Fact and entity extraction from unstructured, natural language texts is already a very challenging problem. But even if a system can identify persons, organisations or other entities, it is very likely that an essential piece of information is missing:
- What is the role of a particular individual?
Example: seller or buyer of a product?
- What are the properties of a particular entity?
Example: features of a car that it sold?
- How are the individuals, companies or other facts related to each other?
For an individual who learned to speak and to read a particular language this kind of tasks are rather simple. But for computers that have no language capabilities and no understanding of the real world concepts, this problem is extremely complex to solve. IT introduced the term “unstructured documents” when referring only to textual information coded in natural language (“plain text”).
Though, languages are not unstructured, only our approaches in current IT, (e.g. statistics and rules).
The talk covers how facts and their relationships that are relevant to an organisation can be detected when combining deep linguistic analysis, a universal semantic hierarchy, artificial intelligence and domain-specific knowledge; for example in contract management scenarios.
Technology Evangelist
ABBYY Europe
https://www.abbyy.com/
Michael Fuchs serves as Technology Evangelist at ABBYY. As an engineer, he was starting his career over 20 years ago in digital publishing and on information management using relational databases.
