NLP helps banks with automated document analysis

No items found.

Processing natural language is among the modern technologies based on artificial intelligence. It helps computers understand human language – spoken or written – and extract the important information from it.

You might also enjoy

Read more

A complex understanding of human language requires an understanding not only of the words, but also of syntax and semantics. Systems for natural speech recognition can be found in common applications such as Google translator, grammar checks in MS Office applications, intelligent conversation interfaces (chatbots) or digital assistants (Cortana, Siri, Alexa, etc.). Trask offers them to. This unique solution understands the Czech language and in live operation serves e.g. one of the largest Czech banks to extract information from notarial documents. It replaces and improves the quality of what used to be manual work and frees up employees for more qualified activities and customer care.

The purpose of natural language processing (NLP) technologies, most of which are based on machine learning, it to decipher human language in the form of speech or text, understand the meaning and extract useful information. The absolute goal is to achieve human-level understanding. Computer systems will then be able to understand written and spoken human expression, draw conclusions, summarize, translate and generate natural language outputs.

Computerized natural language processing can be approached in several ways:

  • Symbolic approach: based on generally valid rules of grammar and the lexicon prepared for the system by humans.
  • Statistical approach: based on machine learning. Works with a vast body of text in which it analyzes the occurrence of language phenomena using mathematical methods. It creates its own rules which it applies to further inputs.
  • Hybrid approach: combines the two foregoing principles. The systems refer to the defined generally valid rules and then modify them for specific purposes based on a statistical analysis of the inputs.

100% Czech NLP solution

This is the Trask Semantic Tool, which we develop in cooperation with the Institute of Formal and Applied Linguistics (ÚFAL) attached to the Charles University Faculty of Mathematics and Physics, a cutting-edge workplace engaged in computerized text processing and preparing large bodies of language data massively used for machine learning. Thanks to this, Trask can work with the most advanced instruments and components for using and mining Czech texts and can also use data from a unique language database providing syntactically and semantically annotated Czech texts.

[.infobox][.infobox-heading]Trask Semantic Tool in practice[.infobox-heading]The Trask NLP system serves one of the largest Czech banks for processing notarial documents. It takes over a laborious and time-consuming activity associated with growing legislative requirements without added value for the bank’s clients. The system automatically extracts key information from the notarial applications – data about the deceased (name, surname, date of birth, personal number, date of death), the notary’s name, case number and addressee of the application – and evaluates the accuracy of the results according to the defined rules. It also allows searching the documents. In the case of uncertainty, it requests human/operator assistance and submits an analysis of the anticipated deficiencies.[.infobox]

In its NLP solution, Trask relies on extensive knowledge of the banking environment and many years of experience with the development, implementation and integration of banking systems. It is built on superior security standards and meets all the valid legal regulations (GDPR) and internal bank rules for handling data, including complete audit traces.

“Thanks to cooperation with Trask, we can apply the results of our research in practice. The Prague dependency body contains a vast number of Czech texts analyzed in detail and is therefore ideal for the needs of advanced applications in computational linguistics.”

— Doc. RNDr. Markéta Lopatková, Ph.D., Director of ÚFAL MFF UK —
Jan Kosina, Head of Intelligent Automation at Trask

Do you want to know more about our solution? Tell us what you’re interested in, and we’ll get back to you as soon as possible. You can begin by contacting the author of this article, Head of Intelligent Automation Jan Kosina, on e-mail

Written by

No items found.