Finding ways to learn data science without relying on sight
Understand the barriers to learning data science faced by screen reader users, including the visually impaired, and look for ways to improve.
Awareness of the issues
With the remarkable development of AI technology, there has been an increasing amount of discussion about the diversity surrounding the technology. However, while there has been a lot of talk about the diversity of data and learning results, I feel that there has not been much discussion about the diversity of the environment in which people learn the technology. I hope that there will be more discussions on accessibility of AI development environment in the future.
Actuality, in the field of data science, which is the basis of AI technology, there are many things that need to be interpreted visually, such as analytical methods and teaching materials. Therefore, data science is a very difficult field for people who are visually impaired and who use screen readers to operate computers without relying on sight.
In this article, I would like to give a brief overview of the accessibility status of the development environment and learning materials surrounding data science, and introduce my improvement projects, especially those related to analysis methods using data graph visualization and analysis environment, which I think are issues.
Accessibility of various development tools
Visual Studio Code, a popular programming editor, has a screen reader optimized mode as a standard feature, and the environment surrounding programming has improved a lot compared to the past, which is a very good trend.On the other hand, in terms of notebook-like analysis environment, which is essential for data science, there are many parts where the accessibility is insufficient compared to general websites, and I think it is still difficult to use. However, Google Colab can be operated to a certain level with some efforts, and there are some improvement signs for the more popular Jupyter, so I think we can expect more improvement in the future.
Visual Studio Code Accessibility
JupyterLab Accessibility Meetings Minutes
Learning materials
Nowadays, many data science books, such as papers and technical articles, are available online and can be read with a screen reader. Also, mathematical formulas, which are essential for data science, can be understood with a screen reader if they are written in MathML. However, equally with general websites, there are still cases where accessibility is insufficient. Particularly in data science, there are many explanations that rely on illustrations, so there tend to be many points that need to be considered for accessibility. I hope that this will be improved as general accessibility technology improves.
Data visualization and other analysis methods
One of the biggest challenges in the accessibility of data science, and one that is unlikely to be solved with improvements in other fields, is the problem of analytical methods that rely on sight, such as graphical visualization of data. In addition, in recent years, as it has become common practice to target large amounts of data, it will be quite difficult to achieve results by simply checking individual raw data sequentially as an alternative. The importance of methods such as data visualization and quick, multifaceted analysis is increasing in business, science, and all other fields. In this context, it is essential to take various ways to analyze data that do not rely on sight in order to prevent the visually impaired from being left behind in the data society.
The first step to Improvement
At first, I am developing tools to improve accessibility at a minimum, with the goal of being able to learn a typical tutorial on data science with a screen reader. By doing so, I believe that as the number of visually impaired people who can learn data science increases little by little, the issues will gradually come to light and the improvement will accelerate.
To be more detailed, I started with Google Colaboratory, which is often used to publish tutorials on data science, and developed a library for sounding out visual changes resulting from operations.
Once I was able to operate the basic operations, I started to work on the main task of graph data sonification.
In the past, when I tried to convert graph data into sound, I received feedback that they wanted to be able to check the data interactively, just like a data scientist does, rather than simply playing back the audio data, so I developed a tool that allows users to check the graph data by tracing it with a touch pad or mouse. This made it possible to compare data in detail and expand the scope of data analysis.
And now that I developed prototypes of the basic data science environment and analysis methods for visually impaired, I can also prepare tutorials on basic machine learning and advanced deep learning.
This is still a prototype version, but I’ d like to continue to improve it by getting various feedbacks. I would also be happy to get collaborators if possible. The following is the link to the pages where various libraries and tutorials are available.
Towards Accessibility for AI
If you know any screen reader users who are interested in learning data science, I hope you will tell them about this project. In addition, this project is still in its early stages of open source release, so I would be very happy to receive any feedback, not just problems and questions, but also ideas and advice.
Perhaps it will be useful to you when you become visually impaired in the future, or perhaps it will trigger a talent that is buried within the visually impaired. I hope to continue my efforts to expand the limits of all these possibilities. It would be very encouraging if you could support me. Thanks!
Note: This is not to say that I think that the current accessibility is sufficient, and I would appreciate it if there were no misunderstandings. I believe that accessibility should ultimately aim for a level of quality that makes it not matter whether a person has a disability or not, but what I am aiming for in this project is a stage where I am trying to explore a mountain path where even a narrow path does not exist in the first place.