\documentclass[11pt]{article} \usepackage{amsmath, amssymb} \usepackage[margin=1in]{geometry}
\title{Aphasia Bridge \ \large A Human-Centered AI Communication System for Anomic Aphasia} \date{}
\begin{document}
\maketitle
\section*{Inspiration}
Aphasia is not a loss of intelligence; it is a loss of access.
People with anomic aphasia know what they want to say but cannot retrieve the words. They rely on fragmented phrases and descriptions, yet the final word remains inaccessible. This creates a gap between cognition and communication.
Clinical therapy uses \textbf{Semantic Feature Analysis (SFA)}, where patients describe attributes of a word (category, function, appearance) to recover it. However, this process is not available in real-time communication.
We asked: what if this structured clinical method could be embedded into an interactive system?
\begin{equation} \text{Fragmented Thought} \rightarrow \text{Structured Meaning} \rightarrow \text{Clear Communication} \end{equation}
\section*{What We Built}
\subsection*{Sentence Reconstruction}
\begin{equation}
\text{want water cold''} \rightarrow \text{I want a cold glass of water''}
\end{equation}
The system reconstructs incomplete input into grammatically complete sentences.
\subsection*{Guided Word-Finding (SFA)}
Users refine meaning through structured prompts: \begin{itemize} \item Category \item Function \item Attributes \end{itemize}
\section*{How We Built It}
\subsection*{System Architecture}
\begin{equation} \text{Input} \rightarrow \text{Speech/Text} \rightarrow \text{AI Reconstruction} \rightarrow \text{Guided Recovery} \rightarrow \text{Output} \end{equation}
\subsection*{Deterministic SFA Engine}
We model word retrieval as a feature-matching problem:
\begin{equation} \text{Score}(w) = \sum_{i=1}^{n} \mathbf{1}(f_i \in w) \end{equation}
where: \begin{itemize} \item $w$ is a candidate word \item $f_i$ are user-selected features \item $\mathbf{1}(\cdot)$ is the indicator function \end{itemize}
\subsection*{Hybrid System}
\begin{equation} \text{System} = \text{AI} + \text{Deterministic Logic} \end{equation}
\section*{Challenges}
\subsection*{Fragmented Input}
User input is often incomplete, ambiguous, and ungrammatical. The system must infer meaning without introducing incorrect assumptions.
\subsection*{Balancing AI and Reliability}
Pure AI approaches were inconsistent. We introduced constraints:
\begin{equation} \text{Reliability} > \text{Model Complexity} \end{equation}
\subsection*{Accessibility Constraints}
We designed for users with cognitive and language impairments, requiring: \begin{itemize} \item Minimal interface complexity \item Large interaction targets \item Clear feedback \end{itemize}
\section*{What We Learned}
\subsection*{Simplicity Outperforms Complexity}
\begin{equation} \text{Performance} \propto \frac{1}{\text{User Friction}} \end{equation}
\subsection*{Structured AI is More Effective}
\begin{equation} \text{Effective AI} = \text{Model} + \text{Constraints} \end{equation}
\section*{Accomplishments}
\begin{itemize} \item Built a real-time working prototype in 48 hours \item Combined AI reasoning with deterministic logic \item Digitized a clinically validated therapy method (SFA) \end{itemize}
\section*{What’s Next}
\subsection*{Personalization}
\begin{equation} \text{User Model} \rightarrow \text{Adaptive Predictions} \end{equation}
\subsection*{Progress Tracking}
\begin{equation} \text{Recovery Rate} = \frac{\text{Successful Retrievals}}{\text{Attempts}} \end{equation}
\section*{Accessibility and HCI Design}
\subsection*{Visual Processing}
\begin{equation} \text{Color} + \text{Size} + \text{Contrast} \rightarrow \text{Immediate Recognition} \end{equation}
\section*{Conclusion}
\begin{equation} \text{Communication Ability} \neq \text{Cognitive Ability} \end{equation}
Aphasia Bridge closes this gap by transforming structured meaning into clear communication.
\end{document}
Log in or sign up for Devpost to join the conversation.