Natural language is complex, and for a machine, it’s difficult to structure. The meaning of words is always formed together with the sentence and the other words in the whole text. The challenge of machine interpretation of natural language is increased by several factors such as synonyms, words with close meanings, the use of sarcasm and even fake news. The meaning of words always depends on the context in which they occur. There is a subjectivity associated with natural language: everyone builds a definition for words through their own experiences. Furthermore, the labor market uses a much different language than the educational world, which makes dialogue on the demand and supply of skills difficult.

In some languages, words can take many different forms due to the possible word suffixes and their combinations. English is somewhat easy to machine-read because the prepositions are already separated from the word stems. A language like Finnish, which uses suffixes instead of prepositions, brings challenges. Due to rich morphology, there are millions of inflected forms of words. Headai’s dynamic language model consists of millions of words. AI has built the vocabulary by reading e.g. news, job postings, tech & economy news and scientific articles in different languages. In this session you will see how, by using this dynamic language model, Headai builds detailed and flexible skill-related digital twins of e.g. I) the working life skill demand, II) the supply of training or education, III) the individual, learner, employee, trainer or other individual and, IV) companies. The solution enables automation of text based research activities.