After having studied both computer science and general linguistics, I sought for solutions by combining knowledge from both fields. As a researcher, I became involved in an Academy of Finland project (1981–84), led by Fred Karlsson, which was investigating automatic recognition of the Finnish language.

Already in the first month Lauri Karttunen, then a professor in Austin, Texas, came to visit. When he heard what I had thought was roughly the target of my work, he invited me to a seminar in Austin later that spring, which was to be attended, in particular, by Ron Kaplan and Martin Kay, who had been working with similar kinds of ideas. I got funding and off I went.

I gained a first-hand picture of their work and insights. The method was well suited to English, but problems arose due to the high incidence of phoneme alternations in Finnish. I worked away at both their ideas and my own until December, when the pieces of the puzzle began to fall into place. Instead of Kaplan and Kay’s cascade model, I succeeded in getting a method to work with parallel rules. Formulating them was just as easy as with traditional rules, but the computer was also able to implement extensive systems of rules in a light way. Thus, the so-called two-level model for morphological analysis was born, and in the following three months I completed the theory along with a preliminary description of the morphology of Finnish words and a computer program that recognised words in practice. The program was language independent, and soon linguists were producing their own two-level descriptions in over a dozen different languages. I concentrated on using the method as the basis for my doctoral dissertation, which took 18 months.