In the soon days of the Covid-19 pandemic the Centers for Disease Control and Prevention exhibitd a single chart to elucidate how measures like mask wearing and collective distancing could ’flatten the curve’ and lessen the peak of infections.
The chart was amplified by news sites and shared on collective media platforms but it frequently lacked a corresponding text description to make it affable for sightless individuals who use a screen reader to navigate the web shutting out many of the 253 million nation worldwide who have visual disabilities.
This choice text is frequently missing from online charts and even when it is included it is frequently unenlightenative or even incorrect according to qualitative data gathered by scientists at MIT.
These investigationers conducted a study with sightless and sighted readers to determine which text is advantageous to include in a chart description which text is not and why. Ultimately they establish that captions for sightless readers should centre on the overall trends and statistics in the chart not its design elements or higher-level insights.
They also formd a conceptual standard that can be used to evaluate a chart description whether the text was engenderd automatically by software or manually by a ethnical creator. Their work could help journalists academics and communicators form descriptions that are more powerful for sightless individuals and lead investigationers as they educe better tools to automatically engender captions.
’Ninety-nine-point-nine percent of images on Twitter lack any kind of description — and that is not hyperbole that is the educeed statistic’ says Alan Lundgard a graduate student in the Computer Science and Artificial Intelligence Laboratory (CSAIL) and lead creator of the paper. ’Having nation manually creator those descriptions seems to be hard for a difference of reasons. Perhaps semiautonomous tools could help with that. But it is searching to do this previous participatory design work to aspect out what is the target for these tools so we are not generating full that is whichever not advantageous to its intended hearers or in the worst case erroneous.’
Lundgard wrote the paper with senior creator Arvind Satyanarayan an helper professor of computer science who leads the Visualization Group in CSAIL. The investigation will be presented at the Institute of Electrical and Electronics Engineers Visualization Conference in October.
To educe the conceptual standard the investigationers planned to initiate by studying graphs featured by common online publications such as FiveThirtyEight and NYTimes.com but they ran into a problem — those charts mainly lacked any textual descriptions. So instead they calm descriptions for these charts from graduate students in an MIT data visualization class and through an online scan then clustered the captions into four categories.
Level 1 descriptions centre on the elements of the chart such as its title myth and colors. Level 2 descriptions draw statistical full like the minimum ultimatum or correlations. Level 3 descriptions cover perceptual translateations of the data like intricate trends or clusters. Level 4 descriptions include mental translateations that go over the data and draw on the creators apprehension.
In a study with sightless and sighted readers the investigationers presented visualizations with descriptions at different levels and asked participants to rate how advantageous they were. While both clusters agreed that level 1 full on its own was not very helpful sighted readers gave level 4 full the highest marks while sightless readers ranked that full among the smallest advantageous.
Survey results revealed that a superiority of sightless readers were earnest that descriptions should not hold an creators editorialization but rather stick to direct facts almost the data. On the other hand most sighted readers preferred a description that told a story almost the data.
’For me a surprising finding almost the lack of advantageousness for the highest-level full is that it ties very closely to feelings almost agency and control as a disabled individual. In our investigation sightless readers specifically didnt want the descriptions to tell them what to ponder almost the data. They want the data to be affable in a way that allows them to translate it for themselves and they want to have the agency to do that translateation’ Lundgard says.
This work could have implications as data scientists last to educe and refine machine learning methods for autogenerating captions and choice text.
’We are not able to do it yet but it is not inconceivable to conceive that in the forthcoming we would be able to automate the creation of some of this higher-level full and build standards that target level 2 or level 3 in our framework. And now we know what the investigation questions are. If we want to exhibit these automated captions what should those captions say? We are able to be a bit more directed in our forthcoming investigation owing we have these four levels’ Satyanarayan says.
In the forthcoming the four-level framework could also help investigationers educe machine learning standards that can automatically hint powerful visualizations as part of the data analysis process or standards that can draw the most advantageous enlightenation from a chart.
This investigation could also enlighten forthcoming work in Satyanarayans cluster that seeks to make interactive visualizations more affable for sightless readers who use a screen reader to approach and translate the enlightenation.
’The question of how to fix that charts and graphs are affable to screen reader users is both a collectively significant equity effect and a challenge that can advance the state-of-the-art in AI’ says Meredith Ringel Morris ruler and highest scientist of the People + AI Research team at Google Research who was not implicated with this study. ’By introducing a framework for conceptualizing intrinsic speech descriptions of enlightenation graphics that is rooted in end-user needs this work helps fix that forthcoming AI investigationers will centre their efforts on problems aligned with end-users values.’
Morris adds: ’Rich intrinsic-speech descriptions of data graphics will not only swell approach to nice enlightenation for nation who are sightless but will also boon a much wider hearers as eyes-free interactions via keen speakers chatbots and other AI-powered agents befit increasingly commonplace.’
This investigation was supported by the National Science Foundation.