Work Style Analysis: Creating a Todos Santos Mam conversation book in 30 days
Backstory
Because I work with many children who speak the Todos Santos Cuchumatán dialect of Mam, my goal was to develop a current, accurate, and accessible Todos Santos Mam conversation guide. As I conceived the project, I wanted to create several practical usage guides for daily life activities, such as for school, work, community, healthcare, dealing with authorities and emergency situations, etc. I sought to:
- Document and preserve a threatened Indigenous language
- Bridge communication gaps between Mam speakers and English/Spanish speakers
- Support immigrant Mam communities, particularly in the U.S., who face systemic barriers in education, healthcare, legal systems, and daily life
- Empower teachers, social workers, and allies with tools to communicate compassionately and effectively
- Create a practical, structured learning resource built on verified data
- Create a workflow for saving endangered languages that could be shared, replicated, or modified
Creating and maintaining the calendar
Calendar considerations aren’t just about time constraints, they have to bend to changing tasks and evolving learnings. It was critical to identify the tasks needed to meet the deadline and achieve the goal. After my first full day of work, it became abundantly clear that there were several time-dependent domains where my project was at risk:
- Organizing both my work and the project materials
- Maintaining the organization
- Knowing / learning what to do
- Working messy (failing to clear file clutter)
- Backing up work
- Trusting AI and incorporating failures (fabrications, lies, guesses, stuff that sounds good, lost data)
- Documenting the work
Interviewing AIs for the Assistant Linguist position
After several false starts with ChatGPT 4, I decided to treat selecting an AI like hiring a person for a job. I decided to conduct job interviews and allow the different prominent AIs to make their cases for why they were the best candidates for becoming Mam linguists. I used the following prompt as the basis for my interviews. I used ChatGPT 4 (under activated Truth Mode, Respect Mode, and Honesty Mode) to help me refine my foundational interview prompt. I then got ChatGPT’s assistance in crafting interview prompts tailored to each of my AI candidates, based on their known strengths and weaknesses.
Gathering source materials and tools
One of my big goals for this project was to keep it low budget. Most of the language source materials were gathered online or from in-person interviews with Todos Santos Mam native speakers. I used an iPhone, which I already had in my possession, a laptop (also already in my possession), and a wide variety of free online tools and apps.
Identifing important prompts, prompt strategies, and prompt styles
Even though AIs have access to a vast repository of information, can search the internet in real time, can read documents that are uploaded to them, can analyze text, and apply their knowledge to make recommendations and even at times deductions in reasoning, great care must be taken to ensure that the output received isn’t corrupted by conflicting programming patterns which seem to be innate to all AIs on the market today. Of course, prompts that set the scene and define what the AIs should do are important, but just as important are prompts that stop the AIs from fabricating, placating, inventing, filling in gaps, guessing, and even misrepresenting factual information.
Joanna Hofer 