Automatic Question Answering
The possibility of training computers to answer questions asked by people automatically is one of the most exciting challenges that the community of researchers in the field of deep learning is currently facing.
The truth is that being able to overcome this challenge can have a great impact on a large number of professional disciplines. The reason is that today almost all types of industries are totally saturated by the large amount of information they are exposed to on a daily basis. In addition, the fact that this information comes in unstructured formats complicates things even more.
However, the latest advances in natural language processing, retrieval information, and of course, deep learning have stimulated the creation of new systems that have a fairly high success rate. In addition, there are two very different kinds of systems depending of their fields of application. One is related to the work with general purpose information. And the other one is related to the work with specific purpose information.
As for working with general purpose information. Today it is possible to answer a large number of questions that have no professional background. For example, a human operator could ask which is the capital of Russia, which river runs through Prague or which is the gross domestic product of China.
As for the specific purpose. For example, a clear case of use is in the legal field, where professionals are exposed to huge amounts of local, regional, national and international legislation that makes their work very difficult. With the new systems for legal question answering [1] their work can be facilitated in a comfortable and simple way.
It is very clear that there are still many aspects that need to be improved. For example, the interpretability of the resulting models. To date, it is possible to obtain a high degree of accuracy in relation to the answers that are sought, however, it is more difficult for a human operator to understand why those answers are sought, since most of the models that are trained follow a paradigm based on black boxes [2].
References
[1] Martinez-Gil J., Freudenthaler B., Tjoa A.M. (2019) Multiple Choice
Question Answering in the Legal Domain Using Reinforced Co-occurrence.
Database and Expert Systems Applications. DEXA 2019.
Lecture Notes in Computer Science, vol 11706. Springer, Cham
[2] Martinez-Gil J., Freudenthaler B., Tjoa A.M. (2019) A General Framework
for Multiple Choice Question Answering Based on Mutual Information and
Reinforced Co-occurrence. Transactions on Large-Scale Data- and Knowledge-Centered Systems XLII.
Lecture Notes in Computer Science, vol 11860. Springer, Berlin,
Heidelberg