The Power of Narrative
Welcome to the first OntoInsights blog post which explains the company’s focus
on narratives.
People have listened to and studied stories for insights into
cultures, customs, values and life. They have used narratives to explain how and
why the world works, and their experiences in it. Stories are humans’ approach
to structuring knowledge and providing insight.
Listeners can easily extract
knowledge from narratives, but for a computer to do this, a combination of
approaches and technologies is needed:
- Natural language processing (NLP) to parse the text
- Semantic (ontological) and linguistic understanding to distill meaning
- Graph technologies to encode the events of the narratives and background knowledge
- Pattern recognition algorithms to discover similarities and differences
- Inference, reasoning and causal analysis to extract knowledge and provide explanations
The following components (in a
Python framework) are assembled to create the needed "combination of approaches
and technologies" discussed above:
- spaCy for NLP
- Event-based ontologies (posted to our GitHub repository) coupled with VerbNet and WordNet for natural language understanding (NLU)
- Stardog graph database and its Python libraries for storage and reasoning against the ontologies
- CIA World Factbook, Wikipedia, GDELT – a global news database and many other sources for accurate, background knowledge
- ATOMIC and ConceptNet for common-sense knowledge
- NetworkX, igraph and other open source libraries such as MatchPy for pattern/graph analysis
- DoWhy, pympy and other open source libraries for causal analysis and inference
The "magic sauce" is not the core algorithms, but in how
they are assembled, and with what semantic and programmatic insights. This is
not about solving the difficult problem of narrative understanding
automatically, but about assisting people and groups to define and refine their
theories about what is happening in the world, based on analyzing narratives. In
addition, since all the narratives and details are retained, none of the
personality or individuality of the stories are lost. The human element is
captured in all its idiosyncratic forms, and is not reduced to numbers or
aggregations for ease of analysis. In this way, peoples’ experiences can be
better understood and the perspectives of “non-authoritative voices” can be
heard – giving a voice to people who have been left out of the mainstream
discussions on issues of poverty, addiction, economic mobility and much more.
OntoInsights’ goals are to provide an infrastructure to automate work such as
conducted at the University of Arizona, by Christina Greene (https://www.sciencedirect.com/science/article/abs/pii/S0016718521000518). In her paper, she states “Droughts are both a biophysical event and a social
event. However, drought planning and response is dominated by physical science
framings of drought while the social dynamics of drought are poorly monitored or
ignored … This can lead to a mismatch between how drought policies identify and
respond to drought versus how droughts are being experienced by people …
Understanding how droughts are understood and experienced by people and
communities is critical to making the human impacts of drought visible.”
Our
hypothesis is that narratives and life stories gathered directly from
individuals and families in environments and situations of interest represent an
unconstrained and rich source of information. This approach improves on standard
surveys because it better captures insight into the complex interdependent
constraints of people’s lives, which can provide a greater understanding into
causes and effects, and into the rationale for limited or successful impact of
current and past programs intended to help. Further, since the approach (by
necessity) engages the community, its findings create a mechanism for meaningful
discussions so that all stakeholders can work to co-develop solutions.
Comments
Post a Comment