When we learn a new language, we quickly realize that natural language (written or spoken) is highly complex. The big players such as Microsoft, Google, and Amazon are developing solutions for the technical processing of natural language. Alexa from Amazon is a good example. Even though we keep seeing successes in this field, the technology is still a long way from being suitable for everyday use. Yet this topic is not only of interest to private individuals. Companies can also benefit from language processing. The potential and added value for companies is enormous. Consider single word recognition in warehousing, for example. The complexity of natural language lies in determining a speaker's intention with a sentence. Intention often depends on small details. There may be multiple intentions in one sentence, or none at all. Relevant information such as details about people, places, or times and general objects have to be technically identified, aside from intentions - in numerous versions and variations.
Microsoft offers a language processing service called LUIS (Language Understanding Intelligent Service). We have refined the Microsoft solution for the identification of intentions in customer visit reports of consultants (often in the CRM environment) for the life sciences sector. A concrete scenario with a limited number of possible intentions was examined, for example a telephone appointment, setting up an opportunity, informing a person, or sending documents/information. These are complemented by relevant information such as the date, time, participants for the telephone appointment, who is to be informed about what, or what is to be sent to whom and when.
In the first step, the input text is translated into English, because language recognition for the English language is the most advanced. While the translation is not one hundred percent correct, it is only for technical processing and therefore almost always adequate to understand the meaning of the sentence or statement. Then, the entire text is analyzed and divided into its components with language identification methods (part of speech tagging). Several sentences, each with an intention, can be extracted from this even if they were formally in one sentence (not separated by periods). Individual sentences are sent to Microsoft LUIS. This tool has learned how to assign intentions to sentences based on training sentences. The tool also returns relevant information.
Corresponding actions (e.g. in CRM) are proposed in the final step based on the respective intentions and their relevant information.
This system can be easily adapted to any language input and analysis scenario. It requires several (~20 per intention) example sentences and the respective information about what to detect. Then a procedure for the concrete application case can be prepared.