Understanding how humans perceive and make sense of inequality

Despite tremendous progress, Artificial Intelligence and Natural Language Processing systems still grapple with two major problems: meaning and understanding. Both are however crucial if we want to build intelligent machines that can help with important decision-making processes and if we want to innovate the ways in which we interact and share knowledge with each other.

While most current approaches take a data-centered view, in which artificial systems try to find useful patterns from huge amounts of data, this installation demonstrates how our more human-centered language technologies work.

Frame Extractor

This installation is part of an ongoing effort of the Sony CSL Paris Language Team at understanding which narratives are constructed and shared by people about inequality. In order to understand how humans perceive and make sense of inequality, we can try to identify how they “frame” particular events by using their language as a window to the mind.

For example, if someone writes a blog post in which they say that we should “fight poverty”, they frame poverty as an enemy that must be combatted. An alternative framing can be found in the sentence “Ben Carson said that poverty is a state of mind”, in which poverty is framed as an individual problem.

The way a society perceives a particular issue has important effects on policy making: if poverty is seen as an enemy of the people, governments may be willing to take far-reaching action. If however poverty is seen as a person’s individual responsibility, a government may be more reluctant to introduce new policies.

Reconstructing semantic frames from text is therefore not only important for understanding how people make sense of complex issues, but also for informing policy makers about the concerns of a society and about the spread of (mis)information.

This installation demonstrates one of the most important frames for constructing such narratives: causal frames. More specifically, you will be able to test a piece of text in Italian, and the Causal Frame Extractor will try to detect causes and effects in the text.