with "I am doing a project." Although this response is grammatically

correct, it is contextually inappropriate. However, the model accurately responds to questions

like "What is the color of the sky?" or "What is the capital of Italy?" within the correct context.

Based on the above, we can affirm that our chatbot model is often able to provide

pertinent textual answers to the received textual questions and, thus, represent an interesting first

solution to the problem we are trying to solve. We cannot say that the imprecisions resulting

from our initial evaluation of the chatbot performance are totally unexpected, as chatbots often

exhibit similar results. Nonetheless, it is our thought that the proposed chatbot model should be

reevaluated on a bigger set of question s and improved, as specified in the “Further

Improvements”section.

Final Application

After training and completing our two models, we developed an application to facilitate

user interaction. Using Streamlit, an open-source Python framework for building interactive data

apps, we equipped our application with several key features. First, it can capture an image from a

live video feed by utilizing a webcam as a data source. Users can click a button to capture a

197