Exploring AI-Based Support in Speech-Language Pathology for Culturally and Linguistically Diverse Children

Research Background and Issues

• Issues and Challenges:
  The author identifies a major issue in the field of speech-language pathology: speech-language pathologists (SLPs) face numerous challenges when providing speech and language interventions for culturally and linguistically diverse (CLD) children. These challenges include, but are not limited to, a lack of representative materials, unreliable machine translation, insufficient support for language variations, and the homogeneity of the SLP professional group (e.g., the vast majority of SLPs are white and monolingual).

• Significance:
  These issues directly impact CLD children's access to equitable and culturally sensitive treatment opportunities and may exacerbate systemic discrimination based on race, language, and ability. In the United States, CLD children constitute a significant proportion of the population, yet they often receive fewer treatment resources compared to white monolingual children.

• Research Motivation and Related Work:
  Advances in AI, particularly generative AI (e.g., ChatGPT), offer potential solutions to these challenges. Previous studies have shown that AI can support SLPs by automating tasks, providing real-time feedback, and creating personalized treatment plans. However, little research has focused on how AI can provide culturally, linguistically, and ability-sensitive responsive support.

Proposed Solution

• Methods and Approach:
  The author explores the potential of AI in the field of speech-language pathology through two studies:

  1. Conducting semi-structured interviews with 15 SLPs to understand current challenges and areas where AI support is desired.
  2. Administering a two-part questionnaire involving 13 SLPs to evaluate the performance of current generative AI tools (e.g., ChatGPT-4o) in generating culturally and linguistically sensitive materials and to identify biases within these outputs.

• Innovative Contributions:
  The primary innovation of this study lies in its focus on leveraging AI to address the historical inequities and insensitivities in practices involving CLD children. By gathering SLP perspectives, the study explores how AI technologies can assist in generating customized assessment, treatment, and family collaboration materials. Additionally, it highlights the inherent biases in AI-generated content and their potential impact on equitable treatment.

• Implementation Steps and Techniques:
  The research was conducted in two phases:

  • Interview Phase: Identifying SLPs' current work practices and challenges.
  • Survey Phase:
    • SLPs designed AI prompts relevant to their practice.
    • AI was used to generate text and image materials based on these prompts.
    • Participants evaluated the accuracy, relevance, safety, and transparency of the generated materials.

Research Findings

• Key Findings:

  1. Challenges Identified: SLPs face issues such as unrepresentative materials, unreliable translation tools, and insufficient support for language variations.
  2. SLP Needs for AI Support: These include personalized material creation, efficient translation tools, and language variation recognition technologies.
  3. Evaluation Results: While generative AI provides a "good starting point" in certain aspects of effectiveness, its outputs lack significant cultural, linguistic, and ability adjustments and are deficient in transparency and explainability.
  4. AI Bias Analysis: Generated outputs exhibit clear biases, including cultural (defaulting to white-centric representations), linguistic (prioritizing English), and ability biases (assuming no disabilities or specific assistive tools).

• Advantages:
  Compared to traditional tools, AI has the potential to quickly generate personalized materials, alleviating some of the time pressures faced by SLPs.

• Limitations and Future Directions:

  1. Limitations:
     • Current AI models lack a deep understanding of SLP-specific practice standards and the cultural, linguistic, and ability contexts of their clients.
     • Biases in training data perpetuate systemic inequities in AI-generated materials.
  2. Future Directions:
     • Develop interactive AI prompt tools tailored specifically for SLPs to improve the quality of generated outputs.
     • Incorporate input from children and families to enhance the adaptability and sensitivity of AI-generated materials.
     • Improve model explainability to increase trustworthiness.

Through this study, the author calls for the development of more inclusive and equitable AI technologies and emphasizes the need to provide SLPs with training to enhance their ability to interact effectively with AI.